2023-03-31 18:51:54,786 INFO [train.py:975] (1/4) Training started 2023-03-31 18:51:54,787 INFO [train.py:985] (1/4) Device: cuda:1 2023-03-31 18:51:54,826 INFO [train.py:994] (1/4) {'best_train_loss': inf, 'best_valid_loss': inf, 'best_train_epoch': -1, 'best_valid_epoch': -1, 'batch_idx_train': 0, 'log_interval': 50, 'reset_interval': 200, 'valid_interval': 3000, 'feature_dim': 80, 'subsampling_factor': 4, 'warm_step': 2000, 'env_info': {'k2-version': '1.23.3', 'k2-build-type': 'Debug', 'k2-with-cuda': True, 'k2-git-sha1': '1c9950559223ec24d187f56bc424c3b43904bed3', 'k2-git-date': 'Thu Jan 26 22:00:26 2023', 'lhotse-version': '1.13.0.dev+git.ca98c73.dirty', 'torch-version': '2.0.0+cu117', 'torch-cuda-available': True, 'torch-cuda-version': '11.7', 'python-version': '3.8', 'icefall-git-branch': 'surt', 'icefall-git-sha1': '51e6a8a-dirty', 'icefall-git-date': 'Fri Mar 17 11:23:13 2023', 'icefall-path': '/exp/draj/mini_scale_2022/icefall', 'k2-path': '/exp/draj/mini_scale_2022/k2/k2/python/k2/__init__.py', 'lhotse-path': '/exp/draj/mini_scale_2022/lhotse/lhotse/__init__.py', 'hostname': 'r2n03', 'IP address': '10.1.2.3'}, 'world_size': 4, 'master_port': 54321, 'tensorboard': True, 'num_epochs': 30, 'start_epoch': 1, 'start_batch': 0, 'exp_dir': PosixPath('pruned_transducer_stateless7_streaming/exp/v2'), 'bpe_model': 'data/lang_bpe_500/bpe.model', 'base_lr': 0.05, 'lr_batches': 5000, 'lr_epochs': 3.5, 'context_size': 2, 'prune_range': 5, 'lm_scale': 0.25, 'am_scale': 0.0, 'simple_loss_scale': 0.5, 'seed': 42, 'print_diagnostics': False, 'inf_check': False, 'save_every_n': 2000, 'keep_last_k': 10, 'average_period': 200, 'use_fp16': True, 'num_encoder_layers': '2,2,2,2,2', 'feedforward_dims': '768,768,768,768,768', 'nhead': '8,8,8,8,8', 'encoder_dims': '256,256,256,256,256', 'attention_dims': '192,192,192,192,192', 'encoder_unmasked_dims': '192,192,192,192,192', 'zipformer_downsampling_factors': '1,2,4,8,2', 'cnn_module_kernels': '31,31,31,31,31', 'decoder_dim': 512, 'joiner_dim': 512, 'short_chunk_size': 50, 'num_left_chunks': 4, 'decode_chunk_len': 32, 'full_libri': True, 'manifest_dir': PosixPath('data/manifests'), 'max_duration': 800, 'bucketing_sampler': True, 'num_buckets': 30, 'concatenate_cuts': False, 'duration_factor': 1.0, 'gap': 1.0, 'on_the_fly_feats': False, 'shuffle': True, 'drop_last': True, 'return_cuts': True, 'num_workers': 2, 'enable_spec_aug': True, 'spec_aug_time_warp_factor': 80, 'enable_musan': True, 'input_strategy': 'PrecomputedFeatures', 'blank_id': 0, 'vocab_size': 500} 2023-03-31 18:51:54,826 INFO [train.py:996] (1/4) About to create model 2023-03-31 18:51:55,680 INFO [zipformer.py:405] (1/4) At encoder stack 4, which has downsampling_factor=2, we will combine the outputs of layers 1 and 3, with downsampling_factors=2 and 8. 2023-03-31 18:51:55,693 INFO [train.py:1000] (1/4) Number of model parameters: 20697573 2023-03-31 18:52:03,003 INFO [train.py:1019] (1/4) Using DDP 2023-03-31 18:52:03,650 INFO [asr_datamodule.py:429] (1/4) About to get the shuffled train-clean-100, train-clean-360 and train-other-500 cuts, combined with their reverberated versions 2023-03-31 18:52:03,691 INFO [asr_datamodule.py:224] (1/4) Enable MUSAN 2023-03-31 18:52:03,691 INFO [asr_datamodule.py:225] (1/4) About to get Musan cuts 2023-03-31 18:52:05,879 INFO [asr_datamodule.py:249] (1/4) Enable SpecAugment 2023-03-31 18:52:05,879 INFO [asr_datamodule.py:250] (1/4) Time warp factor: 80 2023-03-31 18:52:05,879 INFO [asr_datamodule.py:260] (1/4) Num frame mask: 10 2023-03-31 18:52:05,879 INFO [asr_datamodule.py:273] (1/4) About to create train dataset 2023-03-31 18:52:05,879 INFO [asr_datamodule.py:300] (1/4) Using DynamicBucketingSampler. 2023-03-31 18:52:08,162 WARNING [train.py:1073] (1/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 2023-03-31 18:52:08,613 WARNING [train.py:1073] (1/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 2023-03-31 18:52:08,889 INFO [asr_datamodule.py:315] (1/4) About to create train dataloader 2023-03-31 18:52:08,890 INFO [asr_datamodule.py:440] (1/4) About to get dev-clean cuts 2023-03-31 18:52:08,925 INFO [asr_datamodule.py:447] (1/4) About to get dev-other cuts 2023-03-31 18:52:08,952 INFO [asr_datamodule.py:346] (1/4) About to create dev dataset 2023-03-31 18:52:09,395 INFO [asr_datamodule.py:363] (1/4) About to create dev dataloader 2023-03-31 18:52:24,008 WARNING [train.py:1073] (1/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 2023-03-31 18:52:24,461 WARNING [train.py:1073] (1/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 2023-03-31 18:52:36,261 INFO [train.py:903] (1/4) Epoch 1, batch 0, loss[loss=7.307, simple_loss=6.612, pruned_loss=6.939, over 19592.00 frames. ], tot_loss[loss=7.307, simple_loss=6.612, pruned_loss=6.939, over 19592.00 frames. ], batch size: 57, lr: 2.50e-02, grad_scale: 2.0 2023-03-31 18:52:36,262 INFO [train.py:928] (1/4) Computing validation loss 2023-03-31 18:52:49,136 INFO [train.py:937] (1/4) Epoch 1, validation: loss=6.888, simple_loss=6.229, pruned_loss=6.575, over 944034.00 frames. 2023-03-31 18:52:49,137 INFO [train.py:938] (1/4) Maximum memory allocated so far is 10937MB 2023-03-31 18:53:02,994 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 2023-03-31 18:53:58,984 INFO [train.py:903] (1/4) Epoch 1, batch 50, loss[loss=1.39, simple_loss=1.231, pruned_loss=1.422, over 18824.00 frames. ], tot_loss[loss=2.167, simple_loss=1.957, pruned_loss=2.01, over 863774.73 frames. ], batch size: 74, lr: 2.75e-02, grad_scale: 0.125 2023-03-31 18:53:59,782 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=108.63 vs. limit=5.0 2023-03-31 18:54:00,563 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=52.0, num_to_drop=1, layers_to_drop={0} 2023-03-31 18:54:36,583 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=77.0, num_to_drop=1, layers_to_drop={0} 2023-03-31 18:54:41,871 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1_rvb from training. Duration: 27.0318125 2023-03-31 18:55:04,257 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=7.36 vs. limit=2.0 2023-03-31 18:55:11,344 INFO [train.py:903] (1/4) Epoch 1, batch 100, loss[loss=1.151, simple_loss=0.9889, pruned_loss=1.285, over 19535.00 frames. ], tot_loss[loss=1.626, simple_loss=1.447, pruned_loss=1.612, over 1527276.03 frames. ], batch size: 56, lr: 3.00e-02, grad_scale: 0.25 2023-03-31 18:55:11,608 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=101.0, num_to_drop=1, layers_to_drop={1} 2023-03-31 18:55:17,831 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 9.479e+01 1.678e+02 3.237e+02 1.260e+03 8.630e+04, threshold=6.475e+02, percent-clipped=0.0 2023-03-31 18:55:25,911 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 2023-03-31 18:56:18,847 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=2.88 vs. limit=2.0 2023-03-31 18:56:20,070 INFO [train.py:903] (1/4) Epoch 1, batch 150, loss[loss=0.9524, simple_loss=0.8089, pruned_loss=1.037, over 19758.00 frames. ], tot_loss[loss=1.377, simple_loss=1.21, pruned_loss=1.412, over 2042012.87 frames. ], batch size: 51, lr: 3.25e-02, grad_scale: 0.25 2023-03-31 18:57:32,423 INFO [train.py:903] (1/4) Epoch 1, batch 200, loss[loss=1.027, simple_loss=0.8696, pruned_loss=1.054, over 19791.00 frames. ], tot_loss[loss=1.235, simple_loss=1.075, pruned_loss=1.274, over 2448634.22 frames. ], batch size: 56, lr: 3.50e-02, grad_scale: 0.5 2023-03-31 18:57:32,458 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 2023-03-31 18:57:39,435 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 7.067e+01 1.186e+02 1.653e+02 2.090e+02 5.158e+02, threshold=3.307e+02, percent-clipped=0.0 2023-03-31 18:58:43,201 INFO [train.py:903] (1/4) Epoch 1, batch 250, loss[loss=0.9682, simple_loss=0.8141, pruned_loss=0.9637, over 19578.00 frames. ], tot_loss[loss=1.155, simple_loss=0.9981, pruned_loss=1.182, over 2744782.19 frames. ], batch size: 52, lr: 3.75e-02, grad_scale: 0.5 2023-03-31 18:59:06,197 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=5.80 vs. limit=5.0 2023-03-31 18:59:48,220 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=5.19 vs. limit=5.0 2023-03-31 18:59:51,885 INFO [train.py:903] (1/4) Epoch 1, batch 300, loss[loss=0.9232, simple_loss=0.7764, pruned_loss=0.8717, over 18180.00 frames. ], tot_loss[loss=1.096, simple_loss=0.9406, pruned_loss=1.108, over 2987007.86 frames. ], batch size: 83, lr: 4.00e-02, grad_scale: 1.0 2023-03-31 18:59:56,684 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 8.717e+01 1.166e+02 1.521e+02 1.991e+02 3.277e+02, threshold=3.043e+02, percent-clipped=0.0 2023-03-31 18:59:58,411 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=306.0, num_to_drop=1, layers_to_drop={1} 2023-03-31 19:00:09,460 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=314.0, num_to_drop=1, layers_to_drop={1} 2023-03-31 19:00:58,524 INFO [train.py:903] (1/4) Epoch 1, batch 350, loss[loss=0.93, simple_loss=0.7704, pruned_loss=0.8809, over 19654.00 frames. ], tot_loss[loss=1.055, simple_loss=0.8985, pruned_loss=1.052, over 3177969.89 frames. ], batch size: 58, lr: 4.25e-02, grad_scale: 1.0 2023-03-31 19:01:05,464 WARNING [train.py:1073] (1/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1_rvb from training. Duration: 0.9681875 2023-03-31 19:02:08,280 INFO [train.py:903] (1/4) Epoch 1, batch 400, loss[loss=0.8668, simple_loss=0.7112, pruned_loss=0.809, over 19087.00 frames. ], tot_loss[loss=1.025, simple_loss=0.867, pruned_loss=1.006, over 3316833.49 frames. ], batch size: 42, lr: 4.50e-02, grad_scale: 2.0 2023-03-31 19:02:13,389 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 9.539e+01 1.278e+02 1.546e+02 1.978e+02 5.474e+02, threshold=3.091e+02, percent-clipped=7.0 2023-03-31 19:02:13,641 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=405.0, num_to_drop=1, layers_to_drop={1} 2023-03-31 19:02:33,548 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=421.0, num_to_drop=1, layers_to_drop={0} 2023-03-31 19:03:05,053 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=445.0, num_to_drop=1, layers_to_drop={1} 2023-03-31 19:03:07,118 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=6.64 vs. limit=5.0 2023-03-31 19:03:12,892 INFO [train.py:903] (1/4) Epoch 1, batch 450, loss[loss=0.9279, simple_loss=0.7633, pruned_loss=0.8273, over 19842.00 frames. ], tot_loss[loss=1, simple_loss=0.8411, pruned_loss=0.9638, over 3436282.94 frames. ], batch size: 52, lr: 4.75e-02, grad_scale: 2.0 2023-03-31 19:03:17,852 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.7085, 4.3057, 3.7241, 3.4803, 4.0774, 3.6254, 3.8994, 4.3769], device='cuda:1'), covar=tensor([0.0201, 0.0141, 0.0239, 0.0172, 0.0110, 0.0894, 0.0162, 0.0117], device='cuda:1'), in_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0015, 0.0015, 0.0015, 0.0015, 0.0014], device='cuda:1'), out_proj_covar=tensor([9.5444e-06, 9.7004e-06, 9.7803e-06, 9.4442e-06, 9.9968e-06, 9.4626e-06, 9.9635e-06, 9.9348e-06], device='cuda:1') 2023-03-31 19:03:49,664 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 2023-03-31 19:03:51,777 WARNING [train.py:1073] (1/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 2023-03-31 19:04:16,584 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=6.17 vs. limit=5.0 2023-03-31 19:04:19,393 INFO [train.py:903] (1/4) Epoch 1, batch 500, loss[loss=0.8732, simple_loss=0.7238, pruned_loss=0.7377, over 19624.00 frames. ], tot_loss[loss=0.9809, simple_loss=0.8218, pruned_loss=0.9232, over 3520212.27 frames. ], batch size: 50, lr: 4.99e-02, grad_scale: 2.0 2023-03-31 19:04:25,204 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 9.539e+01 1.386e+02 1.860e+02 2.529e+02 4.736e+02, threshold=3.719e+02, percent-clipped=12.0 2023-03-31 19:05:27,842 INFO [train.py:903] (1/4) Epoch 1, batch 550, loss[loss=0.9376, simple_loss=0.7807, pruned_loss=0.7601, over 18718.00 frames. ], tot_loss[loss=0.9629, simple_loss=0.8051, pruned_loss=0.8826, over 3565533.85 frames. ], batch size: 74, lr: 4.98e-02, grad_scale: 2.0 2023-03-31 19:05:28,393 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=2.64 vs. limit=2.0 2023-03-31 19:05:40,551 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=560.0, num_to_drop=1, layers_to_drop={1} 2023-03-31 19:06:03,868 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=580.0, num_to_drop=1, layers_to_drop={0} 2023-03-31 19:06:12,419 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=586.0, num_to_drop=1, layers_to_drop={1} 2023-03-31 19:06:32,555 INFO [train.py:903] (1/4) Epoch 1, batch 600, loss[loss=0.8076, simple_loss=0.6809, pruned_loss=0.6181, over 19853.00 frames. ], tot_loss[loss=0.9411, simple_loss=0.787, pruned_loss=0.8366, over 3628105.98 frames. ], batch size: 52, lr: 4.98e-02, grad_scale: 2.0 2023-03-31 19:06:36,892 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.512e+02 2.910e+02 4.086e+02 6.136e+02 1.097e+03, threshold=8.173e+02, percent-clipped=60.0 2023-03-31 19:06:40,831 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=608.0, num_to_drop=1, layers_to_drop={1} 2023-03-31 19:07:11,500 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 2023-03-31 19:07:38,082 INFO [train.py:903] (1/4) Epoch 1, batch 650, loss[loss=0.8372, simple_loss=0.7112, pruned_loss=0.6153, over 19542.00 frames. ], tot_loss[loss=0.9129, simple_loss=0.7648, pruned_loss=0.7862, over 3675516.72 frames. ], batch size: 56, lr: 4.98e-02, grad_scale: 2.0 2023-03-31 19:07:47,306 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=658.0, num_to_drop=1, layers_to_drop={1} 2023-03-31 19:08:13,404 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=677.0, num_to_drop=1, layers_to_drop={0} 2023-03-31 19:08:41,834 INFO [train.py:903] (1/4) Epoch 1, batch 700, loss[loss=0.7309, simple_loss=0.6267, pruned_loss=0.5152, over 19480.00 frames. ], tot_loss[loss=0.8871, simple_loss=0.7452, pruned_loss=0.7398, over 3709486.30 frames. ], batch size: 49, lr: 4.98e-02, grad_scale: 2.0 2023-03-31 19:08:43,407 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=702.0, num_to_drop=1, layers_to_drop={0} 2023-03-31 19:08:46,605 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.000e+02 5.173e+02 6.580e+02 8.914e+02 3.039e+03, threshold=1.316e+03, percent-clipped=29.0 2023-03-31 19:08:59,142 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.2137, 1.4595, 2.4414, 1.7821, 2.5404, 2.2759, 2.4021, 2.3107], device='cuda:1'), covar=tensor([0.2391, 0.5703, 0.2943, 0.4250, 0.1630, 0.2413, 0.3088, 0.2743], device='cuda:1'), in_proj_covar=tensor([0.0036, 0.0044, 0.0037, 0.0041, 0.0037, 0.0039, 0.0043, 0.0037], device='cuda:1'), out_proj_covar=tensor([2.2808e-05, 3.1594e-05, 2.2624e-05, 2.8300e-05, 2.3633e-05, 2.2850e-05, 2.7749e-05, 2.4723e-05], device='cuda:1') 2023-03-31 19:09:43,577 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=749.0, num_to_drop=1, layers_to_drop={0} 2023-03-31 19:09:45,318 INFO [train.py:903] (1/4) Epoch 1, batch 750, loss[loss=0.8168, simple_loss=0.6967, pruned_loss=0.5727, over 18162.00 frames. ], tot_loss[loss=0.8608, simple_loss=0.7257, pruned_loss=0.6953, over 3724401.58 frames. ], batch size: 83, lr: 4.97e-02, grad_scale: 2.0 2023-03-31 19:10:14,484 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=773.0, num_to_drop=1, layers_to_drop={0} 2023-03-31 19:10:48,972 INFO [train.py:903] (1/4) Epoch 1, batch 800, loss[loss=0.7225, simple_loss=0.6226, pruned_loss=0.4879, over 19592.00 frames. ], tot_loss[loss=0.8352, simple_loss=0.7067, pruned_loss=0.6539, over 3745627.59 frames. ], batch size: 61, lr: 4.97e-02, grad_scale: 4.0 2023-03-31 19:10:53,094 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.110e+02 5.827e+02 7.991e+02 1.030e+03 2.888e+03, threshold=1.598e+03, percent-clipped=14.0 2023-03-31 19:11:01,626 WARNING [train.py:1073] (1/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 2023-03-31 19:11:03,310 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=4.78 vs. limit=5.0 2023-03-31 19:11:08,447 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=816.0, num_to_drop=1, layers_to_drop={0} 2023-03-31 19:11:40,358 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=841.0, num_to_drop=1, layers_to_drop={0} 2023-03-31 19:11:52,164 INFO [train.py:903] (1/4) Epoch 1, batch 850, loss[loss=0.7578, simple_loss=0.6557, pruned_loss=0.5003, over 19616.00 frames. ], tot_loss[loss=0.8125, simple_loss=0.6901, pruned_loss=0.6178, over 3748691.42 frames. ], batch size: 61, lr: 4.96e-02, grad_scale: 4.0 2023-03-31 19:12:06,984 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9724, 0.7910, 1.0493, 1.1723, 1.8002, 1.8163, 1.2395, 1.4915], device='cuda:1'), covar=tensor([0.9893, 2.0870, 2.0057, 1.5953, 0.6676, 1.9937, 1.8155, 1.2857], device='cuda:1'), in_proj_covar=tensor([0.0090, 0.0115, 0.0118, 0.0100, 0.0072, 0.0129, 0.0101, 0.0096], device='cuda:1'), out_proj_covar=tensor([5.8551e-05, 7.7465e-05, 7.7369e-05, 5.9372e-05, 4.2425e-05, 8.5179e-05, 6.3534e-05, 5.8211e-05], device='cuda:1') 2023-03-31 19:12:10,465 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=864.0, num_to_drop=1, layers_to_drop={0} 2023-03-31 19:12:12,314 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=865.0, num_to_drop=1, layers_to_drop={0} 2023-03-31 19:12:43,354 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9_rvb from training. Duration: 25.061125 2023-03-31 19:12:46,975 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0647, 1.4073, 1.1270, 2.0397, 1.1530, 2.0910, 2.0768, 1.4011], device='cuda:1'), covar=tensor([0.4878, 1.1753, 1.3744, 0.4644, 1.7246, 0.4836, 0.5788, 0.9036], device='cuda:1'), in_proj_covar=tensor([0.0036, 0.0052, 0.0063, 0.0038, 0.0068, 0.0041, 0.0042, 0.0044], device='cuda:1'), out_proj_covar=tensor([2.2538e-05, 3.6620e-05, 4.5423e-05, 2.4333e-05, 5.0861e-05, 2.4064e-05, 2.5554e-05, 3.0093e-05], device='cuda:1') 2023-03-31 19:12:54,612 INFO [train.py:903] (1/4) Epoch 1, batch 900, loss[loss=0.6634, simple_loss=0.5795, pruned_loss=0.4242, over 19853.00 frames. ], tot_loss[loss=0.7905, simple_loss=0.6738, pruned_loss=0.5845, over 3760553.25 frames. ], batch size: 52, lr: 4.96e-02, grad_scale: 4.0 2023-03-31 19:12:59,594 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.676e+02 6.072e+02 7.456e+02 9.579e+02 1.181e+04, threshold=1.491e+03, percent-clipped=3.0 2023-03-31 19:13:21,570 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=924.0, num_to_drop=1, layers_to_drop={1} 2023-03-31 19:13:30,380 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=930.0, num_to_drop=1, layers_to_drop={0} 2023-03-31 19:13:53,705 INFO [train.py:903] (1/4) Epoch 1, batch 950, loss[loss=0.6765, simple_loss=0.5929, pruned_loss=0.4253, over 19543.00 frames. ], tot_loss[loss=0.7675, simple_loss=0.6569, pruned_loss=0.5522, over 3786940.23 frames. ], batch size: 54, lr: 4.96e-02, grad_scale: 4.0 2023-03-31 19:13:53,727 WARNING [train.py:1073] (1/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9_rvb from training. Duration: 26.32775 2023-03-31 19:13:56,045 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=952.0, num_to_drop=1, layers_to_drop={1} 2023-03-31 19:14:06,053 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.92 vs. limit=2.0 2023-03-31 19:14:21,085 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=4.92 vs. limit=5.0 2023-03-31 19:14:51,782 INFO [train.py:903] (1/4) Epoch 1, batch 1000, loss[loss=0.6989, simple_loss=0.6128, pruned_loss=0.4351, over 18209.00 frames. ], tot_loss[loss=0.7478, simple_loss=0.6424, pruned_loss=0.5248, over 3802889.71 frames. ], batch size: 83, lr: 4.95e-02, grad_scale: 4.0 2023-03-31 19:14:56,983 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.794e+02 5.980e+02 7.509e+02 1.052e+03 2.029e+03, threshold=1.502e+03, percent-clipped=4.0 2023-03-31 19:15:25,851 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1029.0, num_to_drop=1, layers_to_drop={1} 2023-03-31 19:15:38,851 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1039.0, num_to_drop=1, layers_to_drop={0} 2023-03-31 19:15:41,719 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 2023-03-31 19:15:45,035 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1045.0, num_to_drop=1, layers_to_drop={1} 2023-03-31 19:15:52,670 INFO [train.py:903] (1/4) Epoch 1, batch 1050, loss[loss=0.6844, simple_loss=0.6015, pruned_loss=0.4206, over 19537.00 frames. ], tot_loss[loss=0.729, simple_loss=0.629, pruned_loss=0.4991, over 3813165.64 frames. ], batch size: 54, lr: 4.95e-02, grad_scale: 4.0 2023-03-31 19:15:56,724 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1054.0, num_to_drop=1, layers_to_drop={0} 2023-03-31 19:16:12,550 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1067.0, num_to_drop=1, layers_to_drop={0} 2023-03-31 19:16:20,717 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 2023-03-31 19:16:53,095 INFO [train.py:903] (1/4) Epoch 1, batch 1100, loss[loss=0.7346, simple_loss=0.6295, pruned_loss=0.4675, over 19486.00 frames. ], tot_loss[loss=0.713, simple_loss=0.6173, pruned_loss=0.4779, over 3810250.38 frames. ], batch size: 64, lr: 4.94e-02, grad_scale: 4.0 2023-03-31 19:16:57,404 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.036e+02 7.117e+02 8.563e+02 1.068e+03 2.368e+03, threshold=1.713e+03, percent-clipped=4.0 2023-03-31 19:17:14,947 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1120.0, num_to_drop=1, layers_to_drop={1} 2023-03-31 19:17:44,294 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1145.0, num_to_drop=1, layers_to_drop={1} 2023-03-31 19:17:51,774 INFO [train.py:903] (1/4) Epoch 1, batch 1150, loss[loss=0.5991, simple_loss=0.5247, pruned_loss=0.3647, over 19726.00 frames. ], tot_loss[loss=0.6964, simple_loss=0.6054, pruned_loss=0.4572, over 3803545.10 frames. ], batch size: 47, lr: 4.94e-02, grad_scale: 4.0 2023-03-31 19:18:12,988 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1171.0, num_to_drop=1, layers_to_drop={0} 2023-03-31 19:18:47,615 INFO [train.py:903] (1/4) Epoch 1, batch 1200, loss[loss=0.6781, simple_loss=0.5987, pruned_loss=0.4046, over 19330.00 frames. ], tot_loss[loss=0.6872, simple_loss=0.5986, pruned_loss=0.4436, over 3802456.67 frames. ], batch size: 70, lr: 4.93e-02, grad_scale: 8.0 2023-03-31 19:18:52,224 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.229e+02 7.433e+02 9.314e+02 1.239e+03 3.000e+03, threshold=1.863e+03, percent-clipped=16.0 2023-03-31 19:18:56,103 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=1209.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 19:19:16,424 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 2023-03-31 19:19:42,376 INFO [train.py:903] (1/4) Epoch 1, batch 1250, loss[loss=0.6207, simple_loss=0.5452, pruned_loss=0.371, over 19583.00 frames. ], tot_loss[loss=0.674, simple_loss=0.5899, pruned_loss=0.4268, over 3813708.36 frames. ], batch size: 52, lr: 4.92e-02, grad_scale: 8.0 2023-03-31 19:20:32,299 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1295.0, num_to_drop=1, layers_to_drop={1} 2023-03-31 19:20:39,095 INFO [train.py:903] (1/4) Epoch 1, batch 1300, loss[loss=0.5945, simple_loss=0.5288, pruned_loss=0.3467, over 19425.00 frames. ], tot_loss[loss=0.6613, simple_loss=0.5811, pruned_loss=0.4118, over 3822283.43 frames. ], batch size: 48, lr: 4.92e-02, grad_scale: 8.0 2023-03-31 19:20:39,513 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1301.0, num_to_drop=1, layers_to_drop={1} 2023-03-31 19:20:41,405 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4125, 0.8917, 0.9701, 0.9627, 1.1289, 1.4159, 1.2797, 1.2666], device='cuda:1'), covar=tensor([0.5466, 1.0360, 1.0409, 0.7195, 0.7256, 0.9843, 0.9189, 0.6223], device='cuda:1'), in_proj_covar=tensor([0.0123, 0.0159, 0.0164, 0.0129, 0.0147, 0.0168, 0.0148, 0.0129], device='cuda:1'), out_proj_covar=tensor([8.3574e-05, 1.0621e-04, 1.1064e-04, 8.3685e-05, 9.4480e-05, 1.1128e-04, 9.9269e-05, 8.3402e-05], device='cuda:1') 2023-03-31 19:20:43,725 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.522e+02 7.699e+02 1.048e+03 1.379e+03 4.741e+03, threshold=2.097e+03, percent-clipped=13.0 2023-03-31 19:21:00,020 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1320.0, num_to_drop=1, layers_to_drop={1} 2023-03-31 19:21:03,702 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1323.0, num_to_drop=1, layers_to_drop={0} 2023-03-31 19:21:04,828 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1324.0, num_to_drop=1, layers_to_drop={0} 2023-03-31 19:21:06,822 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1326.0, num_to_drop=1, layers_to_drop={0} 2023-03-31 19:21:33,464 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1348.0, num_to_drop=1, layers_to_drop={0} 2023-03-31 19:21:35,925 INFO [train.py:903] (1/4) Epoch 1, batch 1350, loss[loss=0.5545, simple_loss=0.4991, pruned_loss=0.3164, over 19755.00 frames. ], tot_loss[loss=0.6498, simple_loss=0.573, pruned_loss=0.3986, over 3827585.89 frames. ], batch size: 46, lr: 4.91e-02, grad_scale: 8.0 2023-03-31 19:22:31,171 INFO [train.py:903] (1/4) Epoch 1, batch 1400, loss[loss=0.6623, simple_loss=0.5891, pruned_loss=0.3826, over 17334.00 frames. ], tot_loss[loss=0.6395, simple_loss=0.5663, pruned_loss=0.3863, over 3821449.43 frames. ], batch size: 101, lr: 4.91e-02, grad_scale: 8.0 2023-03-31 19:22:35,196 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.777e+02 7.620e+02 9.515e+02 1.230e+03 4.278e+03, threshold=1.903e+03, percent-clipped=3.0 2023-03-31 19:23:04,500 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.0148, 1.6877, 1.5409, 2.2630, 2.8006, 2.1618, 2.6960, 2.4339], device='cuda:1'), covar=tensor([0.0774, 0.3930, 0.6025, 0.2283, 0.1106, 0.5041, 0.1202, 0.1613], device='cuda:1'), in_proj_covar=tensor([0.0082, 0.0109, 0.0143, 0.0092, 0.0108, 0.0166, 0.0102, 0.0087], device='cuda:1'), out_proj_covar=tensor([4.7813e-05, 7.5708e-05, 9.8948e-05, 6.4597e-05, 6.5640e-05, 1.0900e-04, 6.4754e-05, 5.9698e-05], device='cuda:1') 2023-03-31 19:23:24,304 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_rvb from training. Duration: 25.85 2023-03-31 19:23:25,222 INFO [train.py:903] (1/4) Epoch 1, batch 1450, loss[loss=0.6741, simple_loss=0.597, pruned_loss=0.39, over 19332.00 frames. ], tot_loss[loss=0.6331, simple_loss=0.5622, pruned_loss=0.3779, over 3810254.48 frames. ], batch size: 70, lr: 4.90e-02, grad_scale: 8.0 2023-03-31 19:23:26,461 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1452.0, num_to_drop=1, layers_to_drop={0} 2023-03-31 19:24:18,031 INFO [train.py:903] (1/4) Epoch 1, batch 1500, loss[loss=0.5461, simple_loss=0.4957, pruned_loss=0.3048, over 19420.00 frames. ], tot_loss[loss=0.6252, simple_loss=0.5571, pruned_loss=0.3687, over 3810163.72 frames. ], batch size: 48, lr: 4.89e-02, grad_scale: 8.0 2023-03-31 19:24:23,064 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.214e+02 9.104e+02 1.060e+03 1.370e+03 5.981e+03, threshold=2.119e+03, percent-clipped=12.0 2023-03-31 19:24:35,376 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=1515.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 19:24:39,510 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.4290, 1.5968, 1.5091, 2.3667, 3.0200, 1.6879, 2.8665, 2.8632], device='cuda:1'), covar=tensor([0.0760, 0.3989, 0.5445, 0.2031, 0.1069, 0.6120, 0.1106, 0.1119], device='cuda:1'), in_proj_covar=tensor([0.0087, 0.0121, 0.0152, 0.0102, 0.0117, 0.0185, 0.0112, 0.0094], device='cuda:1'), out_proj_covar=tensor([4.9373e-05, 8.3634e-05, 1.0634e-04, 7.2056e-05, 7.0243e-05, 1.2044e-04, 7.2726e-05, 6.3695e-05], device='cuda:1') 2023-03-31 19:25:14,433 INFO [train.py:903] (1/4) Epoch 1, batch 1550, loss[loss=0.6676, simple_loss=0.5955, pruned_loss=0.3797, over 19739.00 frames. ], tot_loss[loss=0.6167, simple_loss=0.5518, pruned_loss=0.3595, over 3801034.09 frames. ], batch size: 63, lr: 4.89e-02, grad_scale: 8.0 2023-03-31 19:25:28,743 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2023-03-31 19:25:43,582 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1580.0, num_to_drop=1, layers_to_drop={0} 2023-03-31 19:25:47,370 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1584.0, num_to_drop=1, layers_to_drop={0} 2023-03-31 19:26:05,888 INFO [train.py:903] (1/4) Epoch 1, batch 1600, loss[loss=0.5073, simple_loss=0.4801, pruned_loss=0.2674, over 19730.00 frames. ], tot_loss[loss=0.607, simple_loss=0.5455, pruned_loss=0.3498, over 3811866.58 frames. ], batch size: 51, lr: 4.88e-02, grad_scale: 8.0 2023-03-31 19:26:10,815 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 5.811e+02 9.198e+02 1.152e+03 1.497e+03 2.578e+03, threshold=2.303e+03, percent-clipped=3.0 2023-03-31 19:26:11,244 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1605.0, num_to_drop=1, layers_to_drop={1} 2023-03-31 19:26:25,445 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9_rvb from training. Duration: 30.1555625 2023-03-31 19:26:35,448 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1629.0, num_to_drop=1, layers_to_drop={1} 2023-03-31 19:26:36,323 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1630.0, num_to_drop=1, layers_to_drop={1} 2023-03-31 19:26:57,905 INFO [train.py:903] (1/4) Epoch 1, batch 1650, loss[loss=0.4481, simple_loss=0.4279, pruned_loss=0.2335, over 19729.00 frames. ], tot_loss[loss=0.6003, simple_loss=0.5413, pruned_loss=0.3427, over 3810853.67 frames. ], batch size: 45, lr: 4.87e-02, grad_scale: 8.0 2023-03-31 19:27:31,042 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.4424, 1.4281, 1.4141, 2.7005, 3.1729, 1.9710, 2.8652, 2.9171], device='cuda:1'), covar=tensor([0.0396, 0.2893, 0.5012, 0.1298, 0.0469, 0.3330, 0.0607, 0.0710], device='cuda:1'), in_proj_covar=tensor([0.0086, 0.0130, 0.0164, 0.0115, 0.0120, 0.0201, 0.0116, 0.0100], device='cuda:1'), out_proj_covar=tensor([4.8306e-05, 9.0777e-05, 1.1418e-04, 8.1079e-05, 7.2026e-05, 1.3056e-04, 7.4271e-05, 6.7406e-05], device='cuda:1') 2023-03-31 19:27:40,175 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9988, 1.2595, 2.1222, 1.3895, 2.1346, 2.5744, 2.5065, 2.6047], device='cuda:1'), covar=tensor([0.3332, 0.5230, 0.2262, 0.6011, 0.2276, 0.0783, 0.1135, 0.1129], device='cuda:1'), in_proj_covar=tensor([0.0132, 0.0123, 0.0102, 0.0142, 0.0102, 0.0072, 0.0089, 0.0074], device='cuda:1'), out_proj_covar=tensor([8.5116e-05, 8.1632e-05, 6.4274e-05, 9.2415e-05, 6.4919e-05, 3.9783e-05, 5.0504e-05, 4.0153e-05], device='cuda:1') 2023-03-31 19:27:52,350 INFO [train.py:903] (1/4) Epoch 1, batch 1700, loss[loss=0.543, simple_loss=0.5064, pruned_loss=0.2912, over 19786.00 frames. ], tot_loss[loss=0.5906, simple_loss=0.5347, pruned_loss=0.334, over 3820011.67 frames. ], batch size: 56, lr: 4.86e-02, grad_scale: 8.0 2023-03-31 19:27:56,192 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 5.719e+02 9.402e+02 1.223e+03 1.535e+03 2.582e+03, threshold=2.447e+03, percent-clipped=3.0 2023-03-31 19:28:26,633 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 2023-03-31 19:28:29,038 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6691, 1.4474, 1.3243, 2.1126, 1.7555, 2.3074, 1.9913, 1.2015], device='cuda:1'), covar=tensor([0.3219, 0.3605, 0.5419, 0.1833, 0.4355, 0.1423, 0.2660, 0.4714], device='cuda:1'), in_proj_covar=tensor([0.0066, 0.0068, 0.0096, 0.0078, 0.0115, 0.0053, 0.0068, 0.0080], device='cuda:1'), out_proj_covar=tensor([4.5606e-05, 4.6950e-05, 6.7610e-05, 5.6521e-05, 7.9562e-05, 3.1947e-05, 5.1156e-05, 5.6833e-05], device='cuda:1') 2023-03-31 19:28:46,841 INFO [train.py:903] (1/4) Epoch 1, batch 1750, loss[loss=0.6009, simple_loss=0.5462, pruned_loss=0.3309, over 18244.00 frames. ], tot_loss[loss=0.5832, simple_loss=0.5307, pruned_loss=0.3266, over 3817193.99 frames. ], batch size: 83, lr: 4.86e-02, grad_scale: 8.0 2023-03-31 19:29:37,333 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=1796.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 19:29:43,445 INFO [train.py:903] (1/4) Epoch 1, batch 1800, loss[loss=0.4744, simple_loss=0.4537, pruned_loss=0.2471, over 19319.00 frames. ], tot_loss[loss=0.5782, simple_loss=0.5272, pruned_loss=0.3218, over 3799184.18 frames. ], batch size: 44, lr: 4.85e-02, grad_scale: 8.0 2023-03-31 19:29:47,613 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 5.927e+02 9.266e+02 1.209e+03 1.539e+03 2.564e+03, threshold=2.418e+03, percent-clipped=2.0 2023-03-31 19:30:36,184 WARNING [train.py:1073] (1/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9_rvb from training. Duration: 25.0944375 2023-03-31 19:30:40,566 INFO [train.py:903] (1/4) Epoch 1, batch 1850, loss[loss=0.6044, simple_loss=0.544, pruned_loss=0.3346, over 12719.00 frames. ], tot_loss[loss=0.5697, simple_loss=0.5223, pruned_loss=0.3143, over 3803666.65 frames. ], batch size: 135, lr: 4.84e-02, grad_scale: 8.0 2023-03-31 19:30:45,855 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1856.0, num_to_drop=1, layers_to_drop={1} 2023-03-31 19:31:06,928 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1875.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 19:31:11,916 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9_rvb from training. Duration: 27.8166875 2023-03-31 19:31:20,232 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1886.0, num_to_drop=1, layers_to_drop={1} 2023-03-31 19:31:22,094 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1888.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 19:31:36,186 INFO [train.py:903] (1/4) Epoch 1, batch 1900, loss[loss=0.5169, simple_loss=0.4854, pruned_loss=0.2744, over 19405.00 frames. ], tot_loss[loss=0.5643, simple_loss=0.5196, pruned_loss=0.3091, over 3813448.61 frames. ], batch size: 48, lr: 4.83e-02, grad_scale: 8.0 2023-03-31 19:31:40,288 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.070e+02 9.078e+02 1.104e+03 1.499e+03 2.754e+03, threshold=2.207e+03, percent-clipped=2.0 2023-03-31 19:31:47,458 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1911.0, num_to_drop=1, layers_to_drop={1} 2023-03-31 19:31:47,477 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1911.0, num_to_drop=1, layers_to_drop={0} 2023-03-31 19:31:52,298 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 2023-03-31 19:31:56,237 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9_rvb from training. Duration: 27.02225 2023-03-31 19:32:06,347 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=1928.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 19:32:19,668 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 2023-03-31 19:32:31,502 INFO [train.py:903] (1/4) Epoch 1, batch 1950, loss[loss=0.5425, simple_loss=0.5183, pruned_loss=0.2832, over 19367.00 frames. ], tot_loss[loss=0.5623, simple_loss=0.5184, pruned_loss=0.3068, over 3790921.09 frames. ], batch size: 70, lr: 4.83e-02, grad_scale: 8.0 2023-03-31 19:32:57,189 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=1973.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 19:33:13,493 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 2023-03-31 19:33:29,173 INFO [train.py:903] (1/4) Epoch 1, batch 2000, loss[loss=0.4398, simple_loss=0.4299, pruned_loss=0.2249, over 19758.00 frames. ], tot_loss[loss=0.555, simple_loss=0.5143, pruned_loss=0.3008, over 3792896.18 frames. ], batch size: 47, lr: 4.82e-02, grad_scale: 8.0 2023-03-31 19:33:33,532 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 5.917e+02 1.007e+03 1.260e+03 1.703e+03 3.255e+03, threshold=2.521e+03, percent-clipped=11.0 2023-03-31 19:34:19,000 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=2043.0, num_to_drop=1, layers_to_drop={0} 2023-03-31 19:34:25,213 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_rvb from training. Duration: 26.8349375 2023-03-31 19:34:27,589 INFO [train.py:903] (1/4) Epoch 1, batch 2050, loss[loss=0.5522, simple_loss=0.524, pruned_loss=0.2902, over 19604.00 frames. ], tot_loss[loss=0.5484, simple_loss=0.511, pruned_loss=0.2952, over 3797751.54 frames. ], batch size: 57, lr: 4.81e-02, grad_scale: 16.0 2023-03-31 19:34:43,686 WARNING [train.py:1073] (1/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 2023-03-31 19:34:43,723 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 2023-03-31 19:35:06,983 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 2023-03-31 19:35:12,662 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=2088.0, num_to_drop=1, layers_to_drop={1} 2023-03-31 19:35:27,374 INFO [train.py:903] (1/4) Epoch 1, batch 2100, loss[loss=0.5711, simple_loss=0.531, pruned_loss=0.3056, over 18636.00 frames. ], tot_loss[loss=0.5406, simple_loss=0.5071, pruned_loss=0.2888, over 3804675.74 frames. ], batch size: 74, lr: 4.80e-02, grad_scale: 16.0 2023-03-31 19:35:30,829 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([5.5241, 4.8430, 2.3555, 4.9175, 1.8370, 5.7358, 5.1072, 5.2773], device='cuda:1'), covar=tensor([0.0342, 0.0682, 0.2341, 0.0344, 0.2408, 0.0216, 0.0472, 0.0390], device='cuda:1'), in_proj_covar=tensor([0.0142, 0.0177, 0.0179, 0.0122, 0.0195, 0.0108, 0.0123, 0.0112], device='cuda:1'), out_proj_covar=tensor([1.1510e-04, 1.3543e-04, 1.2023e-04, 8.8821e-05, 1.3606e-04, 7.8351e-05, 8.7946e-05, 8.3719e-05], device='cuda:1') 2023-03-31 19:35:31,674 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.951e+02 9.211e+02 1.091e+03 1.524e+03 2.851e+03, threshold=2.182e+03, percent-clipped=6.0 2023-03-31 19:35:56,698 WARNING [train.py:1073] (1/4) Exclude cut with ID 453-131332-0000-131866_sp0.9_rvb from training. Duration: 25.3333125 2023-03-31 19:36:17,092 WARNING [train.py:1073] (1/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9_rvb from training. Duration: 26.6166875 2023-03-31 19:36:24,972 INFO [train.py:903] (1/4) Epoch 1, batch 2150, loss[loss=0.4753, simple_loss=0.4737, pruned_loss=0.2385, over 19669.00 frames. ], tot_loss[loss=0.531, simple_loss=0.5013, pruned_loss=0.2817, over 3815778.78 frames. ], batch size: 58, lr: 4.79e-02, grad_scale: 16.0 2023-03-31 19:36:45,539 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=2167.0, num_to_drop=1, layers_to_drop={1} 2023-03-31 19:37:13,894 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=2192.0, num_to_drop=1, layers_to_drop={0} 2023-03-31 19:37:24,837 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=2200.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 19:37:25,806 INFO [train.py:903] (1/4) Epoch 1, batch 2200, loss[loss=0.5041, simple_loss=0.4962, pruned_loss=0.256, over 19665.00 frames. ], tot_loss[loss=0.526, simple_loss=0.4993, pruned_loss=0.2774, over 3818616.67 frames. ], batch size: 59, lr: 4.78e-02, grad_scale: 16.0 2023-03-31 19:37:31,690 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 5.347e+02 9.332e+02 1.145e+03 1.435e+03 3.303e+03, threshold=2.290e+03, percent-clipped=7.0 2023-03-31 19:37:49,477 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=2219.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 19:37:55,065 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.9732, 4.2114, 2.6159, 4.2865, 1.7849, 5.0783, 4.5984, 4.6067], device='cuda:1'), covar=tensor([0.0394, 0.0849, 0.2427, 0.0491, 0.2592, 0.0303, 0.0527, 0.0689], device='cuda:1'), in_proj_covar=tensor([0.0153, 0.0186, 0.0191, 0.0130, 0.0203, 0.0116, 0.0128, 0.0117], device='cuda:1'), out_proj_covar=tensor([1.2508e-04, 1.4385e-04, 1.2808e-04, 9.5759e-05, 1.4200e-04, 8.4706e-05, 9.1586e-05, 8.7291e-05], device='cuda:1') 2023-03-31 19:38:03,614 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=2232.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 19:38:16,705 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4583, 1.3205, 1.1032, 1.6203, 1.4337, 1.7238, 1.6284, 1.4912], device='cuda:1'), covar=tensor([0.2296, 0.3138, 0.5152, 0.2586, 0.6390, 0.1928, 0.3499, 0.2691], device='cuda:1'), in_proj_covar=tensor([0.0147, 0.0160, 0.0214, 0.0155, 0.0249, 0.0159, 0.0186, 0.0167], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0001, 0.0001], device='cuda:1') 2023-03-31 19:38:27,459 INFO [train.py:903] (1/4) Epoch 1, batch 2250, loss[loss=0.518, simple_loss=0.4944, pruned_loss=0.2707, over 19864.00 frames. ], tot_loss[loss=0.5174, simple_loss=0.4939, pruned_loss=0.2713, over 3820476.44 frames. ], batch size: 52, lr: 4.77e-02, grad_scale: 16.0 2023-03-31 19:39:11,079 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.4602, 1.7084, 1.8435, 2.1472, 3.0286, 1.2272, 2.4028, 3.1739], device='cuda:1'), covar=tensor([0.0317, 0.2551, 0.3370, 0.2197, 0.0513, 0.4283, 0.1075, 0.0570], device='cuda:1'), in_proj_covar=tensor([0.0107, 0.0178, 0.0201, 0.0186, 0.0142, 0.0267, 0.0165, 0.0141], device='cuda:1'), out_proj_covar=tensor([7.0953e-05, 1.2776e-04, 1.4438e-04, 1.4056e-04, 9.5749e-05, 1.7669e-04, 1.2525e-04, 1.0042e-04], device='cuda:1') 2023-03-31 19:39:24,146 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=2299.0, num_to_drop=1, layers_to_drop={0} 2023-03-31 19:39:25,913 INFO [train.py:903] (1/4) Epoch 1, batch 2300, loss[loss=0.4353, simple_loss=0.4397, pruned_loss=0.2155, over 19731.00 frames. ], tot_loss[loss=0.5119, simple_loss=0.4909, pruned_loss=0.2671, over 3823204.60 frames. ], batch size: 51, lr: 4.77e-02, grad_scale: 8.0 2023-03-31 19:39:31,315 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.883e+02 9.458e+02 1.205e+03 1.557e+03 3.326e+03, threshold=2.410e+03, percent-clipped=10.0 2023-03-31 19:39:39,125 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_rvb from training. Duration: 26.205 2023-03-31 19:39:41,688 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=2315.0, num_to_drop=1, layers_to_drop={0} 2023-03-31 19:39:53,919 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=2324.0, num_to_drop=1, layers_to_drop={1} 2023-03-31 19:40:04,979 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=2334.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 19:40:17,267 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=2344.0, num_to_drop=1, layers_to_drop={1} 2023-03-31 19:40:20,640 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=2347.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 19:40:24,828 INFO [train.py:903] (1/4) Epoch 1, batch 2350, loss[loss=0.4883, simple_loss=0.4893, pruned_loss=0.2437, over 19782.00 frames. ], tot_loss[loss=0.5095, simple_loss=0.4902, pruned_loss=0.2649, over 3831592.03 frames. ], batch size: 56, lr: 4.76e-02, grad_scale: 8.0 2023-03-31 19:40:33,243 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=2358.0, num_to_drop=1, layers_to_drop={1} 2023-03-31 19:40:48,369 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=2369.0, num_to_drop=1, layers_to_drop={1} 2023-03-31 19:40:57,588 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.89 vs. limit=2.0 2023-03-31 19:41:07,208 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_rvb from training. Duration: 25.775 2023-03-31 19:41:23,351 WARNING [train.py:1073] (1/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9_rvb from training. Duration: 25.45 2023-03-31 19:41:26,485 INFO [train.py:903] (1/4) Epoch 1, batch 2400, loss[loss=0.5022, simple_loss=0.5081, pruned_loss=0.2482, over 19647.00 frames. ], tot_loss[loss=0.506, simple_loss=0.4886, pruned_loss=0.2621, over 3830833.45 frames. ], batch size: 60, lr: 4.75e-02, grad_scale: 8.0 2023-03-31 19:41:33,165 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 5.460e+02 9.458e+02 1.226e+03 1.613e+03 2.603e+03, threshold=2.451e+03, percent-clipped=4.0 2023-03-31 19:42:26,147 INFO [train.py:903] (1/4) Epoch 1, batch 2450, loss[loss=0.4677, simple_loss=0.4536, pruned_loss=0.2409, over 19488.00 frames. ], tot_loss[loss=0.5023, simple_loss=0.4868, pruned_loss=0.2592, over 3828572.27 frames. ], batch size: 49, lr: 4.74e-02, grad_scale: 8.0 2023-03-31 19:43:24,714 INFO [train.py:903] (1/4) Epoch 1, batch 2500, loss[loss=0.5837, simple_loss=0.5339, pruned_loss=0.3167, over 19466.00 frames. ], tot_loss[loss=0.5005, simple_loss=0.4854, pruned_loss=0.258, over 3821894.69 frames. ], batch size: 64, lr: 4.73e-02, grad_scale: 8.0 2023-03-31 19:43:30,995 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 6.049e+02 1.082e+03 1.390e+03 1.742e+03 4.873e+03, threshold=2.779e+03, percent-clipped=5.0 2023-03-31 19:44:22,064 INFO [train.py:903] (1/4) Epoch 1, batch 2550, loss[loss=0.4602, simple_loss=0.4571, pruned_loss=0.2317, over 19588.00 frames. ], tot_loss[loss=0.4982, simple_loss=0.4843, pruned_loss=0.2563, over 3821050.90 frames. ], batch size: 52, lr: 4.72e-02, grad_scale: 8.0 2023-03-31 19:44:47,134 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=2571.0, num_to_drop=1, layers_to_drop={0} 2023-03-31 19:45:08,998 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=2590.0, num_to_drop=1, layers_to_drop={1} 2023-03-31 19:45:14,139 WARNING [train.py:1073] (1/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 2023-03-31 19:45:15,574 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=2596.0, num_to_drop=1, layers_to_drop={1} 2023-03-31 19:45:21,670 INFO [train.py:903] (1/4) Epoch 1, batch 2600, loss[loss=0.4827, simple_loss=0.4733, pruned_loss=0.246, over 18785.00 frames. ], tot_loss[loss=0.4903, simple_loss=0.4793, pruned_loss=0.2507, over 3817858.70 frames. ], batch size: 74, lr: 4.71e-02, grad_scale: 8.0 2023-03-31 19:45:24,454 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=2603.0, num_to_drop=1, layers_to_drop={0} 2023-03-31 19:45:28,253 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 5.353e+02 9.154e+02 1.259e+03 1.710e+03 2.682e+03, threshold=2.519e+03, percent-clipped=0.0 2023-03-31 19:45:39,542 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=2615.0, num_to_drop=1, layers_to_drop={0} 2023-03-31 19:45:55,210 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=2628.0, num_to_drop=1, layers_to_drop={1} 2023-03-31 19:46:22,934 INFO [train.py:903] (1/4) Epoch 1, batch 2650, loss[loss=0.4487, simple_loss=0.4473, pruned_loss=0.225, over 19608.00 frames. ], tot_loss[loss=0.4866, simple_loss=0.4771, pruned_loss=0.2481, over 3810995.90 frames. ], batch size: 50, lr: 4.70e-02, grad_scale: 8.0 2023-03-31 19:46:39,350 WARNING [train.py:1073] (1/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9_rvb from training. Duration: 27.25 2023-03-31 19:47:23,181 INFO [train.py:903] (1/4) Epoch 1, batch 2700, loss[loss=0.4836, simple_loss=0.4844, pruned_loss=0.2414, over 17184.00 frames. ], tot_loss[loss=0.4811, simple_loss=0.4742, pruned_loss=0.2441, over 3812808.31 frames. ], batch size: 101, lr: 4.69e-02, grad_scale: 8.0 2023-03-31 19:47:24,582 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=2702.0, num_to_drop=1, layers_to_drop={1} 2023-03-31 19:47:25,671 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=2703.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 19:47:29,726 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 5.007e+02 8.490e+02 1.133e+03 1.436e+03 3.154e+03, threshold=2.267e+03, percent-clipped=3.0 2023-03-31 19:48:24,803 INFO [train.py:903] (1/4) Epoch 1, batch 2750, loss[loss=0.4358, simple_loss=0.4557, pruned_loss=0.2079, over 19523.00 frames. ], tot_loss[loss=0.4767, simple_loss=0.4715, pruned_loss=0.241, over 3813883.48 frames. ], batch size: 54, lr: 4.68e-02, grad_scale: 8.0 2023-03-31 19:49:25,709 INFO [train.py:903] (1/4) Epoch 1, batch 2800, loss[loss=0.4808, simple_loss=0.4809, pruned_loss=0.2403, over 17225.00 frames. ], tot_loss[loss=0.4747, simple_loss=0.4707, pruned_loss=0.2394, over 3803816.22 frames. ], batch size: 101, lr: 4.67e-02, grad_scale: 8.0 2023-03-31 19:49:31,050 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.211e+02 1.002e+03 1.265e+03 1.511e+03 4.462e+03, threshold=2.529e+03, percent-clipped=7.0 2023-03-31 19:49:45,690 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=2817.0, num_to_drop=1, layers_to_drop={1} 2023-03-31 19:50:15,412 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=2842.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 19:50:26,209 INFO [train.py:903] (1/4) Epoch 1, batch 2850, loss[loss=0.4826, simple_loss=0.4758, pruned_loss=0.2446, over 19757.00 frames. ], tot_loss[loss=0.4733, simple_loss=0.4696, pruned_loss=0.2385, over 3813363.01 frames. ], batch size: 54, lr: 4.66e-02, grad_scale: 8.0 2023-03-31 19:51:22,261 WARNING [train.py:1073] (1/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 2023-03-31 19:51:25,262 INFO [train.py:903] (1/4) Epoch 1, batch 2900, loss[loss=0.4438, simple_loss=0.4527, pruned_loss=0.2174, over 19591.00 frames. ], tot_loss[loss=0.4695, simple_loss=0.4671, pruned_loss=0.2359, over 3819617.16 frames. ], batch size: 57, lr: 4.65e-02, grad_scale: 8.0 2023-03-31 19:51:30,473 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 6.237e+02 1.045e+03 1.349e+03 1.754e+03 3.463e+03, threshold=2.699e+03, percent-clipped=4.0 2023-03-31 19:52:12,180 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2987, 1.0528, 1.1257, 1.4045, 1.4411, 1.6557, 1.5136, 1.4849], device='cuda:1'), covar=tensor([0.1326, 0.2684, 0.2301, 0.1559, 0.2831, 0.0953, 0.1393, 0.1534], device='cuda:1'), in_proj_covar=tensor([0.0108, 0.0143, 0.0158, 0.0130, 0.0187, 0.0118, 0.0132, 0.0121], device='cuda:1'), out_proj_covar=tensor([7.8904e-05, 1.0216e-04, 1.0859e-04, 9.3535e-05, 1.3275e-04, 8.3586e-05, 9.1483e-05, 8.6552e-05], device='cuda:1') 2023-03-31 19:52:25,021 INFO [train.py:903] (1/4) Epoch 1, batch 2950, loss[loss=0.4407, simple_loss=0.4566, pruned_loss=0.2124, over 19661.00 frames. ], tot_loss[loss=0.4675, simple_loss=0.466, pruned_loss=0.2345, over 3811471.93 frames. ], batch size: 55, lr: 4.64e-02, grad_scale: 8.0 2023-03-31 19:52:30,813 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=2955.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 19:52:32,690 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.85 vs. limit=2.0 2023-03-31 19:53:26,125 INFO [train.py:903] (1/4) Epoch 1, batch 3000, loss[loss=0.4436, simple_loss=0.4581, pruned_loss=0.2145, over 19450.00 frames. ], tot_loss[loss=0.4631, simple_loss=0.4633, pruned_loss=0.2315, over 3820259.06 frames. ], batch size: 64, lr: 4.63e-02, grad_scale: 8.0 2023-03-31 19:53:26,126 INFO [train.py:928] (1/4) Computing validation loss 2023-03-31 19:53:37,095 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2079, 1.1985, 1.2532, 1.1677, 1.1447, 0.7766, 0.5395, 1.2007], device='cuda:1'), covar=tensor([0.1550, 0.0771, 0.0671, 0.1136, 0.1257, 0.1364, 0.2354, 0.1184], device='cuda:1'), in_proj_covar=tensor([0.0200, 0.0127, 0.0119, 0.0151, 0.0129, 0.0170, 0.0197, 0.0173], device='cuda:1'), out_proj_covar=tensor([1.5072e-04, 9.8014e-05, 9.2597e-05, 1.1889e-04, 1.0297e-04, 1.3044e-04, 1.4498e-04, 1.3205e-04], device='cuda:1') 2023-03-31 19:53:38,705 INFO [train.py:937] (1/4) Epoch 1, validation: loss=0.3995, simple_loss=0.4801, pruned_loss=0.1594, over 944034.00 frames. 2023-03-31 19:53:38,706 INFO [train.py:938] (1/4) Maximum memory allocated so far is 16499MB 2023-03-31 19:53:43,179 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_rvb from training. Duration: 29.735 2023-03-31 19:53:45,669 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.188e+02 9.060e+02 1.151e+03 1.550e+03 2.691e+03, threshold=2.303e+03, percent-clipped=0.0 2023-03-31 19:54:23,279 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=3037.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 19:54:35,524 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=3047.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 19:54:40,055 INFO [train.py:903] (1/4) Epoch 1, batch 3050, loss[loss=0.4647, simple_loss=0.4713, pruned_loss=0.2291, over 19778.00 frames. ], tot_loss[loss=0.4598, simple_loss=0.4608, pruned_loss=0.2294, over 3818699.47 frames. ], batch size: 56, lr: 4.62e-02, grad_scale: 8.0 2023-03-31 19:55:07,327 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=3073.0, num_to_drop=1, layers_to_drop={1} 2023-03-31 19:55:36,291 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=3098.0, num_to_drop=1, layers_to_drop={1} 2023-03-31 19:55:37,609 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.68 vs. limit=2.0 2023-03-31 19:55:39,201 INFO [train.py:903] (1/4) Epoch 1, batch 3100, loss[loss=0.4919, simple_loss=0.48, pruned_loss=0.2519, over 19672.00 frames. ], tot_loss[loss=0.4607, simple_loss=0.4613, pruned_loss=0.2301, over 3825696.79 frames. ], batch size: 60, lr: 4.61e-02, grad_scale: 8.0 2023-03-31 19:55:45,832 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 5.197e+02 1.021e+03 1.362e+03 1.815e+03 5.785e+03, threshold=2.723e+03, percent-clipped=14.0 2023-03-31 19:56:41,856 INFO [train.py:903] (1/4) Epoch 1, batch 3150, loss[loss=0.4158, simple_loss=0.4248, pruned_loss=0.2034, over 19802.00 frames. ], tot_loss[loss=0.4608, simple_loss=0.4621, pruned_loss=0.2297, over 3828550.79 frames. ], batch size: 49, lr: 4.60e-02, grad_scale: 8.0 2023-03-31 19:56:54,535 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=3162.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 19:57:08,962 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 2023-03-31 19:57:24,055 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=3186.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 19:57:26,957 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.99 vs. limit=2.0 2023-03-31 19:57:36,131 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.88 vs. limit=2.0 2023-03-31 19:57:40,821 INFO [train.py:903] (1/4) Epoch 1, batch 3200, loss[loss=0.4424, simple_loss=0.4609, pruned_loss=0.212, over 19574.00 frames. ], tot_loss[loss=0.4587, simple_loss=0.461, pruned_loss=0.2282, over 3844104.19 frames. ], batch size: 57, lr: 4.59e-02, grad_scale: 8.0 2023-03-31 19:57:45,653 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6291, 1.2717, 1.4082, 1.0521, 2.2069, 2.2286, 2.0157, 2.0588], device='cuda:1'), covar=tensor([0.1281, 0.1813, 0.1273, 0.2224, 0.0525, 0.0162, 0.0291, 0.0280], device='cuda:1'), in_proj_covar=tensor([0.0253, 0.0215, 0.0203, 0.0259, 0.0181, 0.0103, 0.0129, 0.0107], device='cuda:1'), out_proj_covar=tensor([1.7718e-04, 1.4783e-04, 1.4082e-04, 1.7845e-04, 1.4670e-04, 6.7671e-05, 9.2051e-05, 7.5712e-05], device='cuda:1') 2023-03-31 19:57:46,447 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 5.226e+02 9.158e+02 1.127e+03 1.418e+03 2.574e+03, threshold=2.253e+03, percent-clipped=0.0 2023-03-31 19:58:41,766 INFO [train.py:903] (1/4) Epoch 1, batch 3250, loss[loss=0.4658, simple_loss=0.4723, pruned_loss=0.2296, over 19305.00 frames. ], tot_loss[loss=0.4585, simple_loss=0.4611, pruned_loss=0.2279, over 3842612.81 frames. ], batch size: 66, lr: 4.58e-02, grad_scale: 8.0 2023-03-31 19:59:13,240 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6255, 1.4060, 1.6593, 1.3346, 1.1061, 1.4225, 0.6466, 1.0014], device='cuda:1'), covar=tensor([0.0644, 0.0545, 0.0418, 0.0624, 0.0928, 0.0858, 0.1344, 0.1298], device='cuda:1'), in_proj_covar=tensor([0.0086, 0.0086, 0.0092, 0.0115, 0.0122, 0.0120, 0.0146, 0.0140], device='cuda:1'), out_proj_covar=tensor([6.9507e-05, 6.1082e-05, 7.0318e-05, 8.8108e-05, 9.1278e-05, 8.9758e-05, 1.0871e-04, 1.0688e-04], device='cuda:1') 2023-03-31 19:59:40,379 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=3299.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 19:59:42,431 INFO [train.py:903] (1/4) Epoch 1, batch 3300, loss[loss=0.4492, simple_loss=0.4451, pruned_loss=0.2266, over 19581.00 frames. ], tot_loss[loss=0.4539, simple_loss=0.458, pruned_loss=0.2249, over 3837231.61 frames. ], batch size: 52, lr: 4.57e-02, grad_scale: 8.0 2023-03-31 19:59:42,835 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=3301.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 19:59:48,772 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 6.005e+02 9.991e+02 1.183e+03 1.562e+03 4.237e+03, threshold=2.366e+03, percent-clipped=7.0 2023-03-31 19:59:48,808 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 2023-03-31 19:59:49,095 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=3306.0, num_to_drop=1, layers_to_drop={1} 2023-03-31 20:00:35,188 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([0.9358, 1.2821, 2.2660, 1.6083, 2.8121, 4.2201, 3.2661, 2.6013], device='cuda:1'), covar=tensor([0.2436, 0.1571, 0.1882, 0.2181, 0.1073, 0.0244, 0.0938, 0.1382], device='cuda:1'), in_proj_covar=tensor([0.0179, 0.0144, 0.0166, 0.0194, 0.0172, 0.0095, 0.0164, 0.0171], device='cuda:1'), out_proj_covar=tensor([1.2361e-04, 9.9629e-05, 1.1801e-04, 1.3407e-04, 1.1563e-04, 6.7293e-05, 1.0771e-04, 1.1344e-04], device='cuda:1') 2023-03-31 20:00:43,738 INFO [train.py:903] (1/4) Epoch 1, batch 3350, loss[loss=0.4168, simple_loss=0.4346, pruned_loss=0.1995, over 19483.00 frames. ], tot_loss[loss=0.4505, simple_loss=0.4554, pruned_loss=0.2228, over 3838058.11 frames. ], batch size: 49, lr: 4.56e-02, grad_scale: 8.0 2023-03-31 20:01:22,255 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=3381.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 20:01:46,098 INFO [train.py:903] (1/4) Epoch 1, batch 3400, loss[loss=0.4803, simple_loss=0.4743, pruned_loss=0.2432, over 19613.00 frames. ], tot_loss[loss=0.4511, simple_loss=0.4555, pruned_loss=0.2234, over 3836945.33 frames. ], batch size: 57, lr: 4.55e-02, grad_scale: 8.0 2023-03-31 20:01:52,899 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 5.525e+02 9.967e+02 1.253e+03 1.611e+03 4.007e+03, threshold=2.507e+03, percent-clipped=3.0 2023-03-31 20:02:02,098 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=3414.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 20:02:07,714 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=3418.0, num_to_drop=1, layers_to_drop={0} 2023-03-31 20:02:38,872 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=3443.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 20:02:48,478 INFO [train.py:903] (1/4) Epoch 1, batch 3450, loss[loss=0.4269, simple_loss=0.4443, pruned_loss=0.2048, over 19448.00 frames. ], tot_loss[loss=0.4526, simple_loss=0.4565, pruned_loss=0.2243, over 3810457.99 frames. ], batch size: 64, lr: 4.54e-02, grad_scale: 8.0 2023-03-31 20:02:50,744 WARNING [train.py:1073] (1/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9_rvb from training. Duration: 25.2444375 2023-03-31 20:03:25,943 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2249, 1.0979, 0.9747, 1.3957, 1.2862, 1.2689, 1.2689, 1.2672], device='cuda:1'), covar=tensor([0.1141, 0.1588, 0.1858, 0.1391, 0.1777, 0.1793, 0.2019, 0.1283], device='cuda:1'), in_proj_covar=tensor([0.0206, 0.0255, 0.0259, 0.0280, 0.0363, 0.0246, 0.0323, 0.0225], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-03-31 20:03:43,985 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=3496.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 20:03:50,180 INFO [train.py:903] (1/4) Epoch 1, batch 3500, loss[loss=0.454, simple_loss=0.4719, pruned_loss=0.218, over 19555.00 frames. ], tot_loss[loss=0.4509, simple_loss=0.4558, pruned_loss=0.223, over 3825907.44 frames. ], batch size: 56, lr: 4.53e-02, grad_scale: 8.0 2023-03-31 20:03:56,687 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 5.496e+02 9.703e+02 1.213e+03 1.703e+03 9.610e+03, threshold=2.427e+03, percent-clipped=9.0 2023-03-31 20:04:52,056 INFO [train.py:903] (1/4) Epoch 1, batch 3550, loss[loss=0.4379, simple_loss=0.4517, pruned_loss=0.212, over 19787.00 frames. ], tot_loss[loss=0.4465, simple_loss=0.4529, pruned_loss=0.22, over 3829761.75 frames. ], batch size: 56, lr: 4.51e-02, grad_scale: 8.0 2023-03-31 20:05:00,241 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=3557.0, num_to_drop=1, layers_to_drop={1} 2023-03-31 20:05:07,891 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=3564.0, num_to_drop=1, layers_to_drop={1} 2023-03-31 20:05:30,991 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=3582.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 20:05:44,033 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=5.07 vs. limit=5.0 2023-03-31 20:05:53,927 INFO [train.py:903] (1/4) Epoch 1, batch 3600, loss[loss=0.3539, simple_loss=0.3813, pruned_loss=0.1632, over 16449.00 frames. ], tot_loss[loss=0.4589, simple_loss=0.4604, pruned_loss=0.2287, over 3818352.54 frames. ], batch size: 36, lr: 4.50e-02, grad_scale: 8.0 2023-03-31 20:06:00,964 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 5.467e+02 9.459e+02 1.417e+03 1.964e+03 2.103e+04, threshold=2.834e+03, percent-clipped=17.0 2023-03-31 20:06:55,025 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=3650.0, num_to_drop=1, layers_to_drop={0} 2023-03-31 20:06:55,790 INFO [train.py:903] (1/4) Epoch 1, batch 3650, loss[loss=0.4618, simple_loss=0.4376, pruned_loss=0.243, over 19763.00 frames. ], tot_loss[loss=0.4552, simple_loss=0.4585, pruned_loss=0.2259, over 3822099.68 frames. ], batch size: 46, lr: 4.49e-02, grad_scale: 8.0 2023-03-31 20:07:18,743 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=3670.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 20:07:49,493 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=3694.0, num_to_drop=1, layers_to_drop={1} 2023-03-31 20:07:50,793 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=3695.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 20:07:56,859 INFO [train.py:903] (1/4) Epoch 1, batch 3700, loss[loss=0.4446, simple_loss=0.4404, pruned_loss=0.2245, over 19099.00 frames. ], tot_loss[loss=0.4619, simple_loss=0.4632, pruned_loss=0.2303, over 3813277.64 frames. ], batch size: 42, lr: 4.48e-02, grad_scale: 8.0 2023-03-31 20:08:05,842 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 5.255e+02 1.022e+03 1.666e+03 2.666e+03 1.441e+04, threshold=3.331e+03, percent-clipped=22.0 2023-03-31 20:09:01,163 INFO [train.py:903] (1/4) Epoch 1, batch 3750, loss[loss=0.4832, simple_loss=0.4924, pruned_loss=0.237, over 19663.00 frames. ], tot_loss[loss=0.4562, simple_loss=0.4594, pruned_loss=0.2264, over 3818948.61 frames. ], batch size: 60, lr: 4.47e-02, grad_scale: 8.0 2023-03-31 20:09:02,739 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=3752.0, num_to_drop=1, layers_to_drop={1} 2023-03-31 20:09:19,385 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=3765.0, num_to_drop=1, layers_to_drop={1} 2023-03-31 20:09:34,305 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=3777.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 20:09:37,704 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([0.9413, 1.0756, 0.9195, 0.9422, 1.0197, 0.7316, 0.2774, 1.2271], device='cuda:1'), covar=tensor([0.1006, 0.0641, 0.0932, 0.1098, 0.0911, 0.1472, 0.1968, 0.0936], device='cuda:1'), in_proj_covar=tensor([0.0235, 0.0153, 0.0155, 0.0193, 0.0145, 0.0214, 0.0233, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-03-31 20:09:51,927 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.7686, 1.2041, 1.5542, 1.0056, 2.5370, 3.1192, 2.7500, 2.6548], device='cuda:1'), covar=tensor([0.1854, 0.2843, 0.2059, 0.2894, 0.0635, 0.0189, 0.0283, 0.0392], device='cuda:1'), in_proj_covar=tensor([0.0280, 0.0253, 0.0254, 0.0285, 0.0199, 0.0130, 0.0151, 0.0129], device='cuda:1'), out_proj_covar=tensor([2.0324e-04, 1.8362e-04, 1.8715e-04, 2.0750e-04, 1.7186e-04, 8.9677e-05, 1.1534e-04, 9.8026e-05], device='cuda:1') 2023-03-31 20:09:54,755 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.02 vs. limit=5.0 2023-03-31 20:10:05,359 INFO [train.py:903] (1/4) Epoch 1, batch 3800, loss[loss=0.3939, simple_loss=0.4054, pruned_loss=0.1912, over 19373.00 frames. ], tot_loss[loss=0.4495, simple_loss=0.4552, pruned_loss=0.2219, over 3819876.54 frames. ], batch size: 47, lr: 4.46e-02, grad_scale: 8.0 2023-03-31 20:10:12,575 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 5.844e+02 1.035e+03 1.394e+03 1.973e+03 4.112e+03, threshold=2.788e+03, percent-clipped=1.0 2023-03-31 20:10:41,855 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9_rvb from training. Duration: 29.1166875 2023-03-31 20:11:08,740 INFO [train.py:903] (1/4) Epoch 1, batch 3850, loss[loss=0.4954, simple_loss=0.4751, pruned_loss=0.2578, over 13554.00 frames. ], tot_loss[loss=0.4494, simple_loss=0.4551, pruned_loss=0.2218, over 3801379.37 frames. ], batch size: 136, lr: 4.45e-02, grad_scale: 8.0 2023-03-31 20:12:11,133 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([0.9160, 1.0017, 2.1149, 1.4536, 2.4281, 3.3048, 3.0733, 2.0063], device='cuda:1'), covar=tensor([0.2522, 0.2251, 0.1996, 0.2103, 0.1312, 0.0559, 0.1134, 0.1809], device='cuda:1'), in_proj_covar=tensor([0.0226, 0.0205, 0.0208, 0.0239, 0.0217, 0.0148, 0.0213, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001], device='cuda:1') 2023-03-31 20:12:13,108 INFO [train.py:903] (1/4) Epoch 1, batch 3900, loss[loss=0.4498, simple_loss=0.4566, pruned_loss=0.2215, over 19593.00 frames. ], tot_loss[loss=0.4481, simple_loss=0.4541, pruned_loss=0.221, over 3792573.00 frames. ], batch size: 57, lr: 4.44e-02, grad_scale: 8.0 2023-03-31 20:12:22,003 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 6.917e+02 1.152e+03 1.441e+03 1.935e+03 3.736e+03, threshold=2.883e+03, percent-clipped=2.0 2023-03-31 20:12:23,348 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=3908.0, num_to_drop=1, layers_to_drop={1} 2023-03-31 20:13:08,696 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=3944.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 20:13:18,187 INFO [train.py:903] (1/4) Epoch 1, batch 3950, loss[loss=0.4322, simple_loss=0.457, pruned_loss=0.2037, over 18077.00 frames. ], tot_loss[loss=0.4488, simple_loss=0.4547, pruned_loss=0.2215, over 3810997.76 frames. ], batch size: 83, lr: 4.43e-02, grad_scale: 8.0 2023-03-31 20:13:23,982 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 2023-03-31 20:13:24,176 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.6535, 4.2361, 2.2096, 3.7958, 1.4131, 4.3091, 3.8693, 3.9636], device='cuda:1'), covar=tensor([0.0401, 0.0851, 0.2409, 0.0583, 0.3049, 0.0676, 0.0482, 0.0643], device='cuda:1'), in_proj_covar=tensor([0.0203, 0.0219, 0.0251, 0.0197, 0.0265, 0.0198, 0.0156, 0.0162], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001], device='cuda:1') 2023-03-31 20:14:23,991 INFO [train.py:903] (1/4) Epoch 1, batch 4000, loss[loss=0.3436, simple_loss=0.3692, pruned_loss=0.1591, over 19700.00 frames. ], tot_loss[loss=0.4437, simple_loss=0.4514, pruned_loss=0.218, over 3815695.04 frames. ], batch size: 45, lr: 4.42e-02, grad_scale: 8.0 2023-03-31 20:14:30,939 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 5.942e+02 1.053e+03 1.358e+03 1.948e+03 9.883e+03, threshold=2.717e+03, percent-clipped=12.0 2023-03-31 20:14:49,349 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=4021.0, num_to_drop=1, layers_to_drop={0} 2023-03-31 20:14:52,454 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=4023.0, num_to_drop=1, layers_to_drop={0} 2023-03-31 20:14:54,228 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.23 vs. limit=5.0 2023-03-31 20:15:11,861 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=4038.0, num_to_drop=1, layers_to_drop={0} 2023-03-31 20:15:12,778 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1_rvb from training. Duration: 0.7545625 2023-03-31 20:15:22,680 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=4046.0, num_to_drop=1, layers_to_drop={0} 2023-03-31 20:15:28,264 INFO [train.py:903] (1/4) Epoch 1, batch 4050, loss[loss=0.4485, simple_loss=0.4485, pruned_loss=0.2243, over 19845.00 frames. ], tot_loss[loss=0.4427, simple_loss=0.4503, pruned_loss=0.2175, over 3809615.81 frames. ], batch size: 52, lr: 4.41e-02, grad_scale: 8.0 2023-03-31 20:15:40,280 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2430, 1.2100, 0.9462, 1.0533, 0.8335, 1.0357, 0.2163, 0.5934], device='cuda:1'), covar=tensor([0.0666, 0.0722, 0.0555, 0.0758, 0.1224, 0.0959, 0.1978, 0.1666], device='cuda:1'), in_proj_covar=tensor([0.0113, 0.0117, 0.0125, 0.0146, 0.0170, 0.0159, 0.0188, 0.0182], device='cuda:1'), out_proj_covar=tensor([9.9255e-05, 9.3466e-05, 9.9753e-05, 1.1115e-04, 1.3213e-04, 1.2078e-04, 1.4041e-04, 1.4460e-04], device='cuda:1') 2023-03-31 20:16:32,903 INFO [train.py:903] (1/4) Epoch 1, batch 4100, loss[loss=0.5205, simple_loss=0.5065, pruned_loss=0.2672, over 16985.00 frames. ], tot_loss[loss=0.4416, simple_loss=0.4495, pruned_loss=0.2169, over 3805313.16 frames. ], batch size: 101, lr: 4.40e-02, grad_scale: 8.0 2023-03-31 20:16:41,799 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.649e+02 1.198e+03 1.458e+03 1.833e+03 3.490e+03, threshold=2.915e+03, percent-clipped=3.0 2023-03-31 20:17:08,497 WARNING [train.py:1073] (1/4) Exclude cut with ID 497-129325-0061-9566_sp1.1_rvb from training. Duration: 0.97725 2023-03-31 20:17:38,860 INFO [train.py:903] (1/4) Epoch 1, batch 4150, loss[loss=0.4902, simple_loss=0.4799, pruned_loss=0.2503, over 13634.00 frames. ], tot_loss[loss=0.4391, simple_loss=0.448, pruned_loss=0.2152, over 3789411.88 frames. ], batch size: 136, lr: 4.39e-02, grad_scale: 8.0 2023-03-31 20:17:41,677 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=4153.0, num_to_drop=1, layers_to_drop={0} 2023-03-31 20:17:49,785 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=4159.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 20:18:31,271 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.77 vs. limit=2.0 2023-03-31 20:18:44,276 INFO [train.py:903] (1/4) Epoch 1, batch 4200, loss[loss=0.3237, simple_loss=0.3629, pruned_loss=0.1422, over 19740.00 frames. ], tot_loss[loss=0.4347, simple_loss=0.4452, pruned_loss=0.2121, over 3803331.30 frames. ], batch size: 46, lr: 4.38e-02, grad_scale: 8.0 2023-03-31 20:18:46,642 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 2023-03-31 20:18:51,463 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 5.163e+02 8.919e+02 1.098e+03 1.489e+03 3.268e+03, threshold=2.196e+03, percent-clipped=3.0 2023-03-31 20:19:47,734 INFO [train.py:903] (1/4) Epoch 1, batch 4250, loss[loss=0.4104, simple_loss=0.4301, pruned_loss=0.1954, over 19671.00 frames. ], tot_loss[loss=0.4349, simple_loss=0.4455, pruned_loss=0.2122, over 3814747.85 frames. ], batch size: 58, lr: 4.36e-02, grad_scale: 8.0 2023-03-31 20:19:57,795 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 2023-03-31 20:20:02,063 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9_rvb from training. Duration: 29.816625 2023-03-31 20:20:15,219 WARNING [train.py:1073] (1/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 2023-03-31 20:20:25,098 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=4279.0, num_to_drop=1, layers_to_drop={0} 2023-03-31 20:20:36,582 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=4288.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 20:20:52,965 INFO [train.py:903] (1/4) Epoch 1, batch 4300, loss[loss=0.4446, simple_loss=0.4543, pruned_loss=0.2175, over 19670.00 frames. ], tot_loss[loss=0.4339, simple_loss=0.4446, pruned_loss=0.2116, over 3820725.35 frames. ], batch size: 60, lr: 4.35e-02, grad_scale: 8.0 2023-03-31 20:20:57,860 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=4304.0, num_to_drop=1, layers_to_drop={1} 2023-03-31 20:21:00,201 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=4306.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 20:21:02,053 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 6.021e+02 1.171e+03 1.478e+03 2.100e+03 3.660e+03, threshold=2.957e+03, percent-clipped=20.0 2023-03-31 20:21:46,486 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 2023-03-31 20:21:59,419 INFO [train.py:903] (1/4) Epoch 1, batch 4350, loss[loss=0.3566, simple_loss=0.3741, pruned_loss=0.1696, over 19755.00 frames. ], tot_loss[loss=0.4293, simple_loss=0.4416, pruned_loss=0.2085, over 3819141.90 frames. ], batch size: 46, lr: 4.34e-02, grad_scale: 8.0 2023-03-31 20:22:08,240 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9706, 2.0312, 1.3950, 2.0002, 1.6757, 1.0389, 1.1149, 1.3655], device='cuda:1'), covar=tensor([0.1147, 0.0825, 0.1059, 0.1018, 0.1209, 0.1370, 0.2485, 0.1474], device='cuda:1'), in_proj_covar=tensor([0.0104, 0.0090, 0.0119, 0.0130, 0.0136, 0.0079, 0.0136, 0.0113], device='cuda:1'), out_proj_covar=tensor([7.3150e-05, 6.2377e-05, 7.7721e-05, 8.6248e-05, 8.5480e-05, 4.8552e-05, 9.9747e-05, 7.6768e-05], device='cuda:1') 2023-03-31 20:23:03,132 INFO [train.py:903] (1/4) Epoch 1, batch 4400, loss[loss=0.4334, simple_loss=0.4404, pruned_loss=0.2132, over 19511.00 frames. ], tot_loss[loss=0.429, simple_loss=0.4418, pruned_loss=0.2081, over 3827576.28 frames. ], batch size: 54, lr: 4.33e-02, grad_scale: 8.0 2023-03-31 20:23:05,853 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=4403.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 20:23:11,485 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 5.454e+02 9.391e+02 1.114e+03 1.514e+03 3.216e+03, threshold=2.228e+03, percent-clipped=1.0 2023-03-31 20:23:14,559 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=4409.0, num_to_drop=1, layers_to_drop={1} 2023-03-31 20:23:29,261 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_rvb from training. Duration: 25.285 2023-03-31 20:23:39,400 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 2023-03-31 20:23:46,884 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=4434.0, num_to_drop=1, layers_to_drop={0} 2023-03-31 20:24:06,851 INFO [train.py:903] (1/4) Epoch 1, batch 4450, loss[loss=0.4172, simple_loss=0.4367, pruned_loss=0.1989, over 19668.00 frames. ], tot_loss[loss=0.4295, simple_loss=0.4426, pruned_loss=0.2082, over 3826043.59 frames. ], batch size: 55, lr: 4.32e-02, grad_scale: 8.0 2023-03-31 20:25:09,797 INFO [train.py:903] (1/4) Epoch 1, batch 4500, loss[loss=0.3992, simple_loss=0.4329, pruned_loss=0.1827, over 18068.00 frames. ], tot_loss[loss=0.4264, simple_loss=0.4402, pruned_loss=0.2063, over 3822719.15 frames. ], batch size: 84, lr: 4.31e-02, grad_scale: 8.0 2023-03-31 20:25:12,514 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=4503.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 20:25:18,130 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 5.500e+02 1.049e+03 1.357e+03 1.620e+03 3.962e+03, threshold=2.713e+03, percent-clipped=8.0 2023-03-31 20:26:14,039 INFO [train.py:903] (1/4) Epoch 1, batch 4550, loss[loss=0.5119, simple_loss=0.4995, pruned_loss=0.2621, over 13797.00 frames. ], tot_loss[loss=0.4243, simple_loss=0.439, pruned_loss=0.2048, over 3805142.19 frames. ], batch size: 135, lr: 4.30e-02, grad_scale: 8.0 2023-03-31 20:26:18,724 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.6466, 1.3281, 1.3300, 2.4763, 3.2058, 1.5295, 2.2505, 3.2461], device='cuda:1'), covar=tensor([0.0373, 0.2847, 0.3077, 0.1356, 0.0438, 0.2288, 0.1066, 0.0429], device='cuda:1'), in_proj_covar=tensor([0.0169, 0.0265, 0.0256, 0.0254, 0.0191, 0.0310, 0.0227, 0.0201], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-03-31 20:26:24,210 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9_rvb from training. Duration: 28.72225 2023-03-31 20:26:47,295 WARNING [train.py:1073] (1/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 2023-03-31 20:26:58,296 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3100, 1.4014, 1.0846, 1.4932, 1.3120, 1.5146, 1.2512, 1.4399], device='cuda:1'), covar=tensor([0.1374, 0.2269, 0.2132, 0.1352, 0.2367, 0.0969, 0.1962, 0.1016], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0283, 0.0252, 0.0213, 0.0278, 0.0206, 0.0231, 0.0190], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001], device='cuda:1') 2023-03-31 20:27:16,034 INFO [train.py:903] (1/4) Epoch 1, batch 4600, loss[loss=0.4438, simple_loss=0.4645, pruned_loss=0.2116, over 19752.00 frames. ], tot_loss[loss=0.4251, simple_loss=0.4394, pruned_loss=0.2054, over 3804425.14 frames. ], batch size: 63, lr: 4.29e-02, grad_scale: 4.0 2023-03-31 20:27:24,059 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 6.210e+02 9.691e+02 1.279e+03 1.723e+03 8.130e+03, threshold=2.557e+03, percent-clipped=7.0 2023-03-31 20:27:36,614 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=4618.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 20:27:41,329 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.7191, 1.2298, 1.1174, 1.4701, 1.3736, 1.5869, 1.5129, 1.6207], device='cuda:1'), covar=tensor([0.0799, 0.1643, 0.1632, 0.1488, 0.1898, 0.1602, 0.1935, 0.0936], device='cuda:1'), in_proj_covar=tensor([0.0229, 0.0296, 0.0286, 0.0311, 0.0392, 0.0279, 0.0350, 0.0254], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-03-31 20:28:17,147 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=4650.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 20:28:18,170 INFO [train.py:903] (1/4) Epoch 1, batch 4650, loss[loss=0.3356, simple_loss=0.3666, pruned_loss=0.1524, over 19802.00 frames. ], tot_loss[loss=0.423, simple_loss=0.4382, pruned_loss=0.2039, over 3794985.35 frames. ], batch size: 47, lr: 4.28e-02, grad_scale: 4.0 2023-03-31 20:28:27,882 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=4659.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 20:28:34,604 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_rvb from training. Duration: 0.92 2023-03-31 20:28:44,784 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_rvb from training. Duration: 0.83 2023-03-31 20:28:58,620 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=4684.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 20:29:18,955 INFO [train.py:903] (1/4) Epoch 1, batch 4700, loss[loss=0.4113, simple_loss=0.4231, pruned_loss=0.1998, over 19783.00 frames. ], tot_loss[loss=0.422, simple_loss=0.4376, pruned_loss=0.2032, over 3794812.40 frames. ], batch size: 47, lr: 4.27e-02, grad_scale: 4.0 2023-03-31 20:29:28,009 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.618e+02 9.658e+02 1.202e+03 1.526e+03 2.859e+03, threshold=2.405e+03, percent-clipped=1.0 2023-03-31 20:29:39,197 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9_rvb from training. Duration: 0.92225 2023-03-31 20:30:21,814 INFO [train.py:903] (1/4) Epoch 1, batch 4750, loss[loss=0.4236, simple_loss=0.4454, pruned_loss=0.2009, over 18855.00 frames. ], tot_loss[loss=0.4215, simple_loss=0.4374, pruned_loss=0.2028, over 3790455.65 frames. ], batch size: 74, lr: 4.26e-02, grad_scale: 4.0 2023-03-31 20:30:39,338 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=4765.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 20:31:23,846 INFO [train.py:903] (1/4) Epoch 1, batch 4800, loss[loss=0.321, simple_loss=0.3623, pruned_loss=0.1399, over 19412.00 frames. ], tot_loss[loss=0.4199, simple_loss=0.436, pruned_loss=0.2019, over 3801507.49 frames. ], batch size: 48, lr: 4.25e-02, grad_scale: 8.0 2023-03-31 20:31:32,976 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 7.504e+02 1.038e+03 1.224e+03 1.522e+03 3.175e+03, threshold=2.447e+03, percent-clipped=5.0 2023-03-31 20:32:25,500 INFO [train.py:903] (1/4) Epoch 1, batch 4850, loss[loss=0.3662, simple_loss=0.3847, pruned_loss=0.1739, over 19796.00 frames. ], tot_loss[loss=0.419, simple_loss=0.4347, pruned_loss=0.2016, over 3791559.55 frames. ], batch size: 48, lr: 4.24e-02, grad_scale: 8.0 2023-03-31 20:32:46,766 WARNING [train.py:1073] (1/4) Exclude cut with ID 774-127930-0014-48411_sp1.1_rvb from training. Duration: 0.95 2023-03-31 20:32:54,335 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=4874.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 20:33:06,425 WARNING [train.py:1073] (1/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1_rvb from training. Duration: 0.9409375 2023-03-31 20:33:12,764 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 2023-03-31 20:33:12,785 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 2023-03-31 20:33:23,800 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1_rvb from training. Duration: 25.3818125 2023-03-31 20:33:25,326 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=4899.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 20:33:27,240 INFO [train.py:903] (1/4) Epoch 1, batch 4900, loss[loss=0.4146, simple_loss=0.4418, pruned_loss=0.1937, over 18055.00 frames. ], tot_loss[loss=0.4194, simple_loss=0.4355, pruned_loss=0.2016, over 3777807.24 frames. ], batch size: 83, lr: 4.23e-02, grad_scale: 8.0 2023-03-31 20:33:37,034 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 6.319e+02 9.845e+02 1.167e+03 1.485e+03 2.856e+03, threshold=2.333e+03, percent-clipped=2.0 2023-03-31 20:33:44,811 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 2023-03-31 20:34:20,318 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.2607, 1.5325, 2.0314, 2.0926, 2.6961, 4.3158, 3.8467, 4.3012], device='cuda:1'), covar=tensor([0.1706, 0.2914, 0.2222, 0.2044, 0.0630, 0.0145, 0.0163, 0.0146], device='cuda:1'), in_proj_covar=tensor([0.0309, 0.0283, 0.0296, 0.0308, 0.0199, 0.0129, 0.0167, 0.0125], device='cuda:1'), out_proj_covar=tensor([2.4239e-04, 2.2402e-04, 2.3288e-04, 2.4574e-04, 1.8619e-04, 9.9757e-05, 1.3597e-04, 1.0763e-04], device='cuda:1') 2023-03-31 20:34:29,499 INFO [train.py:903] (1/4) Epoch 1, batch 4950, loss[loss=0.6014, simple_loss=0.5492, pruned_loss=0.3268, over 17292.00 frames. ], tot_loss[loss=0.4182, simple_loss=0.4347, pruned_loss=0.2009, over 3784414.98 frames. ], batch size: 101, lr: 4.21e-02, grad_scale: 8.0 2023-03-31 20:34:40,566 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 2023-03-31 20:35:04,989 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6713, 1.4334, 1.6290, 1.2211, 2.7743, 3.6233, 3.1903, 3.6600], device='cuda:1'), covar=tensor([0.1842, 0.2733, 0.2383, 0.2650, 0.0539, 0.0170, 0.0259, 0.0206], device='cuda:1'), in_proj_covar=tensor([0.0304, 0.0276, 0.0292, 0.0303, 0.0198, 0.0128, 0.0169, 0.0127], device='cuda:1'), out_proj_covar=tensor([2.3959e-04, 2.1881e-04, 2.3047e-04, 2.4149e-04, 1.8403e-04, 9.7999e-05, 1.3699e-04, 1.0993e-04], device='cuda:1') 2023-03-31 20:35:05,870 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 2023-03-31 20:35:31,801 INFO [train.py:903] (1/4) Epoch 1, batch 5000, loss[loss=0.4384, simple_loss=0.4587, pruned_loss=0.209, over 19671.00 frames. ], tot_loss[loss=0.4187, simple_loss=0.4354, pruned_loss=0.201, over 3781073.92 frames. ], batch size: 60, lr: 4.20e-02, grad_scale: 8.0 2023-03-31 20:35:35,554 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_rvb from training. Duration: 27.14 2023-03-31 20:35:40,124 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 5.692e+02 8.720e+02 1.063e+03 1.451e+03 3.452e+03, threshold=2.125e+03, percent-clipped=4.0 2023-03-31 20:35:46,916 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 2023-03-31 20:35:56,107 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=5021.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 20:36:27,167 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=5046.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 20:36:33,277 INFO [train.py:903] (1/4) Epoch 1, batch 5050, loss[loss=0.3968, simple_loss=0.4283, pruned_loss=0.1827, over 19308.00 frames. ], tot_loss[loss=0.4134, simple_loss=0.4317, pruned_loss=0.1976, over 3784420.70 frames. ], batch size: 66, lr: 4.19e-02, grad_scale: 8.0 2023-03-31 20:37:02,514 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9_rvb from training. Duration: 26.62775 2023-03-31 20:37:09,240 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.77 vs. limit=2.0 2023-03-31 20:37:34,521 INFO [train.py:903] (1/4) Epoch 1, batch 5100, loss[loss=0.517, simple_loss=0.4967, pruned_loss=0.2687, over 13467.00 frames. ], tot_loss[loss=0.4127, simple_loss=0.4312, pruned_loss=0.1971, over 3785906.82 frames. ], batch size: 136, lr: 4.18e-02, grad_scale: 8.0 2023-03-31 20:37:39,760 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 2023-03-31 20:37:43,121 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 5.551e+02 1.065e+03 1.254e+03 1.490e+03 3.647e+03, threshold=2.509e+03, percent-clipped=6.0 2023-03-31 20:37:43,157 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_rvb from training. Duration: 25.035 2023-03-31 20:37:47,692 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_rvb from training. Duration: 27.92 2023-03-31 20:38:36,973 INFO [train.py:903] (1/4) Epoch 1, batch 5150, loss[loss=0.4607, simple_loss=0.4564, pruned_loss=0.2325, over 13440.00 frames. ], tot_loss[loss=0.4105, simple_loss=0.4293, pruned_loss=0.1958, over 3791462.67 frames. ], batch size: 135, lr: 4.17e-02, grad_scale: 8.0 2023-03-31 20:38:46,441 WARNING [train.py:1073] (1/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 2023-03-31 20:39:20,323 WARNING [train.py:1073] (1/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1_rvb from training. Duration: 0.9681875 2023-03-31 20:39:37,966 INFO [train.py:903] (1/4) Epoch 1, batch 5200, loss[loss=0.3731, simple_loss=0.3936, pruned_loss=0.1763, over 19344.00 frames. ], tot_loss[loss=0.4128, simple_loss=0.4308, pruned_loss=0.1974, over 3799894.63 frames. ], batch size: 47, lr: 4.16e-02, grad_scale: 8.0 2023-03-31 20:39:45,881 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.640e+02 1.028e+03 1.252e+03 1.630e+03 4.880e+03, threshold=2.504e+03, percent-clipped=1.0 2023-03-31 20:39:50,384 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 2023-03-31 20:40:32,315 WARNING [train.py:1073] (1/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 2023-03-31 20:40:39,294 INFO [train.py:903] (1/4) Epoch 1, batch 5250, loss[loss=0.3615, simple_loss=0.3927, pruned_loss=0.1652, over 19768.00 frames. ], tot_loss[loss=0.4119, simple_loss=0.4303, pruned_loss=0.1968, over 3802107.43 frames. ], batch size: 49, lr: 4.15e-02, grad_scale: 8.0 2023-03-31 20:40:52,182 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.44 vs. limit=5.0 2023-03-31 20:41:39,595 INFO [train.py:903] (1/4) Epoch 1, batch 5300, loss[loss=0.3769, simple_loss=0.3867, pruned_loss=0.1835, over 19696.00 frames. ], tot_loss[loss=0.4135, simple_loss=0.4309, pruned_loss=0.198, over 3786681.54 frames. ], batch size: 45, lr: 4.14e-02, grad_scale: 8.0 2023-03-31 20:41:48,685 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 5.962e+02 9.394e+02 1.191e+03 1.647e+03 4.206e+03, threshold=2.383e+03, percent-clipped=5.0 2023-03-31 20:41:53,358 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 2023-03-31 20:42:41,823 INFO [train.py:903] (1/4) Epoch 1, batch 5350, loss[loss=0.3582, simple_loss=0.3813, pruned_loss=0.1675, over 19065.00 frames. ], tot_loss[loss=0.4128, simple_loss=0.4304, pruned_loss=0.1976, over 3794358.51 frames. ], batch size: 42, lr: 4.13e-02, grad_scale: 8.0 2023-03-31 20:43:13,704 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9_rvb from training. Duration: 26.438875 2023-03-31 20:43:43,593 INFO [train.py:903] (1/4) Epoch 1, batch 5400, loss[loss=0.3244, simple_loss=0.3705, pruned_loss=0.1391, over 19384.00 frames. ], tot_loss[loss=0.4089, simple_loss=0.4278, pruned_loss=0.195, over 3807349.29 frames. ], batch size: 48, lr: 4.12e-02, grad_scale: 8.0 2023-03-31 20:43:51,070 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 6.073e+02 9.364e+02 1.084e+03 1.611e+03 4.795e+03, threshold=2.168e+03, percent-clipped=7.0 2023-03-31 20:44:44,635 INFO [train.py:903] (1/4) Epoch 1, batch 5450, loss[loss=0.3966, simple_loss=0.4035, pruned_loss=0.1949, over 19409.00 frames. ], tot_loss[loss=0.4073, simple_loss=0.4269, pruned_loss=0.1938, over 3813598.36 frames. ], batch size: 48, lr: 4.11e-02, grad_scale: 8.0 2023-03-31 20:45:05,113 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 2023-03-31 20:45:46,527 INFO [train.py:903] (1/4) Epoch 1, batch 5500, loss[loss=0.4794, simple_loss=0.4829, pruned_loss=0.238, over 19763.00 frames. ], tot_loss[loss=0.4075, simple_loss=0.4273, pruned_loss=0.1939, over 3804334.01 frames. ], batch size: 56, lr: 4.10e-02, grad_scale: 8.0 2023-03-31 20:45:54,008 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.842e+02 9.504e+02 1.107e+03 1.412e+03 4.004e+03, threshold=2.214e+03, percent-clipped=7.0 2023-03-31 20:46:08,673 WARNING [train.py:1073] (1/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9_rvb from training. Duration: 27.47775 2023-03-31 20:46:46,623 INFO [train.py:903] (1/4) Epoch 1, batch 5550, loss[loss=0.334, simple_loss=0.3665, pruned_loss=0.1507, over 19727.00 frames. ], tot_loss[loss=0.4062, simple_loss=0.4262, pruned_loss=0.1932, over 3810644.35 frames. ], batch size: 46, lr: 4.09e-02, grad_scale: 8.0 2023-03-31 20:46:53,465 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 2023-03-31 20:46:54,449 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.78 vs. limit=5.0 2023-03-31 20:47:35,046 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.0942, 1.2886, 0.8923, 0.9682, 1.1164, 0.8650, 0.3908, 1.3420], device='cuda:1'), covar=tensor([0.0898, 0.0527, 0.1108, 0.0811, 0.0696, 0.1519, 0.1547, 0.0714], device='cuda:1'), in_proj_covar=tensor([0.0249, 0.0150, 0.0185, 0.0222, 0.0157, 0.0246, 0.0249, 0.0236], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-03-31 20:47:42,845 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 2023-03-31 20:47:47,483 INFO [train.py:903] (1/4) Epoch 1, batch 5600, loss[loss=0.401, simple_loss=0.4304, pruned_loss=0.1859, over 19765.00 frames. ], tot_loss[loss=0.4055, simple_loss=0.4258, pruned_loss=0.1926, over 3817057.57 frames. ], batch size: 56, lr: 4.08e-02, grad_scale: 8.0 2023-03-31 20:47:56,550 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 5.862e+02 1.009e+03 1.185e+03 1.400e+03 2.216e+03, threshold=2.370e+03, percent-clipped=2.0 2023-03-31 20:48:48,825 INFO [train.py:903] (1/4) Epoch 1, batch 5650, loss[loss=0.4417, simple_loss=0.4612, pruned_loss=0.2111, over 19732.00 frames. ], tot_loss[loss=0.4092, simple_loss=0.4283, pruned_loss=0.1951, over 3802967.29 frames. ], batch size: 63, lr: 4.07e-02, grad_scale: 8.0 2023-03-31 20:49:09,682 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=5668.0, num_to_drop=1, layers_to_drop={0} 2023-03-31 20:49:32,788 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9_rvb from training. Duration: 28.0944375 2023-03-31 20:49:49,715 INFO [train.py:903] (1/4) Epoch 1, batch 5700, loss[loss=0.4574, simple_loss=0.4678, pruned_loss=0.2235, over 18802.00 frames. ], tot_loss[loss=0.4101, simple_loss=0.429, pruned_loss=0.1956, over 3796701.70 frames. ], batch size: 74, lr: 4.06e-02, grad_scale: 8.0 2023-03-31 20:49:57,492 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 6.193e+02 1.084e+03 1.385e+03 1.754e+03 4.325e+03, threshold=2.770e+03, percent-clipped=14.0 2023-03-31 20:50:51,576 INFO [train.py:903] (1/4) Epoch 1, batch 5750, loss[loss=0.4221, simple_loss=0.4428, pruned_loss=0.2007, over 18181.00 frames. ], tot_loss[loss=0.4067, simple_loss=0.4266, pruned_loss=0.1934, over 3807550.26 frames. ], batch size: 83, lr: 4.05e-02, grad_scale: 8.0 2023-03-31 20:50:51,595 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9_rvb from training. Duration: 33.038875 2023-03-31 20:50:59,667 WARNING [train.py:1073] (1/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 2023-03-31 20:51:05,161 WARNING [train.py:1073] (1/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 2023-03-31 20:51:18,556 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=5773.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 20:51:21,937 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.8724, 4.2761, 5.5800, 5.1874, 1.7560, 4.9035, 4.6047, 4.6523], device='cuda:1'), covar=tensor([0.0204, 0.0420, 0.0273, 0.0172, 0.2847, 0.0173, 0.0246, 0.0722], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0226, 0.0276, 0.0190, 0.0370, 0.0139, 0.0203, 0.0290], device='cuda:1'), out_proj_covar=tensor([1.2034e-04, 1.4622e-04, 1.7960e-04, 1.1123e-04, 2.0182e-04, 9.0813e-05, 1.2534e-04, 1.6646e-04], device='cuda:1') 2023-03-31 20:51:52,608 INFO [train.py:903] (1/4) Epoch 1, batch 5800, loss[loss=0.3823, simple_loss=0.4157, pruned_loss=0.1744, over 19659.00 frames. ], tot_loss[loss=0.404, simple_loss=0.4248, pruned_loss=0.1916, over 3822824.57 frames. ], batch size: 58, lr: 4.04e-02, grad_scale: 8.0 2023-03-31 20:51:59,111 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.8654, 1.4310, 1.7138, 2.2353, 1.5936, 2.2075, 2.7789, 2.2904], device='cuda:1'), covar=tensor([0.0840, 0.2132, 0.1978, 0.1666, 0.2732, 0.1662, 0.1879, 0.1209], device='cuda:1'), in_proj_covar=tensor([0.0246, 0.0298, 0.0305, 0.0315, 0.0391, 0.0272, 0.0351, 0.0259], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-03-31 20:52:02,181 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 6.310e+02 8.969e+02 1.169e+03 1.352e+03 2.735e+03, threshold=2.337e+03, percent-clipped=0.0 2023-03-31 20:52:53,453 INFO [train.py:903] (1/4) Epoch 1, batch 5850, loss[loss=0.3267, simple_loss=0.3634, pruned_loss=0.145, over 19763.00 frames. ], tot_loss[loss=0.4024, simple_loss=0.4235, pruned_loss=0.1907, over 3818144.03 frames. ], batch size: 48, lr: 4.03e-02, grad_scale: 8.0 2023-03-31 20:53:23,806 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=5876.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 20:53:55,080 INFO [train.py:903] (1/4) Epoch 1, batch 5900, loss[loss=0.3432, simple_loss=0.3874, pruned_loss=0.1495, over 19592.00 frames. ], tot_loss[loss=0.3988, simple_loss=0.4209, pruned_loss=0.1883, over 3809432.95 frames. ], batch size: 52, lr: 4.02e-02, grad_scale: 8.0 2023-03-31 20:53:58,584 WARNING [train.py:1073] (1/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 2023-03-31 20:54:03,105 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 5.633e+02 8.668e+02 1.127e+03 1.397e+03 3.736e+03, threshold=2.255e+03, percent-clipped=4.0 2023-03-31 20:54:21,183 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9_rvb from training. Duration: 27.511125 2023-03-31 20:54:45,250 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([5.6615, 5.0726, 3.3979, 4.6162, 1.5987, 5.4741, 4.9160, 5.3490], device='cuda:1'), covar=tensor([0.0554, 0.1197, 0.1942, 0.0589, 0.3608, 0.0673, 0.0567, 0.0447], device='cuda:1'), in_proj_covar=tensor([0.0246, 0.0244, 0.0280, 0.0226, 0.0294, 0.0228, 0.0177, 0.0190], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-03-31 20:54:55,343 INFO [train.py:903] (1/4) Epoch 1, batch 5950, loss[loss=0.4766, simple_loss=0.4696, pruned_loss=0.2418, over 13987.00 frames. ], tot_loss[loss=0.3991, simple_loss=0.4215, pruned_loss=0.1883, over 3817026.85 frames. ], batch size: 136, lr: 4.01e-02, grad_scale: 8.0 2023-03-31 20:55:47,584 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0 2023-03-31 20:55:57,493 INFO [train.py:903] (1/4) Epoch 1, batch 6000, loss[loss=0.4279, simple_loss=0.4502, pruned_loss=0.2028, over 18816.00 frames. ], tot_loss[loss=0.3978, simple_loss=0.4202, pruned_loss=0.1878, over 3808889.93 frames. ], batch size: 74, lr: 4.00e-02, grad_scale: 8.0 2023-03-31 20:55:57,493 INFO [train.py:928] (1/4) Computing validation loss 2023-03-31 20:56:10,586 INFO [train.py:937] (1/4) Epoch 1, validation: loss=0.2784, simple_loss=0.3626, pruned_loss=0.09714, over 944034.00 frames. 2023-03-31 20:56:10,587 INFO [train.py:938] (1/4) Maximum memory allocated so far is 17726MB 2023-03-31 20:56:19,579 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 5.849e+02 9.012e+02 1.240e+03 1.620e+03 2.952e+03, threshold=2.480e+03, percent-clipped=5.0 2023-03-31 20:56:19,913 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6008.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 20:56:24,072 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6012.0, num_to_drop=1, layers_to_drop={0} 2023-03-31 20:57:10,741 INFO [train.py:903] (1/4) Epoch 1, batch 6050, loss[loss=0.3891, simple_loss=0.4131, pruned_loss=0.1825, over 19621.00 frames. ], tot_loss[loss=0.3993, simple_loss=0.4212, pruned_loss=0.1887, over 3816629.04 frames. ], batch size: 50, lr: 3.99e-02, grad_scale: 8.0 2023-03-31 20:57:23,117 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.6543, 4.1827, 2.3483, 3.5510, 1.3260, 4.1281, 3.6835, 3.8392], device='cuda:1'), covar=tensor([0.0641, 0.1282, 0.2511, 0.0725, 0.3878, 0.0809, 0.0696, 0.0681], device='cuda:1'), in_proj_covar=tensor([0.0240, 0.0243, 0.0272, 0.0226, 0.0297, 0.0229, 0.0173, 0.0192], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-03-31 20:57:24,343 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6061.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 20:57:27,990 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0146, 1.8816, 1.3591, 1.5331, 1.3306, 1.5774, 0.1971, 0.9696], device='cuda:1'), covar=tensor([0.0610, 0.0538, 0.0375, 0.0584, 0.1088, 0.0708, 0.1636, 0.1269], device='cuda:1'), in_proj_covar=tensor([0.0160, 0.0166, 0.0162, 0.0207, 0.0236, 0.0211, 0.0229, 0.0230], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-03-31 20:58:12,899 INFO [train.py:903] (1/4) Epoch 1, batch 6100, loss[loss=0.4256, simple_loss=0.4349, pruned_loss=0.2081, over 19693.00 frames. ], tot_loss[loss=0.3982, simple_loss=0.4207, pruned_loss=0.1879, over 3814552.33 frames. ], batch size: 53, lr: 3.98e-02, grad_scale: 8.0 2023-03-31 20:58:20,960 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.479e+02 9.509e+02 1.169e+03 1.489e+03 2.977e+03, threshold=2.338e+03, percent-clipped=4.0 2023-03-31 20:58:32,472 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6117.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 20:58:44,729 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6127.0, num_to_drop=1, layers_to_drop={1} 2023-03-31 20:58:47,275 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.95 vs. limit=2.0 2023-03-31 20:59:13,351 INFO [train.py:903] (1/4) Epoch 1, batch 6150, loss[loss=0.3969, simple_loss=0.422, pruned_loss=0.1859, over 18176.00 frames. ], tot_loss[loss=0.398, simple_loss=0.4207, pruned_loss=0.1876, over 3816146.70 frames. ], batch size: 83, lr: 3.97e-02, grad_scale: 8.0 2023-03-31 20:59:15,696 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6153.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 20:59:42,658 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9_rvb from training. Duration: 31.02225 2023-03-31 20:59:50,712 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4243, 2.0191, 1.6889, 1.6724, 1.7482, 0.8333, 0.8137, 1.7524], device='cuda:1'), covar=tensor([0.1040, 0.0444, 0.1049, 0.0631, 0.0867, 0.1546, 0.1361, 0.0780], device='cuda:1'), in_proj_covar=tensor([0.0262, 0.0150, 0.0215, 0.0230, 0.0160, 0.0251, 0.0258, 0.0245], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-03-31 20:59:58,475 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6188.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:00:02,020 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.1482, 1.5352, 1.3920, 2.2838, 1.4949, 2.2467, 1.9202, 2.1652], device='cuda:1'), covar=tensor([0.0794, 0.1589, 0.1735, 0.1266, 0.2285, 0.1200, 0.1910, 0.0922], device='cuda:1'), in_proj_covar=tensor([0.0251, 0.0306, 0.0300, 0.0323, 0.0391, 0.0278, 0.0343, 0.0263], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-03-31 21:00:13,785 INFO [train.py:903] (1/4) Epoch 1, batch 6200, loss[loss=0.4141, simple_loss=0.4381, pruned_loss=0.195, over 17364.00 frames. ], tot_loss[loss=0.3975, simple_loss=0.4206, pruned_loss=0.1872, over 3830012.50 frames. ], batch size: 101, lr: 3.96e-02, grad_scale: 8.0 2023-03-31 21:00:22,726 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.993e+02 9.585e+02 1.181e+03 1.511e+03 2.920e+03, threshold=2.362e+03, percent-clipped=2.0 2023-03-31 21:00:31,011 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6721, 1.0367, 1.4006, 0.7976, 2.6167, 2.7594, 2.4716, 2.5485], device='cuda:1'), covar=tensor([0.1688, 0.3158, 0.2730, 0.2816, 0.0441, 0.0175, 0.0318, 0.0318], device='cuda:1'), in_proj_covar=tensor([0.0311, 0.0287, 0.0316, 0.0304, 0.0202, 0.0117, 0.0180, 0.0128], device='cuda:1'), out_proj_covar=tensor([2.5590e-04, 2.4325e-04, 2.6154e-04, 2.5763e-04, 1.9470e-04, 9.8021e-05, 1.5082e-04, 1.1786e-04], device='cuda:1') 2023-03-31 21:00:38,390 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6220.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:00:39,677 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6221.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:00:49,036 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6229.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:00:52,525 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6232.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:01:16,023 INFO [train.py:903] (1/4) Epoch 1, batch 6250, loss[loss=0.403, simple_loss=0.4275, pruned_loss=0.1893, over 19664.00 frames. ], tot_loss[loss=0.3965, simple_loss=0.4201, pruned_loss=0.1865, over 3824266.94 frames. ], batch size: 53, lr: 3.95e-02, grad_scale: 8.0 2023-03-31 21:01:46,721 WARNING [train.py:1073] (1/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9_rvb from training. Duration: 25.988875 2023-03-31 21:01:56,423 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6284.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:02:18,553 INFO [train.py:903] (1/4) Epoch 1, batch 6300, loss[loss=0.362, simple_loss=0.3946, pruned_loss=0.1647, over 19579.00 frames. ], tot_loss[loss=0.3942, simple_loss=0.4188, pruned_loss=0.1848, over 3825732.04 frames. ], batch size: 52, lr: 3.94e-02, grad_scale: 8.0 2023-03-31 21:02:26,555 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.984e+02 8.812e+02 1.125e+03 1.363e+03 2.149e+03, threshold=2.249e+03, percent-clipped=0.0 2023-03-31 21:02:41,813 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.3271, 2.0315, 2.0520, 2.9162, 1.9235, 2.5777, 1.9735, 1.9000], device='cuda:1'), covar=tensor([0.0721, 0.0695, 0.0505, 0.0420, 0.0918, 0.0418, 0.1311, 0.0815], device='cuda:1'), in_proj_covar=tensor([0.0199, 0.0179, 0.0194, 0.0246, 0.0245, 0.0139, 0.0273, 0.0205], device='cuda:1'), out_proj_covar=tensor([1.4865e-04, 1.3274e-04, 1.2952e-04, 1.6485e-04, 1.5998e-04, 9.6092e-05, 1.9448e-04, 1.4389e-04], device='cuda:1') 2023-03-31 21:02:58,278 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6335.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:03:17,634 INFO [train.py:903] (1/4) Epoch 1, batch 6350, loss[loss=0.3566, simple_loss=0.4043, pruned_loss=0.1545, over 19770.00 frames. ], tot_loss[loss=0.3982, simple_loss=0.4214, pruned_loss=0.1875, over 3833720.64 frames. ], batch size: 54, lr: 3.93e-02, grad_scale: 8.0 2023-03-31 21:03:18,875 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6352.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:03:58,098 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6383.0, num_to_drop=1, layers_to_drop={1} 2023-03-31 21:04:18,897 INFO [train.py:903] (1/4) Epoch 1, batch 6400, loss[loss=0.4027, simple_loss=0.4367, pruned_loss=0.1844, over 19383.00 frames. ], tot_loss[loss=0.3978, simple_loss=0.4217, pruned_loss=0.1869, over 3829366.40 frames. ], batch size: 70, lr: 3.92e-02, grad_scale: 8.0 2023-03-31 21:04:24,482 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6405.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:04:27,871 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 5.880e+02 9.359e+02 1.206e+03 1.547e+03 5.333e+03, threshold=2.412e+03, percent-clipped=7.0 2023-03-31 21:04:28,285 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6408.0, num_to_drop=1, layers_to_drop={0} 2023-03-31 21:05:19,056 INFO [train.py:903] (1/4) Epoch 1, batch 6450, loss[loss=0.4214, simple_loss=0.4245, pruned_loss=0.2092, over 19755.00 frames. ], tot_loss[loss=0.3989, simple_loss=0.4225, pruned_loss=0.1877, over 3807886.86 frames. ], batch size: 47, lr: 3.91e-02, grad_scale: 8.0 2023-03-31 21:05:39,886 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6467.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:05:55,625 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6480.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:06:01,919 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6485.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:06:05,060 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9_rvb from training. Duration: 28.638875 2023-03-31 21:06:05,477 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6488.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:06:07,528 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.7543, 2.1216, 2.2942, 2.6960, 4.1396, 1.3238, 2.1934, 4.0787], device='cuda:1'), covar=tensor([0.0244, 0.2389, 0.2498, 0.1643, 0.0298, 0.2382, 0.1172, 0.0358], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0287, 0.0272, 0.0267, 0.0219, 0.0309, 0.0239, 0.0233], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-03-31 21:06:15,203 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6497.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:06:21,207 INFO [train.py:903] (1/4) Epoch 1, batch 6500, loss[loss=0.3395, simple_loss=0.3806, pruned_loss=0.1492, over 19589.00 frames. ], tot_loss[loss=0.3973, simple_loss=0.4211, pruned_loss=0.1867, over 3809625.36 frames. ], batch size: 52, lr: 3.90e-02, grad_scale: 8.0 2023-03-31 21:06:25,586 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1_rvb from training. Duration: 0.836375 2023-03-31 21:06:28,779 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 5.341e+02 9.704e+02 1.201e+03 1.443e+03 2.205e+03, threshold=2.402e+03, percent-clipped=0.0 2023-03-31 21:06:35,119 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6513.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:06:43,872 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6520.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:06:58,405 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6532.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:06:59,669 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.6055, 4.2023, 5.2740, 4.9516, 2.0194, 4.5474, 4.2659, 4.5910], device='cuda:1'), covar=tensor([0.0228, 0.0478, 0.0355, 0.0185, 0.2898, 0.0205, 0.0335, 0.0730], device='cuda:1'), in_proj_covar=tensor([0.0210, 0.0251, 0.0304, 0.0219, 0.0394, 0.0150, 0.0228, 0.0326], device='cuda:1'), out_proj_covar=tensor([1.3459e-04, 1.5947e-04, 1.9683e-04, 1.2751e-04, 2.1279e-04, 9.5569e-05, 1.3716e-04, 1.8500e-04], device='cuda:1') 2023-03-31 21:07:21,792 INFO [train.py:903] (1/4) Epoch 1, batch 6550, loss[loss=0.4543, simple_loss=0.4655, pruned_loss=0.2215, over 19610.00 frames. ], tot_loss[loss=0.3957, simple_loss=0.4197, pruned_loss=0.1858, over 3786893.96 frames. ], batch size: 57, lr: 3.89e-02, grad_scale: 8.0 2023-03-31 21:07:37,340 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3301, 1.9829, 1.5371, 1.5371, 1.6632, 0.8793, 0.7923, 1.7908], device='cuda:1'), covar=tensor([0.1316, 0.0497, 0.1285, 0.0761, 0.0975, 0.1727, 0.1559, 0.0866], device='cuda:1'), in_proj_covar=tensor([0.0262, 0.0157, 0.0232, 0.0238, 0.0165, 0.0269, 0.0257, 0.0251], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-03-31 21:07:39,358 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6565.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:07:48,412 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6573.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:08:10,395 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6591.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:08:22,316 INFO [train.py:903] (1/4) Epoch 1, batch 6600, loss[loss=0.3492, simple_loss=0.3894, pruned_loss=0.1545, over 19848.00 frames. ], tot_loss[loss=0.3912, simple_loss=0.4165, pruned_loss=0.1829, over 3800825.91 frames. ], batch size: 52, lr: 3.89e-02, grad_scale: 16.0 2023-03-31 21:08:31,046 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.089e+02 8.736e+02 1.082e+03 1.231e+03 3.386e+03, threshold=2.164e+03, percent-clipped=2.0 2023-03-31 21:08:36,141 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6612.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:08:40,707 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6616.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:08:54,950 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.6677, 4.2976, 2.3659, 3.9267, 1.4295, 4.1500, 3.8180, 4.2186], device='cuda:1'), covar=tensor([0.0536, 0.1078, 0.1829, 0.0617, 0.2983, 0.0728, 0.0619, 0.0491], device='cuda:1'), in_proj_covar=tensor([0.0249, 0.0243, 0.0276, 0.0241, 0.0292, 0.0236, 0.0183, 0.0196], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-03-31 21:08:56,073 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6628.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:09:19,946 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6647.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:09:20,018 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6647.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:09:24,251 INFO [train.py:903] (1/4) Epoch 1, batch 6650, loss[loss=0.3427, simple_loss=0.3841, pruned_loss=0.1506, over 19844.00 frames. ], tot_loss[loss=0.3906, simple_loss=0.4162, pruned_loss=0.1825, over 3806677.44 frames. ], batch size: 52, lr: 3.88e-02, grad_scale: 4.0 2023-03-31 21:09:59,890 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6680.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:10:08,149 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.1190, 1.4498, 1.6081, 2.1353, 2.9633, 1.4286, 2.0298, 2.7619], device='cuda:1'), covar=tensor([0.0314, 0.2295, 0.2291, 0.1315, 0.0282, 0.2118, 0.1102, 0.0431], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0291, 0.0274, 0.0266, 0.0220, 0.0316, 0.0246, 0.0236], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-03-31 21:10:09,254 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6688.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:10:23,709 INFO [train.py:903] (1/4) Epoch 1, batch 6700, loss[loss=0.4039, simple_loss=0.4298, pruned_loss=0.189, over 19404.00 frames. ], tot_loss[loss=0.3941, simple_loss=0.4188, pruned_loss=0.1847, over 3801708.77 frames. ], batch size: 70, lr: 3.87e-02, grad_scale: 4.0 2023-03-31 21:10:32,322 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.5293, 1.6012, 1.8578, 2.2392, 3.2068, 1.5009, 2.1925, 3.2067], device='cuda:1'), covar=tensor([0.0328, 0.2486, 0.2454, 0.1605, 0.0364, 0.2313, 0.1220, 0.0369], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0286, 0.0272, 0.0264, 0.0219, 0.0311, 0.0242, 0.0233], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-03-31 21:10:34,583 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.7756, 3.4536, 2.0290, 3.1129, 1.4592, 3.3118, 3.0572, 3.2647], device='cuda:1'), covar=tensor([0.0726, 0.1249, 0.2545, 0.0843, 0.3552, 0.1041, 0.0732, 0.0869], device='cuda:1'), in_proj_covar=tensor([0.0253, 0.0245, 0.0280, 0.0241, 0.0299, 0.0235, 0.0183, 0.0198], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-03-31 21:10:35,446 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.729e+02 8.354e+02 1.101e+03 1.693e+03 1.016e+04, threshold=2.202e+03, percent-clipped=16.0 2023-03-31 21:10:51,582 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6723.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:11:14,953 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6743.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:11:20,754 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6748.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:11:23,798 INFO [train.py:903] (1/4) Epoch 1, batch 6750, loss[loss=0.4772, simple_loss=0.4667, pruned_loss=0.2438, over 13426.00 frames. ], tot_loss[loss=0.3919, simple_loss=0.4177, pruned_loss=0.1831, over 3811036.49 frames. ], batch size: 137, lr: 3.86e-02, grad_scale: 4.0 2023-03-31 21:11:38,467 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.0383, 1.0762, 1.7169, 1.2194, 2.4406, 2.3622, 2.6052, 1.3066], device='cuda:1'), covar=tensor([0.1697, 0.2039, 0.1459, 0.1753, 0.0941, 0.0870, 0.1014, 0.1842], device='cuda:1'), in_proj_covar=tensor([0.0297, 0.0299, 0.0287, 0.0301, 0.0325, 0.0269, 0.0370, 0.0320], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-03-31 21:11:51,635 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6776.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:12:18,951 INFO [train.py:903] (1/4) Epoch 1, batch 6800, loss[loss=0.377, simple_loss=0.4084, pruned_loss=0.1728, over 19772.00 frames. ], tot_loss[loss=0.3933, simple_loss=0.418, pruned_loss=0.1843, over 3793435.54 frames. ], batch size: 54, lr: 3.85e-02, grad_scale: 8.0 2023-03-31 21:12:19,295 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6801.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:12:28,857 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 5.487e+02 9.125e+02 1.072e+03 1.412e+03 3.162e+03, threshold=2.143e+03, percent-clipped=4.0 2023-03-31 21:12:43,734 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6824.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:13:03,380 WARNING [train.py:1073] (1/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 2023-03-31 21:13:04,526 WARNING [train.py:1073] (1/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 2023-03-31 21:13:06,751 INFO [train.py:903] (1/4) Epoch 2, batch 0, loss[loss=0.397, simple_loss=0.42, pruned_loss=0.187, over 19687.00 frames. ], tot_loss[loss=0.397, simple_loss=0.42, pruned_loss=0.187, over 19687.00 frames. ], batch size: 53, lr: 3.77e-02, grad_scale: 8.0 2023-03-31 21:13:06,752 INFO [train.py:928] (1/4) Computing validation loss 2023-03-31 21:13:18,415 INFO [train.py:937] (1/4) Epoch 2, validation: loss=0.2802, simple_loss=0.3637, pruned_loss=0.09835, over 944034.00 frames. 2023-03-31 21:13:18,416 INFO [train.py:938] (1/4) Maximum memory allocated so far is 17726MB 2023-03-31 21:13:18,576 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6829.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:13:28,813 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 2023-03-31 21:14:06,569 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6868.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:14:20,895 INFO [train.py:903] (1/4) Epoch 2, batch 50, loss[loss=0.3684, simple_loss=0.3877, pruned_loss=0.1745, over 19757.00 frames. ], tot_loss[loss=0.3884, simple_loss=0.4136, pruned_loss=0.1816, over 859170.51 frames. ], batch size: 47, lr: 3.76e-02, grad_scale: 8.0 2023-03-31 21:14:37,429 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6893.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:14:41,561 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6896.0, num_to_drop=1, layers_to_drop={0} 2023-03-31 21:14:49,461 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6903.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:14:54,389 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1_rvb from training. Duration: 27.0318125 2023-03-31 21:14:57,929 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 6.488e+02 9.076e+02 1.150e+03 1.515e+03 2.802e+03, threshold=2.301e+03, percent-clipped=3.0 2023-03-31 21:15:19,042 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6927.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:15:20,147 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6928.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:15:20,721 INFO [train.py:903] (1/4) Epoch 2, batch 100, loss[loss=0.469, simple_loss=0.4743, pruned_loss=0.2318, over 18221.00 frames. ], tot_loss[loss=0.3898, simple_loss=0.4152, pruned_loss=0.1822, over 1522420.68 frames. ], batch size: 83, lr: 3.75e-02, grad_scale: 8.0 2023-03-31 21:15:29,257 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6936.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:15:32,200 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 2023-03-31 21:15:33,671 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6939.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:15:39,418 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6944.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:15:39,486 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6944.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:15:43,891 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.03 vs. limit=5.0 2023-03-31 21:16:01,530 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6961.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:16:10,910 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6969.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:16:23,040 INFO [train.py:903] (1/4) Epoch 2, batch 150, loss[loss=0.2712, simple_loss=0.3276, pruned_loss=0.1075, over 19803.00 frames. ], tot_loss[loss=0.3895, simple_loss=0.4151, pruned_loss=0.182, over 2036124.77 frames. ], batch size: 48, lr: 3.74e-02, grad_scale: 4.0 2023-03-31 21:16:38,557 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6991.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:16:49,257 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6999.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:16:53,972 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0167, 1.9408, 1.8307, 1.5315, 1.5455, 1.4794, 0.1804, 1.1357], device='cuda:1'), covar=tensor([0.0646, 0.0520, 0.0262, 0.0527, 0.0857, 0.0651, 0.1344, 0.1049], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0186, 0.0180, 0.0221, 0.0259, 0.0233, 0.0234, 0.0245], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-03-31 21:17:03,027 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 5.128e+02 7.870e+02 9.855e+02 1.288e+03 4.108e+03, threshold=1.971e+03, percent-clipped=4.0 2023-03-31 21:17:19,695 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7024.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:17:24,042 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 2023-03-31 21:17:25,199 INFO [train.py:903] (1/4) Epoch 2, batch 200, loss[loss=0.3784, simple_loss=0.4095, pruned_loss=0.1737, over 19611.00 frames. ], tot_loss[loss=0.3818, simple_loss=0.4096, pruned_loss=0.177, over 2450044.05 frames. ], batch size: 57, lr: 3.73e-02, grad_scale: 4.0 2023-03-31 21:18:11,617 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7066.0, num_to_drop=1, layers_to_drop={1} 2023-03-31 21:18:29,136 INFO [train.py:903] (1/4) Epoch 2, batch 250, loss[loss=0.3674, simple_loss=0.3867, pruned_loss=0.174, over 19782.00 frames. ], tot_loss[loss=0.3825, simple_loss=0.4104, pruned_loss=0.1773, over 2762459.85 frames. ], batch size: 49, lr: 3.72e-02, grad_scale: 4.0 2023-03-31 21:19:03,855 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7106.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:19:09,453 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.258e+02 7.747e+02 9.740e+02 1.157e+03 2.695e+03, threshold=1.948e+03, percent-clipped=1.0 2023-03-31 21:19:17,176 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 2023-03-31 21:19:33,822 INFO [train.py:903] (1/4) Epoch 2, batch 300, loss[loss=0.4056, simple_loss=0.4348, pruned_loss=0.1882, over 19498.00 frames. ], tot_loss[loss=0.3778, simple_loss=0.4076, pruned_loss=0.174, over 2993667.54 frames. ], batch size: 64, lr: 3.72e-02, grad_scale: 4.0 2023-03-31 21:20:35,435 INFO [train.py:903] (1/4) Epoch 2, batch 350, loss[loss=0.3296, simple_loss=0.381, pruned_loss=0.1391, over 19561.00 frames. ], tot_loss[loss=0.383, simple_loss=0.4112, pruned_loss=0.1774, over 3179411.35 frames. ], batch size: 52, lr: 3.71e-02, grad_scale: 4.0 2023-03-31 21:20:37,904 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7181.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:20:39,817 WARNING [train.py:1073] (1/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1_rvb from training. Duration: 0.9681875 2023-03-31 21:20:55,913 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7195.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:21:01,711 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7200.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:21:15,795 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.882e+02 9.781e+02 1.245e+03 1.512e+03 3.081e+03, threshold=2.489e+03, percent-clipped=8.0 2023-03-31 21:21:26,537 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7220.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:21:32,341 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7225.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:21:37,136 INFO [train.py:903] (1/4) Epoch 2, batch 400, loss[loss=0.4526, simple_loss=0.4662, pruned_loss=0.2195, over 19366.00 frames. ], tot_loss[loss=0.3835, simple_loss=0.4114, pruned_loss=0.1778, over 3331808.28 frames. ], batch size: 66, lr: 3.70e-02, grad_scale: 8.0 2023-03-31 21:21:51,114 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=7240.0, num_to_drop=1, layers_to_drop={1} 2023-03-31 21:22:30,121 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=7271.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:22:40,123 INFO [train.py:903] (1/4) Epoch 2, batch 450, loss[loss=0.3663, simple_loss=0.3975, pruned_loss=0.1676, over 19750.00 frames. ], tot_loss[loss=0.384, simple_loss=0.412, pruned_loss=0.1779, over 3438970.04 frames. ], batch size: 51, lr: 3.69e-02, grad_scale: 8.0 2023-03-31 21:22:58,508 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7293.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:23:14,126 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 2023-03-31 21:23:15,288 WARNING [train.py:1073] (1/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 2023-03-31 21:23:19,468 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 5.877e+02 8.763e+02 1.192e+03 1.491e+03 2.950e+03, threshold=2.384e+03, percent-clipped=4.0 2023-03-31 21:23:43,126 INFO [train.py:903] (1/4) Epoch 2, batch 500, loss[loss=0.3749, simple_loss=0.4086, pruned_loss=0.1706, over 18702.00 frames. ], tot_loss[loss=0.3845, simple_loss=0.4123, pruned_loss=0.1783, over 3523912.78 frames. ], batch size: 74, lr: 3.68e-02, grad_scale: 8.0 2023-03-31 21:24:14,633 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7355.0, num_to_drop=1, layers_to_drop={0} 2023-03-31 21:24:23,957 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7362.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:24:45,278 INFO [train.py:903] (1/4) Epoch 2, batch 550, loss[loss=0.3609, simple_loss=0.385, pruned_loss=0.1684, over 19732.00 frames. ], tot_loss[loss=0.3849, simple_loss=0.4124, pruned_loss=0.1787, over 3591293.67 frames. ], batch size: 45, lr: 3.67e-02, grad_scale: 8.0 2023-03-31 21:24:53,919 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7386.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:24:55,088 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7387.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:25:24,046 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=7410.0, num_to_drop=1, layers_to_drop={0} 2023-03-31 21:25:26,117 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.968e+02 9.206e+02 1.127e+03 1.377e+03 2.659e+03, threshold=2.254e+03, percent-clipped=2.0 2023-03-31 21:25:43,337 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7425.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:25:47,649 INFO [train.py:903] (1/4) Epoch 2, batch 600, loss[loss=0.4151, simple_loss=0.438, pruned_loss=0.1961, over 19186.00 frames. ], tot_loss[loss=0.3835, simple_loss=0.4115, pruned_loss=0.1777, over 3654752.63 frames. ], batch size: 69, lr: 3.66e-02, grad_scale: 8.0 2023-03-31 21:26:09,116 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0 2023-03-31 21:26:29,910 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 2023-03-31 21:26:36,178 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1282, 1.1689, 0.9814, 1.0926, 0.9488, 1.1644, 0.0360, 0.5413], device='cuda:1'), covar=tensor([0.0473, 0.0597, 0.0307, 0.0348, 0.0984, 0.0501, 0.1126, 0.1030], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0198, 0.0191, 0.0226, 0.0267, 0.0240, 0.0241, 0.0254], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-03-31 21:26:41,799 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6768, 1.2769, 1.2218, 1.8089, 1.2705, 1.6448, 1.8388, 1.8022], device='cuda:1'), covar=tensor([0.1076, 0.1611, 0.1908, 0.1371, 0.2008, 0.1336, 0.1579, 0.0945], device='cuda:1'), in_proj_covar=tensor([0.0263, 0.0306, 0.0308, 0.0330, 0.0383, 0.0277, 0.0342, 0.0260], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-03-31 21:26:50,577 INFO [train.py:903] (1/4) Epoch 2, batch 650, loss[loss=0.4052, simple_loss=0.4363, pruned_loss=0.187, over 19796.00 frames. ], tot_loss[loss=0.3835, simple_loss=0.4115, pruned_loss=0.1777, over 3679748.10 frames. ], batch size: 56, lr: 3.66e-02, grad_scale: 8.0 2023-03-31 21:26:59,956 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([5.1332, 4.5120, 5.7850, 5.5594, 1.7237, 5.2051, 4.6713, 5.1483], device='cuda:1'), covar=tensor([0.0190, 0.0398, 0.0303, 0.0157, 0.2905, 0.0178, 0.0250, 0.0637], device='cuda:1'), in_proj_covar=tensor([0.0215, 0.0253, 0.0321, 0.0217, 0.0392, 0.0153, 0.0228, 0.0330], device='cuda:1'), out_proj_covar=tensor([1.3334e-04, 1.5696e-04, 2.0104e-04, 1.2492e-04, 2.1041e-04, 9.8423e-05, 1.3335e-04, 1.8358e-04], device='cuda:1') 2023-03-31 21:27:30,826 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.799e+02 8.519e+02 1.041e+03 1.431e+03 3.840e+03, threshold=2.082e+03, percent-clipped=3.0 2023-03-31 21:27:49,382 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=7525.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:27:49,567 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7525.0, num_to_drop=1, layers_to_drop={1} 2023-03-31 21:27:53,956 INFO [train.py:903] (1/4) Epoch 2, batch 700, loss[loss=0.3293, simple_loss=0.3681, pruned_loss=0.1453, over 19737.00 frames. ], tot_loss[loss=0.3852, simple_loss=0.4126, pruned_loss=0.1789, over 3702764.88 frames. ], batch size: 46, lr: 3.65e-02, grad_scale: 8.0 2023-03-31 21:28:56,978 INFO [train.py:903] (1/4) Epoch 2, batch 750, loss[loss=0.3852, simple_loss=0.4218, pruned_loss=0.1743, over 19580.00 frames. ], tot_loss[loss=0.3866, simple_loss=0.414, pruned_loss=0.1796, over 3739940.09 frames. ], batch size: 61, lr: 3.64e-02, grad_scale: 8.0 2023-03-31 21:29:35,962 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 5.266e+02 8.676e+02 1.032e+03 1.220e+03 3.020e+03, threshold=2.064e+03, percent-clipped=5.0 2023-03-31 21:29:36,423 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7611.0, num_to_drop=1, layers_to_drop={1} 2023-03-31 21:29:58,614 INFO [train.py:903] (1/4) Epoch 2, batch 800, loss[loss=0.3843, simple_loss=0.4138, pruned_loss=0.1774, over 17376.00 frames. ], tot_loss[loss=0.3839, simple_loss=0.4123, pruned_loss=0.1777, over 3766331.01 frames. ], batch size: 101, lr: 3.63e-02, grad_scale: 8.0 2023-03-31 21:30:07,144 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7636.0, num_to_drop=1, layers_to_drop={1} 2023-03-31 21:30:09,184 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=7637.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:30:12,802 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7640.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:30:15,239 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7642.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:30:15,376 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7642.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:30:16,154 WARNING [train.py:1073] (1/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 2023-03-31 21:30:41,310 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.1317, 2.7131, 2.1839, 2.5273, 1.8648, 1.8390, 0.3748, 1.9774], device='cuda:1'), covar=tensor([0.0564, 0.0354, 0.0325, 0.0445, 0.0889, 0.0764, 0.1460, 0.1026], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0195, 0.0189, 0.0232, 0.0272, 0.0233, 0.0243, 0.0253], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-03-31 21:30:46,815 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7667.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:31:01,249 INFO [train.py:903] (1/4) Epoch 2, batch 850, loss[loss=0.4121, simple_loss=0.4417, pruned_loss=0.1913, over 19754.00 frames. ], tot_loss[loss=0.3808, simple_loss=0.4099, pruned_loss=0.1758, over 3778021.43 frames. ], batch size: 63, lr: 3.62e-02, grad_scale: 8.0 2023-03-31 21:31:36,340 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0505, 1.4710, 1.4305, 1.9958, 1.4862, 2.0701, 2.1022, 2.0702], device='cuda:1'), covar=tensor([0.0890, 0.1585, 0.1683, 0.1362, 0.2087, 0.1082, 0.1563, 0.0882], device='cuda:1'), in_proj_covar=tensor([0.0259, 0.0304, 0.0303, 0.0328, 0.0377, 0.0271, 0.0337, 0.0254], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-03-31 21:31:41,593 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 5.436e+02 9.016e+02 1.057e+03 1.450e+03 5.160e+03, threshold=2.114e+03, percent-clipped=6.0 2023-03-31 21:31:56,577 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9_rvb from training. Duration: 25.061125 2023-03-31 21:32:02,495 INFO [train.py:903] (1/4) Epoch 2, batch 900, loss[loss=0.3985, simple_loss=0.4196, pruned_loss=0.1887, over 19530.00 frames. ], tot_loss[loss=0.3811, simple_loss=0.4097, pruned_loss=0.1762, over 3773559.66 frames. ], batch size: 54, lr: 3.61e-02, grad_scale: 4.0 2023-03-31 21:32:32,753 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7752.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:32:43,824 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7761.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:32:53,073 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=7769.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:33:06,312 INFO [train.py:903] (1/4) Epoch 2, batch 950, loss[loss=0.3766, simple_loss=0.4118, pruned_loss=0.1707, over 19576.00 frames. ], tot_loss[loss=0.3817, simple_loss=0.4104, pruned_loss=0.1765, over 3779026.22 frames. ], batch size: 61, lr: 3.61e-02, grad_scale: 4.0 2023-03-31 21:33:09,107 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7781.0, num_to_drop=1, layers_to_drop={0} 2023-03-31 21:33:10,928 WARNING [train.py:1073] (1/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9_rvb from training. Duration: 26.32775 2023-03-31 21:33:40,006 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7806.0, num_to_drop=1, layers_to_drop={1} 2023-03-31 21:33:46,539 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 5.660e+02 8.698e+02 1.089e+03 1.494e+03 2.916e+03, threshold=2.178e+03, percent-clipped=6.0 2023-03-31 21:33:58,936 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7820.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:34:09,629 INFO [train.py:903] (1/4) Epoch 2, batch 1000, loss[loss=0.4056, simple_loss=0.4324, pruned_loss=0.1895, over 18807.00 frames. ], tot_loss[loss=0.3814, simple_loss=0.4106, pruned_loss=0.1761, over 3787738.71 frames. ], batch size: 74, lr: 3.60e-02, grad_scale: 4.0 2023-03-31 21:34:29,513 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0628, 1.9787, 1.7409, 2.9948, 1.8847, 3.4046, 2.3088, 1.8903], device='cuda:1'), covar=tensor([0.0996, 0.0769, 0.0576, 0.0483, 0.1092, 0.0206, 0.1086, 0.0820], device='cuda:1'), in_proj_covar=tensor([0.0260, 0.0231, 0.0251, 0.0323, 0.0323, 0.0172, 0.0339, 0.0269], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-03-31 21:34:58,586 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0 2023-03-31 21:35:04,952 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 2023-03-31 21:35:11,833 INFO [train.py:903] (1/4) Epoch 2, batch 1050, loss[loss=0.3544, simple_loss=0.3914, pruned_loss=0.1587, over 19611.00 frames. ], tot_loss[loss=0.3777, simple_loss=0.4081, pruned_loss=0.1737, over 3802405.76 frames. ], batch size: 52, lr: 3.59e-02, grad_scale: 4.0 2023-03-31 21:35:18,877 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7884.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:35:33,750 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7896.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:35:46,815 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 2023-03-31 21:35:53,415 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 5.580e+02 8.878e+02 9.952e+02 1.228e+03 3.126e+03, threshold=1.990e+03, percent-clipped=5.0 2023-03-31 21:36:05,095 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3648, 0.9899, 1.2835, 0.6782, 2.9002, 2.8624, 2.5734, 2.9531], device='cuda:1'), covar=tensor([0.1680, 0.3064, 0.3058, 0.2579, 0.0317, 0.0149, 0.0290, 0.0181], device='cuda:1'), in_proj_covar=tensor([0.0299, 0.0282, 0.0324, 0.0295, 0.0197, 0.0110, 0.0186, 0.0115], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0001, 0.0002, 0.0001], device='cuda:1') 2023-03-31 21:36:05,151 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7921.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:36:14,161 INFO [train.py:903] (1/4) Epoch 2, batch 1100, loss[loss=0.4675, simple_loss=0.4713, pruned_loss=0.2319, over 19330.00 frames. ], tot_loss[loss=0.3784, simple_loss=0.4084, pruned_loss=0.1742, over 3814339.60 frames. ], batch size: 70, lr: 3.58e-02, grad_scale: 4.0 2023-03-31 21:37:16,835 INFO [train.py:903] (1/4) Epoch 2, batch 1150, loss[loss=0.4496, simple_loss=0.4386, pruned_loss=0.2303, over 19368.00 frames. ], tot_loss[loss=0.3785, simple_loss=0.408, pruned_loss=0.1745, over 3806675.63 frames. ], batch size: 47, lr: 3.57e-02, grad_scale: 4.0 2023-03-31 21:37:20,372 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5269, 1.1438, 1.5031, 0.7134, 2.7967, 2.7807, 2.5838, 2.8626], device='cuda:1'), covar=tensor([0.1428, 0.2767, 0.2537, 0.2506, 0.0309, 0.0181, 0.0302, 0.0191], device='cuda:1'), in_proj_covar=tensor([0.0296, 0.0282, 0.0324, 0.0295, 0.0197, 0.0109, 0.0185, 0.0114], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0001, 0.0002, 0.0001], device='cuda:1') 2023-03-31 21:37:26,687 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=7986.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:37:54,521 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=8008.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:37:58,747 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.862e+02 8.681e+02 1.035e+03 1.273e+03 2.854e+03, threshold=2.070e+03, percent-clipped=5.0 2023-03-31 21:38:06,515 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.31 vs. limit=5.0 2023-03-31 21:38:21,517 INFO [train.py:903] (1/4) Epoch 2, batch 1200, loss[loss=0.3329, simple_loss=0.3739, pruned_loss=0.146, over 19749.00 frames. ], tot_loss[loss=0.3775, simple_loss=0.4075, pruned_loss=0.1738, over 3811733.89 frames. ], batch size: 51, lr: 3.56e-02, grad_scale: 8.0 2023-03-31 21:38:26,445 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=8033.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:38:52,436 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 2023-03-31 21:39:22,658 INFO [train.py:903] (1/4) Epoch 2, batch 1250, loss[loss=0.4358, simple_loss=0.4527, pruned_loss=0.2094, over 18767.00 frames. ], tot_loss[loss=0.3778, simple_loss=0.4072, pruned_loss=0.1742, over 3806576.31 frames. ], batch size: 74, lr: 3.56e-02, grad_scale: 8.0 2023-03-31 21:39:50,803 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=8101.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:39:57,189 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=8105.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:40:05,139 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 5.516e+02 8.392e+02 1.041e+03 1.254e+03 3.427e+03, threshold=2.083e+03, percent-clipped=3.0 2023-03-31 21:40:25,457 INFO [train.py:903] (1/4) Epoch 2, batch 1300, loss[loss=0.3497, simple_loss=0.3796, pruned_loss=0.16, over 19731.00 frames. ], tot_loss[loss=0.3758, simple_loss=0.4057, pruned_loss=0.1729, over 3809534.46 frames. ], batch size: 51, lr: 3.55e-02, grad_scale: 8.0 2023-03-31 21:40:39,826 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=8140.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:41:09,967 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=8164.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:41:11,407 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=8165.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:41:28,090 INFO [train.py:903] (1/4) Epoch 2, batch 1350, loss[loss=0.3724, simple_loss=0.4215, pruned_loss=0.1616, over 19779.00 frames. ], tot_loss[loss=0.3752, simple_loss=0.4048, pruned_loss=0.1728, over 3809627.88 frames. ], batch size: 56, lr: 3.54e-02, grad_scale: 8.0 2023-03-31 21:41:32,933 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0270, 1.5255, 1.4335, 2.0694, 1.4645, 1.8688, 1.8917, 1.7321], device='cuda:1'), covar=tensor([0.0792, 0.1252, 0.1539, 0.1241, 0.1999, 0.1230, 0.1428, 0.1011], device='cuda:1'), in_proj_covar=tensor([0.0262, 0.0292, 0.0303, 0.0320, 0.0370, 0.0270, 0.0335, 0.0250], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-03-31 21:42:08,818 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 5.809e+02 9.088e+02 1.109e+03 1.527e+03 2.312e+03, threshold=2.218e+03, percent-clipped=6.0 2023-03-31 21:42:19,637 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=8220.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:42:30,881 INFO [train.py:903] (1/4) Epoch 2, batch 1400, loss[loss=0.3367, simple_loss=0.3791, pruned_loss=0.1471, over 19856.00 frames. ], tot_loss[loss=0.3767, simple_loss=0.4066, pruned_loss=0.1733, over 3811853.04 frames. ], batch size: 52, lr: 3.53e-02, grad_scale: 8.0 2023-03-31 21:43:14,084 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 2023-03-31 21:43:32,694 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_rvb from training. Duration: 25.85 2023-03-31 21:43:33,842 INFO [train.py:903] (1/4) Epoch 2, batch 1450, loss[loss=0.3526, simple_loss=0.3963, pruned_loss=0.1544, over 19658.00 frames. ], tot_loss[loss=0.3768, simple_loss=0.4069, pruned_loss=0.1734, over 3798489.62 frames. ], batch size: 58, lr: 3.53e-02, grad_scale: 8.0 2023-03-31 21:43:34,141 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=8279.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:43:38,615 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=8283.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:44:15,429 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 5.216e+02 8.724e+02 1.078e+03 1.353e+03 2.729e+03, threshold=2.156e+03, percent-clipped=3.0 2023-03-31 21:44:26,444 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.0740, 1.1983, 0.9097, 1.0520, 1.1341, 0.8110, 0.4153, 1.2699], device='cuda:1'), covar=tensor([0.0707, 0.0649, 0.1146, 0.0561, 0.0639, 0.1348, 0.1086, 0.0648], device='cuda:1'), in_proj_covar=tensor([0.0272, 0.0178, 0.0265, 0.0241, 0.0177, 0.0287, 0.0259, 0.0258], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-03-31 21:44:35,645 INFO [train.py:903] (1/4) Epoch 2, batch 1500, loss[loss=0.4115, simple_loss=0.4321, pruned_loss=0.1954, over 19702.00 frames. ], tot_loss[loss=0.3772, simple_loss=0.4069, pruned_loss=0.1738, over 3801736.62 frames. ], batch size: 59, lr: 3.52e-02, grad_scale: 8.0 2023-03-31 21:45:10,891 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=8357.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:45:38,212 INFO [train.py:903] (1/4) Epoch 2, batch 1550, loss[loss=0.3638, simple_loss=0.4043, pruned_loss=0.1617, over 19530.00 frames. ], tot_loss[loss=0.3754, simple_loss=0.4058, pruned_loss=0.1725, over 3800761.49 frames. ], batch size: 56, lr: 3.51e-02, grad_scale: 8.0 2023-03-31 21:45:42,362 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=8382.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:45:51,885 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.96 vs. limit=2.0 2023-03-31 21:46:19,242 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 6.712e+02 9.525e+02 1.175e+03 1.582e+03 3.285e+03, threshold=2.351e+03, percent-clipped=5.0 2023-03-31 21:46:41,075 INFO [train.py:903] (1/4) Epoch 2, batch 1600, loss[loss=0.3122, simple_loss=0.3605, pruned_loss=0.132, over 19392.00 frames. ], tot_loss[loss=0.3751, simple_loss=0.4057, pruned_loss=0.1723, over 3794063.67 frames. ], batch size: 47, lr: 3.50e-02, grad_scale: 8.0 2023-03-31 21:47:02,066 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9_rvb from training. Duration: 30.1555625 2023-03-31 21:47:39,687 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=8476.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:47:42,841 INFO [train.py:903] (1/4) Epoch 2, batch 1650, loss[loss=0.343, simple_loss=0.404, pruned_loss=0.141, over 19665.00 frames. ], tot_loss[loss=0.3757, simple_loss=0.4066, pruned_loss=0.1724, over 3793339.66 frames. ], batch size: 58, lr: 3.49e-02, grad_scale: 8.0 2023-03-31 21:48:07,776 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.3322, 2.4088, 2.4164, 2.3561, 2.3727, 2.2394, 0.3733, 1.9302], device='cuda:1'), covar=tensor([0.0460, 0.0568, 0.0312, 0.0487, 0.0628, 0.0630, 0.1366, 0.0887], device='cuda:1'), in_proj_covar=tensor([0.0197, 0.0205, 0.0195, 0.0231, 0.0269, 0.0240, 0.0245, 0.0249], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-03-31 21:48:10,108 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=8501.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:48:23,195 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.971e+02 8.791e+02 1.048e+03 1.403e+03 4.696e+03, threshold=2.096e+03, percent-clipped=2.0 2023-03-31 21:48:44,116 INFO [train.py:903] (1/4) Epoch 2, batch 1700, loss[loss=0.3561, simple_loss=0.376, pruned_loss=0.1681, over 19336.00 frames. ], tot_loss[loss=0.3781, simple_loss=0.4079, pruned_loss=0.1741, over 3804652.89 frames. ], batch size: 44, lr: 3.49e-02, grad_scale: 8.0 2023-03-31 21:48:52,071 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=8535.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:49:23,164 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=8560.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:49:23,914 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 2023-03-31 21:49:46,288 INFO [train.py:903] (1/4) Epoch 2, batch 1750, loss[loss=0.3256, simple_loss=0.3754, pruned_loss=0.1379, over 19612.00 frames. ], tot_loss[loss=0.3771, simple_loss=0.4074, pruned_loss=0.1734, over 3802149.58 frames. ], batch size: 50, lr: 3.48e-02, grad_scale: 8.0 2023-03-31 21:50:27,274 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.711e+02 8.589e+02 1.062e+03 1.367e+03 2.706e+03, threshold=2.124e+03, percent-clipped=6.0 2023-03-31 21:50:47,001 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=8627.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:50:48,941 INFO [train.py:903] (1/4) Epoch 2, batch 1800, loss[loss=0.4182, simple_loss=0.4426, pruned_loss=0.1969, over 19779.00 frames. ], tot_loss[loss=0.3751, simple_loss=0.406, pruned_loss=0.1721, over 3792541.63 frames. ], batch size: 54, lr: 3.47e-02, grad_scale: 8.0 2023-03-31 21:51:10,780 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.56 vs. limit=2.0 2023-03-31 21:51:48,878 WARNING [train.py:1073] (1/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9_rvb from training. Duration: 25.0944375 2023-03-31 21:51:52,354 INFO [train.py:903] (1/4) Epoch 2, batch 1850, loss[loss=0.4483, simple_loss=0.45, pruned_loss=0.2233, over 13581.00 frames. ], tot_loss[loss=0.3736, simple_loss=0.4049, pruned_loss=0.1712, over 3786374.72 frames. ], batch size: 136, lr: 3.46e-02, grad_scale: 8.0 2023-03-31 21:52:26,738 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9_rvb from training. Duration: 27.8166875 2023-03-31 21:52:32,412 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 5.683e+02 8.545e+02 1.022e+03 1.402e+03 2.945e+03, threshold=2.044e+03, percent-clipped=4.0 2023-03-31 21:52:53,440 INFO [train.py:903] (1/4) Epoch 2, batch 1900, loss[loss=0.3881, simple_loss=0.407, pruned_loss=0.1846, over 19760.00 frames. ], tot_loss[loss=0.3726, simple_loss=0.4041, pruned_loss=0.1705, over 3791633.54 frames. ], batch size: 51, lr: 3.46e-02, grad_scale: 8.0 2023-03-31 21:53:10,049 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=8742.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:53:13,037 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 2023-03-31 21:53:19,069 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9_rvb from training. Duration: 27.02225 2023-03-31 21:53:30,857 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4717, 1.1212, 1.4101, 0.8312, 2.6824, 2.8003, 2.5954, 2.8558], device='cuda:1'), covar=tensor([0.1464, 0.2743, 0.2776, 0.2184, 0.0346, 0.0151, 0.0296, 0.0196], device='cuda:1'), in_proj_covar=tensor([0.0301, 0.0285, 0.0331, 0.0294, 0.0200, 0.0109, 0.0192, 0.0117], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0001, 0.0002, 0.0001], device='cuda:1') 2023-03-31 21:53:44,519 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 2023-03-31 21:53:56,617 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=8778.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:53:57,531 INFO [train.py:903] (1/4) Epoch 2, batch 1950, loss[loss=0.3547, simple_loss=0.3938, pruned_loss=0.1578, over 19716.00 frames. ], tot_loss[loss=0.3709, simple_loss=0.4034, pruned_loss=0.1692, over 3809296.13 frames. ], batch size: 59, lr: 3.45e-02, grad_scale: 8.0 2023-03-31 21:54:08,606 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.3208, 1.6209, 1.3079, 2.0948, 1.8581, 2.0930, 2.0414, 2.0324], device='cuda:1'), covar=tensor([0.0751, 0.1466, 0.1765, 0.1343, 0.1681, 0.1207, 0.1790, 0.0924], device='cuda:1'), in_proj_covar=tensor([0.0261, 0.0301, 0.0306, 0.0329, 0.0372, 0.0274, 0.0345, 0.0257], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-03-31 21:54:18,053 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=2.15 vs. limit=2.0 2023-03-31 21:54:38,566 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 6.007e+02 8.441e+02 1.013e+03 1.280e+03 2.038e+03, threshold=2.026e+03, percent-clipped=0.0 2023-03-31 21:54:43,184 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.60 vs. limit=2.0 2023-03-31 21:55:00,983 INFO [train.py:903] (1/4) Epoch 2, batch 2000, loss[loss=0.3942, simple_loss=0.4273, pruned_loss=0.1806, over 19526.00 frames. ], tot_loss[loss=0.3707, simple_loss=0.4029, pruned_loss=0.1693, over 3791638.69 frames. ], batch size: 56, lr: 3.44e-02, grad_scale: 8.0 2023-03-31 21:55:08,315 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.1527, 1.4842, 1.7031, 2.0927, 2.0762, 2.3496, 2.8546, 1.8661], device='cuda:1'), covar=tensor([0.1083, 0.2209, 0.1711, 0.1594, 0.1945, 0.1397, 0.1490, 0.1411], device='cuda:1'), in_proj_covar=tensor([0.0262, 0.0298, 0.0301, 0.0328, 0.0372, 0.0272, 0.0340, 0.0258], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-03-31 21:56:00,565 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_rvb from training. Duration: 26.8349375 2023-03-31 21:56:01,579 INFO [train.py:903] (1/4) Epoch 2, batch 2050, loss[loss=0.4851, simple_loss=0.4702, pruned_loss=0.25, over 19782.00 frames. ], tot_loss[loss=0.3727, simple_loss=0.4041, pruned_loss=0.1706, over 3802855.69 frames. ], batch size: 56, lr: 3.43e-02, grad_scale: 8.0 2023-03-31 21:56:19,575 WARNING [train.py:1073] (1/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 2023-03-31 21:56:20,763 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 2023-03-31 21:56:41,673 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 2023-03-31 21:56:44,109 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 5.978e+02 1.029e+03 1.194e+03 1.427e+03 4.040e+03, threshold=2.389e+03, percent-clipped=7.0 2023-03-31 21:57:05,503 INFO [train.py:903] (1/4) Epoch 2, batch 2100, loss[loss=0.3512, simple_loss=0.3905, pruned_loss=0.156, over 19574.00 frames. ], tot_loss[loss=0.3726, simple_loss=0.4041, pruned_loss=0.1705, over 3809731.09 frames. ], batch size: 61, lr: 3.43e-02, grad_scale: 8.0 2023-03-31 21:57:25,723 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=8945.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:57:36,780 WARNING [train.py:1073] (1/4) Exclude cut with ID 453-131332-0000-131866_sp0.9_rvb from training. Duration: 25.3333125 2023-03-31 21:57:53,987 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=2.02 vs. limit=2.0 2023-03-31 21:57:59,426 WARNING [train.py:1073] (1/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9_rvb from training. Duration: 26.6166875 2023-03-31 21:58:07,608 INFO [train.py:903] (1/4) Epoch 2, batch 2150, loss[loss=0.3142, simple_loss=0.348, pruned_loss=0.1401, over 19790.00 frames. ], tot_loss[loss=0.3699, simple_loss=0.4023, pruned_loss=0.1687, over 3814506.00 frames. ], batch size: 48, lr: 3.42e-02, grad_scale: 8.0 2023-03-31 21:58:32,397 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=8998.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:58:39,151 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.7972, 1.4040, 1.2713, 1.8230, 1.4768, 1.6551, 1.3536, 1.5826], device='cuda:1'), covar=tensor([0.0804, 0.1843, 0.1468, 0.1111, 0.1699, 0.0687, 0.1252, 0.0729], device='cuda:1'), in_proj_covar=tensor([0.0249, 0.0382, 0.0288, 0.0262, 0.0329, 0.0262, 0.0282, 0.0221], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-03-31 21:58:40,128 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=9004.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:58:49,978 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.755e+02 7.392e+02 9.171e+02 1.181e+03 2.165e+03, threshold=1.834e+03, percent-clipped=0.0 2023-03-31 21:59:04,384 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=9023.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 21:59:11,991 INFO [train.py:903] (1/4) Epoch 2, batch 2200, loss[loss=0.3821, simple_loss=0.4208, pruned_loss=0.1716, over 19684.00 frames. ], tot_loss[loss=0.3707, simple_loss=0.4036, pruned_loss=0.1689, over 3816127.70 frames. ], batch size: 55, lr: 3.41e-02, grad_scale: 8.0 2023-03-31 21:59:14,443 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=9031.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 22:00:13,975 INFO [train.py:903] (1/4) Epoch 2, batch 2250, loss[loss=0.4086, simple_loss=0.4421, pruned_loss=0.1876, over 19442.00 frames. ], tot_loss[loss=0.3701, simple_loss=0.4033, pruned_loss=0.1684, over 3819723.04 frames. ], batch size: 70, lr: 3.41e-02, grad_scale: 8.0 2023-03-31 22:00:39,598 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.2524, 2.2308, 1.8821, 2.7085, 2.0310, 2.6224, 3.0468, 2.3853], device='cuda:1'), covar=tensor([0.0621, 0.1285, 0.1549, 0.1366, 0.1742, 0.1200, 0.1267, 0.0902], device='cuda:1'), in_proj_covar=tensor([0.0261, 0.0300, 0.0297, 0.0328, 0.0367, 0.0271, 0.0335, 0.0256], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-03-31 22:00:56,138 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.993e+02 8.727e+02 9.943e+02 1.293e+03 2.077e+03, threshold=1.989e+03, percent-clipped=4.0 2023-03-31 22:01:08,140 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=9122.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 22:01:10,659 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=5.40 vs. limit=5.0 2023-03-31 22:01:17,886 INFO [train.py:903] (1/4) Epoch 2, batch 2300, loss[loss=0.362, simple_loss=0.404, pruned_loss=0.16, over 19462.00 frames. ], tot_loss[loss=0.3704, simple_loss=0.4037, pruned_loss=0.1685, over 3820461.75 frames. ], batch size: 64, lr: 3.40e-02, grad_scale: 8.0 2023-03-31 22:01:31,582 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_rvb from training. Duration: 26.205 2023-03-31 22:02:19,095 INFO [train.py:903] (1/4) Epoch 2, batch 2350, loss[loss=0.3862, simple_loss=0.4218, pruned_loss=0.1753, over 19512.00 frames. ], tot_loss[loss=0.3702, simple_loss=0.4034, pruned_loss=0.1684, over 3827423.88 frames. ], batch size: 64, lr: 3.39e-02, grad_scale: 8.0 2023-03-31 22:03:00,734 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.988e+02 9.117e+02 1.090e+03 1.432e+03 2.529e+03, threshold=2.180e+03, percent-clipped=5.0 2023-03-31 22:03:00,811 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_rvb from training. Duration: 25.775 2023-03-31 22:03:18,257 WARNING [train.py:1073] (1/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9_rvb from training. Duration: 25.45 2023-03-31 22:03:22,887 INFO [train.py:903] (1/4) Epoch 2, batch 2400, loss[loss=0.3796, simple_loss=0.4185, pruned_loss=0.1703, over 19681.00 frames. ], tot_loss[loss=0.3678, simple_loss=0.4015, pruned_loss=0.167, over 3834403.61 frames. ], batch size: 59, lr: 3.38e-02, grad_scale: 8.0 2023-03-31 22:03:32,282 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=9237.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 22:04:24,300 INFO [train.py:903] (1/4) Epoch 2, batch 2450, loss[loss=0.3914, simple_loss=0.4251, pruned_loss=0.1788, over 18150.00 frames. ], tot_loss[loss=0.3692, simple_loss=0.4024, pruned_loss=0.168, over 3838686.08 frames. ], batch size: 83, lr: 3.38e-02, grad_scale: 8.0 2023-03-31 22:04:38,156 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=9289.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 22:04:50,698 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.52 vs. limit=2.0 2023-03-31 22:05:06,391 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.471e+02 8.791e+02 1.096e+03 1.484e+03 3.289e+03, threshold=2.192e+03, percent-clipped=7.0 2023-03-31 22:05:13,396 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([0.5299, 0.8221, 0.6010, 0.6753, 0.7390, 0.5032, 0.2130, 0.8156], device='cuda:1'), covar=tensor([0.0589, 0.0359, 0.0835, 0.0423, 0.0466, 0.1016, 0.0890, 0.0474], device='cuda:1'), in_proj_covar=tensor([0.0282, 0.0184, 0.0280, 0.0250, 0.0191, 0.0293, 0.0263, 0.0268], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-03-31 22:05:27,994 INFO [train.py:903] (1/4) Epoch 2, batch 2500, loss[loss=0.3596, simple_loss=0.4001, pruned_loss=0.1596, over 19666.00 frames. ], tot_loss[loss=0.3691, simple_loss=0.4021, pruned_loss=0.168, over 3832258.80 frames. ], batch size: 60, lr: 3.37e-02, grad_scale: 8.0 2023-03-31 22:05:52,442 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=9348.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 22:06:25,078 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=9375.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 22:06:29,684 INFO [train.py:903] (1/4) Epoch 2, batch 2550, loss[loss=0.3275, simple_loss=0.3624, pruned_loss=0.1463, over 19486.00 frames. ], tot_loss[loss=0.3683, simple_loss=0.4018, pruned_loss=0.1674, over 3831397.46 frames. ], batch size: 49, lr: 3.36e-02, grad_scale: 8.0 2023-03-31 22:07:01,788 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=9404.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 22:07:11,556 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.591e+02 8.134e+02 9.492e+02 1.275e+03 2.544e+03, threshold=1.898e+03, percent-clipped=3.0 2023-03-31 22:07:26,384 WARNING [train.py:1073] (1/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 2023-03-31 22:07:33,270 INFO [train.py:903] (1/4) Epoch 2, batch 2600, loss[loss=0.3555, simple_loss=0.3998, pruned_loss=0.1556, over 19580.00 frames. ], tot_loss[loss=0.3678, simple_loss=0.4016, pruned_loss=0.167, over 3836683.86 frames. ], batch size: 61, lr: 3.36e-02, grad_scale: 8.0 2023-03-31 22:07:36,914 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=9432.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 22:08:14,519 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=9463.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 22:08:34,690 INFO [train.py:903] (1/4) Epoch 2, batch 2650, loss[loss=0.3614, simple_loss=0.4024, pruned_loss=0.1602, over 19781.00 frames. ], tot_loss[loss=0.367, simple_loss=0.401, pruned_loss=0.1665, over 3842174.43 frames. ], batch size: 56, lr: 3.35e-02, grad_scale: 8.0 2023-03-31 22:08:49,063 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=9490.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 22:08:52,740 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=9493.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 22:08:53,507 WARNING [train.py:1073] (1/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9_rvb from training. Duration: 27.25 2023-03-31 22:09:16,684 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 5.258e+02 8.887e+02 1.023e+03 1.383e+03 3.476e+03, threshold=2.047e+03, percent-clipped=7.0 2023-03-31 22:09:23,987 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=9518.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 22:09:36,965 INFO [train.py:903] (1/4) Epoch 2, batch 2700, loss[loss=0.4118, simple_loss=0.4404, pruned_loss=0.1916, over 19787.00 frames. ], tot_loss[loss=0.3689, simple_loss=0.4023, pruned_loss=0.1678, over 3817902.08 frames. ], batch size: 56, lr: 3.34e-02, grad_scale: 8.0 2023-03-31 22:09:58,953 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.72 vs. limit=5.0 2023-03-31 22:10:39,553 INFO [train.py:903] (1/4) Epoch 2, batch 2750, loss[loss=0.3351, simple_loss=0.381, pruned_loss=0.1446, over 19785.00 frames. ], tot_loss[loss=0.368, simple_loss=0.4015, pruned_loss=0.1673, over 3808852.03 frames. ], batch size: 54, lr: 3.34e-02, grad_scale: 8.0 2023-03-31 22:11:08,792 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=9602.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 22:11:20,990 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 5.046e+02 9.047e+02 1.065e+03 1.297e+03 2.590e+03, threshold=2.130e+03, percent-clipped=3.0 2023-03-31 22:11:43,324 INFO [train.py:903] (1/4) Epoch 2, batch 2800, loss[loss=0.4096, simple_loss=0.4456, pruned_loss=0.1868, over 19605.00 frames. ], tot_loss[loss=0.3682, simple_loss=0.4025, pruned_loss=0.1669, over 3821285.38 frames. ], batch size: 57, lr: 3.33e-02, grad_scale: 8.0 2023-03-31 22:12:20,933 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=9660.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 22:12:41,335 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=9676.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 22:12:44,607 INFO [train.py:903] (1/4) Epoch 2, batch 2850, loss[loss=0.2846, simple_loss=0.3453, pruned_loss=0.1119, over 19761.00 frames. ], tot_loss[loss=0.3679, simple_loss=0.4024, pruned_loss=0.1667, over 3813492.07 frames. ], batch size: 51, lr: 3.32e-02, grad_scale: 8.0 2023-03-31 22:12:51,798 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=9685.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 22:13:26,587 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 6.408e+02 8.898e+02 1.125e+03 1.384e+03 2.599e+03, threshold=2.251e+03, percent-clipped=6.0 2023-03-31 22:13:35,203 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=9719.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 22:13:43,464 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.6175, 3.8306, 4.1231, 3.9599, 1.5035, 3.6177, 3.3273, 3.5603], device='cuda:1'), covar=tensor([0.0364, 0.0450, 0.0386, 0.0328, 0.2885, 0.0205, 0.0400, 0.1005], device='cuda:1'), in_proj_covar=tensor([0.0268, 0.0281, 0.0375, 0.0263, 0.0421, 0.0183, 0.0260, 0.0379], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0001, 0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1') 2023-03-31 22:13:45,582 WARNING [train.py:1073] (1/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 2023-03-31 22:13:46,840 INFO [train.py:903] (1/4) Epoch 2, batch 2900, loss[loss=0.3228, simple_loss=0.3507, pruned_loss=0.1475, over 19718.00 frames. ], tot_loss[loss=0.366, simple_loss=0.4008, pruned_loss=0.1656, over 3816697.75 frames. ], batch size: 46, lr: 3.31e-02, grad_scale: 16.0 2023-03-31 22:14:06,881 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=9744.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 22:14:09,003 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=9746.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 22:14:23,218 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9020, 1.4905, 1.2039, 1.6710, 1.4345, 1.9120, 2.0751, 1.8706], device='cuda:1'), covar=tensor([0.0915, 0.1417, 0.1830, 0.1662, 0.2118, 0.1066, 0.1330, 0.0904], device='cuda:1'), in_proj_covar=tensor([0.0270, 0.0307, 0.0303, 0.0332, 0.0361, 0.0273, 0.0331, 0.0266], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-03-31 22:14:27,818 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=9761.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 22:14:39,344 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=9771.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 22:14:45,602 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=9776.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 22:14:49,535 INFO [train.py:903] (1/4) Epoch 2, batch 2950, loss[loss=0.4028, simple_loss=0.4153, pruned_loss=0.1952, over 19391.00 frames. ], tot_loss[loss=0.3644, simple_loss=0.3994, pruned_loss=0.1647, over 3830385.43 frames. ], batch size: 48, lr: 3.31e-02, grad_scale: 8.0 2023-03-31 22:14:52,367 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.46 vs. limit=2.0 2023-03-31 22:15:31,703 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 5.533e+02 8.475e+02 1.131e+03 1.413e+03 3.215e+03, threshold=2.262e+03, percent-clipped=4.0 2023-03-31 22:15:53,078 INFO [train.py:903] (1/4) Epoch 2, batch 3000, loss[loss=0.3281, simple_loss=0.3638, pruned_loss=0.1462, over 19769.00 frames. ], tot_loss[loss=0.3633, simple_loss=0.3988, pruned_loss=0.1639, over 3827110.36 frames. ], batch size: 47, lr: 3.30e-02, grad_scale: 4.0 2023-03-31 22:15:53,078 INFO [train.py:928] (1/4) Computing validation loss 2023-03-31 22:16:06,234 INFO [train.py:937] (1/4) Epoch 2, validation: loss=0.2513, simple_loss=0.3423, pruned_loss=0.08019, over 944034.00 frames. 2023-03-31 22:16:06,236 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18649MB 2023-03-31 22:16:12,116 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_rvb from training. Duration: 29.735 2023-03-31 22:16:54,068 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 2023-03-31 22:17:08,086 INFO [train.py:903] (1/4) Epoch 2, batch 3050, loss[loss=0.3549, simple_loss=0.3804, pruned_loss=0.1646, over 19770.00 frames. ], tot_loss[loss=0.3615, simple_loss=0.3971, pruned_loss=0.1629, over 3843416.39 frames. ], batch size: 47, lr: 3.29e-02, grad_scale: 4.0 2023-03-31 22:17:22,798 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=9891.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 22:17:50,737 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 5.558e+02 8.806e+02 1.158e+03 1.402e+03 3.282e+03, threshold=2.315e+03, percent-clipped=2.0 2023-03-31 22:18:10,040 INFO [train.py:903] (1/4) Epoch 2, batch 3100, loss[loss=0.3503, simple_loss=0.3875, pruned_loss=0.1565, over 19853.00 frames. ], tot_loss[loss=0.3651, simple_loss=0.3993, pruned_loss=0.1654, over 3819621.03 frames. ], batch size: 52, lr: 3.29e-02, grad_scale: 4.0 2023-03-31 22:18:29,700 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=9946.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 22:19:11,240 INFO [train.py:903] (1/4) Epoch 2, batch 3150, loss[loss=0.4158, simple_loss=0.4329, pruned_loss=0.1993, over 19356.00 frames. ], tot_loss[loss=0.3636, simple_loss=0.3981, pruned_loss=0.1646, over 3830043.30 frames. ], batch size: 70, lr: 3.28e-02, grad_scale: 4.0 2023-03-31 22:19:42,639 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 2023-03-31 22:19:56,512 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.945e+02 7.345e+02 9.404e+02 1.247e+03 3.615e+03, threshold=1.881e+03, percent-clipped=3.0 2023-03-31 22:20:03,835 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=10020.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 22:20:15,207 INFO [train.py:903] (1/4) Epoch 2, batch 3200, loss[loss=0.3589, simple_loss=0.4061, pruned_loss=0.1559, over 19600.00 frames. ], tot_loss[loss=0.3629, simple_loss=0.3977, pruned_loss=0.1641, over 3829689.91 frames. ], batch size: 57, lr: 3.27e-02, grad_scale: 8.0 2023-03-31 22:20:50,873 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=10057.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 22:20:55,701 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=10061.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 22:21:19,932 INFO [train.py:903] (1/4) Epoch 2, batch 3250, loss[loss=0.5004, simple_loss=0.4978, pruned_loss=0.2516, over 19507.00 frames. ], tot_loss[loss=0.3601, simple_loss=0.3959, pruned_loss=0.1621, over 3846718.88 frames. ], batch size: 64, lr: 3.27e-02, grad_scale: 8.0 2023-03-31 22:21:21,254 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=10080.0, num_to_drop=1, layers_to_drop={0} 2023-03-31 22:21:50,580 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=10105.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 22:22:02,472 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.757e+02 8.486e+02 1.037e+03 1.291e+03 3.604e+03, threshold=2.074e+03, percent-clipped=6.0 2023-03-31 22:22:20,718 INFO [train.py:903] (1/4) Epoch 2, batch 3300, loss[loss=0.3392, simple_loss=0.3839, pruned_loss=0.1473, over 19520.00 frames. ], tot_loss[loss=0.3624, simple_loss=0.3979, pruned_loss=0.1635, over 3843812.94 frames. ], batch size: 54, lr: 3.26e-02, grad_scale: 8.0 2023-03-31 22:22:25,202 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 2023-03-31 22:22:27,873 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=10135.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 22:22:42,693 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=10147.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 22:23:14,945 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=10172.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 22:23:18,679 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 2023-03-31 22:23:22,721 INFO [train.py:903] (1/4) Epoch 2, batch 3350, loss[loss=0.3886, simple_loss=0.4179, pruned_loss=0.1796, over 19770.00 frames. ], tot_loss[loss=0.3616, simple_loss=0.3969, pruned_loss=0.1631, over 3833640.25 frames. ], batch size: 56, lr: 3.26e-02, grad_scale: 8.0 2023-03-31 22:24:07,628 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 5.516e+02 8.361e+02 9.909e+02 1.198e+03 2.844e+03, threshold=1.982e+03, percent-clipped=3.0 2023-03-31 22:24:14,942 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=10220.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 22:24:25,934 INFO [train.py:903] (1/4) Epoch 2, batch 3400, loss[loss=0.355, simple_loss=0.3949, pruned_loss=0.1576, over 19671.00 frames. ], tot_loss[loss=0.3595, simple_loss=0.3957, pruned_loss=0.1617, over 3832080.79 frames. ], batch size: 55, lr: 3.25e-02, grad_scale: 8.0 2023-03-31 22:24:47,983 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1209, 1.1199, 1.5174, 1.2367, 1.9576, 1.9085, 2.0824, 0.5590], device='cuda:1'), covar=tensor([0.1212, 0.1697, 0.0996, 0.1222, 0.0630, 0.0801, 0.0662, 0.1490], device='cuda:1'), in_proj_covar=tensor([0.0355, 0.0375, 0.0347, 0.0355, 0.0410, 0.0332, 0.0475, 0.0360], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-03-31 22:25:29,363 INFO [train.py:903] (1/4) Epoch 2, batch 3450, loss[loss=0.3068, simple_loss=0.3607, pruned_loss=0.1264, over 19733.00 frames. ], tot_loss[loss=0.3618, simple_loss=0.3967, pruned_loss=0.1634, over 3822717.92 frames. ], batch size: 51, lr: 3.24e-02, grad_scale: 4.0 2023-03-31 22:25:32,672 WARNING [train.py:1073] (1/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9_rvb from training. Duration: 25.2444375 2023-03-31 22:26:13,338 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 5.771e+02 9.409e+02 1.166e+03 1.453e+03 2.796e+03, threshold=2.333e+03, percent-clipped=9.0 2023-03-31 22:26:17,028 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=10317.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 22:26:31,595 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.67 vs. limit=2.0 2023-03-31 22:26:31,777 INFO [train.py:903] (1/4) Epoch 2, batch 3500, loss[loss=0.2757, simple_loss=0.3274, pruned_loss=0.112, over 19748.00 frames. ], tot_loss[loss=0.3599, simple_loss=0.3954, pruned_loss=0.1622, over 3827593.27 frames. ], batch size: 45, lr: 3.24e-02, grad_scale: 4.0 2023-03-31 22:26:47,019 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=10342.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 22:27:33,762 INFO [train.py:903] (1/4) Epoch 2, batch 3550, loss[loss=0.3623, simple_loss=0.3978, pruned_loss=0.1633, over 18141.00 frames. ], tot_loss[loss=0.3625, simple_loss=0.3974, pruned_loss=0.1638, over 3823961.02 frames. ], batch size: 83, lr: 3.23e-02, grad_scale: 4.0 2023-03-31 22:27:49,487 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=10391.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 22:28:04,141 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=10401.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 22:28:20,153 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 5.354e+02 7.776e+02 1.013e+03 1.369e+03 3.978e+03, threshold=2.027e+03, percent-clipped=2.0 2023-03-31 22:28:21,549 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=10416.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 22:28:30,587 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=10424.0, num_to_drop=1, layers_to_drop={0} 2023-03-31 22:28:35,010 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=10427.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 22:28:37,136 INFO [train.py:903] (1/4) Epoch 2, batch 3600, loss[loss=0.3571, simple_loss=0.3895, pruned_loss=0.1624, over 19837.00 frames. ], tot_loss[loss=0.3619, simple_loss=0.3971, pruned_loss=0.1634, over 3818951.71 frames. ], batch size: 52, lr: 3.22e-02, grad_scale: 8.0 2023-03-31 22:29:26,943 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.51 vs. limit=2.0 2023-03-31 22:29:32,580 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=10472.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 22:29:37,345 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=10476.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 22:29:41,347 INFO [train.py:903] (1/4) Epoch 2, batch 3650, loss[loss=0.3031, simple_loss=0.3506, pruned_loss=0.1278, over 19390.00 frames. ], tot_loss[loss=0.3601, simple_loss=0.3962, pruned_loss=0.162, over 3828222.94 frames. ], batch size: 48, lr: 3.22e-02, grad_scale: 8.0 2023-03-31 22:30:09,140 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=10501.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 22:30:11,812 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 2023-03-31 22:30:26,008 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.209e+02 8.454e+02 1.072e+03 1.396e+03 2.688e+03, threshold=2.143e+03, percent-clipped=6.0 2023-03-31 22:30:27,515 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=10516.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 22:30:45,121 INFO [train.py:903] (1/4) Epoch 2, batch 3700, loss[loss=0.336, simple_loss=0.3764, pruned_loss=0.1478, over 19622.00 frames. ], tot_loss[loss=0.3615, simple_loss=0.397, pruned_loss=0.163, over 3828861.92 frames. ], batch size: 50, lr: 3.21e-02, grad_scale: 8.0 2023-03-31 22:30:57,263 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=10539.0, num_to_drop=1, layers_to_drop={0} 2023-03-31 22:30:59,693 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6774, 1.6099, 1.6547, 2.2623, 1.5834, 2.1135, 1.8637, 1.4926], device='cuda:1'), covar=tensor([0.0844, 0.0705, 0.0433, 0.0374, 0.0773, 0.0269, 0.0883, 0.0724], device='cuda:1'), in_proj_covar=tensor([0.0346, 0.0314, 0.0328, 0.0423, 0.0394, 0.0231, 0.0427, 0.0336], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0003, 0.0002], device='cuda:1') 2023-03-31 22:31:47,573 INFO [train.py:903] (1/4) Epoch 2, batch 3750, loss[loss=0.3208, simple_loss=0.3796, pruned_loss=0.131, over 19798.00 frames. ], tot_loss[loss=0.3578, simple_loss=0.3948, pruned_loss=0.1604, over 3836045.75 frames. ], batch size: 56, lr: 3.20e-02, grad_scale: 8.0 2023-03-31 22:32:33,531 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 5.274e+02 8.986e+02 1.057e+03 1.326e+03 2.585e+03, threshold=2.114e+03, percent-clipped=3.0 2023-03-31 22:32:50,174 INFO [train.py:903] (1/4) Epoch 2, batch 3800, loss[loss=0.4934, simple_loss=0.4847, pruned_loss=0.2511, over 19470.00 frames. ], tot_loss[loss=0.3593, simple_loss=0.396, pruned_loss=0.1613, over 3825111.17 frames. ], batch size: 64, lr: 3.20e-02, grad_scale: 8.0 2023-03-31 22:33:09,668 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2313, 0.9061, 0.9975, 1.4706, 0.9994, 1.1556, 1.3977, 1.1068], device='cuda:1'), covar=tensor([0.0964, 0.1782, 0.1510, 0.0962, 0.1309, 0.1318, 0.1051, 0.1147], device='cuda:1'), in_proj_covar=tensor([0.0267, 0.0311, 0.0302, 0.0325, 0.0357, 0.0277, 0.0324, 0.0259], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-03-31 22:33:23,955 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9_rvb from training. Duration: 29.1166875 2023-03-31 22:33:51,757 INFO [train.py:903] (1/4) Epoch 2, batch 3850, loss[loss=0.4155, simple_loss=0.4421, pruned_loss=0.1945, over 19299.00 frames. ], tot_loss[loss=0.3595, simple_loss=0.3964, pruned_loss=0.1613, over 3829732.06 frames. ], batch size: 66, lr: 3.19e-02, grad_scale: 8.0 2023-03-31 22:34:37,544 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 5.106e+02 8.604e+02 1.077e+03 1.442e+03 2.985e+03, threshold=2.155e+03, percent-clipped=6.0 2023-03-31 22:34:40,397 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.04 vs. limit=5.0 2023-03-31 22:34:41,605 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=2.04 vs. limit=2.0 2023-03-31 22:34:55,297 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 2023-03-31 22:34:56,855 INFO [train.py:903] (1/4) Epoch 2, batch 3900, loss[loss=0.4025, simple_loss=0.4387, pruned_loss=0.1831, over 19478.00 frames. ], tot_loss[loss=0.3607, simple_loss=0.3977, pruned_loss=0.1618, over 3820087.10 frames. ], batch size: 64, lr: 3.19e-02, grad_scale: 8.0 2023-03-31 22:35:36,279 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=10762.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 22:35:49,462 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=10771.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 22:35:50,828 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=10772.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 22:35:58,742 INFO [train.py:903] (1/4) Epoch 2, batch 3950, loss[loss=0.3548, simple_loss=0.3977, pruned_loss=0.156, over 19598.00 frames. ], tot_loss[loss=0.3586, simple_loss=0.3963, pruned_loss=0.1605, over 3831702.78 frames. ], batch size: 61, lr: 3.18e-02, grad_scale: 8.0 2023-03-31 22:36:04,546 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 2023-03-31 22:36:06,222 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3366, 1.2947, 2.0230, 1.5638, 3.0920, 3.3928, 3.5171, 1.2855], device='cuda:1'), covar=tensor([0.1249, 0.1919, 0.1189, 0.1180, 0.0883, 0.0659, 0.1040, 0.2051], device='cuda:1'), in_proj_covar=tensor([0.0351, 0.0379, 0.0351, 0.0349, 0.0409, 0.0330, 0.0482, 0.0368], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-03-31 22:36:14,864 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.98 vs. limit=2.0 2023-03-31 22:36:15,292 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=10792.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 22:36:16,546 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.7482, 1.5790, 1.6905, 1.7239, 2.9004, 4.4269, 4.4062, 4.6610], device='cuda:1'), covar=tensor([0.1486, 0.2606, 0.2754, 0.1985, 0.0480, 0.0095, 0.0136, 0.0088], device='cuda:1'), in_proj_covar=tensor([0.0287, 0.0279, 0.0324, 0.0282, 0.0196, 0.0105, 0.0188, 0.0110], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0001, 0.0002, 0.0001], device='cuda:1') 2023-03-31 22:36:18,920 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=10795.0, num_to_drop=1, layers_to_drop={1} 2023-03-31 22:36:21,056 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=10797.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 22:36:44,828 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 5.316e+02 7.972e+02 1.008e+03 1.220e+03 2.629e+03, threshold=2.016e+03, percent-clipped=1.0 2023-03-31 22:36:46,286 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=10816.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 22:36:51,323 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=10820.0, num_to_drop=1, layers_to_drop={1} 2023-03-31 22:37:01,759 INFO [train.py:903] (1/4) Epoch 2, batch 4000, loss[loss=0.3406, simple_loss=0.3804, pruned_loss=0.1504, over 19841.00 frames. ], tot_loss[loss=0.3572, simple_loss=0.395, pruned_loss=0.1596, over 3809135.73 frames. ], batch size: 52, lr: 3.17e-02, grad_scale: 8.0 2023-03-31 22:37:49,244 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1_rvb from training. Duration: 0.7545625 2023-03-31 22:37:59,127 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.6355, 1.6348, 1.5946, 2.0459, 3.2012, 1.4354, 2.1148, 3.1815], device='cuda:1'), covar=tensor([0.0309, 0.2393, 0.2367, 0.1542, 0.0413, 0.2014, 0.1083, 0.0435], device='cuda:1'), in_proj_covar=tensor([0.0219, 0.0296, 0.0278, 0.0274, 0.0255, 0.0309, 0.0248, 0.0257], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-03-31 22:38:04,673 INFO [train.py:903] (1/4) Epoch 2, batch 4050, loss[loss=0.3669, simple_loss=0.4012, pruned_loss=0.1663, over 13555.00 frames. ], tot_loss[loss=0.3574, simple_loss=0.3948, pruned_loss=0.1599, over 3796100.62 frames. ], batch size: 136, lr: 3.17e-02, grad_scale: 8.0 2023-03-31 22:38:15,732 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=10886.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 22:38:20,803 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 2023-03-31 22:38:35,400 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=10903.0, num_to_drop=1, layers_to_drop={0} 2023-03-31 22:38:49,942 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 5.194e+02 9.186e+02 1.101e+03 1.338e+03 4.215e+03, threshold=2.201e+03, percent-clipped=7.0 2023-03-31 22:39:10,611 INFO [train.py:903] (1/4) Epoch 2, batch 4100, loss[loss=0.3485, simple_loss=0.3917, pruned_loss=0.1527, over 19694.00 frames. ], tot_loss[loss=0.3554, simple_loss=0.3932, pruned_loss=0.1588, over 3785573.21 frames. ], batch size: 59, lr: 3.16e-02, grad_scale: 8.0 2023-03-31 22:39:13,300 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=10931.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 22:39:39,786 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([0.6106, 0.9037, 0.6696, 0.7374, 0.8109, 0.5385, 0.3043, 0.7823], device='cuda:1'), covar=tensor([0.0522, 0.0345, 0.0732, 0.0419, 0.0367, 0.0847, 0.0692, 0.0403], device='cuda:1'), in_proj_covar=tensor([0.0280, 0.0198, 0.0284, 0.0249, 0.0196, 0.0291, 0.0270, 0.0260], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-03-31 22:39:45,325 WARNING [train.py:1073] (1/4) Exclude cut with ID 497-129325-0061-9566_sp1.1_rvb from training. Duration: 0.97725 2023-03-31 22:39:58,418 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9031, 1.4775, 1.6253, 1.6037, 2.7277, 4.5198, 4.4578, 4.8423], device='cuda:1'), covar=tensor([0.1364, 0.2663, 0.2783, 0.1997, 0.0528, 0.0091, 0.0116, 0.0068], device='cuda:1'), in_proj_covar=tensor([0.0289, 0.0279, 0.0329, 0.0285, 0.0195, 0.0106, 0.0188, 0.0108], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0001, 0.0002, 0.0001], device='cuda:1') 2023-03-31 22:40:13,127 INFO [train.py:903] (1/4) Epoch 2, batch 4150, loss[loss=0.3466, simple_loss=0.3979, pruned_loss=0.1476, over 19305.00 frames. ], tot_loss[loss=0.3546, simple_loss=0.3928, pruned_loss=0.1582, over 3792538.21 frames. ], batch size: 66, lr: 3.16e-02, grad_scale: 8.0 2023-03-31 22:40:59,340 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.653e+02 7.731e+02 1.007e+03 1.258e+03 2.097e+03, threshold=2.015e+03, percent-clipped=0.0 2023-03-31 22:41:07,856 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.7993, 2.0419, 2.2825, 2.3668, 3.6215, 1.9059, 2.7055, 3.3566], device='cuda:1'), covar=tensor([0.0392, 0.1959, 0.1690, 0.1205, 0.0306, 0.1650, 0.1060, 0.0395], device='cuda:1'), in_proj_covar=tensor([0.0223, 0.0304, 0.0288, 0.0279, 0.0254, 0.0318, 0.0259, 0.0266], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-03-31 22:41:15,495 INFO [train.py:903] (1/4) Epoch 2, batch 4200, loss[loss=0.344, simple_loss=0.3798, pruned_loss=0.1541, over 19847.00 frames. ], tot_loss[loss=0.3548, simple_loss=0.3931, pruned_loss=0.1582, over 3794251.63 frames. ], batch size: 52, lr: 3.15e-02, grad_scale: 8.0 2023-03-31 22:41:18,919 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 2023-03-31 22:41:43,744 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 2023-03-31 22:42:18,002 INFO [train.py:903] (1/4) Epoch 2, batch 4250, loss[loss=0.3991, simple_loss=0.4076, pruned_loss=0.1953, over 19833.00 frames. ], tot_loss[loss=0.3557, simple_loss=0.3931, pruned_loss=0.1591, over 3801006.64 frames. ], batch size: 52, lr: 3.14e-02, grad_scale: 8.0 2023-03-31 22:42:26,299 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.1511, 1.9446, 1.8250, 2.7943, 2.0295, 2.9496, 2.3236, 1.7487], device='cuda:1'), covar=tensor([0.0888, 0.0743, 0.0511, 0.0478, 0.0907, 0.0222, 0.0937, 0.0813], device='cuda:1'), in_proj_covar=tensor([0.0354, 0.0322, 0.0342, 0.0446, 0.0410, 0.0242, 0.0445, 0.0353], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0001, 0.0003, 0.0002], device='cuda:1') 2023-03-31 22:42:35,133 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9_rvb from training. Duration: 29.816625 2023-03-31 22:42:45,399 WARNING [train.py:1073] (1/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 2023-03-31 22:42:52,693 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=11106.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 22:43:03,987 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 5.611e+02 8.485e+02 1.107e+03 1.406e+03 3.284e+03, threshold=2.214e+03, percent-clipped=7.0 2023-03-31 22:43:21,882 INFO [train.py:903] (1/4) Epoch 2, batch 4300, loss[loss=0.2916, simple_loss=0.3438, pruned_loss=0.1197, over 19377.00 frames. ], tot_loss[loss=0.3543, simple_loss=0.3922, pruned_loss=0.1582, over 3799439.36 frames. ], batch size: 48, lr: 3.14e-02, grad_scale: 8.0 2023-03-31 22:43:30,408 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=11136.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 22:43:37,635 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=11142.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 22:44:09,025 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=11167.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 22:44:16,407 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 2023-03-31 22:44:23,250 INFO [train.py:903] (1/4) Epoch 2, batch 4350, loss[loss=0.3548, simple_loss=0.3997, pruned_loss=0.1549, over 18859.00 frames. ], tot_loss[loss=0.3527, simple_loss=0.391, pruned_loss=0.1572, over 3804982.94 frames. ], batch size: 74, lr: 3.13e-02, grad_scale: 8.0 2023-03-31 22:44:32,889 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=11187.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 22:44:39,332 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=11192.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 22:45:04,735 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=11212.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 22:45:09,091 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.029e+02 7.841e+02 9.559e+02 1.160e+03 2.939e+03, threshold=1.912e+03, percent-clipped=2.0 2023-03-31 22:45:15,625 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=11221.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 22:45:24,371 INFO [train.py:903] (1/4) Epoch 2, batch 4400, loss[loss=0.3854, simple_loss=0.4217, pruned_loss=0.1746, over 19662.00 frames. ], tot_loss[loss=0.351, simple_loss=0.3898, pruned_loss=0.1561, over 3817225.24 frames. ], batch size: 59, lr: 3.13e-02, grad_scale: 8.0 2023-03-31 22:45:48,406 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=11247.0, num_to_drop=1, layers_to_drop={0} 2023-03-31 22:45:50,511 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_rvb from training. Duration: 25.285 2023-03-31 22:45:54,217 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=11251.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 22:46:00,715 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 2023-03-31 22:46:21,482 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.54 vs. limit=2.0 2023-03-31 22:46:27,505 INFO [train.py:903] (1/4) Epoch 2, batch 4450, loss[loss=0.3571, simple_loss=0.4001, pruned_loss=0.157, over 19525.00 frames. ], tot_loss[loss=0.3497, simple_loss=0.3887, pruned_loss=0.1553, over 3820060.47 frames. ], batch size: 54, lr: 3.12e-02, grad_scale: 8.0 2023-03-31 22:47:14,189 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 5.163e+02 8.820e+02 1.081e+03 1.372e+03 2.333e+03, threshold=2.162e+03, percent-clipped=5.0 2023-03-31 22:47:31,602 INFO [train.py:903] (1/4) Epoch 2, batch 4500, loss[loss=0.4342, simple_loss=0.4441, pruned_loss=0.2121, over 13520.00 frames. ], tot_loss[loss=0.3512, simple_loss=0.3895, pruned_loss=0.1565, over 3795329.65 frames. ], batch size: 136, lr: 3.12e-02, grad_scale: 8.0 2023-03-31 22:48:00,561 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.7009, 1.5142, 1.4305, 2.0136, 1.7996, 1.5719, 1.6431, 1.8105], device='cuda:1'), covar=tensor([0.0846, 0.1888, 0.1339, 0.0998, 0.1188, 0.0673, 0.0991, 0.0605], device='cuda:1'), in_proj_covar=tensor([0.0248, 0.0376, 0.0286, 0.0258, 0.0311, 0.0260, 0.0275, 0.0224], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-03-31 22:48:12,960 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=11362.0, num_to_drop=1, layers_to_drop={1} 2023-03-31 22:48:34,556 INFO [train.py:903] (1/4) Epoch 2, batch 4550, loss[loss=0.266, simple_loss=0.3231, pruned_loss=0.1045, over 19757.00 frames. ], tot_loss[loss=0.3503, simple_loss=0.389, pruned_loss=0.1558, over 3795513.28 frames. ], batch size: 45, lr: 3.11e-02, grad_scale: 8.0 2023-03-31 22:48:37,411 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9456, 1.5422, 1.8847, 1.7908, 2.8902, 4.0614, 4.3557, 4.8692], device='cuda:1'), covar=tensor([0.1429, 0.2733, 0.2585, 0.1948, 0.0472, 0.0205, 0.0145, 0.0088], device='cuda:1'), in_proj_covar=tensor([0.0284, 0.0277, 0.0324, 0.0281, 0.0191, 0.0106, 0.0189, 0.0109], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0001, 0.0002, 0.0001], device='cuda:1') 2023-03-31 22:48:45,435 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9_rvb from training. Duration: 28.72225 2023-03-31 22:49:08,222 WARNING [train.py:1073] (1/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 2023-03-31 22:49:21,603 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.316e+02 7.788e+02 1.003e+03 1.220e+03 2.356e+03, threshold=2.005e+03, percent-clipped=1.0 2023-03-31 22:49:31,376 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.7715, 4.5824, 5.3998, 5.2917, 1.9340, 4.9079, 4.3901, 4.8039], device='cuda:1'), covar=tensor([0.0311, 0.0452, 0.0284, 0.0181, 0.2903, 0.0151, 0.0256, 0.0678], device='cuda:1'), in_proj_covar=tensor([0.0301, 0.0293, 0.0408, 0.0294, 0.0436, 0.0189, 0.0268, 0.0400], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1') 2023-03-31 22:49:37,101 INFO [train.py:903] (1/4) Epoch 2, batch 4600, loss[loss=0.3535, simple_loss=0.4003, pruned_loss=0.1533, over 19599.00 frames. ], tot_loss[loss=0.3474, simple_loss=0.387, pruned_loss=0.1539, over 3815373.49 frames. ], batch size: 57, lr: 3.10e-02, grad_scale: 8.0 2023-03-31 22:50:37,505 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=11477.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 22:50:39,644 INFO [train.py:903] (1/4) Epoch 2, batch 4650, loss[loss=0.3221, simple_loss=0.3569, pruned_loss=0.1437, over 19069.00 frames. ], tot_loss[loss=0.3499, simple_loss=0.3889, pruned_loss=0.1554, over 3812299.31 frames. ], batch size: 42, lr: 3.10e-02, grad_scale: 8.0 2023-03-31 22:50:59,949 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_rvb from training. Duration: 0.92 2023-03-31 22:51:02,638 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.8053, 1.8429, 1.6812, 2.3014, 1.8591, 2.6587, 2.6045, 2.1859], device='cuda:1'), covar=tensor([0.0699, 0.1299, 0.1435, 0.1460, 0.1699, 0.0933, 0.1365, 0.0873], device='cuda:1'), in_proj_covar=tensor([0.0265, 0.0310, 0.0296, 0.0325, 0.0352, 0.0269, 0.0326, 0.0263], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1') 2023-03-31 22:51:09,695 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=11502.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 22:51:10,450 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_rvb from training. Duration: 0.83 2023-03-31 22:51:15,583 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=11507.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 22:51:26,519 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 5.039e+02 8.212e+02 1.126e+03 1.409e+03 2.689e+03, threshold=2.252e+03, percent-clipped=6.0 2023-03-31 22:51:42,948 INFO [train.py:903] (1/4) Epoch 2, batch 4700, loss[loss=0.3999, simple_loss=0.4256, pruned_loss=0.1871, over 19598.00 frames. ], tot_loss[loss=0.3496, simple_loss=0.3892, pruned_loss=0.155, over 3821494.88 frames. ], batch size: 61, lr: 3.09e-02, grad_scale: 8.0 2023-03-31 22:51:48,018 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=11532.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 22:51:52,267 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=11536.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 22:52:05,041 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9_rvb from training. Duration: 0.92225 2023-03-31 22:52:08,750 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=11550.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 22:52:46,155 INFO [train.py:903] (1/4) Epoch 2, batch 4750, loss[loss=0.3083, simple_loss=0.3526, pruned_loss=0.132, over 19388.00 frames. ], tot_loss[loss=0.35, simple_loss=0.3895, pruned_loss=0.1552, over 3815706.33 frames. ], batch size: 48, lr: 3.09e-02, grad_scale: 8.0 2023-03-31 22:53:23,995 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 2023-03-31 22:53:32,502 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.603e+02 7.433e+02 1.012e+03 1.332e+03 3.283e+03, threshold=2.025e+03, percent-clipped=2.0 2023-03-31 22:53:35,149 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=11618.0, num_to_drop=1, layers_to_drop={1} 2023-03-31 22:53:36,118 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.8775, 3.5584, 2.2497, 3.1688, 1.3150, 3.3864, 3.1306, 3.2629], device='cuda:1'), covar=tensor([0.0671, 0.1040, 0.1883, 0.0733, 0.3067, 0.0848, 0.0705, 0.0822], device='cuda:1'), in_proj_covar=tensor([0.0286, 0.0281, 0.0311, 0.0256, 0.0320, 0.0279, 0.0220, 0.0246], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-03-31 22:53:45,440 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.0927, 1.5052, 1.6381, 2.3129, 1.9953, 2.5298, 2.8830, 2.2987], device='cuda:1'), covar=tensor([0.0449, 0.1274, 0.1304, 0.1249, 0.1432, 0.0902, 0.1167, 0.0746], device='cuda:1'), in_proj_covar=tensor([0.0254, 0.0306, 0.0292, 0.0321, 0.0345, 0.0264, 0.0320, 0.0252], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-03-31 22:53:47,416 INFO [train.py:903] (1/4) Epoch 2, batch 4800, loss[loss=0.3652, simple_loss=0.4053, pruned_loss=0.1626, over 19657.00 frames. ], tot_loss[loss=0.3501, simple_loss=0.3895, pruned_loss=0.1553, over 3814818.85 frames. ], batch size: 60, lr: 3.08e-02, grad_scale: 8.0 2023-03-31 22:53:53,497 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 2023-03-31 22:54:04,647 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=11643.0, num_to_drop=1, layers_to_drop={1} 2023-03-31 22:54:15,083 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=11651.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 22:54:19,416 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=11653.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 22:54:49,928 INFO [train.py:903] (1/4) Epoch 2, batch 4850, loss[loss=0.3838, simple_loss=0.42, pruned_loss=0.1738, over 19743.00 frames. ], tot_loss[loss=0.3527, simple_loss=0.3914, pruned_loss=0.157, over 3807579.89 frames. ], batch size: 63, lr: 3.08e-02, grad_scale: 8.0 2023-03-31 22:55:14,996 WARNING [train.py:1073] (1/4) Exclude cut with ID 774-127930-0014-48411_sp1.1_rvb from training. Duration: 0.95 2023-03-31 22:55:34,589 WARNING [train.py:1073] (1/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1_rvb from training. Duration: 0.9409375 2023-03-31 22:55:36,934 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.407e+02 7.585e+02 1.008e+03 1.288e+03 2.592e+03, threshold=2.016e+03, percent-clipped=4.0 2023-03-31 22:55:38,523 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=11717.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 22:55:40,653 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 2023-03-31 22:55:41,808 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 2023-03-31 22:55:42,123 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9530, 1.3321, 1.2011, 1.8220, 1.4877, 1.9278, 2.2121, 1.6833], device='cuda:1'), covar=tensor([0.0778, 0.1482, 0.1602, 0.1434, 0.1588, 0.1108, 0.1284, 0.0918], device='cuda:1'), in_proj_covar=tensor([0.0260, 0.0310, 0.0293, 0.0321, 0.0350, 0.0270, 0.0321, 0.0255], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1') 2023-03-31 22:55:51,139 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1_rvb from training. Duration: 25.3818125 2023-03-31 22:55:53,220 INFO [train.py:903] (1/4) Epoch 2, batch 4900, loss[loss=0.3419, simple_loss=0.3792, pruned_loss=0.1523, over 19471.00 frames. ], tot_loss[loss=0.3509, simple_loss=0.3904, pruned_loss=0.1557, over 3820926.81 frames. ], batch size: 49, lr: 3.07e-02, grad_scale: 8.0 2023-03-31 22:56:13,257 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 2023-03-31 22:56:25,972 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 2023-03-31 22:56:55,832 INFO [train.py:903] (1/4) Epoch 2, batch 4950, loss[loss=0.307, simple_loss=0.3435, pruned_loss=0.1353, over 19338.00 frames. ], tot_loss[loss=0.3515, simple_loss=0.3907, pruned_loss=0.1561, over 3829793.76 frames. ], batch size: 44, lr: 3.06e-02, grad_scale: 8.0 2023-03-31 22:57:12,065 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 2023-03-31 22:57:37,558 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 2023-03-31 22:57:41,841 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 5.086e+02 9.257e+02 1.136e+03 1.412e+03 3.441e+03, threshold=2.272e+03, percent-clipped=4.0 2023-03-31 22:57:57,804 INFO [train.py:903] (1/4) Epoch 2, batch 5000, loss[loss=0.3266, simple_loss=0.3658, pruned_loss=0.1438, over 19332.00 frames. ], tot_loss[loss=0.353, simple_loss=0.3919, pruned_loss=0.1571, over 3815097.49 frames. ], batch size: 44, lr: 3.06e-02, grad_scale: 8.0 2023-03-31 22:58:04,641 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=11834.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 22:58:07,425 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_rvb from training. Duration: 27.14 2023-03-31 22:58:16,656 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 2023-03-31 22:58:42,531 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=11864.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 22:59:00,547 INFO [train.py:903] (1/4) Epoch 2, batch 5050, loss[loss=0.3117, simple_loss=0.3502, pruned_loss=0.1366, over 19699.00 frames. ], tot_loss[loss=0.3534, simple_loss=0.3915, pruned_loss=0.1576, over 3820056.85 frames. ], batch size: 45, lr: 3.05e-02, grad_scale: 8.0 2023-03-31 22:59:19,025 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=11894.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 22:59:36,730 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=11907.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 22:59:38,657 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9_rvb from training. Duration: 26.62775 2023-03-31 22:59:48,010 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 5.479e+02 8.836e+02 1.119e+03 1.460e+03 3.605e+03, threshold=2.237e+03, percent-clipped=7.0 2023-03-31 22:59:48,451 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.2683, 2.0514, 1.4647, 1.8064, 1.4472, 1.5628, 0.2429, 1.2010], device='cuda:1'), covar=tensor([0.0292, 0.0271, 0.0271, 0.0224, 0.0611, 0.0420, 0.0697, 0.0514], device='cuda:1'), in_proj_covar=tensor([0.0213, 0.0230, 0.0230, 0.0244, 0.0307, 0.0263, 0.0260, 0.0259], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-03-31 23:00:03,340 INFO [train.py:903] (1/4) Epoch 2, batch 5100, loss[loss=0.3284, simple_loss=0.3847, pruned_loss=0.1361, over 19487.00 frames. ], tot_loss[loss=0.3514, simple_loss=0.3903, pruned_loss=0.1563, over 3821190.68 frames. ], batch size: 64, lr: 3.05e-02, grad_scale: 8.0 2023-03-31 23:00:08,038 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=11932.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 23:00:16,232 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=11938.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 23:00:16,322 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5122, 1.0620, 1.0889, 1.6585, 1.3143, 1.6834, 1.8309, 1.3422], device='cuda:1'), covar=tensor([0.0863, 0.1501, 0.1613, 0.1143, 0.1304, 0.0921, 0.1053, 0.1035], device='cuda:1'), in_proj_covar=tensor([0.0263, 0.0306, 0.0294, 0.0325, 0.0343, 0.0263, 0.0315, 0.0257], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1') 2023-03-31 23:00:19,213 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 2023-03-31 23:00:21,237 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 2023-03-31 23:00:22,858 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_rvb from training. Duration: 25.035 2023-03-31 23:00:26,277 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_rvb from training. Duration: 27.92 2023-03-31 23:00:43,217 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([5.1312, 4.7665, 5.7839, 5.5397, 1.8463, 5.1459, 4.6305, 5.1666], device='cuda:1'), covar=tensor([0.0341, 0.0399, 0.0312, 0.0203, 0.2818, 0.0124, 0.0312, 0.0639], device='cuda:1'), in_proj_covar=tensor([0.0311, 0.0297, 0.0398, 0.0295, 0.0436, 0.0197, 0.0276, 0.0407], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1') 2023-03-31 23:00:50,278 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=11966.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 23:01:07,394 INFO [train.py:903] (1/4) Epoch 2, batch 5150, loss[loss=0.3428, simple_loss=0.3906, pruned_loss=0.1475, over 19788.00 frames. ], tot_loss[loss=0.3514, simple_loss=0.3903, pruned_loss=0.1563, over 3818986.52 frames. ], batch size: 56, lr: 3.04e-02, grad_scale: 8.0 2023-03-31 23:01:20,935 WARNING [train.py:1073] (1/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 2023-03-31 23:01:29,904 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=11997.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 23:01:36,653 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=12002.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 23:01:44,883 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12009.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 23:01:54,415 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.835e+02 8.254e+02 9.989e+02 1.287e+03 2.673e+03, threshold=1.998e+03, percent-clipped=4.0 2023-03-31 23:01:56,786 WARNING [train.py:1073] (1/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1_rvb from training. Duration: 0.9681875 2023-03-31 23:02:10,382 INFO [train.py:903] (1/4) Epoch 2, batch 5200, loss[loss=0.3272, simple_loss=0.3712, pruned_loss=0.1416, over 19591.00 frames. ], tot_loss[loss=0.3512, simple_loss=0.3904, pruned_loss=0.156, over 3819171.04 frames. ], batch size: 52, lr: 3.04e-02, grad_scale: 8.0 2023-03-31 23:02:16,670 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([5.2516, 4.8921, 6.0270, 5.6268, 2.0111, 5.5482, 4.8619, 5.3113], device='cuda:1'), covar=tensor([0.0323, 0.0424, 0.0302, 0.0231, 0.3135, 0.0103, 0.0302, 0.0799], device='cuda:1'), in_proj_covar=tensor([0.0316, 0.0299, 0.0405, 0.0301, 0.0438, 0.0200, 0.0280, 0.0411], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1') 2023-03-31 23:02:25,414 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 2023-03-31 23:02:52,147 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=12061.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 23:03:11,210 WARNING [train.py:1073] (1/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 2023-03-31 23:03:13,777 INFO [train.py:903] (1/4) Epoch 2, batch 5250, loss[loss=0.2866, simple_loss=0.33, pruned_loss=0.1216, over 19317.00 frames. ], tot_loss[loss=0.3502, simple_loss=0.3895, pruned_loss=0.1555, over 3821012.84 frames. ], batch size: 44, lr: 3.03e-02, grad_scale: 8.0 2023-03-31 23:03:32,756 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=12094.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 23:03:55,634 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12112.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 23:03:59,693 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 5.115e+02 8.666e+02 1.057e+03 1.421e+03 4.195e+03, threshold=2.115e+03, percent-clipped=5.0 2023-03-31 23:04:14,787 INFO [train.py:903] (1/4) Epoch 2, batch 5300, loss[loss=0.3109, simple_loss=0.3681, pruned_loss=0.1269, over 19539.00 frames. ], tot_loss[loss=0.3511, simple_loss=0.3902, pruned_loss=0.156, over 3809327.69 frames. ], batch size: 56, lr: 3.03e-02, grad_scale: 8.0 2023-03-31 23:04:35,580 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 2023-03-31 23:04:58,629 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 2023-03-31 23:05:00,389 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([6.0179, 5.3797, 2.7802, 4.8208, 0.9301, 5.4256, 5.2249, 5.4666], device='cuda:1'), covar=tensor([0.0386, 0.0867, 0.1777, 0.0520, 0.3743, 0.0642, 0.0361, 0.0428], device='cuda:1'), in_proj_covar=tensor([0.0307, 0.0291, 0.0320, 0.0270, 0.0339, 0.0289, 0.0230, 0.0259], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-03-31 23:05:13,792 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12176.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 23:05:15,925 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=12178.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 23:05:16,933 INFO [train.py:903] (1/4) Epoch 2, batch 5350, loss[loss=0.3654, simple_loss=0.4025, pruned_loss=0.1642, over 19113.00 frames. ], tot_loss[loss=0.3517, simple_loss=0.3903, pruned_loss=0.1566, over 3817147.78 frames. ], batch size: 69, lr: 3.02e-02, grad_scale: 8.0 2023-03-31 23:05:53,523 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9_rvb from training. Duration: 26.438875 2023-03-31 23:05:53,656 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=12208.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 23:06:03,422 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 5.241e+02 9.122e+02 1.169e+03 1.506e+03 3.477e+03, threshold=2.338e+03, percent-clipped=13.0 2023-03-31 23:06:20,779 INFO [train.py:903] (1/4) Epoch 2, batch 5400, loss[loss=0.2986, simple_loss=0.3593, pruned_loss=0.1189, over 19658.00 frames. ], tot_loss[loss=0.3523, simple_loss=0.3907, pruned_loss=0.1569, over 3815950.30 frames. ], batch size: 53, lr: 3.02e-02, grad_scale: 8.0 2023-03-31 23:07:06,441 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=12265.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 23:07:23,451 INFO [train.py:903] (1/4) Epoch 2, batch 5450, loss[loss=0.355, simple_loss=0.4005, pruned_loss=0.1547, over 19459.00 frames. ], tot_loss[loss=0.3516, simple_loss=0.3905, pruned_loss=0.1563, over 3815707.18 frames. ], batch size: 64, lr: 3.01e-02, grad_scale: 8.0 2023-03-31 23:07:27,096 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=12282.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 23:07:36,440 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=12290.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 23:07:39,770 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12293.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 23:08:01,552 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=12310.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 23:08:08,445 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.315e+02 7.539e+02 9.464e+02 1.137e+03 1.898e+03, threshold=1.893e+03, percent-clipped=0.0 2023-03-31 23:08:17,872 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12323.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 23:08:23,992 INFO [train.py:903] (1/4) Epoch 2, batch 5500, loss[loss=0.3339, simple_loss=0.3779, pruned_loss=0.1449, over 19592.00 frames. ], tot_loss[loss=0.3516, simple_loss=0.3909, pruned_loss=0.1562, over 3826392.77 frames. ], batch size: 52, lr: 3.01e-02, grad_scale: 8.0 2023-03-31 23:08:38,828 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2888, 1.2496, 2.2330, 1.5283, 2.9255, 3.0929, 3.4715, 1.5629], device='cuda:1'), covar=tensor([0.1345, 0.2164, 0.1191, 0.1249, 0.0972, 0.0829, 0.1190, 0.1965], device='cuda:1'), in_proj_covar=tensor([0.0368, 0.0403, 0.0369, 0.0366, 0.0437, 0.0354, 0.0514, 0.0392], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-03-31 23:08:47,596 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=12346.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 23:08:49,885 WARNING [train.py:1073] (1/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9_rvb from training. Duration: 27.47775 2023-03-31 23:09:14,242 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=12368.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 23:09:17,007 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.78 vs. limit=5.0 2023-03-31 23:09:28,525 INFO [train.py:903] (1/4) Epoch 2, batch 5550, loss[loss=0.3483, simple_loss=0.3927, pruned_loss=0.152, over 19668.00 frames. ], tot_loss[loss=0.3513, simple_loss=0.3906, pruned_loss=0.1559, over 3807306.86 frames. ], batch size: 60, lr: 3.00e-02, grad_scale: 8.0 2023-03-31 23:09:36,184 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 2023-03-31 23:09:46,135 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=12393.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 23:09:50,681 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12397.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 23:10:05,526 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.9750, 4.1536, 4.5615, 4.4506, 1.5220, 3.9676, 3.8092, 4.0349], device='cuda:1'), covar=tensor([0.0404, 0.0420, 0.0317, 0.0215, 0.2921, 0.0190, 0.0292, 0.0794], device='cuda:1'), in_proj_covar=tensor([0.0313, 0.0300, 0.0404, 0.0303, 0.0447, 0.0198, 0.0281, 0.0410], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1') 2023-03-31 23:10:14,304 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.827e+02 8.577e+02 1.018e+03 1.198e+03 2.956e+03, threshold=2.037e+03, percent-clipped=3.0 2023-03-31 23:10:25,907 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 2023-03-31 23:10:27,073 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12425.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 23:10:31,331 INFO [train.py:903] (1/4) Epoch 2, batch 5600, loss[loss=0.3757, simple_loss=0.4153, pruned_loss=0.168, over 19091.00 frames. ], tot_loss[loss=0.3507, simple_loss=0.3903, pruned_loss=0.1555, over 3823708.83 frames. ], batch size: 69, lr: 3.00e-02, grad_scale: 8.0 2023-03-31 23:10:35,294 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=12432.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 23:10:41,859 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=12438.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 23:11:06,543 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=12457.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 23:11:12,018 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12461.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 23:11:16,296 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=12464.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 23:11:33,642 INFO [train.py:903] (1/4) Epoch 2, batch 5650, loss[loss=0.309, simple_loss=0.3636, pruned_loss=0.1272, over 19692.00 frames. ], tot_loss[loss=0.3495, simple_loss=0.3891, pruned_loss=0.1549, over 3817243.65 frames. ], batch size: 53, lr: 2.99e-02, grad_scale: 8.0 2023-03-31 23:12:19,586 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 5.393e+02 8.152e+02 1.040e+03 1.340e+03 2.595e+03, threshold=2.080e+03, percent-clipped=4.0 2023-03-31 23:12:21,925 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9_rvb from training. Duration: 28.0944375 2023-03-31 23:12:32,531 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([5.0316, 2.0549, 2.0948, 2.8534, 4.4217, 1.8014, 2.2418, 4.4116], device='cuda:1'), covar=tensor([0.0172, 0.1909, 0.1830, 0.1047, 0.0292, 0.1626, 0.0964, 0.0274], device='cuda:1'), in_proj_covar=tensor([0.0219, 0.0295, 0.0276, 0.0263, 0.0256, 0.0304, 0.0253, 0.0262], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-03-31 23:12:35,558 INFO [train.py:903] (1/4) Epoch 2, batch 5700, loss[loss=0.3586, simple_loss=0.4062, pruned_loss=0.1555, over 19656.00 frames. ], tot_loss[loss=0.3521, simple_loss=0.3907, pruned_loss=0.1568, over 3828561.37 frames. ], batch size: 55, lr: 2.98e-02, grad_scale: 8.0 2023-03-31 23:12:42,951 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2683, 1.0404, 1.0007, 1.4237, 1.0522, 1.2950, 1.2926, 1.2665], device='cuda:1'), covar=tensor([0.0845, 0.1327, 0.1438, 0.0889, 0.1198, 0.1018, 0.1117, 0.0900], device='cuda:1'), in_proj_covar=tensor([0.0261, 0.0301, 0.0290, 0.0322, 0.0345, 0.0272, 0.0312, 0.0260], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1') 2023-03-31 23:13:02,395 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=12549.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 23:13:06,799 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12553.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 23:13:32,315 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=12574.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 23:13:37,679 INFO [train.py:903] (1/4) Epoch 2, batch 5750, loss[loss=0.318, simple_loss=0.3629, pruned_loss=0.1365, over 19754.00 frames. ], tot_loss[loss=0.3499, simple_loss=0.3896, pruned_loss=0.1551, over 3834124.61 frames. ], batch size: 51, lr: 2.98e-02, grad_scale: 8.0 2023-03-31 23:13:38,102 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=12579.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 23:13:42,211 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9_rvb from training. Duration: 33.038875 2023-03-31 23:13:50,271 WARNING [train.py:1073] (1/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 2023-03-31 23:13:56,868 WARNING [train.py:1073] (1/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 2023-03-31 23:14:10,454 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=12604.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 23:14:23,857 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.880e+02 8.083e+02 9.516e+02 1.322e+03 3.330e+03, threshold=1.903e+03, percent-clipped=5.0 2023-03-31 23:14:40,961 INFO [train.py:903] (1/4) Epoch 2, batch 5800, loss[loss=0.3164, simple_loss=0.36, pruned_loss=0.1364, over 19390.00 frames. ], tot_loss[loss=0.3504, simple_loss=0.39, pruned_loss=0.1554, over 3838139.48 frames. ], batch size: 48, lr: 2.97e-02, grad_scale: 8.0 2023-03-31 23:14:58,358 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9394, 2.0069, 1.9747, 2.3993, 2.2266, 2.0663, 1.7329, 2.2372], device='cuda:1'), covar=tensor([0.0663, 0.1335, 0.0908, 0.0676, 0.0947, 0.0456, 0.0875, 0.0459], device='cuda:1'), in_proj_covar=tensor([0.0258, 0.0376, 0.0281, 0.0252, 0.0319, 0.0260, 0.0277, 0.0229], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-03-31 23:15:10,978 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=12653.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 23:15:42,684 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=12678.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 23:15:43,508 INFO [train.py:903] (1/4) Epoch 2, batch 5850, loss[loss=0.3973, simple_loss=0.4278, pruned_loss=0.1833, over 19594.00 frames. ], tot_loss[loss=0.349, simple_loss=0.389, pruned_loss=0.1545, over 3831931.13 frames. ], batch size: 61, lr: 2.97e-02, grad_scale: 8.0 2023-03-31 23:15:46,262 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=12681.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 23:16:18,363 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=12706.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 23:16:30,352 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 5.479e+02 7.980e+02 9.827e+02 1.217e+03 2.781e+03, threshold=1.965e+03, percent-clipped=6.0 2023-03-31 23:16:31,914 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=12717.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 23:16:46,098 INFO [train.py:903] (1/4) Epoch 2, batch 5900, loss[loss=0.3491, simple_loss=0.3993, pruned_loss=0.1495, over 19498.00 frames. ], tot_loss[loss=0.3498, simple_loss=0.3895, pruned_loss=0.155, over 3826518.49 frames. ], batch size: 64, lr: 2.96e-02, grad_scale: 8.0 2023-03-31 23:16:49,455 WARNING [train.py:1073] (1/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 2023-03-31 23:17:03,736 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=12742.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 23:17:11,481 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9_rvb from training. Duration: 27.511125 2023-03-31 23:17:49,031 INFO [train.py:903] (1/4) Epoch 2, batch 5950, loss[loss=0.3769, simple_loss=0.407, pruned_loss=0.1734, over 17504.00 frames. ], tot_loss[loss=0.3495, simple_loss=0.3895, pruned_loss=0.1547, over 3820425.72 frames. ], batch size: 101, lr: 2.96e-02, grad_scale: 8.0 2023-03-31 23:17:52,908 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=12781.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 23:18:25,412 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=12808.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 23:18:26,825 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=12809.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 23:18:34,471 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.856e+02 9.090e+02 1.139e+03 1.452e+03 3.383e+03, threshold=2.279e+03, percent-clipped=8.0 2023-03-31 23:18:51,784 INFO [train.py:903] (1/4) Epoch 2, batch 6000, loss[loss=0.3455, simple_loss=0.3975, pruned_loss=0.1468, over 19664.00 frames. ], tot_loss[loss=0.3488, simple_loss=0.3893, pruned_loss=0.1542, over 3819598.77 frames. ], batch size: 58, lr: 2.95e-02, grad_scale: 8.0 2023-03-31 23:18:51,784 INFO [train.py:928] (1/4) Computing validation loss 2023-03-31 23:19:06,008 INFO [train.py:937] (1/4) Epoch 2, validation: loss=0.246, simple_loss=0.337, pruned_loss=0.07745, over 944034.00 frames. 2023-03-31 23:19:06,010 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18649MB 2023-03-31 23:19:13,340 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=12834.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 23:19:43,052 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.0099, 1.0412, 1.4043, 1.1046, 1.8663, 1.8041, 1.9400, 0.6292], device='cuda:1'), covar=tensor([0.1486, 0.2123, 0.1158, 0.1493, 0.0796, 0.1033, 0.0825, 0.1876], device='cuda:1'), in_proj_covar=tensor([0.0364, 0.0401, 0.0360, 0.0361, 0.0436, 0.0347, 0.0507, 0.0381], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-03-31 23:20:08,251 INFO [train.py:903] (1/4) Epoch 2, batch 6050, loss[loss=0.3162, simple_loss=0.3681, pruned_loss=0.1322, over 19571.00 frames. ], tot_loss[loss=0.3508, simple_loss=0.3904, pruned_loss=0.1556, over 3825782.21 frames. ], batch size: 52, lr: 2.95e-02, grad_scale: 4.0 2023-03-31 23:20:56,514 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 5.060e+02 7.840e+02 9.937e+02 1.323e+03 8.220e+03, threshold=1.987e+03, percent-clipped=9.0 2023-03-31 23:21:03,484 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12923.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 23:21:10,258 INFO [train.py:903] (1/4) Epoch 2, batch 6100, loss[loss=0.331, simple_loss=0.3796, pruned_loss=0.1412, over 19666.00 frames. ], tot_loss[loss=0.3489, simple_loss=0.3889, pruned_loss=0.1544, over 3832313.08 frames. ], batch size: 60, lr: 2.94e-02, grad_scale: 4.0 2023-03-31 23:22:11,867 INFO [train.py:903] (1/4) Epoch 2, batch 6150, loss[loss=0.3034, simple_loss=0.3603, pruned_loss=0.1233, over 19834.00 frames. ], tot_loss[loss=0.349, simple_loss=0.3888, pruned_loss=0.1546, over 3823884.57 frames. ], batch size: 52, lr: 2.94e-02, grad_scale: 4.0 2023-03-31 23:22:28,848 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5259, 1.1237, 1.1061, 1.7532, 1.2695, 1.6242, 1.5782, 1.3927], device='cuda:1'), covar=tensor([0.0861, 0.1395, 0.1559, 0.0995, 0.1282, 0.0969, 0.1290, 0.1009], device='cuda:1'), in_proj_covar=tensor([0.0271, 0.0305, 0.0294, 0.0327, 0.0345, 0.0274, 0.0311, 0.0271], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1') 2023-03-31 23:22:40,948 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2897, 1.2026, 1.8691, 1.5131, 2.4605, 2.5864, 2.7187, 1.1878], device='cuda:1'), covar=tensor([0.1404, 0.2159, 0.1289, 0.1288, 0.1013, 0.0931, 0.1269, 0.2111], device='cuda:1'), in_proj_covar=tensor([0.0368, 0.0400, 0.0368, 0.0367, 0.0442, 0.0353, 0.0519, 0.0388], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-03-31 23:22:42,915 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9_rvb from training. Duration: 31.02225 2023-03-31 23:23:00,772 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 5.553e+02 7.749e+02 1.029e+03 1.287e+03 3.235e+03, threshold=2.059e+03, percent-clipped=7.0 2023-03-31 23:23:13,348 INFO [train.py:903] (1/4) Epoch 2, batch 6200, loss[loss=0.3469, simple_loss=0.3867, pruned_loss=0.1535, over 19772.00 frames. ], tot_loss[loss=0.3474, simple_loss=0.3876, pruned_loss=0.1536, over 3830166.36 frames. ], batch size: 54, lr: 2.93e-02, grad_scale: 4.0 2023-03-31 23:24:15,393 INFO [train.py:903] (1/4) Epoch 2, batch 6250, loss[loss=0.4393, simple_loss=0.441, pruned_loss=0.2187, over 13248.00 frames. ], tot_loss[loss=0.3463, simple_loss=0.3866, pruned_loss=0.153, over 3826204.49 frames. ], batch size: 136, lr: 2.93e-02, grad_scale: 4.0 2023-03-31 23:24:47,225 WARNING [train.py:1073] (1/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9_rvb from training. Duration: 25.988875 2023-03-31 23:25:04,357 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 5.358e+02 8.552e+02 1.023e+03 1.333e+03 3.705e+03, threshold=2.046e+03, percent-clipped=2.0 2023-03-31 23:25:13,038 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=13125.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 23:25:17,708 INFO [train.py:903] (1/4) Epoch 2, batch 6300, loss[loss=0.3248, simple_loss=0.3831, pruned_loss=0.1333, over 19356.00 frames. ], tot_loss[loss=0.3466, simple_loss=0.3866, pruned_loss=0.1533, over 3821485.59 frames. ], batch size: 66, lr: 2.92e-02, grad_scale: 4.0 2023-03-31 23:26:19,875 INFO [train.py:903] (1/4) Epoch 2, batch 6350, loss[loss=0.2451, simple_loss=0.3021, pruned_loss=0.09405, over 19712.00 frames. ], tot_loss[loss=0.3435, simple_loss=0.3843, pruned_loss=0.1513, over 3807868.44 frames. ], batch size: 45, lr: 2.92e-02, grad_scale: 4.0 2023-03-31 23:26:20,309 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=13179.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 23:26:49,675 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=13204.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 23:27:06,689 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 5.673e+02 8.539e+02 1.071e+03 1.407e+03 4.202e+03, threshold=2.141e+03, percent-clipped=6.0 2023-03-31 23:27:19,358 INFO [train.py:903] (1/4) Epoch 2, batch 6400, loss[loss=0.4104, simple_loss=0.4279, pruned_loss=0.1964, over 18732.00 frames. ], tot_loss[loss=0.3434, simple_loss=0.3843, pruned_loss=0.1512, over 3815702.97 frames. ], batch size: 74, lr: 2.92e-02, grad_scale: 8.0 2023-03-31 23:27:33,749 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=13240.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 23:28:22,168 INFO [train.py:903] (1/4) Epoch 2, batch 6450, loss[loss=0.3485, simple_loss=0.3869, pruned_loss=0.1551, over 19489.00 frames. ], tot_loss[loss=0.342, simple_loss=0.383, pruned_loss=0.1505, over 3823126.46 frames. ], batch size: 64, lr: 2.91e-02, grad_scale: 8.0 2023-03-31 23:29:09,605 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9_rvb from training. Duration: 28.638875 2023-03-31 23:29:10,634 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.859e+02 7.874e+02 9.907e+02 1.203e+03 2.411e+03, threshold=1.981e+03, percent-clipped=4.0 2023-03-31 23:29:24,102 INFO [train.py:903] (1/4) Epoch 2, batch 6500, loss[loss=0.3001, simple_loss=0.3568, pruned_loss=0.1217, over 19831.00 frames. ], tot_loss[loss=0.3432, simple_loss=0.3843, pruned_loss=0.151, over 3830142.87 frames. ], batch size: 52, lr: 2.91e-02, grad_scale: 8.0 2023-03-31 23:29:30,893 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1_rvb from training. Duration: 0.836375 2023-03-31 23:29:41,462 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.7703, 1.0888, 1.4072, 1.0463, 2.5292, 3.3891, 3.1653, 3.4654], device='cuda:1'), covar=tensor([0.1218, 0.2685, 0.2745, 0.2138, 0.0447, 0.0105, 0.0218, 0.0113], device='cuda:1'), in_proj_covar=tensor([0.0278, 0.0273, 0.0319, 0.0274, 0.0186, 0.0108, 0.0191, 0.0110], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0001, 0.0002, 0.0001], device='cuda:1') 2023-03-31 23:30:26,693 INFO [train.py:903] (1/4) Epoch 2, batch 6550, loss[loss=0.2915, simple_loss=0.344, pruned_loss=0.1195, over 19615.00 frames. ], tot_loss[loss=0.3412, simple_loss=0.3829, pruned_loss=0.1497, over 3819299.02 frames. ], batch size: 50, lr: 2.90e-02, grad_scale: 8.0 2023-03-31 23:31:14,493 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.467e+02 7.924e+02 9.507e+02 1.218e+03 2.525e+03, threshold=1.901e+03, percent-clipped=3.0 2023-03-31 23:31:27,086 INFO [train.py:903] (1/4) Epoch 2, batch 6600, loss[loss=0.4018, simple_loss=0.4279, pruned_loss=0.1879, over 19522.00 frames. ], tot_loss[loss=0.3426, simple_loss=0.3838, pruned_loss=0.1507, over 3831277.92 frames. ], batch size: 54, lr: 2.90e-02, grad_scale: 8.0 2023-03-31 23:32:15,794 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 2023-03-31 23:32:20,370 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.78 vs. limit=5.0 2023-03-31 23:32:29,104 INFO [train.py:903] (1/4) Epoch 2, batch 6650, loss[loss=0.3893, simple_loss=0.4181, pruned_loss=0.1803, over 19788.00 frames. ], tot_loss[loss=0.345, simple_loss=0.3854, pruned_loss=0.1523, over 3819750.69 frames. ], batch size: 56, lr: 2.89e-02, grad_scale: 8.0 2023-03-31 23:32:38,046 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1574, 1.1536, 1.9052, 1.3194, 2.4991, 2.2645, 2.6783, 1.1316], device='cuda:1'), covar=tensor([0.1468, 0.2303, 0.1257, 0.1316, 0.0827, 0.0975, 0.0963, 0.1917], device='cuda:1'), in_proj_covar=tensor([0.0382, 0.0415, 0.0375, 0.0369, 0.0440, 0.0359, 0.0525, 0.0391], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-03-31 23:32:50,443 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=13496.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 23:33:17,438 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 5.558e+02 9.096e+02 1.181e+03 1.471e+03 3.411e+03, threshold=2.361e+03, percent-clipped=10.0 2023-03-31 23:33:21,166 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=13521.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 23:33:29,660 INFO [train.py:903] (1/4) Epoch 2, batch 6700, loss[loss=0.3413, simple_loss=0.3871, pruned_loss=0.1478, over 19516.00 frames. ], tot_loss[loss=0.3439, simple_loss=0.3846, pruned_loss=0.1516, over 3831368.93 frames. ], batch size: 64, lr: 2.89e-02, grad_scale: 8.0 2023-03-31 23:34:27,765 INFO [train.py:903] (1/4) Epoch 2, batch 6750, loss[loss=0.3902, simple_loss=0.4171, pruned_loss=0.1817, over 19597.00 frames. ], tot_loss[loss=0.3447, simple_loss=0.3855, pruned_loss=0.152, over 3826433.34 frames. ], batch size: 61, lr: 2.88e-02, grad_scale: 8.0 2023-03-31 23:34:33,435 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.7641, 1.7062, 2.0097, 2.5482, 4.3106, 1.3798, 2.4671, 4.2390], device='cuda:1'), covar=tensor([0.0230, 0.2471, 0.2177, 0.1432, 0.0361, 0.2202, 0.0962, 0.0332], device='cuda:1'), in_proj_covar=tensor([0.0234, 0.0303, 0.0284, 0.0274, 0.0275, 0.0320, 0.0263, 0.0271], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-03-31 23:35:12,680 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.052e+02 8.174e+02 9.996e+02 1.293e+03 2.664e+03, threshold=1.999e+03, percent-clipped=1.0 2023-03-31 23:35:25,232 INFO [train.py:903] (1/4) Epoch 2, batch 6800, loss[loss=0.3436, simple_loss=0.398, pruned_loss=0.1446, over 19523.00 frames. ], tot_loss[loss=0.3469, simple_loss=0.3873, pruned_loss=0.1533, over 3810698.81 frames. ], batch size: 56, lr: 2.88e-02, grad_scale: 8.0 2023-03-31 23:36:10,334 WARNING [train.py:1073] (1/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 2023-03-31 23:36:11,469 WARNING [train.py:1073] (1/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 2023-03-31 23:36:13,723 INFO [train.py:903] (1/4) Epoch 3, batch 0, loss[loss=0.3454, simple_loss=0.3681, pruned_loss=0.1613, over 19750.00 frames. ], tot_loss[loss=0.3454, simple_loss=0.3681, pruned_loss=0.1613, over 19750.00 frames. ], batch size: 46, lr: 2.73e-02, grad_scale: 8.0 2023-03-31 23:36:13,724 INFO [train.py:928] (1/4) Computing validation loss 2023-03-31 23:36:24,488 INFO [train.py:937] (1/4) Epoch 3, validation: loss=0.241, simple_loss=0.3346, pruned_loss=0.07374, over 944034.00 frames. 2023-03-31 23:36:24,489 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18649MB 2023-03-31 23:36:37,424 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 2023-03-31 23:37:25,523 INFO [train.py:903] (1/4) Epoch 3, batch 50, loss[loss=0.2776, simple_loss=0.3319, pruned_loss=0.1117, over 19769.00 frames. ], tot_loss[loss=0.3369, simple_loss=0.3794, pruned_loss=0.1472, over 851938.49 frames. ], batch size: 47, lr: 2.73e-02, grad_scale: 8.0 2023-03-31 23:37:38,309 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.119e+02 7.787e+02 9.326e+02 1.115e+03 3.182e+03, threshold=1.865e+03, percent-clipped=5.0 2023-03-31 23:37:49,194 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 2023-03-31 23:37:58,967 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1_rvb from training. Duration: 27.0318125 2023-03-31 23:38:23,959 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=13755.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 23:38:25,858 INFO [train.py:903] (1/4) Epoch 3, batch 100, loss[loss=0.3243, simple_loss=0.3608, pruned_loss=0.1438, over 19777.00 frames. ], tot_loss[loss=0.3369, simple_loss=0.3807, pruned_loss=0.1465, over 1504901.87 frames. ], batch size: 49, lr: 2.72e-02, grad_scale: 8.0 2023-03-31 23:38:35,109 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 2023-03-31 23:39:27,419 INFO [train.py:903] (1/4) Epoch 3, batch 150, loss[loss=0.2804, simple_loss=0.3342, pruned_loss=0.1133, over 19377.00 frames. ], tot_loss[loss=0.3418, simple_loss=0.3844, pruned_loss=0.1496, over 2021395.17 frames. ], batch size: 48, lr: 2.72e-02, grad_scale: 8.0 2023-03-31 23:39:30,164 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.4548, 1.0652, 1.3230, 1.4073, 2.1930, 1.2920, 1.9103, 2.1445], device='cuda:1'), covar=tensor([0.0418, 0.2347, 0.2045, 0.1277, 0.0534, 0.1368, 0.0705, 0.0574], device='cuda:1'), in_proj_covar=tensor([0.0231, 0.0303, 0.0287, 0.0275, 0.0273, 0.0318, 0.0265, 0.0271], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-03-31 23:39:40,067 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.782e+02 7.521e+02 1.009e+03 1.351e+03 3.530e+03, threshold=2.018e+03, percent-clipped=10.0 2023-03-31 23:40:28,867 INFO [train.py:903] (1/4) Epoch 3, batch 200, loss[loss=0.3951, simple_loss=0.4178, pruned_loss=0.1862, over 19667.00 frames. ], tot_loss[loss=0.3389, simple_loss=0.3824, pruned_loss=0.1477, over 2430099.76 frames. ], batch size: 58, lr: 2.71e-02, grad_scale: 8.0 2023-03-31 23:40:28,910 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 2023-03-31 23:41:28,987 INFO [train.py:903] (1/4) Epoch 3, batch 250, loss[loss=0.2986, simple_loss=0.3544, pruned_loss=0.1214, over 19683.00 frames. ], tot_loss[loss=0.3431, simple_loss=0.385, pruned_loss=0.1506, over 2737408.59 frames. ], batch size: 53, lr: 2.71e-02, grad_scale: 8.0 2023-03-31 23:41:44,255 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.886e+02 8.729e+02 1.056e+03 1.304e+03 3.760e+03, threshold=2.113e+03, percent-clipped=6.0 2023-03-31 23:41:49,459 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 2023-03-31 23:42:33,073 INFO [train.py:903] (1/4) Epoch 3, batch 300, loss[loss=0.3195, simple_loss=0.3554, pruned_loss=0.1418, over 19785.00 frames. ], tot_loss[loss=0.3402, simple_loss=0.3828, pruned_loss=0.1488, over 2985583.95 frames. ], batch size: 47, lr: 2.71e-02, grad_scale: 8.0 2023-03-31 23:43:34,497 INFO [train.py:903] (1/4) Epoch 3, batch 350, loss[loss=0.3078, simple_loss=0.3582, pruned_loss=0.1287, over 19670.00 frames. ], tot_loss[loss=0.3388, simple_loss=0.3812, pruned_loss=0.1482, over 3170047.03 frames. ], batch size: 53, lr: 2.70e-02, grad_scale: 8.0 2023-03-31 23:43:40,034 WARNING [train.py:1073] (1/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1_rvb from training. Duration: 0.9681875 2023-03-31 23:43:46,961 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.582e+02 7.628e+02 9.853e+02 1.217e+03 3.369e+03, threshold=1.971e+03, percent-clipped=3.0 2023-03-31 23:44:34,932 INFO [train.py:903] (1/4) Epoch 3, batch 400, loss[loss=0.3771, simple_loss=0.4193, pruned_loss=0.1675, over 19525.00 frames. ], tot_loss[loss=0.3391, simple_loss=0.3815, pruned_loss=0.1484, over 3323361.82 frames. ], batch size: 56, lr: 2.70e-02, grad_scale: 8.0 2023-03-31 23:45:17,155 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=14090.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 23:45:22,985 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2317, 1.2421, 1.8648, 1.4080, 2.5629, 2.3651, 2.9305, 1.2163], device='cuda:1'), covar=tensor([0.1177, 0.1781, 0.1002, 0.1042, 0.0824, 0.0835, 0.0974, 0.1733], device='cuda:1'), in_proj_covar=tensor([0.0384, 0.0417, 0.0385, 0.0374, 0.0455, 0.0366, 0.0536, 0.0393], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-03-31 23:45:27,294 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=14099.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 23:45:36,305 INFO [train.py:903] (1/4) Epoch 3, batch 450, loss[loss=0.3416, simple_loss=0.3846, pruned_loss=0.1493, over 19348.00 frames. ], tot_loss[loss=0.3368, simple_loss=0.3801, pruned_loss=0.1467, over 3438705.05 frames. ], batch size: 66, lr: 2.69e-02, grad_scale: 8.0 2023-03-31 23:45:52,383 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.440e+02 8.213e+02 1.015e+03 1.206e+03 3.609e+03, threshold=2.029e+03, percent-clipped=6.0 2023-03-31 23:45:55,489 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.84 vs. limit=5.0 2023-03-31 23:45:55,957 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=14121.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 23:45:59,453 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.0259, 1.0639, 1.8308, 1.3924, 2.4512, 2.0279, 2.6486, 1.0108], device='cuda:1'), covar=tensor([0.1537, 0.2308, 0.1146, 0.1258, 0.0904, 0.1112, 0.1063, 0.2062], device='cuda:1'), in_proj_covar=tensor([0.0384, 0.0420, 0.0385, 0.0370, 0.0455, 0.0368, 0.0533, 0.0391], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-03-31 23:46:10,162 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 2023-03-31 23:46:11,152 WARNING [train.py:1073] (1/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 2023-03-31 23:46:14,946 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=14138.0, num_to_drop=1, layers_to_drop={0} 2023-03-31 23:46:38,894 INFO [train.py:903] (1/4) Epoch 3, batch 500, loss[loss=0.3308, simple_loss=0.3658, pruned_loss=0.1479, over 19674.00 frames. ], tot_loss[loss=0.3391, simple_loss=0.3815, pruned_loss=0.1484, over 3524044.84 frames. ], batch size: 53, lr: 2.69e-02, grad_scale: 8.0 2023-03-31 23:47:38,989 INFO [train.py:903] (1/4) Epoch 3, batch 550, loss[loss=0.3097, simple_loss=0.3533, pruned_loss=0.133, over 19341.00 frames. ], tot_loss[loss=0.339, simple_loss=0.3815, pruned_loss=0.1482, over 3594694.30 frames. ], batch size: 47, lr: 2.68e-02, grad_scale: 8.0 2023-03-31 23:47:47,487 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=14214.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 23:47:51,334 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.716e+02 8.063e+02 9.949e+02 1.307e+03 2.222e+03, threshold=1.990e+03, percent-clipped=3.0 2023-03-31 23:48:38,812 INFO [train.py:903] (1/4) Epoch 3, batch 600, loss[loss=0.3169, simple_loss=0.365, pruned_loss=0.1344, over 19735.00 frames. ], tot_loss[loss=0.3387, simple_loss=0.3816, pruned_loss=0.1479, over 3643562.24 frames. ], batch size: 51, lr: 2.68e-02, grad_scale: 8.0 2023-03-31 23:49:16,712 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 2023-03-31 23:49:39,181 INFO [train.py:903] (1/4) Epoch 3, batch 650, loss[loss=0.339, simple_loss=0.3814, pruned_loss=0.1483, over 19780.00 frames. ], tot_loss[loss=0.3379, simple_loss=0.3812, pruned_loss=0.1473, over 3686016.46 frames. ], batch size: 54, lr: 2.68e-02, grad_scale: 8.0 2023-03-31 23:49:48,981 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.91 vs. limit=5.0 2023-03-31 23:49:54,623 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 5.103e+02 8.257e+02 1.098e+03 1.322e+03 3.191e+03, threshold=2.197e+03, percent-clipped=10.0 2023-03-31 23:50:04,329 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9423, 1.4263, 1.3656, 1.8465, 1.6435, 1.6024, 1.5070, 1.6622], device='cuda:1'), covar=tensor([0.0645, 0.1345, 0.1172, 0.0703, 0.0901, 0.0502, 0.0814, 0.0636], device='cuda:1'), in_proj_covar=tensor([0.0266, 0.0380, 0.0290, 0.0255, 0.0321, 0.0271, 0.0278, 0.0242], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-03-31 23:50:41,498 INFO [train.py:903] (1/4) Epoch 3, batch 700, loss[loss=0.3101, simple_loss=0.3556, pruned_loss=0.1323, over 19721.00 frames. ], tot_loss[loss=0.3369, simple_loss=0.3801, pruned_loss=0.1468, over 3720611.88 frames. ], batch size: 51, lr: 2.67e-02, grad_scale: 8.0 2023-03-31 23:51:43,784 INFO [train.py:903] (1/4) Epoch 3, batch 750, loss[loss=0.295, simple_loss=0.3421, pruned_loss=0.1239, over 19467.00 frames. ], tot_loss[loss=0.3365, simple_loss=0.3801, pruned_loss=0.1465, over 3735242.79 frames. ], batch size: 49, lr: 2.67e-02, grad_scale: 8.0 2023-03-31 23:51:56,450 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.299e+02 7.798e+02 9.551e+02 1.191e+03 2.807e+03, threshold=1.910e+03, percent-clipped=6.0 2023-03-31 23:52:11,652 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=14431.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 23:52:14,915 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=14434.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 23:52:44,747 INFO [train.py:903] (1/4) Epoch 3, batch 800, loss[loss=0.3176, simple_loss=0.3644, pruned_loss=0.1354, over 19607.00 frames. ], tot_loss[loss=0.3365, simple_loss=0.3803, pruned_loss=0.1464, over 3754242.35 frames. ], batch size: 50, lr: 2.66e-02, grad_scale: 8.0 2023-03-31 23:52:53,415 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 2023-03-31 23:52:53,873 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=14465.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 23:52:54,933 WARNING [train.py:1073] (1/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 2023-03-31 23:52:59,786 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=14470.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 23:53:04,268 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([0.7696, 1.2694, 0.9794, 0.8527, 1.1828, 0.8195, 0.2312, 1.0989], device='cuda:1'), covar=tensor([0.0634, 0.0553, 0.0984, 0.0591, 0.0452, 0.1172, 0.1019, 0.0519], device='cuda:1'), in_proj_covar=tensor([0.0282, 0.0219, 0.0310, 0.0254, 0.0204, 0.0316, 0.0271, 0.0257], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-03-31 23:53:15,249 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=14482.0, num_to_drop=1, layers_to_drop={0} 2023-03-31 23:53:30,978 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=14495.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 23:53:42,697 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=14505.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 23:53:44,740 INFO [train.py:903] (1/4) Epoch 3, batch 850, loss[loss=0.4422, simple_loss=0.4436, pruned_loss=0.2204, over 12994.00 frames. ], tot_loss[loss=0.3393, simple_loss=0.3819, pruned_loss=0.1484, over 3756727.42 frames. ], batch size: 136, lr: 2.66e-02, grad_scale: 8.0 2023-03-31 23:53:58,374 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.292e+02 8.633e+02 1.105e+03 1.534e+03 3.114e+03, threshold=2.210e+03, percent-clipped=11.0 2023-03-31 23:54:11,047 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.1524, 1.2338, 1.3716, 1.9830, 2.6469, 1.8429, 1.9766, 2.6298], device='cuda:1'), covar=tensor([0.0463, 0.2818, 0.2328, 0.1311, 0.0661, 0.1639, 0.1068, 0.0629], device='cuda:1'), in_proj_covar=tensor([0.0235, 0.0302, 0.0290, 0.0277, 0.0270, 0.0319, 0.0260, 0.0269], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-03-31 23:54:32,152 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9_rvb from training. Duration: 25.061125 2023-03-31 23:54:32,565 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.7572, 1.8262, 1.3286, 1.3851, 1.2650, 1.3859, 0.1991, 0.6977], device='cuda:1'), covar=tensor([0.0223, 0.0212, 0.0154, 0.0162, 0.0483, 0.0230, 0.0427, 0.0376], device='cuda:1'), in_proj_covar=tensor([0.0232, 0.0238, 0.0239, 0.0251, 0.0311, 0.0265, 0.0255, 0.0259], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-03-31 23:54:35,854 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=14549.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 23:54:45,564 INFO [train.py:903] (1/4) Epoch 3, batch 900, loss[loss=0.3627, simple_loss=0.3978, pruned_loss=0.1638, over 13821.00 frames. ], tot_loss[loss=0.3396, simple_loss=0.3826, pruned_loss=0.1482, over 3763109.74 frames. ], batch size: 136, lr: 2.66e-02, grad_scale: 8.0 2023-03-31 23:55:15,227 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=14580.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 23:55:33,859 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=14597.0, num_to_drop=1, layers_to_drop={0} 2023-03-31 23:55:46,731 INFO [train.py:903] (1/4) Epoch 3, batch 950, loss[loss=0.3312, simple_loss=0.3874, pruned_loss=0.1375, over 19516.00 frames. ], tot_loss[loss=0.3373, simple_loss=0.3816, pruned_loss=0.1466, over 3784060.85 frames. ], batch size: 64, lr: 2.65e-02, grad_scale: 4.0 2023-03-31 23:55:46,741 WARNING [train.py:1073] (1/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9_rvb from training. Duration: 26.32775 2023-03-31 23:56:00,994 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.858e+02 7.381e+02 9.246e+02 1.263e+03 4.500e+03, threshold=1.849e+03, percent-clipped=5.0 2023-03-31 23:56:46,894 INFO [train.py:903] (1/4) Epoch 3, batch 1000, loss[loss=0.3293, simple_loss=0.3857, pruned_loss=0.1365, over 19516.00 frames. ], tot_loss[loss=0.3363, simple_loss=0.3805, pruned_loss=0.146, over 3795784.32 frames. ], batch size: 54, lr: 2.65e-02, grad_scale: 4.0 2023-03-31 23:57:10,651 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9301, 1.8503, 1.8079, 2.6833, 1.7538, 2.5799, 2.3473, 1.8398], device='cuda:1'), covar=tensor([0.0790, 0.0643, 0.0422, 0.0337, 0.0730, 0.0205, 0.0649, 0.0608], device='cuda:1'), in_proj_covar=tensor([0.0415, 0.0390, 0.0397, 0.0517, 0.0468, 0.0301, 0.0489, 0.0400], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-03-31 23:57:38,691 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 2023-03-31 23:57:47,677 INFO [train.py:903] (1/4) Epoch 3, batch 1050, loss[loss=0.3786, simple_loss=0.4227, pruned_loss=0.1673, over 19372.00 frames. ], tot_loss[loss=0.3353, simple_loss=0.3797, pruned_loss=0.1454, over 3805689.55 frames. ], batch size: 66, lr: 2.64e-02, grad_scale: 4.0 2023-03-31 23:58:01,062 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.125e+02 7.305e+02 8.951e+02 1.118e+03 2.421e+03, threshold=1.790e+03, percent-clipped=2.0 2023-03-31 23:58:17,596 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 2023-03-31 23:58:48,367 INFO [train.py:903] (1/4) Epoch 3, batch 1100, loss[loss=0.2907, simple_loss=0.3335, pruned_loss=0.124, over 19745.00 frames. ], tot_loss[loss=0.3351, simple_loss=0.3793, pruned_loss=0.1455, over 3791513.54 frames. ], batch size: 45, lr: 2.64e-02, grad_scale: 4.0 2023-03-31 23:58:49,809 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=14758.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 23:59:10,910 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=14775.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 23:59:11,733 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.64 vs. limit=2.0 2023-03-31 23:59:47,759 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=14805.0, num_to_drop=0, layers_to_drop=set() 2023-03-31 23:59:49,456 INFO [train.py:903] (1/4) Epoch 3, batch 1150, loss[loss=0.3836, simple_loss=0.4126, pruned_loss=0.1774, over 17274.00 frames. ], tot_loss[loss=0.3351, simple_loss=0.3793, pruned_loss=0.1454, over 3785701.18 frames. ], batch size: 101, lr: 2.64e-02, grad_scale: 4.0 2023-03-31 23:59:50,857 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.2096, 2.8857, 1.7252, 2.6711, 0.9699, 2.7815, 2.5963, 2.7400], device='cuda:1'), covar=tensor([0.0986, 0.1700, 0.2254, 0.0948, 0.3899, 0.1209, 0.0858, 0.1112], device='cuda:1'), in_proj_covar=tensor([0.0294, 0.0286, 0.0318, 0.0260, 0.0338, 0.0284, 0.0233, 0.0266], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 00:00:03,474 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=2.02 vs. limit=2.0 2023-04-01 00:00:03,884 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.960e+02 7.465e+02 1.021e+03 1.238e+03 3.548e+03, threshold=2.043e+03, percent-clipped=7.0 2023-04-01 00:00:16,791 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=14830.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 00:00:23,624 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=14836.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 00:00:40,912 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=14849.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 00:00:45,704 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=14853.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 00:00:49,934 INFO [train.py:903] (1/4) Epoch 3, batch 1200, loss[loss=0.2988, simple_loss=0.3439, pruned_loss=0.1269, over 19361.00 frames. ], tot_loss[loss=0.3337, simple_loss=0.3784, pruned_loss=0.1446, over 3798740.41 frames. ], batch size: 47, lr: 2.63e-02, grad_scale: 8.0 2023-04-01 00:00:55,788 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=14861.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 00:01:15,108 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=14878.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 00:01:17,848 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 2023-04-01 00:01:31,260 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=14890.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 00:01:51,491 INFO [train.py:903] (1/4) Epoch 3, batch 1250, loss[loss=0.2622, simple_loss=0.3134, pruned_loss=0.1055, over 19757.00 frames. ], tot_loss[loss=0.3313, simple_loss=0.3764, pruned_loss=0.1432, over 3811993.77 frames. ], batch size: 47, lr: 2.63e-02, grad_scale: 8.0 2023-04-01 00:02:05,938 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.325e+02 7.701e+02 1.002e+03 1.250e+03 2.941e+03, threshold=2.004e+03, percent-clipped=3.0 2023-04-01 00:02:26,242 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.06 vs. limit=5.0 2023-04-01 00:02:53,118 INFO [train.py:903] (1/4) Epoch 3, batch 1300, loss[loss=0.3276, simple_loss=0.3775, pruned_loss=0.1389, over 19378.00 frames. ], tot_loss[loss=0.3299, simple_loss=0.3752, pruned_loss=0.1422, over 3820195.41 frames. ], batch size: 70, lr: 2.62e-02, grad_scale: 8.0 2023-04-01 00:03:02,025 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=14964.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 00:03:54,923 INFO [train.py:903] (1/4) Epoch 3, batch 1350, loss[loss=0.3226, simple_loss=0.373, pruned_loss=0.1361, over 19664.00 frames. ], tot_loss[loss=0.3323, simple_loss=0.3767, pruned_loss=0.144, over 3817698.31 frames. ], batch size: 58, lr: 2.62e-02, grad_scale: 8.0 2023-04-01 00:04:10,656 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 5.789e+02 8.351e+02 9.883e+02 1.225e+03 3.360e+03, threshold=1.977e+03, percent-clipped=2.0 2023-04-01 00:04:57,762 INFO [train.py:903] (1/4) Epoch 3, batch 1400, loss[loss=0.3939, simple_loss=0.4199, pruned_loss=0.184, over 19585.00 frames. ], tot_loss[loss=0.3312, simple_loss=0.376, pruned_loss=0.1432, over 3829194.48 frames. ], batch size: 61, lr: 2.62e-02, grad_scale: 8.0 2023-04-01 00:05:53,742 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=15102.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 00:05:59,109 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_rvb from training. Duration: 25.85 2023-04-01 00:06:00,262 INFO [train.py:903] (1/4) Epoch 3, batch 1450, loss[loss=0.3139, simple_loss=0.3702, pruned_loss=0.1288, over 19528.00 frames. ], tot_loss[loss=0.3304, simple_loss=0.3758, pruned_loss=0.1425, over 3834532.92 frames. ], batch size: 54, lr: 2.61e-02, grad_scale: 8.0 2023-04-01 00:06:13,780 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.668e+02 7.923e+02 9.351e+02 1.150e+03 2.880e+03, threshold=1.870e+03, percent-clipped=3.0 2023-04-01 00:06:49,262 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=15146.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 00:07:01,532 INFO [train.py:903] (1/4) Epoch 3, batch 1500, loss[loss=0.3339, simple_loss=0.3888, pruned_loss=0.1395, over 18942.00 frames. ], tot_loss[loss=0.3297, simple_loss=0.3751, pruned_loss=0.1422, over 3832009.79 frames. ], batch size: 74, lr: 2.61e-02, grad_scale: 8.0 2023-04-01 00:07:20,416 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=15171.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 00:07:26,939 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.6993, 4.1773, 2.4182, 3.8103, 1.5009, 3.9184, 3.7927, 4.0190], device='cuda:1'), covar=tensor([0.0469, 0.1043, 0.1959, 0.0649, 0.3580, 0.0815, 0.0624, 0.0666], device='cuda:1'), in_proj_covar=tensor([0.0310, 0.0293, 0.0326, 0.0273, 0.0352, 0.0291, 0.0242, 0.0269], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 00:07:44,440 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.74 vs. limit=2.0 2023-04-01 00:08:03,684 INFO [train.py:903] (1/4) Epoch 3, batch 1550, loss[loss=0.2849, simple_loss=0.3447, pruned_loss=0.1125, over 19596.00 frames. ], tot_loss[loss=0.3305, simple_loss=0.3761, pruned_loss=0.1425, over 3833591.87 frames. ], batch size: 52, lr: 2.60e-02, grad_scale: 8.0 2023-04-01 00:08:12,780 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=15213.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 00:08:18,377 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=15217.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 00:08:20,276 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.312e+02 7.825e+02 9.432e+02 1.149e+03 3.008e+03, threshold=1.886e+03, percent-clipped=3.0 2023-04-01 00:08:21,906 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=15220.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 00:08:52,246 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=15245.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 00:09:08,811 INFO [train.py:903] (1/4) Epoch 3, batch 1600, loss[loss=0.3683, simple_loss=0.4177, pruned_loss=0.1595, over 18768.00 frames. ], tot_loss[loss=0.3304, simple_loss=0.3761, pruned_loss=0.1423, over 3835043.00 frames. ], batch size: 74, lr: 2.60e-02, grad_scale: 8.0 2023-04-01 00:09:32,825 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9_rvb from training. Duration: 30.1555625 2023-04-01 00:10:10,174 INFO [train.py:903] (1/4) Epoch 3, batch 1650, loss[loss=0.3935, simple_loss=0.4184, pruned_loss=0.1843, over 18835.00 frames. ], tot_loss[loss=0.3303, simple_loss=0.3759, pruned_loss=0.1424, over 3827688.63 frames. ], batch size: 74, lr: 2.60e-02, grad_scale: 8.0 2023-04-01 00:10:24,984 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.775e+02 8.096e+02 9.310e+02 1.117e+03 2.889e+03, threshold=1.862e+03, percent-clipped=6.0 2023-04-01 00:10:43,139 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6086, 1.6095, 1.4429, 2.1996, 1.5284, 1.9698, 1.8169, 1.1897], device='cuda:1'), covar=tensor([0.1271, 0.1061, 0.0885, 0.0517, 0.1017, 0.0382, 0.1407, 0.1305], device='cuda:1'), in_proj_covar=tensor([0.0430, 0.0404, 0.0411, 0.0544, 0.0480, 0.0314, 0.0505, 0.0409], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 00:11:11,888 INFO [train.py:903] (1/4) Epoch 3, batch 1700, loss[loss=0.3324, simple_loss=0.3772, pruned_loss=0.1438, over 19633.00 frames. ], tot_loss[loss=0.3311, simple_loss=0.3766, pruned_loss=0.1429, over 3822721.07 frames. ], batch size: 61, lr: 2.59e-02, grad_scale: 8.0 2023-04-01 00:11:50,341 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 2023-04-01 00:12:13,187 INFO [train.py:903] (1/4) Epoch 3, batch 1750, loss[loss=0.3039, simple_loss=0.3479, pruned_loss=0.1299, over 19398.00 frames. ], tot_loss[loss=0.3302, simple_loss=0.3758, pruned_loss=0.1423, over 3825187.76 frames. ], batch size: 48, lr: 2.59e-02, grad_scale: 8.0 2023-04-01 00:12:26,539 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.87 vs. limit=5.0 2023-04-01 00:12:30,188 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.268e+02 8.359e+02 1.065e+03 1.276e+03 4.198e+03, threshold=2.129e+03, percent-clipped=6.0 2023-04-01 00:13:17,301 INFO [train.py:903] (1/4) Epoch 3, batch 1800, loss[loss=0.3141, simple_loss=0.3596, pruned_loss=0.1343, over 19861.00 frames. ], tot_loss[loss=0.3301, simple_loss=0.3759, pruned_loss=0.1422, over 3834180.37 frames. ], batch size: 52, lr: 2.58e-02, grad_scale: 8.0 2023-04-01 00:13:37,270 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=15473.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 00:14:07,862 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=15498.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 00:14:14,379 WARNING [train.py:1073] (1/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9_rvb from training. Duration: 25.0944375 2023-04-01 00:14:17,659 INFO [train.py:903] (1/4) Epoch 3, batch 1850, loss[loss=0.3001, simple_loss=0.3538, pruned_loss=0.1232, over 19733.00 frames. ], tot_loss[loss=0.3287, simple_loss=0.3749, pruned_loss=0.1413, over 3828128.11 frames. ], batch size: 51, lr: 2.58e-02, grad_scale: 8.0 2023-04-01 00:14:32,117 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.736e+02 7.448e+02 9.407e+02 1.169e+03 2.273e+03, threshold=1.881e+03, percent-clipped=1.0 2023-04-01 00:14:50,378 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9_rvb from training. Duration: 27.8166875 2023-04-01 00:15:18,051 INFO [train.py:903] (1/4) Epoch 3, batch 1900, loss[loss=0.3159, simple_loss=0.3506, pruned_loss=0.1406, over 19311.00 frames. ], tot_loss[loss=0.3287, simple_loss=0.3749, pruned_loss=0.1412, over 3824482.57 frames. ], batch size: 44, lr: 2.58e-02, grad_scale: 8.0 2023-04-01 00:15:18,208 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=15557.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 00:15:22,371 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 2023-04-01 00:15:36,168 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 2023-04-01 00:15:42,573 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9_rvb from training. Duration: 27.02225 2023-04-01 00:16:04,587 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 2023-04-01 00:16:19,662 INFO [train.py:903] (1/4) Epoch 3, batch 1950, loss[loss=0.3107, simple_loss=0.3456, pruned_loss=0.1378, over 19737.00 frames. ], tot_loss[loss=0.3285, simple_loss=0.375, pruned_loss=0.141, over 3821348.58 frames. ], batch size: 48, lr: 2.57e-02, grad_scale: 8.0 2023-04-01 00:16:20,354 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 2023-04-01 00:16:36,885 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.560e+02 7.789e+02 9.617e+02 1.296e+03 2.448e+03, threshold=1.923e+03, percent-clipped=3.0 2023-04-01 00:17:10,323 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.53 vs. limit=2.0 2023-04-01 00:17:22,855 INFO [train.py:903] (1/4) Epoch 3, batch 2000, loss[loss=0.3615, simple_loss=0.3982, pruned_loss=0.1623, over 18120.00 frames. ], tot_loss[loss=0.3274, simple_loss=0.3743, pruned_loss=0.1403, over 3831245.00 frames. ], batch size: 83, lr: 2.57e-02, grad_scale: 8.0 2023-04-01 00:17:41,420 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=15672.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 00:18:20,124 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_rvb from training. Duration: 26.8349375 2023-04-01 00:18:21,670 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6790, 1.4307, 1.2713, 1.7643, 1.4933, 1.5796, 1.5022, 1.7593], device='cuda:1'), covar=tensor([0.0990, 0.1806, 0.1578, 0.0889, 0.1262, 0.0579, 0.0993, 0.0679], device='cuda:1'), in_proj_covar=tensor([0.0259, 0.0388, 0.0289, 0.0257, 0.0324, 0.0263, 0.0276, 0.0239], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-01 00:18:23,586 INFO [train.py:903] (1/4) Epoch 3, batch 2050, loss[loss=0.3757, simple_loss=0.4182, pruned_loss=0.1666, over 19110.00 frames. ], tot_loss[loss=0.3293, simple_loss=0.3757, pruned_loss=0.1415, over 3828269.98 frames. ], batch size: 69, lr: 2.57e-02, grad_scale: 8.0 2023-04-01 00:18:38,231 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.734e+02 7.410e+02 9.269e+02 1.172e+03 2.915e+03, threshold=1.854e+03, percent-clipped=8.0 2023-04-01 00:18:38,294 WARNING [train.py:1073] (1/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 2023-04-01 00:18:39,623 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 2023-04-01 00:18:57,731 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.9533, 4.0645, 4.5110, 4.4610, 1.5328, 4.1066, 3.7047, 4.0337], device='cuda:1'), covar=tensor([0.0487, 0.0495, 0.0422, 0.0289, 0.3267, 0.0230, 0.0362, 0.0870], device='cuda:1'), in_proj_covar=tensor([0.0364, 0.0335, 0.0460, 0.0346, 0.0474, 0.0238, 0.0305, 0.0439], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1') 2023-04-01 00:18:59,542 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 2023-04-01 00:19:25,007 INFO [train.py:903] (1/4) Epoch 3, batch 2100, loss[loss=0.3193, simple_loss=0.3734, pruned_loss=0.1326, over 19382.00 frames. ], tot_loss[loss=0.3283, simple_loss=0.3747, pruned_loss=0.1409, over 3819442.19 frames. ], batch size: 70, lr: 2.56e-02, grad_scale: 8.0 2023-04-01 00:19:42,757 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0836, 1.9843, 1.8905, 3.2020, 2.3488, 3.6515, 3.1291, 1.8200], device='cuda:1'), covar=tensor([0.1043, 0.0800, 0.0483, 0.0515, 0.0874, 0.0185, 0.0610, 0.0740], device='cuda:1'), in_proj_covar=tensor([0.0440, 0.0409, 0.0418, 0.0563, 0.0493, 0.0328, 0.0520, 0.0420], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 00:19:52,237 WARNING [train.py:1073] (1/4) Exclude cut with ID 453-131332-0000-131866_sp0.9_rvb from training. Duration: 25.3333125 2023-04-01 00:20:13,817 WARNING [train.py:1073] (1/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9_rvb from training. Duration: 26.6166875 2023-04-01 00:20:25,853 INFO [train.py:903] (1/4) Epoch 3, batch 2150, loss[loss=0.3667, simple_loss=0.4134, pruned_loss=0.16, over 18152.00 frames. ], tot_loss[loss=0.3271, simple_loss=0.3739, pruned_loss=0.1401, over 3824485.06 frames. ], batch size: 83, lr: 2.56e-02, grad_scale: 8.0 2023-04-01 00:20:42,350 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.336e+02 7.356e+02 9.022e+02 1.284e+03 2.686e+03, threshold=1.804e+03, percent-clipped=4.0 2023-04-01 00:21:28,850 INFO [train.py:903] (1/4) Epoch 3, batch 2200, loss[loss=0.3022, simple_loss=0.3606, pruned_loss=0.1219, over 19752.00 frames. ], tot_loss[loss=0.3275, simple_loss=0.3739, pruned_loss=0.1406, over 3840141.38 frames. ], batch size: 63, lr: 2.55e-02, grad_scale: 8.0 2023-04-01 00:22:23,678 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=15901.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 00:22:30,393 INFO [train.py:903] (1/4) Epoch 3, batch 2250, loss[loss=0.2643, simple_loss=0.3244, pruned_loss=0.1022, over 19790.00 frames. ], tot_loss[loss=0.327, simple_loss=0.3735, pruned_loss=0.1402, over 3853202.84 frames. ], batch size: 49, lr: 2.55e-02, grad_scale: 8.0 2023-04-01 00:22:33,905 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.1693, 1.0338, 1.3063, 1.2491, 1.9623, 1.0798, 1.5436, 1.9104], device='cuda:1'), covar=tensor([0.0465, 0.2196, 0.1930, 0.1261, 0.0602, 0.1624, 0.0995, 0.0569], device='cuda:1'), in_proj_covar=tensor([0.0239, 0.0300, 0.0295, 0.0278, 0.0280, 0.0322, 0.0266, 0.0276], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 00:22:44,787 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.843e+02 7.012e+02 9.226e+02 1.146e+03 2.721e+03, threshold=1.845e+03, percent-clipped=4.0 2023-04-01 00:22:55,678 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=15928.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 00:23:26,930 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=15953.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 00:23:31,837 INFO [train.py:903] (1/4) Epoch 3, batch 2300, loss[loss=0.4015, simple_loss=0.4273, pruned_loss=0.1878, over 19596.00 frames. ], tot_loss[loss=0.3267, simple_loss=0.3735, pruned_loss=0.1399, over 3852004.60 frames. ], batch size: 57, lr: 2.55e-02, grad_scale: 8.0 2023-04-01 00:23:44,547 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_rvb from training. Duration: 26.205 2023-04-01 00:24:33,426 INFO [train.py:903] (1/4) Epoch 3, batch 2350, loss[loss=0.3722, simple_loss=0.4087, pruned_loss=0.1678, over 19528.00 frames. ], tot_loss[loss=0.3287, simple_loss=0.3748, pruned_loss=0.1413, over 3826141.89 frames. ], batch size: 56, lr: 2.54e-02, grad_scale: 8.0 2023-04-01 00:24:48,804 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.022e+02 7.519e+02 9.138e+02 1.115e+03 3.205e+03, threshold=1.828e+03, percent-clipped=8.0 2023-04-01 00:24:49,967 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=16019.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 00:25:08,174 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6368, 1.5907, 1.7936, 1.5720, 2.7697, 4.1386, 4.2287, 4.5375], device='cuda:1'), covar=tensor([0.1451, 0.2478, 0.2719, 0.1784, 0.0501, 0.0126, 0.0137, 0.0083], device='cuda:1'), in_proj_covar=tensor([0.0277, 0.0281, 0.0330, 0.0276, 0.0196, 0.0106, 0.0197, 0.0115], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0001, 0.0002, 0.0001], device='cuda:1') 2023-04-01 00:25:15,413 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_rvb from training. Duration: 25.775 2023-04-01 00:25:31,127 WARNING [train.py:1073] (1/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9_rvb from training. Duration: 25.45 2023-04-01 00:25:34,316 INFO [train.py:903] (1/4) Epoch 3, batch 2400, loss[loss=0.3832, simple_loss=0.4151, pruned_loss=0.1757, over 19593.00 frames. ], tot_loss[loss=0.3297, simple_loss=0.3756, pruned_loss=0.1419, over 3829538.60 frames. ], batch size: 57, lr: 2.54e-02, grad_scale: 8.0 2023-04-01 00:26:33,067 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=16103.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 00:26:35,417 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.8221, 0.8758, 1.4224, 1.2446, 2.6541, 3.3708, 3.3416, 3.7508], device='cuda:1'), covar=tensor([0.1413, 0.3964, 0.3883, 0.2315, 0.0510, 0.0171, 0.0259, 0.0126], device='cuda:1'), in_proj_covar=tensor([0.0273, 0.0278, 0.0325, 0.0271, 0.0194, 0.0102, 0.0194, 0.0113], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0001, 0.0002, 0.0001], device='cuda:1') 2023-04-01 00:26:37,322 INFO [train.py:903] (1/4) Epoch 3, batch 2450, loss[loss=0.2454, simple_loss=0.3073, pruned_loss=0.0917, over 19746.00 frames. ], tot_loss[loss=0.3282, simple_loss=0.3744, pruned_loss=0.141, over 3841817.54 frames. ], batch size: 46, lr: 2.54e-02, grad_scale: 8.0 2023-04-01 00:26:51,593 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.658e+02 8.374e+02 9.822e+02 1.305e+03 3.634e+03, threshold=1.964e+03, percent-clipped=9.0 2023-04-01 00:26:58,746 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.6405, 4.1530, 2.3386, 3.7266, 1.1975, 4.0474, 3.7757, 3.9777], device='cuda:1'), covar=tensor([0.0508, 0.1166, 0.1793, 0.0582, 0.3375, 0.0644, 0.0549, 0.0605], device='cuda:1'), in_proj_covar=tensor([0.0302, 0.0291, 0.0319, 0.0258, 0.0335, 0.0283, 0.0237, 0.0266], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 00:27:38,194 INFO [train.py:903] (1/4) Epoch 3, batch 2500, loss[loss=0.4078, simple_loss=0.4351, pruned_loss=0.1902, over 19387.00 frames. ], tot_loss[loss=0.3274, simple_loss=0.3742, pruned_loss=0.1404, over 3845912.33 frames. ], batch size: 70, lr: 2.53e-02, grad_scale: 8.0 2023-04-01 00:27:50,910 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=16167.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 00:28:07,405 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.40 vs. limit=5.0 2023-04-01 00:28:40,050 INFO [train.py:903] (1/4) Epoch 3, batch 2550, loss[loss=0.3163, simple_loss=0.3743, pruned_loss=0.1291, over 19658.00 frames. ], tot_loss[loss=0.3277, simple_loss=0.3739, pruned_loss=0.1408, over 3848091.29 frames. ], batch size: 53, lr: 2.53e-02, grad_scale: 8.0 2023-04-01 00:28:56,248 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.521e+02 7.641e+02 9.209e+02 1.283e+03 2.881e+03, threshold=1.842e+03, percent-clipped=1.0 2023-04-01 00:29:06,035 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2187, 1.1855, 1.8107, 1.3537, 2.3544, 2.3201, 2.5643, 0.8384], device='cuda:1'), covar=tensor([0.1376, 0.2231, 0.1150, 0.1245, 0.0877, 0.0978, 0.1044, 0.2146], device='cuda:1'), in_proj_covar=tensor([0.0405, 0.0439, 0.0407, 0.0385, 0.0471, 0.0394, 0.0564, 0.0405], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 00:29:09,277 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.6578, 3.9261, 4.2570, 4.1272, 1.4082, 3.8081, 3.5090, 3.7527], device='cuda:1'), covar=tensor([0.0682, 0.0588, 0.0475, 0.0361, 0.3571, 0.0286, 0.0444, 0.1060], device='cuda:1'), in_proj_covar=tensor([0.0362, 0.0335, 0.0454, 0.0344, 0.0470, 0.0241, 0.0297, 0.0436], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1') 2023-04-01 00:29:28,549 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=16245.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 00:29:36,513 WARNING [train.py:1073] (1/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 2023-04-01 00:29:44,017 INFO [train.py:903] (1/4) Epoch 3, batch 2600, loss[loss=0.3781, simple_loss=0.4151, pruned_loss=0.1706, over 19589.00 frames. ], tot_loss[loss=0.3279, simple_loss=0.3738, pruned_loss=0.141, over 3837532.41 frames. ], batch size: 61, lr: 2.53e-02, grad_scale: 8.0 2023-04-01 00:30:07,653 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 2023-04-01 00:30:46,335 INFO [train.py:903] (1/4) Epoch 3, batch 2650, loss[loss=0.3297, simple_loss=0.3801, pruned_loss=0.1396, over 19784.00 frames. ], tot_loss[loss=0.3249, simple_loss=0.3719, pruned_loss=0.1389, over 3848407.34 frames. ], batch size: 56, lr: 2.52e-02, grad_scale: 8.0 2023-04-01 00:31:00,229 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.667e+02 7.570e+02 8.863e+02 1.283e+03 4.568e+03, threshold=1.773e+03, percent-clipped=9.0 2023-04-01 00:31:06,931 WARNING [train.py:1073] (1/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9_rvb from training. Duration: 27.25 2023-04-01 00:31:29,695 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 2023-04-01 00:31:44,569 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 2023-04-01 00:31:47,142 INFO [train.py:903] (1/4) Epoch 3, batch 2700, loss[loss=0.3083, simple_loss=0.3436, pruned_loss=0.1365, over 19747.00 frames. ], tot_loss[loss=0.3253, simple_loss=0.3722, pruned_loss=0.1392, over 3832131.03 frames. ], batch size: 46, lr: 2.52e-02, grad_scale: 8.0 2023-04-01 00:31:51,946 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=16360.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 00:31:54,875 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=16363.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 00:32:26,182 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=16389.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 00:32:47,343 INFO [train.py:903] (1/4) Epoch 3, batch 2750, loss[loss=0.3091, simple_loss=0.355, pruned_loss=0.1316, over 19580.00 frames. ], tot_loss[loss=0.3265, simple_loss=0.3729, pruned_loss=0.14, over 3827579.19 frames. ], batch size: 52, lr: 2.52e-02, grad_scale: 8.0 2023-04-01 00:33:01,693 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.986e+02 7.672e+02 9.838e+02 1.209e+03 2.463e+03, threshold=1.968e+03, percent-clipped=5.0 2023-04-01 00:33:07,380 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.3024, 3.6774, 3.9698, 4.0565, 1.5218, 3.6871, 3.4288, 3.2836], device='cuda:1'), covar=tensor([0.1127, 0.0801, 0.0727, 0.0685, 0.4097, 0.0534, 0.0563, 0.1490], device='cuda:1'), in_proj_covar=tensor([0.0365, 0.0332, 0.0463, 0.0346, 0.0476, 0.0242, 0.0299, 0.0446], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1') 2023-04-01 00:33:07,472 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6593, 1.1726, 1.3892, 1.1204, 2.7075, 3.3290, 3.1124, 3.5715], device='cuda:1'), covar=tensor([0.1131, 0.2611, 0.2681, 0.1869, 0.0407, 0.0106, 0.0209, 0.0081], device='cuda:1'), in_proj_covar=tensor([0.0269, 0.0277, 0.0324, 0.0269, 0.0195, 0.0104, 0.0194, 0.0109], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0001, 0.0002, 0.0001], device='cuda:1') 2023-04-01 00:33:35,972 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=16447.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 00:33:46,965 INFO [train.py:903] (1/4) Epoch 3, batch 2800, loss[loss=0.3267, simple_loss=0.3646, pruned_loss=0.1444, over 19420.00 frames. ], tot_loss[loss=0.3263, simple_loss=0.3731, pruned_loss=0.1397, over 3822615.84 frames. ], batch size: 48, lr: 2.51e-02, grad_scale: 8.0 2023-04-01 00:34:13,547 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=16478.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 00:34:48,737 INFO [train.py:903] (1/4) Epoch 3, batch 2850, loss[loss=0.3162, simple_loss=0.3745, pruned_loss=0.129, over 19689.00 frames. ], tot_loss[loss=0.3274, simple_loss=0.3744, pruned_loss=0.1402, over 3827921.04 frames. ], batch size: 59, lr: 2.51e-02, grad_scale: 8.0 2023-04-01 00:34:54,288 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=16511.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 00:35:03,282 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.663e+02 7.862e+02 1.093e+03 1.433e+03 3.382e+03, threshold=2.185e+03, percent-clipped=3.0 2023-04-01 00:35:47,689 WARNING [train.py:1073] (1/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 2023-04-01 00:35:49,902 INFO [train.py:903] (1/4) Epoch 3, batch 2900, loss[loss=0.4034, simple_loss=0.4312, pruned_loss=0.1877, over 19107.00 frames. ], tot_loss[loss=0.3298, simple_loss=0.3758, pruned_loss=0.1419, over 3826501.24 frames. ], batch size: 69, lr: 2.51e-02, grad_scale: 8.0 2023-04-01 00:35:57,104 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=16562.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 00:36:44,347 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=2.02 vs. limit=2.0 2023-04-01 00:36:51,759 INFO [train.py:903] (1/4) Epoch 3, batch 2950, loss[loss=0.2873, simple_loss=0.3412, pruned_loss=0.1167, over 19385.00 frames. ], tot_loss[loss=0.3275, simple_loss=0.3741, pruned_loss=0.1405, over 3829283.69 frames. ], batch size: 48, lr: 2.50e-02, grad_scale: 16.0 2023-04-01 00:37:02,809 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=16616.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 00:37:07,332 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.542e+02 7.271e+02 9.405e+02 1.170e+03 2.853e+03, threshold=1.881e+03, percent-clipped=4.0 2023-04-01 00:37:15,841 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=16626.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 00:37:33,959 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=16641.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 00:37:52,195 INFO [train.py:903] (1/4) Epoch 3, batch 3000, loss[loss=0.3392, simple_loss=0.3805, pruned_loss=0.1489, over 19771.00 frames. ], tot_loss[loss=0.3271, simple_loss=0.3739, pruned_loss=0.1402, over 3839817.43 frames. ], batch size: 54, lr: 2.50e-02, grad_scale: 16.0 2023-04-01 00:37:52,195 INFO [train.py:928] (1/4) Computing validation loss 2023-04-01 00:38:05,259 INFO [train.py:937] (1/4) Epoch 3, validation: loss=0.231, simple_loss=0.3246, pruned_loss=0.06867, over 944034.00 frames. 2023-04-01 00:38:05,260 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18649MB 2023-04-01 00:38:08,696 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_rvb from training. Duration: 29.735 2023-04-01 00:39:07,682 INFO [train.py:903] (1/4) Epoch 3, batch 3050, loss[loss=0.2986, simple_loss=0.3408, pruned_loss=0.1282, over 19783.00 frames. ], tot_loss[loss=0.3263, simple_loss=0.3733, pruned_loss=0.1396, over 3831696.85 frames. ], batch size: 48, lr: 2.49e-02, grad_scale: 16.0 2023-04-01 00:39:22,588 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.950e+02 7.860e+02 1.014e+03 1.267e+03 1.851e+03, threshold=2.027e+03, percent-clipped=0.0 2023-04-01 00:39:38,693 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=16733.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 00:39:40,011 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=16734.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 00:40:08,118 INFO [train.py:903] (1/4) Epoch 3, batch 3100, loss[loss=0.3024, simple_loss=0.355, pruned_loss=0.1249, over 19490.00 frames. ], tot_loss[loss=0.3271, simple_loss=0.3739, pruned_loss=0.1401, over 3816871.23 frames. ], batch size: 49, lr: 2.49e-02, grad_scale: 16.0 2023-04-01 00:40:11,518 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=16759.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 00:41:08,250 INFO [train.py:903] (1/4) Epoch 3, batch 3150, loss[loss=0.2687, simple_loss=0.3209, pruned_loss=0.1083, over 18631.00 frames. ], tot_loss[loss=0.3276, simple_loss=0.3743, pruned_loss=0.1405, over 3819184.53 frames. ], batch size: 41, lr: 2.49e-02, grad_scale: 16.0 2023-04-01 00:41:10,839 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2341, 1.2254, 1.6091, 1.1068, 2.6100, 3.3164, 3.2958, 3.5833], device='cuda:1'), covar=tensor([0.1385, 0.2503, 0.2478, 0.1910, 0.0436, 0.0106, 0.0201, 0.0082], device='cuda:1'), in_proj_covar=tensor([0.0275, 0.0279, 0.0329, 0.0270, 0.0195, 0.0106, 0.0201, 0.0112], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0001, 0.0002, 0.0001], device='cuda:1') 2023-04-01 00:41:21,861 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=16818.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 00:41:22,518 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.096e+02 7.784e+02 9.820e+02 1.270e+03 2.923e+03, threshold=1.964e+03, percent-clipped=2.0 2023-04-01 00:41:31,804 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 2023-04-01 00:41:51,889 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=16843.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 00:41:57,390 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=16848.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 00:42:07,316 INFO [train.py:903] (1/4) Epoch 3, batch 3200, loss[loss=0.3348, simple_loss=0.3826, pruned_loss=0.1435, over 18839.00 frames. ], tot_loss[loss=0.3295, simple_loss=0.3758, pruned_loss=0.1416, over 3822176.05 frames. ], batch size: 74, lr: 2.48e-02, grad_scale: 8.0 2023-04-01 00:42:39,246 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=16882.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 00:42:43,594 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=16886.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 00:43:08,573 INFO [train.py:903] (1/4) Epoch 3, batch 3250, loss[loss=0.3933, simple_loss=0.4288, pruned_loss=0.1789, over 19326.00 frames. ], tot_loss[loss=0.3287, simple_loss=0.3753, pruned_loss=0.1411, over 3830570.41 frames. ], batch size: 66, lr: 2.48e-02, grad_scale: 8.0 2023-04-01 00:43:08,971 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=16907.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 00:43:24,381 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 5.206e+02 8.507e+02 1.021e+03 1.288e+03 2.328e+03, threshold=2.042e+03, percent-clipped=1.0 2023-04-01 00:43:29,328 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=16924.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 00:44:06,547 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9867, 1.8577, 1.6804, 2.8155, 1.8435, 2.9180, 2.4595, 1.9097], device='cuda:1'), covar=tensor([0.0971, 0.0792, 0.0492, 0.0390, 0.0860, 0.0227, 0.0774, 0.0643], device='cuda:1'), in_proj_covar=tensor([0.0450, 0.0429, 0.0428, 0.0567, 0.0506, 0.0340, 0.0527, 0.0428], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 00:44:09,581 INFO [train.py:903] (1/4) Epoch 3, batch 3300, loss[loss=0.3274, simple_loss=0.3816, pruned_loss=0.1366, over 18687.00 frames. ], tot_loss[loss=0.3274, simple_loss=0.3743, pruned_loss=0.1403, over 3843963.95 frames. ], batch size: 74, lr: 2.48e-02, grad_scale: 8.0 2023-04-01 00:44:16,139 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 2023-04-01 00:45:10,104 INFO [train.py:903] (1/4) Epoch 3, batch 3350, loss[loss=0.2926, simple_loss=0.3563, pruned_loss=0.1145, over 19680.00 frames. ], tot_loss[loss=0.3259, simple_loss=0.373, pruned_loss=0.1394, over 3833387.96 frames. ], batch size: 58, lr: 2.47e-02, grad_scale: 8.0 2023-04-01 00:45:24,558 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.093e+02 7.847e+02 9.419e+02 1.175e+03 3.710e+03, threshold=1.884e+03, percent-clipped=5.0 2023-04-01 00:45:31,276 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.7691, 4.1065, 4.4380, 4.3308, 1.6384, 3.8687, 3.5832, 3.8688], device='cuda:1'), covar=tensor([0.0599, 0.0555, 0.0498, 0.0381, 0.3461, 0.0346, 0.0428, 0.1095], device='cuda:1'), in_proj_covar=tensor([0.0384, 0.0345, 0.0485, 0.0368, 0.0494, 0.0252, 0.0314, 0.0469], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1') 2023-04-01 00:45:42,946 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.0479, 1.0200, 1.3742, 1.1764, 1.7743, 1.6978, 1.8817, 0.5465], device='cuda:1'), covar=tensor([0.1434, 0.2305, 0.1185, 0.1265, 0.0936, 0.1192, 0.0978, 0.2112], device='cuda:1'), in_proj_covar=tensor([0.0407, 0.0447, 0.0409, 0.0386, 0.0492, 0.0391, 0.0570, 0.0409], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 00:46:10,036 INFO [train.py:903] (1/4) Epoch 3, batch 3400, loss[loss=0.2793, simple_loss=0.3493, pruned_loss=0.1047, over 19511.00 frames. ], tot_loss[loss=0.3269, simple_loss=0.3739, pruned_loss=0.14, over 3840821.81 frames. ], batch size: 56, lr: 2.47e-02, grad_scale: 8.0 2023-04-01 00:47:08,469 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=17104.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 00:47:12,305 INFO [train.py:903] (1/4) Epoch 3, batch 3450, loss[loss=0.3142, simple_loss=0.3547, pruned_loss=0.1369, over 19392.00 frames. ], tot_loss[loss=0.3254, simple_loss=0.3731, pruned_loss=0.1388, over 3847346.36 frames. ], batch size: 48, lr: 2.47e-02, grad_scale: 8.0 2023-04-01 00:47:14,347 WARNING [train.py:1073] (1/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9_rvb from training. Duration: 25.2444375 2023-04-01 00:47:28,181 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.582e+02 8.544e+02 1.014e+03 1.278e+03 1.988e+03, threshold=2.028e+03, percent-clipped=3.0 2023-04-01 00:47:39,736 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=17129.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 00:48:12,438 INFO [train.py:903] (1/4) Epoch 3, batch 3500, loss[loss=0.3393, simple_loss=0.4067, pruned_loss=0.136, over 19625.00 frames. ], tot_loss[loss=0.3282, simple_loss=0.3753, pruned_loss=0.1406, over 3836259.29 frames. ], batch size: 57, lr: 2.46e-02, grad_scale: 8.0 2023-04-01 00:48:18,656 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.53 vs. limit=2.0 2023-04-01 00:49:12,365 INFO [train.py:903] (1/4) Epoch 3, batch 3550, loss[loss=0.285, simple_loss=0.3381, pruned_loss=0.1159, over 19587.00 frames. ], tot_loss[loss=0.3273, simple_loss=0.3745, pruned_loss=0.1401, over 3831472.99 frames. ], batch size: 52, lr: 2.46e-02, grad_scale: 8.0 2023-04-01 00:49:12,640 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=17207.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 00:49:26,839 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.377e+02 8.429e+02 1.068e+03 1.302e+03 2.755e+03, threshold=2.137e+03, percent-clipped=4.0 2023-04-01 00:49:40,878 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=17230.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 00:50:11,406 INFO [train.py:903] (1/4) Epoch 3, batch 3600, loss[loss=0.2731, simple_loss=0.3265, pruned_loss=0.1099, over 19781.00 frames. ], tot_loss[loss=0.3258, simple_loss=0.3737, pruned_loss=0.1389, over 3831276.81 frames. ], batch size: 47, lr: 2.46e-02, grad_scale: 8.0 2023-04-01 00:50:24,837 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=17268.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 00:51:11,855 INFO [train.py:903] (1/4) Epoch 3, batch 3650, loss[loss=0.3515, simple_loss=0.3899, pruned_loss=0.1565, over 13099.00 frames. ], tot_loss[loss=0.3269, simple_loss=0.3746, pruned_loss=0.1396, over 3817887.14 frames. ], batch size: 136, lr: 2.45e-02, grad_scale: 8.0 2023-04-01 00:51:27,523 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 5.083e+02 7.894e+02 9.345e+02 1.119e+03 1.949e+03, threshold=1.869e+03, percent-clipped=0.0 2023-04-01 00:51:32,199 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.1143, 1.6675, 2.0343, 2.7270, 2.0395, 2.4097, 2.0275, 2.8234], device='cuda:1'), covar=tensor([0.0796, 0.1647, 0.1093, 0.0688, 0.1082, 0.0412, 0.0891, 0.0490], device='cuda:1'), in_proj_covar=tensor([0.0252, 0.0371, 0.0281, 0.0248, 0.0313, 0.0262, 0.0274, 0.0238], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 00:51:33,450 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5556, 2.2424, 1.6259, 1.8505, 2.1080, 1.1345, 1.1876, 1.6716], device='cuda:1'), covar=tensor([0.0824, 0.0472, 0.0992, 0.0496, 0.0452, 0.1148, 0.0848, 0.0511], device='cuda:1'), in_proj_covar=tensor([0.0285, 0.0233, 0.0316, 0.0265, 0.0229, 0.0310, 0.0280, 0.0261], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 00:51:57,309 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=17345.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 00:52:12,709 INFO [train.py:903] (1/4) Epoch 3, batch 3700, loss[loss=0.3643, simple_loss=0.401, pruned_loss=0.1638, over 19768.00 frames. ], tot_loss[loss=0.3258, simple_loss=0.3741, pruned_loss=0.1387, over 3833890.19 frames. ], batch size: 63, lr: 2.45e-02, grad_scale: 8.0 2023-04-01 00:52:39,382 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=17380.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 00:52:42,829 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=17383.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 00:53:05,565 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=17401.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 00:53:12,837 INFO [train.py:903] (1/4) Epoch 3, batch 3750, loss[loss=0.3484, simple_loss=0.3944, pruned_loss=0.1512, over 19619.00 frames. ], tot_loss[loss=0.3256, simple_loss=0.3737, pruned_loss=0.1387, over 3836510.10 frames. ], batch size: 57, lr: 2.45e-02, grad_scale: 8.0 2023-04-01 00:53:27,566 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.653e+02 9.192e+02 1.060e+03 1.489e+03 3.397e+03, threshold=2.120e+03, percent-clipped=7.0 2023-04-01 00:54:12,635 INFO [train.py:903] (1/4) Epoch 3, batch 3800, loss[loss=0.3216, simple_loss=0.3723, pruned_loss=0.1355, over 19768.00 frames. ], tot_loss[loss=0.3254, simple_loss=0.3737, pruned_loss=0.1385, over 3829525.14 frames. ], batch size: 54, lr: 2.44e-02, grad_scale: 8.0 2023-04-01 00:54:45,638 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9_rvb from training. Duration: 29.1166875 2023-04-01 00:55:11,732 INFO [train.py:903] (1/4) Epoch 3, batch 3850, loss[loss=0.3694, simple_loss=0.4068, pruned_loss=0.166, over 19532.00 frames. ], tot_loss[loss=0.3266, simple_loss=0.3744, pruned_loss=0.1394, over 3824411.09 frames. ], batch size: 54, lr: 2.44e-02, grad_scale: 8.0 2023-04-01 00:55:28,331 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 5.037e+02 7.944e+02 9.720e+02 1.209e+03 3.103e+03, threshold=1.944e+03, percent-clipped=2.0 2023-04-01 00:56:04,068 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=17551.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 00:56:12,746 INFO [train.py:903] (1/4) Epoch 3, batch 3900, loss[loss=0.2707, simple_loss=0.3213, pruned_loss=0.11, over 19713.00 frames. ], tot_loss[loss=0.3265, simple_loss=0.3742, pruned_loss=0.1394, over 3831613.10 frames. ], batch size: 46, lr: 2.44e-02, grad_scale: 8.0 2023-04-01 00:57:05,249 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=17601.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 00:57:11,566 INFO [train.py:903] (1/4) Epoch 3, batch 3950, loss[loss=0.3232, simple_loss=0.3761, pruned_loss=0.1352, over 19659.00 frames. ], tot_loss[loss=0.3265, simple_loss=0.3746, pruned_loss=0.1393, over 3829105.26 frames. ], batch size: 60, lr: 2.44e-02, grad_scale: 8.0 2023-04-01 00:57:14,031 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.0634, 1.1387, 1.7545, 1.3157, 2.4492, 2.2606, 2.6499, 0.9552], device='cuda:1'), covar=tensor([0.1453, 0.2244, 0.1214, 0.1262, 0.0811, 0.0989, 0.0910, 0.2075], device='cuda:1'), in_proj_covar=tensor([0.0415, 0.0456, 0.0412, 0.0388, 0.0490, 0.0402, 0.0575, 0.0411], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 00:57:18,154 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 2023-04-01 00:57:27,262 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.216e+02 7.211e+02 9.089e+02 1.148e+03 2.193e+03, threshold=1.818e+03, percent-clipped=2.0 2023-04-01 00:57:27,665 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4433, 2.4575, 1.5637, 1.6902, 1.9921, 1.2049, 0.8356, 1.8351], device='cuda:1'), covar=tensor([0.0925, 0.0368, 0.1112, 0.0501, 0.0502, 0.1242, 0.1009, 0.0396], device='cuda:1'), in_proj_covar=tensor([0.0285, 0.0235, 0.0314, 0.0262, 0.0224, 0.0312, 0.0279, 0.0263], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 00:57:34,436 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=17626.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 00:57:51,678 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=17639.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 00:57:59,696 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.2413, 2.1830, 1.9403, 3.6477, 2.3014, 4.0800, 3.3572, 1.8565], device='cuda:1'), covar=tensor([0.1213, 0.0921, 0.0514, 0.0488, 0.1127, 0.0189, 0.0797, 0.0829], device='cuda:1'), in_proj_covar=tensor([0.0477, 0.0444, 0.0447, 0.0587, 0.0524, 0.0358, 0.0544, 0.0443], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 00:58:12,125 INFO [train.py:903] (1/4) Epoch 3, batch 4000, loss[loss=0.3701, simple_loss=0.4109, pruned_loss=0.1647, over 19582.00 frames. ], tot_loss[loss=0.326, simple_loss=0.3741, pruned_loss=0.1389, over 3816190.23 frames. ], batch size: 61, lr: 2.43e-02, grad_scale: 8.0 2023-04-01 00:58:20,399 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=17664.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 00:58:22,658 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=17666.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 00:58:32,269 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0 2023-04-01 00:58:59,486 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1_rvb from training. Duration: 0.7545625 2023-04-01 00:59:11,470 INFO [train.py:903] (1/4) Epoch 3, batch 4050, loss[loss=0.3851, simple_loss=0.4092, pruned_loss=0.1805, over 13376.00 frames. ], tot_loss[loss=0.3254, simple_loss=0.3735, pruned_loss=0.1386, over 3797479.79 frames. ], batch size: 136, lr: 2.43e-02, grad_scale: 8.0 2023-04-01 00:59:14,520 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.86 vs. limit=5.0 2023-04-01 00:59:25,529 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=17718.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 00:59:28,911 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 5.065e+02 7.398e+02 9.459e+02 1.250e+03 4.446e+03, threshold=1.892e+03, percent-clipped=10.0 2023-04-01 00:59:32,671 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=17724.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 00:59:57,742 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=17745.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 01:00:01,022 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=17748.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 01:00:12,217 INFO [train.py:903] (1/4) Epoch 3, batch 4100, loss[loss=0.2539, simple_loss=0.306, pruned_loss=0.1009, over 19734.00 frames. ], tot_loss[loss=0.3264, simple_loss=0.3735, pruned_loss=0.1397, over 3791766.87 frames. ], batch size: 46, lr: 2.43e-02, grad_scale: 4.0 2023-04-01 01:00:38,305 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.8217, 1.6730, 1.6706, 2.4820, 1.8138, 2.7119, 2.7996, 2.6248], device='cuda:1'), covar=tensor([0.0622, 0.1159, 0.1319, 0.1297, 0.1366, 0.0854, 0.1114, 0.0651], device='cuda:1'), in_proj_covar=tensor([0.0258, 0.0278, 0.0270, 0.0311, 0.0314, 0.0260, 0.0282, 0.0243], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:1') 2023-04-01 01:00:48,457 WARNING [train.py:1073] (1/4) Exclude cut with ID 497-129325-0061-9566_sp1.1_rvb from training. Duration: 0.97725 2023-04-01 01:01:11,675 INFO [train.py:903] (1/4) Epoch 3, batch 4150, loss[loss=0.3126, simple_loss=0.3686, pruned_loss=0.1283, over 19668.00 frames. ], tot_loss[loss=0.3235, simple_loss=0.3712, pruned_loss=0.1379, over 3800539.91 frames. ], batch size: 60, lr: 2.42e-02, grad_scale: 4.0 2023-04-01 01:01:28,529 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.446e+02 7.784e+02 9.706e+02 1.186e+03 3.618e+03, threshold=1.941e+03, percent-clipped=3.0 2023-04-01 01:01:45,058 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 2023-04-01 01:01:49,964 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=17839.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 01:02:10,744 INFO [train.py:903] (1/4) Epoch 3, batch 4200, loss[loss=0.3174, simple_loss=0.3679, pruned_loss=0.1334, over 19680.00 frames. ], tot_loss[loss=0.3233, simple_loss=0.3715, pruned_loss=0.1376, over 3800614.57 frames. ], batch size: 53, lr: 2.42e-02, grad_scale: 4.0 2023-04-01 01:02:14,267 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=17860.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 01:02:14,912 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 2023-04-01 01:03:09,267 INFO [train.py:903] (1/4) Epoch 3, batch 4250, loss[loss=0.3379, simple_loss=0.3762, pruned_loss=0.1498, over 19581.00 frames. ], tot_loss[loss=0.325, simple_loss=0.3728, pruned_loss=0.1387, over 3816675.12 frames. ], batch size: 52, lr: 2.42e-02, grad_scale: 4.0 2023-04-01 01:03:26,775 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 5.782e+02 8.334e+02 9.808e+02 1.259e+03 2.577e+03, threshold=1.962e+03, percent-clipped=5.0 2023-04-01 01:03:26,829 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9_rvb from training. Duration: 29.816625 2023-04-01 01:03:28,251 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=17922.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 01:03:37,987 WARNING [train.py:1073] (1/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 2023-04-01 01:03:57,614 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=17947.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 01:04:08,379 INFO [train.py:903] (1/4) Epoch 3, batch 4300, loss[loss=0.3007, simple_loss=0.3504, pruned_loss=0.1255, over 19586.00 frames. ], tot_loss[loss=0.3254, simple_loss=0.3728, pruned_loss=0.139, over 3814024.67 frames. ], batch size: 52, lr: 2.41e-02, grad_scale: 4.0 2023-04-01 01:04:35,031 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=17978.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 01:04:57,242 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3297, 2.2509, 1.5496, 1.5574, 2.0294, 1.0293, 0.9437, 1.5577], device='cuda:1'), covar=tensor([0.0957, 0.0541, 0.1156, 0.0741, 0.0660, 0.1429, 0.0917, 0.0606], device='cuda:1'), in_proj_covar=tensor([0.0284, 0.0239, 0.0320, 0.0259, 0.0218, 0.0315, 0.0277, 0.0260], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 01:05:06,032 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 2023-04-01 01:05:10,744 INFO [train.py:903] (1/4) Epoch 3, batch 4350, loss[loss=0.2652, simple_loss=0.338, pruned_loss=0.09626, over 19851.00 frames. ], tot_loss[loss=0.3232, simple_loss=0.371, pruned_loss=0.1377, over 3820523.86 frames. ], batch size: 52, lr: 2.41e-02, grad_scale: 4.0 2023-04-01 01:05:27,029 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.929e+02 7.530e+02 9.485e+02 1.282e+03 2.824e+03, threshold=1.897e+03, percent-clipped=4.0 2023-04-01 01:06:09,595 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.1919, 2.1279, 1.9575, 2.8803, 2.0638, 2.9061, 2.7498, 1.9845], device='cuda:1'), covar=tensor([0.0986, 0.0756, 0.0478, 0.0501, 0.0884, 0.0248, 0.0768, 0.0719], device='cuda:1'), in_proj_covar=tensor([0.0473, 0.0445, 0.0445, 0.0593, 0.0519, 0.0361, 0.0542, 0.0446], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 01:06:11,119 INFO [train.py:903] (1/4) Epoch 3, batch 4400, loss[loss=0.3101, simple_loss=0.3533, pruned_loss=0.1335, over 19715.00 frames. ], tot_loss[loss=0.3226, simple_loss=0.3702, pruned_loss=0.1374, over 3833128.88 frames. ], batch size: 51, lr: 2.41e-02, grad_scale: 8.0 2023-04-01 01:06:17,186 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=18062.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 01:06:32,864 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_rvb from training. Duration: 25.285 2023-04-01 01:06:43,601 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 2023-04-01 01:06:53,791 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=18092.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 01:06:57,450 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=18095.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 01:07:10,934 INFO [train.py:903] (1/4) Epoch 3, batch 4450, loss[loss=0.3226, simple_loss=0.3853, pruned_loss=0.13, over 19308.00 frames. ], tot_loss[loss=0.3211, simple_loss=0.3696, pruned_loss=0.1363, over 3823308.11 frames. ], batch size: 66, lr: 2.40e-02, grad_scale: 8.0 2023-04-01 01:07:21,267 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=18116.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 01:07:26,614 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=18120.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 01:07:28,236 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.891e+02 7.625e+02 9.248e+02 1.171e+03 2.408e+03, threshold=1.850e+03, percent-clipped=4.0 2023-04-01 01:07:53,374 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=18141.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 01:08:11,163 INFO [train.py:903] (1/4) Epoch 3, batch 4500, loss[loss=0.3333, simple_loss=0.3852, pruned_loss=0.1407, over 19668.00 frames. ], tot_loss[loss=0.3195, simple_loss=0.3687, pruned_loss=0.1351, over 3826236.28 frames. ], batch size: 58, lr: 2.40e-02, grad_scale: 4.0 2023-04-01 01:08:37,483 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=18177.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 01:08:54,078 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=18192.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 01:08:55,158 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=18193.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 01:09:11,759 INFO [train.py:903] (1/4) Epoch 3, batch 4550, loss[loss=0.3203, simple_loss=0.3738, pruned_loss=0.1334, over 19691.00 frames. ], tot_loss[loss=0.3177, simple_loss=0.3667, pruned_loss=0.1343, over 3835990.33 frames. ], batch size: 53, lr: 2.40e-02, grad_scale: 4.0 2023-04-01 01:09:12,129 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=18207.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 01:09:19,321 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9_rvb from training. Duration: 28.72225 2023-04-01 01:09:29,444 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.087e+02 7.616e+02 9.596e+02 1.201e+03 2.125e+03, threshold=1.919e+03, percent-clipped=4.0 2023-04-01 01:09:42,206 WARNING [train.py:1073] (1/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 2023-04-01 01:10:11,981 INFO [train.py:903] (1/4) Epoch 3, batch 4600, loss[loss=0.348, simple_loss=0.3909, pruned_loss=0.1526, over 19511.00 frames. ], tot_loss[loss=0.3188, simple_loss=0.3675, pruned_loss=0.1351, over 3826693.93 frames. ], batch size: 64, lr: 2.39e-02, grad_scale: 4.0 2023-04-01 01:11:08,586 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0068, 2.0682, 1.4576, 1.5093, 1.3896, 1.6796, 0.1966, 0.8098], device='cuda:1'), covar=tensor([0.0233, 0.0197, 0.0154, 0.0180, 0.0453, 0.0211, 0.0437, 0.0373], device='cuda:1'), in_proj_covar=tensor([0.0252, 0.0251, 0.0245, 0.0266, 0.0321, 0.0257, 0.0253, 0.0260], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 01:11:11,633 INFO [train.py:903] (1/4) Epoch 3, batch 4650, loss[loss=0.3597, simple_loss=0.4048, pruned_loss=0.1573, over 19698.00 frames. ], tot_loss[loss=0.3198, simple_loss=0.3684, pruned_loss=0.1356, over 3822338.29 frames. ], batch size: 59, lr: 2.39e-02, grad_scale: 4.0 2023-04-01 01:11:27,784 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_rvb from training. Duration: 0.92 2023-04-01 01:11:28,862 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.638e+02 7.851e+02 9.796e+02 1.308e+03 3.825e+03, threshold=1.959e+03, percent-clipped=6.0 2023-04-01 01:11:29,087 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=18322.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 01:11:38,744 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_rvb from training. Duration: 0.83 2023-04-01 01:11:54,698 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.2566, 2.1698, 1.8646, 3.3360, 2.3436, 3.6158, 3.0624, 2.0022], device='cuda:1'), covar=tensor([0.1164, 0.0881, 0.0547, 0.0528, 0.1075, 0.0211, 0.0774, 0.0772], device='cuda:1'), in_proj_covar=tensor([0.0476, 0.0451, 0.0449, 0.0593, 0.0524, 0.0356, 0.0549, 0.0446], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 01:12:10,876 INFO [train.py:903] (1/4) Epoch 3, batch 4700, loss[loss=0.3266, simple_loss=0.3538, pruned_loss=0.1498, over 19394.00 frames. ], tot_loss[loss=0.3218, simple_loss=0.3698, pruned_loss=0.1369, over 3805498.43 frames. ], batch size: 48, lr: 2.39e-02, grad_scale: 4.0 2023-04-01 01:12:33,352 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9_rvb from training. Duration: 0.92225 2023-04-01 01:13:07,535 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0013, 1.9951, 1.7359, 2.9463, 1.9666, 2.8720, 2.4406, 1.6512], device='cuda:1'), covar=tensor([0.1127, 0.0902, 0.0542, 0.0503, 0.1026, 0.0270, 0.0918, 0.0857], device='cuda:1'), in_proj_covar=tensor([0.0480, 0.0450, 0.0449, 0.0599, 0.0525, 0.0361, 0.0552, 0.0449], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 01:13:13,550 INFO [train.py:903] (1/4) Epoch 3, batch 4750, loss[loss=0.2979, simple_loss=0.3392, pruned_loss=0.1283, over 19307.00 frames. ], tot_loss[loss=0.3213, simple_loss=0.3692, pruned_loss=0.1368, over 3817463.61 frames. ], batch size: 44, lr: 2.39e-02, grad_scale: 4.0 2023-04-01 01:13:31,290 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 5.455e+02 7.810e+02 9.309e+02 1.222e+03 2.382e+03, threshold=1.862e+03, percent-clipped=4.0 2023-04-01 01:13:44,338 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=18433.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 01:13:48,805 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=18437.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 01:14:15,102 INFO [train.py:903] (1/4) Epoch 3, batch 4800, loss[loss=0.2551, simple_loss=0.3279, pruned_loss=0.09111, over 19661.00 frames. ], tot_loss[loss=0.3209, simple_loss=0.3688, pruned_loss=0.1365, over 3823189.01 frames. ], batch size: 53, lr: 2.38e-02, grad_scale: 8.0 2023-04-01 01:14:16,620 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=18458.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 01:14:19,800 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.8441, 3.4597, 2.1212, 3.1092, 1.1775, 3.1931, 3.0793, 3.2346], device='cuda:1'), covar=tensor([0.0761, 0.1318, 0.2230, 0.0945, 0.3773, 0.1026, 0.0817, 0.1106], device='cuda:1'), in_proj_covar=tensor([0.0318, 0.0288, 0.0339, 0.0278, 0.0344, 0.0291, 0.0246, 0.0286], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 01:14:22,344 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=18463.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 01:14:53,517 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=18488.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 01:15:16,092 INFO [train.py:903] (1/4) Epoch 3, batch 4850, loss[loss=0.3042, simple_loss=0.3417, pruned_loss=0.1334, over 19753.00 frames. ], tot_loss[loss=0.3177, simple_loss=0.3663, pruned_loss=0.1345, over 3822005.57 frames. ], batch size: 46, lr: 2.38e-02, grad_scale: 8.0 2023-04-01 01:15:17,542 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=18508.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 01:15:34,581 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.685e+02 7.295e+02 9.130e+02 1.063e+03 1.681e+03, threshold=1.826e+03, percent-clipped=0.0 2023-04-01 01:15:38,187 WARNING [train.py:1073] (1/4) Exclude cut with ID 774-127930-0014-48411_sp1.1_rvb from training. Duration: 0.95 2023-04-01 01:15:40,626 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=18526.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 01:15:52,381 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=18536.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 01:15:53,573 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=18537.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 01:16:00,126 WARNING [train.py:1073] (1/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1_rvb from training. Duration: 0.9409375 2023-04-01 01:16:02,865 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0209, 2.0348, 1.6697, 2.9488, 2.1698, 3.0847, 2.7237, 1.8727], device='cuda:1'), covar=tensor([0.1037, 0.0750, 0.0534, 0.0477, 0.0868, 0.0244, 0.0775, 0.0741], device='cuda:1'), in_proj_covar=tensor([0.0489, 0.0458, 0.0460, 0.0610, 0.0538, 0.0372, 0.0558, 0.0460], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 01:16:04,601 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 2023-04-01 01:16:05,767 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 2023-04-01 01:16:14,590 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1_rvb from training. Duration: 25.3818125 2023-04-01 01:16:15,641 INFO [train.py:903] (1/4) Epoch 3, batch 4900, loss[loss=0.2788, simple_loss=0.3347, pruned_loss=0.1114, over 19350.00 frames. ], tot_loss[loss=0.3177, simple_loss=0.3665, pruned_loss=0.1344, over 3829259.31 frames. ], batch size: 47, lr: 2.38e-02, grad_scale: 8.0 2023-04-01 01:16:34,639 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 2023-04-01 01:17:14,799 INFO [train.py:903] (1/4) Epoch 3, batch 4950, loss[loss=0.3491, simple_loss=0.3925, pruned_loss=0.1528, over 19727.00 frames. ], tot_loss[loss=0.3204, simple_loss=0.3688, pruned_loss=0.1361, over 3830079.65 frames. ], batch size: 63, lr: 2.37e-02, grad_scale: 4.0 2023-04-01 01:17:26,350 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2426, 2.2260, 1.5414, 1.4141, 1.9684, 1.0120, 0.7876, 1.5145], device='cuda:1'), covar=tensor([0.0800, 0.0400, 0.0961, 0.0581, 0.0416, 0.1220, 0.0879, 0.0493], device='cuda:1'), in_proj_covar=tensor([0.0289, 0.0240, 0.0314, 0.0260, 0.0220, 0.0314, 0.0284, 0.0259], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 01:17:30,516 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 2023-04-01 01:17:34,976 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 5.825e+02 8.688e+02 1.059e+03 1.337e+03 3.400e+03, threshold=2.119e+03, percent-clipped=10.0 2023-04-01 01:17:52,113 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 2023-04-01 01:18:09,817 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=18651.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 01:18:10,916 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=18652.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 01:18:17,089 INFO [train.py:903] (1/4) Epoch 3, batch 5000, loss[loss=0.336, simple_loss=0.3719, pruned_loss=0.15, over 19280.00 frames. ], tot_loss[loss=0.3194, simple_loss=0.368, pruned_loss=0.1354, over 3825891.67 frames. ], batch size: 44, lr: 2.37e-02, grad_scale: 4.0 2023-04-01 01:18:21,592 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_rvb from training. Duration: 27.14 2023-04-01 01:18:32,451 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 2023-04-01 01:18:59,268 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=18692.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 01:19:00,347 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=18693.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 01:19:16,429 INFO [train.py:903] (1/4) Epoch 3, batch 5050, loss[loss=0.3156, simple_loss=0.3776, pruned_loss=0.1268, over 19542.00 frames. ], tot_loss[loss=0.3228, simple_loss=0.3704, pruned_loss=0.1376, over 3808973.92 frames. ], batch size: 56, lr: 2.37e-02, grad_scale: 4.0 2023-04-01 01:19:29,205 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=18718.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 01:19:35,295 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.108e+02 7.776e+02 1.028e+03 1.229e+03 3.550e+03, threshold=2.057e+03, percent-clipped=2.0 2023-04-01 01:19:44,363 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9358, 1.9395, 2.2631, 2.1971, 2.9743, 3.5000, 3.6066, 3.7080], device='cuda:1'), covar=tensor([0.1048, 0.1858, 0.1889, 0.1254, 0.0508, 0.0192, 0.0154, 0.0109], device='cuda:1'), in_proj_covar=tensor([0.0273, 0.0279, 0.0322, 0.0266, 0.0199, 0.0108, 0.0201, 0.0118], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0001, 0.0002, 0.0001], device='cuda:1') 2023-04-01 01:19:48,428 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9_rvb from training. Duration: 26.62775 2023-04-01 01:20:16,939 INFO [train.py:903] (1/4) Epoch 3, batch 5100, loss[loss=0.3208, simple_loss=0.3708, pruned_loss=0.1354, over 19668.00 frames. ], tot_loss[loss=0.3236, simple_loss=0.372, pruned_loss=0.1376, over 3806251.23 frames. ], batch size: 55, lr: 2.36e-02, grad_scale: 4.0 2023-04-01 01:20:24,654 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 2023-04-01 01:20:27,878 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_rvb from training. Duration: 25.035 2023-04-01 01:20:32,464 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_rvb from training. Duration: 27.92 2023-04-01 01:21:19,959 INFO [train.py:903] (1/4) Epoch 3, batch 5150, loss[loss=0.3049, simple_loss=0.3665, pruned_loss=0.1216, over 19685.00 frames. ], tot_loss[loss=0.3216, simple_loss=0.371, pruned_loss=0.1361, over 3808815.10 frames. ], batch size: 53, lr: 2.36e-02, grad_scale: 4.0 2023-04-01 01:21:31,908 WARNING [train.py:1073] (1/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 2023-04-01 01:21:40,054 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.229e+02 6.924e+02 8.047e+02 1.080e+03 1.932e+03, threshold=1.609e+03, percent-clipped=0.0 2023-04-01 01:22:03,173 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1318, 1.1104, 1.7013, 1.2437, 2.5125, 2.5057, 2.7954, 1.0139], device='cuda:1'), covar=tensor([0.1802, 0.2648, 0.1517, 0.1658, 0.1118, 0.1084, 0.1139, 0.2380], device='cuda:1'), in_proj_covar=tensor([0.0420, 0.0463, 0.0430, 0.0390, 0.0506, 0.0412, 0.0585, 0.0414], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 01:22:07,001 WARNING [train.py:1073] (1/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1_rvb from training. Duration: 0.9681875 2023-04-01 01:22:14,771 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=18852.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 01:22:20,356 INFO [train.py:903] (1/4) Epoch 3, batch 5200, loss[loss=0.285, simple_loss=0.3375, pruned_loss=0.1162, over 19468.00 frames. ], tot_loss[loss=0.3205, simple_loss=0.3703, pruned_loss=0.1353, over 3826525.50 frames. ], batch size: 49, lr: 2.36e-02, grad_scale: 8.0 2023-04-01 01:22:37,382 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 2023-04-01 01:22:37,525 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=18870.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 01:23:21,224 WARNING [train.py:1073] (1/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 2023-04-01 01:23:23,224 INFO [train.py:903] (1/4) Epoch 3, batch 5250, loss[loss=0.4379, simple_loss=0.4431, pruned_loss=0.2163, over 13131.00 frames. ], tot_loss[loss=0.3199, simple_loss=0.3694, pruned_loss=0.1352, over 3817167.93 frames. ], batch size: 135, lr: 2.36e-02, grad_scale: 4.0 2023-04-01 01:23:23,680 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=18907.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 01:23:24,855 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=18908.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 01:23:42,383 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.727e+02 7.388e+02 9.793e+02 1.246e+03 4.620e+03, threshold=1.959e+03, percent-clipped=9.0 2023-04-01 01:23:53,535 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=18932.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 01:23:54,590 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=18933.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 01:24:12,780 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=18948.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 01:24:22,195 INFO [train.py:903] (1/4) Epoch 3, batch 5300, loss[loss=0.3319, simple_loss=0.3821, pruned_loss=0.1409, over 19470.00 frames. ], tot_loss[loss=0.3205, simple_loss=0.3701, pruned_loss=0.1355, over 3813590.11 frames. ], batch size: 64, lr: 2.35e-02, grad_scale: 4.0 2023-04-01 01:24:35,252 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=18967.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 01:24:39,530 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 2023-04-01 01:24:56,456 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=18985.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 01:25:06,616 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5923, 1.3979, 2.0856, 1.6085, 2.8171, 2.5042, 2.9147, 1.7442], device='cuda:1'), covar=tensor([0.1049, 0.1950, 0.0969, 0.1033, 0.0655, 0.0783, 0.0720, 0.1494], device='cuda:1'), in_proj_covar=tensor([0.0417, 0.0466, 0.0428, 0.0392, 0.0507, 0.0408, 0.0581, 0.0418], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 01:25:22,315 INFO [train.py:903] (1/4) Epoch 3, batch 5350, loss[loss=0.3019, simple_loss=0.3393, pruned_loss=0.1322, over 19761.00 frames. ], tot_loss[loss=0.3209, simple_loss=0.3701, pruned_loss=0.1358, over 3817825.11 frames. ], batch size: 48, lr: 2.35e-02, grad_scale: 4.0 2023-04-01 01:25:42,779 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.207e+02 7.837e+02 1.011e+03 1.314e+03 3.062e+03, threshold=2.023e+03, percent-clipped=6.0 2023-04-01 01:25:51,174 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2415, 1.0239, 1.3226, 0.4295, 2.4614, 2.3941, 2.0745, 2.4360], device='cuda:1'), covar=tensor([0.1161, 0.2782, 0.2800, 0.2167, 0.0351, 0.0167, 0.0354, 0.0182], device='cuda:1'), in_proj_covar=tensor([0.0273, 0.0281, 0.0328, 0.0266, 0.0198, 0.0109, 0.0201, 0.0121], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0001, 0.0002, 0.0001], device='cuda:1') 2023-04-01 01:25:55,447 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9_rvb from training. Duration: 26.438875 2023-04-01 01:25:56,694 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=19036.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 01:26:20,407 INFO [train.py:903] (1/4) Epoch 3, batch 5400, loss[loss=0.3077, simple_loss=0.357, pruned_loss=0.1292, over 19471.00 frames. ], tot_loss[loss=0.322, simple_loss=0.3706, pruned_loss=0.1367, over 3816707.46 frames. ], batch size: 49, lr: 2.35e-02, grad_scale: 4.0 2023-04-01 01:27:22,187 INFO [train.py:903] (1/4) Epoch 3, batch 5450, loss[loss=0.3071, simple_loss=0.3671, pruned_loss=0.1235, over 19334.00 frames. ], tot_loss[loss=0.3208, simple_loss=0.3702, pruned_loss=0.1358, over 3830544.01 frames. ], batch size: 66, lr: 2.34e-02, grad_scale: 4.0 2023-04-01 01:27:41,223 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 5.092e+02 7.941e+02 9.480e+02 1.159e+03 2.761e+03, threshold=1.896e+03, percent-clipped=1.0 2023-04-01 01:28:10,077 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=19147.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 01:28:14,878 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=19151.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 01:28:21,131 INFO [train.py:903] (1/4) Epoch 3, batch 5500, loss[loss=0.3724, simple_loss=0.4072, pruned_loss=0.1688, over 19542.00 frames. ], tot_loss[loss=0.3211, simple_loss=0.3698, pruned_loss=0.1362, over 3831350.35 frames. ], batch size: 54, lr: 2.34e-02, grad_scale: 4.0 2023-04-01 01:28:45,303 WARNING [train.py:1073] (1/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9_rvb from training. Duration: 27.47775 2023-04-01 01:29:20,073 INFO [train.py:903] (1/4) Epoch 3, batch 5550, loss[loss=0.3035, simple_loss=0.3526, pruned_loss=0.1272, over 19730.00 frames. ], tot_loss[loss=0.3219, simple_loss=0.3699, pruned_loss=0.137, over 3816973.27 frames. ], batch size: 51, lr: 2.34e-02, grad_scale: 4.0 2023-04-01 01:29:27,432 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 2023-04-01 01:29:37,979 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5241, 0.9331, 1.5545, 0.9807, 2.7597, 3.4537, 3.4154, 3.8226], device='cuda:1'), covar=tensor([0.1315, 0.3017, 0.2749, 0.2101, 0.0404, 0.0119, 0.0190, 0.0089], device='cuda:1'), in_proj_covar=tensor([0.0271, 0.0281, 0.0326, 0.0266, 0.0197, 0.0108, 0.0202, 0.0118], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0001, 0.0002, 0.0001], device='cuda:1') 2023-04-01 01:29:41,412 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=19223.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 01:29:42,151 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.073e+02 7.947e+02 9.800e+02 1.342e+03 2.993e+03, threshold=1.960e+03, percent-clipped=6.0 2023-04-01 01:30:02,049 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=19241.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 01:30:09,698 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=19248.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 01:30:15,624 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 2023-04-01 01:30:17,154 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([5.1431, 1.7109, 1.9561, 2.3575, 4.6717, 1.6196, 2.6240, 4.5512], device='cuda:1'), covar=tensor([0.0205, 0.2160, 0.2050, 0.1310, 0.0303, 0.1843, 0.0944, 0.0274], device='cuda:1'), in_proj_covar=tensor([0.0257, 0.0307, 0.0300, 0.0277, 0.0289, 0.0314, 0.0272, 0.0278], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-01 01:30:21,461 INFO [train.py:903] (1/4) Epoch 3, batch 5600, loss[loss=0.3167, simple_loss=0.3831, pruned_loss=0.1252, over 19675.00 frames. ], tot_loss[loss=0.3203, simple_loss=0.3691, pruned_loss=0.1358, over 3830020.52 frames. ], batch size: 55, lr: 2.34e-02, grad_scale: 8.0 2023-04-01 01:30:33,698 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=19266.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 01:31:03,988 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=19292.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 01:31:22,079 INFO [train.py:903] (1/4) Epoch 3, batch 5650, loss[loss=0.3707, simple_loss=0.4068, pruned_loss=0.1673, over 19668.00 frames. ], tot_loss[loss=0.3211, simple_loss=0.3695, pruned_loss=0.1364, over 3821731.03 frames. ], batch size: 55, lr: 2.33e-02, grad_scale: 8.0 2023-04-01 01:31:41,038 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.993e+02 7.403e+02 9.102e+02 1.185e+03 3.385e+03, threshold=1.820e+03, percent-clipped=3.0 2023-04-01 01:32:09,525 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9_rvb from training. Duration: 28.0944375 2023-04-01 01:32:21,083 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9076, 1.7074, 1.3574, 1.8608, 1.5405, 1.6603, 1.4595, 1.7233], device='cuda:1'), covar=tensor([0.0745, 0.1311, 0.1218, 0.0826, 0.1073, 0.0527, 0.0903, 0.0639], device='cuda:1'), in_proj_covar=tensor([0.0259, 0.0379, 0.0283, 0.0251, 0.0313, 0.0263, 0.0269, 0.0245], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 01:32:21,815 INFO [train.py:903] (1/4) Epoch 3, batch 5700, loss[loss=0.3055, simple_loss=0.3614, pruned_loss=0.1248, over 19630.00 frames. ], tot_loss[loss=0.3213, simple_loss=0.3699, pruned_loss=0.1364, over 3826687.58 frames. ], batch size: 50, lr: 2.33e-02, grad_scale: 8.0 2023-04-01 01:32:50,542 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9492, 1.9194, 1.7415, 2.8955, 1.9048, 2.8117, 2.4781, 1.7413], device='cuda:1'), covar=tensor([0.1127, 0.0863, 0.0561, 0.0441, 0.1048, 0.0280, 0.0893, 0.0856], device='cuda:1'), in_proj_covar=tensor([0.0499, 0.0471, 0.0470, 0.0628, 0.0547, 0.0387, 0.0563, 0.0461], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 01:33:22,521 INFO [train.py:903] (1/4) Epoch 3, batch 5750, loss[loss=0.3111, simple_loss=0.3702, pruned_loss=0.126, over 19647.00 frames. ], tot_loss[loss=0.3198, simple_loss=0.3689, pruned_loss=0.1353, over 3821615.41 frames. ], batch size: 58, lr: 2.33e-02, grad_scale: 8.0 2023-04-01 01:33:22,866 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=19407.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 01:33:22,936 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=19407.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 01:33:24,192 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.3907, 2.1277, 1.9759, 3.5255, 2.2325, 3.9103, 3.3910, 1.9134], device='cuda:1'), covar=tensor([0.1251, 0.0943, 0.0544, 0.0565, 0.1177, 0.0234, 0.0785, 0.0852], device='cuda:1'), in_proj_covar=tensor([0.0491, 0.0465, 0.0464, 0.0622, 0.0540, 0.0383, 0.0553, 0.0455], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 01:33:24,900 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9_rvb from training. Duration: 33.038875 2023-04-01 01:33:30,118 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=19412.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 01:33:34,052 WARNING [train.py:1073] (1/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 2023-04-01 01:33:39,377 WARNING [train.py:1073] (1/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 2023-04-01 01:33:43,561 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.926e+02 7.525e+02 9.538e+02 1.231e+03 3.556e+03, threshold=1.908e+03, percent-clipped=6.0 2023-04-01 01:33:53,845 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=19432.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 01:34:22,888 INFO [train.py:903] (1/4) Epoch 3, batch 5800, loss[loss=0.3195, simple_loss=0.3818, pruned_loss=0.1287, over 19662.00 frames. ], tot_loss[loss=0.3223, simple_loss=0.3709, pruned_loss=0.1369, over 3815220.24 frames. ], batch size: 55, lr: 2.32e-02, grad_scale: 4.0 2023-04-01 01:34:24,441 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3794, 1.1026, 1.5128, 0.9287, 2.4800, 3.1970, 2.8333, 3.4280], device='cuda:1'), covar=tensor([0.1323, 0.2616, 0.2599, 0.2012, 0.0432, 0.0101, 0.0253, 0.0106], device='cuda:1'), in_proj_covar=tensor([0.0265, 0.0276, 0.0319, 0.0266, 0.0196, 0.0107, 0.0199, 0.0118], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0001, 0.0002, 0.0001], device='cuda:1') 2023-04-01 01:35:03,993 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=19491.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 01:35:09,079 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.64 vs. limit=2.0 2023-04-01 01:35:23,137 INFO [train.py:903] (1/4) Epoch 3, batch 5850, loss[loss=0.3475, simple_loss=0.3978, pruned_loss=0.1486, over 19684.00 frames. ], tot_loss[loss=0.3223, simple_loss=0.3712, pruned_loss=0.1367, over 3816951.91 frames. ], batch size: 59, lr: 2.32e-02, grad_scale: 4.0 2023-04-01 01:35:37,152 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9138, 1.8844, 1.6912, 2.7076, 1.9397, 2.6868, 2.3758, 1.7880], device='cuda:1'), covar=tensor([0.1106, 0.0864, 0.0539, 0.0514, 0.1006, 0.0315, 0.0958, 0.0788], device='cuda:1'), in_proj_covar=tensor([0.0500, 0.0473, 0.0470, 0.0632, 0.0550, 0.0390, 0.0571, 0.0462], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 01:35:43,378 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.468e+02 8.354e+02 1.035e+03 1.319e+03 5.609e+03, threshold=2.070e+03, percent-clipped=8.0 2023-04-01 01:36:18,436 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.84 vs. limit=2.0 2023-04-01 01:36:23,587 INFO [train.py:903] (1/4) Epoch 3, batch 5900, loss[loss=0.3516, simple_loss=0.3924, pruned_loss=0.1554, over 19299.00 frames. ], tot_loss[loss=0.3212, simple_loss=0.3703, pruned_loss=0.1361, over 3808297.63 frames. ], batch size: 66, lr: 2.32e-02, grad_scale: 4.0 2023-04-01 01:36:26,962 WARNING [train.py:1073] (1/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 2023-04-01 01:36:46,202 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9_rvb from training. Duration: 27.511125 2023-04-01 01:37:21,778 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=19606.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 01:37:22,518 INFO [train.py:903] (1/4) Epoch 3, batch 5950, loss[loss=0.3518, simple_loss=0.394, pruned_loss=0.1548, over 19779.00 frames. ], tot_loss[loss=0.3187, simple_loss=0.3678, pruned_loss=0.1348, over 3800499.49 frames. ], batch size: 56, lr: 2.32e-02, grad_scale: 4.0 2023-04-01 01:37:45,644 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.616e+02 6.894e+02 9.030e+02 1.163e+03 3.004e+03, threshold=1.806e+03, percent-clipped=5.0 2023-04-01 01:37:59,453 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=19636.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 01:38:24,425 INFO [train.py:903] (1/4) Epoch 3, batch 6000, loss[loss=0.3764, simple_loss=0.3951, pruned_loss=0.1788, over 13546.00 frames. ], tot_loss[loss=0.3193, simple_loss=0.3686, pruned_loss=0.135, over 3796403.60 frames. ], batch size: 136, lr: 2.31e-02, grad_scale: 8.0 2023-04-01 01:38:24,425 INFO [train.py:928] (1/4) Computing validation loss 2023-04-01 01:38:37,333 INFO [train.py:937] (1/4) Epoch 3, validation: loss=0.2218, simple_loss=0.3182, pruned_loss=0.06273, over 944034.00 frames. 2023-04-01 01:38:37,334 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18649MB 2023-04-01 01:38:45,541 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=19663.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 01:39:15,734 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=19688.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 01:39:37,220 INFO [train.py:903] (1/4) Epoch 3, batch 6050, loss[loss=0.3716, simple_loss=0.3933, pruned_loss=0.1749, over 19467.00 frames. ], tot_loss[loss=0.3173, simple_loss=0.3668, pruned_loss=0.1339, over 3806574.91 frames. ], batch size: 49, lr: 2.31e-02, grad_scale: 8.0 2023-04-01 01:39:59,298 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 5.085e+02 7.768e+02 9.451e+02 1.306e+03 2.772e+03, threshold=1.890e+03, percent-clipped=6.0 2023-04-01 01:40:37,795 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=19756.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 01:40:38,850 INFO [train.py:903] (1/4) Epoch 3, batch 6100, loss[loss=0.2602, simple_loss=0.3215, pruned_loss=0.09942, over 19693.00 frames. ], tot_loss[loss=0.3178, simple_loss=0.3672, pruned_loss=0.1342, over 3799437.36 frames. ], batch size: 53, lr: 2.31e-02, grad_scale: 8.0 2023-04-01 01:41:19,574 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 2023-04-01 01:41:38,885 INFO [train.py:903] (1/4) Epoch 3, batch 6150, loss[loss=0.349, simple_loss=0.3762, pruned_loss=0.1609, over 19808.00 frames. ], tot_loss[loss=0.3211, simple_loss=0.3696, pruned_loss=0.1363, over 3805097.39 frames. ], batch size: 48, lr: 2.30e-02, grad_scale: 8.0 2023-04-01 01:42:01,434 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.829e+02 8.602e+02 1.123e+03 1.510e+03 2.312e+03, threshold=2.246e+03, percent-clipped=7.0 2023-04-01 01:42:04,016 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6421, 1.2830, 1.2401, 1.7108, 1.4525, 1.4755, 1.3507, 1.5891], device='cuda:1'), covar=tensor([0.0782, 0.1357, 0.1263, 0.0750, 0.1028, 0.0509, 0.0897, 0.0607], device='cuda:1'), in_proj_covar=tensor([0.0257, 0.0388, 0.0286, 0.0251, 0.0319, 0.0262, 0.0274, 0.0239], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 01:42:04,686 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9_rvb from training. Duration: 31.02225 2023-04-01 01:42:38,569 INFO [train.py:903] (1/4) Epoch 3, batch 6200, loss[loss=0.3055, simple_loss=0.3674, pruned_loss=0.1218, over 19670.00 frames. ], tot_loss[loss=0.3219, simple_loss=0.3703, pruned_loss=0.1367, over 3802274.72 frames. ], batch size: 58, lr: 2.30e-02, grad_scale: 8.0 2023-04-01 01:42:46,183 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=19862.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 01:42:57,072 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=19871.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 01:43:15,887 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=19887.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 01:43:39,797 INFO [train.py:903] (1/4) Epoch 3, batch 6250, loss[loss=0.362, simple_loss=0.4019, pruned_loss=0.161, over 19121.00 frames. ], tot_loss[loss=0.322, simple_loss=0.3708, pruned_loss=0.1366, over 3806044.56 frames. ], batch size: 69, lr: 2.30e-02, grad_scale: 8.0 2023-04-01 01:44:01,767 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.496e+02 7.100e+02 9.208e+02 1.133e+03 3.490e+03, threshold=1.842e+03, percent-clipped=3.0 2023-04-01 01:44:09,641 WARNING [train.py:1073] (1/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9_rvb from training. Duration: 25.988875 2023-04-01 01:44:40,501 INFO [train.py:903] (1/4) Epoch 3, batch 6300, loss[loss=0.3089, simple_loss=0.3688, pruned_loss=0.1245, over 19573.00 frames. ], tot_loss[loss=0.3212, simple_loss=0.3706, pruned_loss=0.1359, over 3805583.14 frames. ], batch size: 67, lr: 2.30e-02, grad_scale: 8.0 2023-04-01 01:44:57,893 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.09 vs. limit=5.0 2023-04-01 01:45:02,069 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2934, 1.0984, 2.1075, 1.6056, 3.1313, 2.9715, 3.5058, 1.5150], device='cuda:1'), covar=tensor([0.1521, 0.2582, 0.1396, 0.1234, 0.1034, 0.1057, 0.1175, 0.2340], device='cuda:1'), in_proj_covar=tensor([0.0422, 0.0473, 0.0437, 0.0401, 0.0523, 0.0423, 0.0595, 0.0427], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 01:45:07,282 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=19980.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 01:45:10,846 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 2023-04-01 01:45:42,100 INFO [train.py:903] (1/4) Epoch 3, batch 6350, loss[loss=0.334, simple_loss=0.387, pruned_loss=0.1405, over 19659.00 frames. ], tot_loss[loss=0.3193, simple_loss=0.3689, pruned_loss=0.1348, over 3800256.60 frames. ], batch size: 55, lr: 2.29e-02, grad_scale: 8.0 2023-04-01 01:46:03,318 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.926e+02 7.424e+02 9.293e+02 1.158e+03 2.968e+03, threshold=1.859e+03, percent-clipped=5.0 2023-04-01 01:46:20,207 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=20038.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 01:46:20,731 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 2023-04-01 01:46:42,148 INFO [train.py:903] (1/4) Epoch 3, batch 6400, loss[loss=0.265, simple_loss=0.3185, pruned_loss=0.1058, over 19732.00 frames. ], tot_loss[loss=0.3208, simple_loss=0.3701, pruned_loss=0.1358, over 3792065.53 frames. ], batch size: 51, lr: 2.29e-02, grad_scale: 8.0 2023-04-01 01:47:29,294 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=20095.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 01:47:32,635 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.7770, 1.3873, 1.7377, 1.6962, 2.8549, 4.6054, 4.6021, 5.0323], device='cuda:1'), covar=tensor([0.1245, 0.2605, 0.2633, 0.1703, 0.0419, 0.0128, 0.0119, 0.0068], device='cuda:1'), in_proj_covar=tensor([0.0267, 0.0275, 0.0321, 0.0264, 0.0192, 0.0110, 0.0200, 0.0121], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0001, 0.0002, 0.0001], device='cuda:1') 2023-04-01 01:47:43,490 INFO [train.py:903] (1/4) Epoch 3, batch 6450, loss[loss=0.3599, simple_loss=0.4003, pruned_loss=0.1598, over 19660.00 frames. ], tot_loss[loss=0.319, simple_loss=0.3686, pruned_loss=0.1347, over 3796892.39 frames. ], batch size: 58, lr: 2.29e-02, grad_scale: 8.0 2023-04-01 01:47:43,788 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=20107.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 01:48:05,562 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 5.028e+02 7.833e+02 9.380e+02 1.145e+03 2.427e+03, threshold=1.876e+03, percent-clipped=3.0 2023-04-01 01:48:08,342 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=20127.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 01:48:12,611 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=20130.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 01:48:29,094 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9_rvb from training. Duration: 28.638875 2023-04-01 01:48:39,345 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=20152.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 01:48:44,547 INFO [train.py:903] (1/4) Epoch 3, batch 6500, loss[loss=0.3153, simple_loss=0.3707, pruned_loss=0.1299, over 19521.00 frames. ], tot_loss[loss=0.3174, simple_loss=0.3673, pruned_loss=0.1337, over 3811262.77 frames. ], batch size: 54, lr: 2.29e-02, grad_scale: 8.0 2023-04-01 01:48:52,292 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1_rvb from training. Duration: 0.836375 2023-04-01 01:49:45,351 INFO [train.py:903] (1/4) Epoch 3, batch 6550, loss[loss=0.3771, simple_loss=0.4063, pruned_loss=0.1739, over 19539.00 frames. ], tot_loss[loss=0.3183, simple_loss=0.368, pruned_loss=0.1343, over 3799813.10 frames. ], batch size: 56, lr: 2.28e-02, grad_scale: 8.0 2023-04-01 01:50:03,598 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=20222.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 01:50:07,448 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.125e+02 7.085e+02 9.596e+02 1.224e+03 2.377e+03, threshold=1.919e+03, percent-clipped=5.0 2023-04-01 01:50:46,685 INFO [train.py:903] (1/4) Epoch 3, batch 6600, loss[loss=0.2926, simple_loss=0.3549, pruned_loss=0.1152, over 19782.00 frames. ], tot_loss[loss=0.3173, simple_loss=0.3675, pruned_loss=0.1336, over 3816019.91 frames. ], batch size: 54, lr: 2.28e-02, grad_scale: 8.0 2023-04-01 01:51:20,946 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.2963, 0.7195, 1.1023, 1.2724, 1.8751, 0.9924, 1.7485, 1.8633], device='cuda:1'), covar=tensor([0.0820, 0.3523, 0.3020, 0.1857, 0.1126, 0.2267, 0.1240, 0.0976], device='cuda:1'), in_proj_covar=tensor([0.0255, 0.0302, 0.0299, 0.0277, 0.0288, 0.0310, 0.0273, 0.0284], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 01:51:47,898 INFO [train.py:903] (1/4) Epoch 3, batch 6650, loss[loss=0.3146, simple_loss=0.3688, pruned_loss=0.1301, over 19541.00 frames. ], tot_loss[loss=0.3143, simple_loss=0.365, pruned_loss=0.1318, over 3813180.30 frames. ], batch size: 56, lr: 2.28e-02, grad_scale: 8.0 2023-04-01 01:51:57,028 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4848, 2.3594, 1.4694, 1.6473, 2.1292, 1.0897, 1.0644, 1.6735], device='cuda:1'), covar=tensor([0.0856, 0.0412, 0.1149, 0.0541, 0.0370, 0.1254, 0.0756, 0.0461], device='cuda:1'), in_proj_covar=tensor([0.0285, 0.0245, 0.0314, 0.0251, 0.0212, 0.0314, 0.0279, 0.0249], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 01:52:10,191 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.507e+02 7.672e+02 8.992e+02 1.144e+03 3.524e+03, threshold=1.798e+03, percent-clipped=4.0 2023-04-01 01:52:42,295 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=20351.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 01:52:49,535 INFO [train.py:903] (1/4) Epoch 3, batch 6700, loss[loss=0.2663, simple_loss=0.3281, pruned_loss=0.1022, over 19484.00 frames. ], tot_loss[loss=0.3166, simple_loss=0.3666, pruned_loss=0.1332, over 3785346.37 frames. ], batch size: 49, lr: 2.28e-02, grad_scale: 8.0 2023-04-01 01:52:50,835 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.5874, 1.8732, 1.8975, 2.4231, 2.2148, 2.6102, 2.7987, 2.3325], device='cuda:1'), covar=tensor([0.0701, 0.1066, 0.1159, 0.1222, 0.1215, 0.0722, 0.1101, 0.0701], device='cuda:1'), in_proj_covar=tensor([0.0249, 0.0281, 0.0267, 0.0305, 0.0312, 0.0246, 0.0275, 0.0244], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:1') 2023-04-01 01:53:11,819 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=20376.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 01:53:17,972 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=20382.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 01:53:44,487 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=20406.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 01:53:46,032 INFO [train.py:903] (1/4) Epoch 3, batch 6750, loss[loss=0.3128, simple_loss=0.3666, pruned_loss=0.1295, over 18134.00 frames. ], tot_loss[loss=0.3181, simple_loss=0.3676, pruned_loss=0.1343, over 3791271.53 frames. ], batch size: 83, lr: 2.27e-02, grad_scale: 8.0 2023-04-01 01:54:05,351 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.786e+02 7.667e+02 1.002e+03 1.269e+03 2.908e+03, threshold=2.004e+03, percent-clipped=6.0 2023-04-01 01:54:17,387 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=20436.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 01:54:34,785 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=20451.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 01:54:41,604 INFO [train.py:903] (1/4) Epoch 3, batch 6800, loss[loss=0.2922, simple_loss=0.349, pruned_loss=0.1177, over 19837.00 frames. ], tot_loss[loss=0.3163, simple_loss=0.3662, pruned_loss=0.1332, over 3796617.56 frames. ], batch size: 52, lr: 2.27e-02, grad_scale: 8.0 2023-04-01 01:55:00,654 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=20474.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 01:55:25,667 WARNING [train.py:1073] (1/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 2023-04-01 01:55:26,115 WARNING [train.py:1073] (1/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 2023-04-01 01:55:29,043 INFO [train.py:903] (1/4) Epoch 4, batch 0, loss[loss=0.3371, simple_loss=0.3824, pruned_loss=0.1459, over 19697.00 frames. ], tot_loss[loss=0.3371, simple_loss=0.3824, pruned_loss=0.1459, over 19697.00 frames. ], batch size: 53, lr: 2.12e-02, grad_scale: 8.0 2023-04-01 01:55:29,043 INFO [train.py:928] (1/4) Computing validation loss 2023-04-01 01:55:40,524 INFO [train.py:937] (1/4) Epoch 4, validation: loss=0.2245, simple_loss=0.3205, pruned_loss=0.06426, over 944034.00 frames. 2023-04-01 01:55:40,525 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18649MB 2023-04-01 01:55:51,821 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.9558, 2.7501, 1.9747, 2.1204, 1.8850, 2.0723, 0.6621, 2.0526], device='cuda:1'), covar=tensor([0.0197, 0.0205, 0.0208, 0.0261, 0.0396, 0.0321, 0.0462, 0.0363], device='cuda:1'), in_proj_covar=tensor([0.0256, 0.0256, 0.0255, 0.0280, 0.0334, 0.0274, 0.0258, 0.0270], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 01:55:53,633 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 2023-04-01 01:55:55,144 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=20497.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 01:56:07,943 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.90 vs. limit=5.0 2023-04-01 01:56:27,869 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.490e+02 6.855e+02 8.790e+02 1.166e+03 2.960e+03, threshold=1.758e+03, percent-clipped=3.0 2023-04-01 01:56:41,007 INFO [train.py:903] (1/4) Epoch 4, batch 50, loss[loss=0.3208, simple_loss=0.3801, pruned_loss=0.1308, over 19573.00 frames. ], tot_loss[loss=0.3162, simple_loss=0.3654, pruned_loss=0.1335, over 863796.41 frames. ], batch size: 61, lr: 2.12e-02, grad_scale: 8.0 2023-04-01 01:57:01,516 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3977, 1.4835, 1.5846, 1.2962, 2.6004, 3.3691, 3.3549, 3.6962], device='cuda:1'), covar=tensor([0.1236, 0.2204, 0.2344, 0.1653, 0.0391, 0.0125, 0.0177, 0.0085], device='cuda:1'), in_proj_covar=tensor([0.0264, 0.0274, 0.0316, 0.0261, 0.0187, 0.0104, 0.0193, 0.0117], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0001, 0.0002, 0.0001], device='cuda:1') 2023-04-01 01:57:14,493 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1_rvb from training. Duration: 27.0318125 2023-04-01 01:57:15,795 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=20566.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 01:57:16,001 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=20566.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 01:57:36,910 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.5012, 3.9049, 4.0935, 4.0511, 1.3733, 3.6705, 3.3936, 3.6873], device='cuda:1'), covar=tensor([0.0740, 0.0508, 0.0455, 0.0351, 0.3587, 0.0290, 0.0412, 0.0869], device='cuda:1'), in_proj_covar=tensor([0.0409, 0.0370, 0.0492, 0.0388, 0.0506, 0.0271, 0.0329, 0.0475], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1') 2023-04-01 01:57:40,114 INFO [train.py:903] (1/4) Epoch 4, batch 100, loss[loss=0.3177, simple_loss=0.3736, pruned_loss=0.1309, over 17241.00 frames. ], tot_loss[loss=0.3174, simple_loss=0.3673, pruned_loss=0.1338, over 1524315.59 frames. ], batch size: 101, lr: 2.12e-02, grad_scale: 8.0 2023-04-01 01:57:46,143 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=20589.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 01:57:52,769 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 2023-04-01 01:58:05,917 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=20606.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 01:58:16,534 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3848, 1.0534, 1.6859, 1.2870, 2.8482, 3.5296, 3.5167, 3.8691], device='cuda:1'), covar=tensor([0.1372, 0.2857, 0.2523, 0.1790, 0.0395, 0.0105, 0.0218, 0.0094], device='cuda:1'), in_proj_covar=tensor([0.0264, 0.0275, 0.0318, 0.0262, 0.0189, 0.0106, 0.0196, 0.0117], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0001, 0.0002, 0.0001], device='cuda:1') 2023-04-01 01:58:29,319 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.104e+02 7.763e+02 9.275e+02 1.175e+03 2.763e+03, threshold=1.855e+03, percent-clipped=7.0 2023-04-01 01:58:41,723 INFO [train.py:903] (1/4) Epoch 4, batch 150, loss[loss=0.2725, simple_loss=0.342, pruned_loss=0.1015, over 19779.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3618, pruned_loss=0.1298, over 2032942.88 frames. ], batch size: 56, lr: 2.11e-02, grad_scale: 8.0 2023-04-01 01:59:35,932 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=20681.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 01:59:40,168 INFO [train.py:903] (1/4) Epoch 4, batch 200, loss[loss=0.3113, simple_loss=0.3669, pruned_loss=0.1278, over 19466.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3626, pruned_loss=0.1294, over 2445663.68 frames. ], batch size: 64, lr: 2.11e-02, grad_scale: 8.0 2023-04-01 01:59:41,285 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 2023-04-01 02:00:28,681 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.823e+02 7.097e+02 9.184e+02 1.257e+03 2.857e+03, threshold=1.837e+03, percent-clipped=5.0 2023-04-01 02:00:39,435 INFO [train.py:903] (1/4) Epoch 4, batch 250, loss[loss=0.3084, simple_loss=0.3691, pruned_loss=0.1239, over 19671.00 frames. ], tot_loss[loss=0.3139, simple_loss=0.3645, pruned_loss=0.1317, over 2742675.71 frames. ], batch size: 58, lr: 2.11e-02, grad_scale: 8.0 2023-04-01 02:00:41,857 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.3086, 2.1546, 2.1628, 3.2074, 2.5415, 3.3696, 3.0671, 2.0353], device='cuda:1'), covar=tensor([0.1170, 0.0901, 0.0477, 0.0540, 0.0882, 0.0230, 0.0666, 0.0797], device='cuda:1'), in_proj_covar=tensor([0.0514, 0.0483, 0.0479, 0.0647, 0.0551, 0.0401, 0.0579, 0.0482], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 02:00:57,790 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=20750.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:01:02,409 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=20753.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:01:32,721 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=20778.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:01:34,829 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=20780.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:01:35,025 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([0.9728, 1.3640, 1.0736, 1.0357, 1.2635, 0.7810, 0.6560, 1.2278], device='cuda:1'), covar=tensor([0.0507, 0.0426, 0.0626, 0.0407, 0.0326, 0.0822, 0.0557, 0.0264], device='cuda:1'), in_proj_covar=tensor([0.0276, 0.0247, 0.0309, 0.0248, 0.0214, 0.0309, 0.0277, 0.0248], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 02:01:41,111 INFO [train.py:903] (1/4) Epoch 4, batch 300, loss[loss=0.2924, simple_loss=0.3395, pruned_loss=0.1227, over 19803.00 frames. ], tot_loss[loss=0.3135, simple_loss=0.364, pruned_loss=0.1315, over 2969314.80 frames. ], batch size: 48, lr: 2.11e-02, grad_scale: 8.0 2023-04-01 02:02:25,241 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=20822.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:02:29,141 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.541e+02 7.559e+02 9.012e+02 1.206e+03 2.235e+03, threshold=1.802e+03, percent-clipped=6.0 2023-04-01 02:02:30,503 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.1898, 1.5118, 1.7906, 2.6391, 1.8559, 2.4453, 2.6927, 2.5087], device='cuda:1'), covar=tensor([0.0751, 0.1132, 0.1091, 0.0996, 0.1203, 0.0714, 0.0966, 0.0612], device='cuda:1'), in_proj_covar=tensor([0.0246, 0.0275, 0.0267, 0.0298, 0.0308, 0.0249, 0.0268, 0.0243], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:1') 2023-04-01 02:02:40,181 INFO [train.py:903] (1/4) Epoch 4, batch 350, loss[loss=0.3321, simple_loss=0.3842, pruned_loss=0.14, over 19658.00 frames. ], tot_loss[loss=0.3134, simple_loss=0.364, pruned_loss=0.1314, over 3169033.71 frames. ], batch size: 60, lr: 2.10e-02, grad_scale: 8.0 2023-04-01 02:02:45,647 WARNING [train.py:1073] (1/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1_rvb from training. Duration: 0.9681875 2023-04-01 02:02:52,693 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=20845.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:02:54,956 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=20847.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:02:59,331 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.5491, 3.8267, 4.0663, 4.0806, 1.4186, 3.6795, 3.2988, 3.6666], device='cuda:1'), covar=tensor([0.0782, 0.0509, 0.0492, 0.0393, 0.3576, 0.0303, 0.0466, 0.1023], device='cuda:1'), in_proj_covar=tensor([0.0420, 0.0375, 0.0508, 0.0391, 0.0514, 0.0278, 0.0333, 0.0484], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1') 2023-04-01 02:03:18,361 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=20865.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:03:23,900 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=20870.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:03:40,719 INFO [train.py:903] (1/4) Epoch 4, batch 400, loss[loss=0.3608, simple_loss=0.4001, pruned_loss=0.1607, over 19473.00 frames. ], tot_loss[loss=0.313, simple_loss=0.3639, pruned_loss=0.1311, over 3321743.21 frames. ], batch size: 64, lr: 2.10e-02, grad_scale: 8.0 2023-04-01 02:03:52,356 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=20895.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:04:18,388 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.5556, 1.1769, 1.3846, 1.8776, 3.0981, 1.2970, 2.0752, 3.0727], device='cuda:1'), covar=tensor([0.0302, 0.2275, 0.2103, 0.1139, 0.0455, 0.1738, 0.0963, 0.0383], device='cuda:1'), in_proj_covar=tensor([0.0266, 0.0304, 0.0303, 0.0277, 0.0295, 0.0313, 0.0279, 0.0281], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 02:04:27,878 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.312e+02 7.529e+02 9.870e+02 1.265e+03 2.610e+03, threshold=1.974e+03, percent-clipped=3.0 2023-04-01 02:04:39,088 INFO [train.py:903] (1/4) Epoch 4, batch 450, loss[loss=0.3194, simple_loss=0.3661, pruned_loss=0.1363, over 19666.00 frames. ], tot_loss[loss=0.311, simple_loss=0.3627, pruned_loss=0.1297, over 3443276.43 frames. ], batch size: 53, lr: 2.10e-02, grad_scale: 8.0 2023-04-01 02:04:41,827 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=20937.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:04:44,742 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.41 vs. limit=2.0 2023-04-01 02:04:58,386 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=20950.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:05:12,736 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 2023-04-01 02:05:13,096 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=20962.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:05:13,939 WARNING [train.py:1073] (1/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 2023-04-01 02:05:40,644 INFO [train.py:903] (1/4) Epoch 4, batch 500, loss[loss=0.2607, simple_loss=0.3295, pruned_loss=0.096, over 19740.00 frames. ], tot_loss[loss=0.3099, simple_loss=0.3622, pruned_loss=0.1288, over 3531338.77 frames. ], batch size: 51, lr: 2.10e-02, grad_scale: 8.0 2023-04-01 02:06:27,513 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.944e+02 7.370e+02 9.144e+02 1.191e+03 3.185e+03, threshold=1.829e+03, percent-clipped=4.0 2023-04-01 02:06:40,256 INFO [train.py:903] (1/4) Epoch 4, batch 550, loss[loss=0.3487, simple_loss=0.3895, pruned_loss=0.154, over 19539.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.3622, pruned_loss=0.1294, over 3588139.52 frames. ], batch size: 54, lr: 2.10e-02, grad_scale: 8.0 2023-04-01 02:07:12,079 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2060, 1.1819, 1.8582, 1.4235, 2.7139, 2.2621, 2.8973, 1.0391], device='cuda:1'), covar=tensor([0.1527, 0.2531, 0.1325, 0.1325, 0.0910, 0.1181, 0.1092, 0.2419], device='cuda:1'), in_proj_covar=tensor([0.0422, 0.0471, 0.0437, 0.0396, 0.0512, 0.0413, 0.0596, 0.0421], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 02:07:17,092 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=21065.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:07:23,277 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=21070.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:07:37,586 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.4966, 4.1275, 2.2993, 3.7010, 1.4357, 3.7898, 3.6510, 3.9368], device='cuda:1'), covar=tensor([0.0506, 0.0961, 0.2140, 0.0622, 0.3276, 0.0805, 0.0697, 0.0707], device='cuda:1'), in_proj_covar=tensor([0.0321, 0.0295, 0.0338, 0.0273, 0.0339, 0.0292, 0.0251, 0.0288], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 02:07:39,553 INFO [train.py:903] (1/4) Epoch 4, batch 600, loss[loss=0.3055, simple_loss=0.3648, pruned_loss=0.1231, over 19666.00 frames. ], tot_loss[loss=0.3115, simple_loss=0.3631, pruned_loss=0.13, over 3639744.34 frames. ], batch size: 53, lr: 2.09e-02, grad_scale: 8.0 2023-04-01 02:08:15,024 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.8362, 1.3500, 1.2949, 2.0215, 1.5451, 2.0288, 2.1205, 1.9367], device='cuda:1'), covar=tensor([0.0812, 0.1231, 0.1336, 0.1081, 0.1155, 0.0756, 0.1024, 0.0724], device='cuda:1'), in_proj_covar=tensor([0.0247, 0.0272, 0.0267, 0.0296, 0.0300, 0.0249, 0.0263, 0.0243], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1') 2023-04-01 02:08:19,255 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 2023-04-01 02:08:23,045 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=21121.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:08:27,137 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.199e+02 8.023e+02 1.001e+03 1.305e+03 2.804e+03, threshold=2.003e+03, percent-clipped=3.0 2023-04-01 02:08:39,443 INFO [train.py:903] (1/4) Epoch 4, batch 650, loss[loss=0.2815, simple_loss=0.3489, pruned_loss=0.107, over 18130.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3609, pruned_loss=0.1283, over 3699564.08 frames. ], batch size: 84, lr: 2.09e-02, grad_scale: 8.0 2023-04-01 02:08:40,616 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=21136.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:08:52,182 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21146.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:09:00,259 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=21151.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:09:16,963 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.3939, 1.1121, 1.3563, 1.8208, 3.0171, 1.2034, 2.0781, 3.1625], device='cuda:1'), covar=tensor([0.0376, 0.2451, 0.2158, 0.1322, 0.0485, 0.2101, 0.1074, 0.0371], device='cuda:1'), in_proj_covar=tensor([0.0265, 0.0303, 0.0299, 0.0277, 0.0294, 0.0318, 0.0282, 0.0284], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 02:09:29,085 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21176.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:09:38,634 INFO [train.py:903] (1/4) Epoch 4, batch 700, loss[loss=0.2961, simple_loss=0.3453, pruned_loss=0.1235, over 19780.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3603, pruned_loss=0.1276, over 3722213.30 frames. ], batch size: 47, lr: 2.09e-02, grad_scale: 8.0 2023-04-01 02:10:26,834 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.035e+02 7.118e+02 8.929e+02 1.162e+03 2.438e+03, threshold=1.786e+03, percent-clipped=3.0 2023-04-01 02:10:40,518 INFO [train.py:903] (1/4) Epoch 4, batch 750, loss[loss=0.2666, simple_loss=0.3146, pruned_loss=0.1093, over 19737.00 frames. ], tot_loss[loss=0.3076, simple_loss=0.3601, pruned_loss=0.1275, over 3757347.84 frames. ], batch size: 47, lr: 2.09e-02, grad_scale: 8.0 2023-04-01 02:10:44,231 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.7129, 1.3700, 1.1354, 1.4807, 1.3930, 1.4530, 1.2301, 1.5577], device='cuda:1'), covar=tensor([0.0849, 0.1401, 0.1499, 0.0896, 0.1123, 0.0571, 0.0997, 0.0696], device='cuda:1'), in_proj_covar=tensor([0.0255, 0.0374, 0.0279, 0.0247, 0.0313, 0.0259, 0.0262, 0.0232], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 02:10:48,865 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1483, 1.0523, 1.4954, 1.2419, 1.7894, 1.7674, 1.9350, 0.4626], device='cuda:1'), covar=tensor([0.1563, 0.2539, 0.1265, 0.1361, 0.0945, 0.1262, 0.0957, 0.2331], device='cuda:1'), in_proj_covar=tensor([0.0424, 0.0476, 0.0438, 0.0396, 0.0517, 0.0414, 0.0597, 0.0425], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 02:11:40,042 INFO [train.py:903] (1/4) Epoch 4, batch 800, loss[loss=0.2922, simple_loss=0.3524, pruned_loss=0.116, over 19660.00 frames. ], tot_loss[loss=0.307, simple_loss=0.3597, pruned_loss=0.1271, over 3760622.74 frames. ], batch size: 53, lr: 2.08e-02, grad_scale: 8.0 2023-04-01 02:11:56,116 WARNING [train.py:1073] (1/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 2023-04-01 02:12:25,406 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=21321.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:12:30,558 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.379e+02 8.113e+02 1.001e+03 1.207e+03 2.017e+03, threshold=2.002e+03, percent-clipped=2.0 2023-04-01 02:12:35,446 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=21330.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:12:37,644 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([6.1361, 5.3892, 3.0084, 4.7645, 1.2703, 5.2155, 5.2388, 5.5131], device='cuda:1'), covar=tensor([0.0353, 0.1040, 0.1707, 0.0611, 0.3898, 0.0621, 0.0526, 0.0541], device='cuda:1'), in_proj_covar=tensor([0.0318, 0.0291, 0.0339, 0.0267, 0.0339, 0.0292, 0.0248, 0.0285], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 02:12:41,604 INFO [train.py:903] (1/4) Epoch 4, batch 850, loss[loss=0.2598, simple_loss=0.3203, pruned_loss=0.09969, over 19780.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3583, pruned_loss=0.1258, over 3776017.40 frames. ], batch size: 46, lr: 2.08e-02, grad_scale: 8.0 2023-04-01 02:12:54,280 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=21346.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:12:54,387 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21346.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:13:30,942 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9_rvb from training. Duration: 25.061125 2023-04-01 02:13:39,981 INFO [train.py:903] (1/4) Epoch 4, batch 900, loss[loss=0.3667, simple_loss=0.3967, pruned_loss=0.1684, over 13107.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.3591, pruned_loss=0.1263, over 3775275.10 frames. ], batch size: 138, lr: 2.08e-02, grad_scale: 8.0 2023-04-01 02:14:14,506 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=21414.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:14:28,709 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.494e+02 7.656e+02 9.192e+02 1.106e+03 2.022e+03, threshold=1.838e+03, percent-clipped=1.0 2023-04-01 02:14:40,651 INFO [train.py:903] (1/4) Epoch 4, batch 950, loss[loss=0.2813, simple_loss=0.3541, pruned_loss=0.1042, over 19705.00 frames. ], tot_loss[loss=0.3066, simple_loss=0.3598, pruned_loss=0.1267, over 3781089.90 frames. ], batch size: 59, lr: 2.08e-02, grad_scale: 8.0 2023-04-01 02:14:43,011 WARNING [train.py:1073] (1/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9_rvb from training. Duration: 26.32775 2023-04-01 02:15:35,419 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=21480.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:15:40,656 INFO [train.py:903] (1/4) Epoch 4, batch 1000, loss[loss=0.3392, simple_loss=0.3782, pruned_loss=0.1501, over 19588.00 frames. ], tot_loss[loss=0.3068, simple_loss=0.3598, pruned_loss=0.1269, over 3788145.08 frames. ], batch size: 52, lr: 2.07e-02, grad_scale: 8.0 2023-04-01 02:16:10,654 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.8466, 4.3331, 2.4044, 3.8963, 1.1631, 3.8086, 3.8607, 4.0444], device='cuda:1'), covar=tensor([0.0458, 0.1093, 0.1997, 0.0671, 0.4056, 0.0892, 0.0702, 0.0832], device='cuda:1'), in_proj_covar=tensor([0.0326, 0.0298, 0.0345, 0.0275, 0.0347, 0.0299, 0.0257, 0.0290], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 02:16:29,779 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 5.128e+02 7.436e+02 9.242e+02 1.292e+03 2.692e+03, threshold=1.848e+03, percent-clipped=7.0 2023-04-01 02:16:33,307 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 2023-04-01 02:16:33,595 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=21529.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:16:40,202 INFO [train.py:903] (1/4) Epoch 4, batch 1050, loss[loss=0.2983, simple_loss=0.3531, pruned_loss=0.1217, over 19603.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3604, pruned_loss=0.127, over 3812027.89 frames. ], batch size: 52, lr: 2.07e-02, grad_scale: 8.0 2023-04-01 02:16:41,963 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 2023-04-01 02:17:14,216 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 2023-04-01 02:17:40,182 INFO [train.py:903] (1/4) Epoch 4, batch 1100, loss[loss=0.2738, simple_loss=0.3382, pruned_loss=0.1048, over 19779.00 frames. ], tot_loss[loss=0.307, simple_loss=0.3598, pruned_loss=0.1271, over 3805588.89 frames. ], batch size: 54, lr: 2.07e-02, grad_scale: 8.0 2023-04-01 02:17:48,981 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 2023-04-01 02:17:52,945 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=21595.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:18:10,462 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1416, 0.9777, 1.2680, 0.8693, 2.2886, 2.9485, 2.8803, 3.2113], device='cuda:1'), covar=tensor([0.1733, 0.4142, 0.4108, 0.2420, 0.0570, 0.0232, 0.0324, 0.0157], device='cuda:1'), in_proj_covar=tensor([0.0264, 0.0278, 0.0319, 0.0261, 0.0190, 0.0109, 0.0198, 0.0121], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0001, 0.0002, 0.0001], device='cuda:1') 2023-04-01 02:18:15,012 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=21613.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:18:30,558 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.767e+02 7.921e+02 9.622e+02 1.275e+03 2.981e+03, threshold=1.924e+03, percent-clipped=6.0 2023-04-01 02:18:36,426 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=21631.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:18:43,698 INFO [train.py:903] (1/4) Epoch 4, batch 1150, loss[loss=0.3015, simple_loss=0.3618, pruned_loss=0.1206, over 18763.00 frames. ], tot_loss[loss=0.3095, simple_loss=0.3623, pruned_loss=0.1283, over 3807504.44 frames. ], batch size: 74, lr: 2.07e-02, grad_scale: 8.0 2023-04-01 02:19:30,211 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 2023-04-01 02:19:30,749 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=21674.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:19:45,550 INFO [train.py:903] (1/4) Epoch 4, batch 1200, loss[loss=0.285, simple_loss=0.3501, pruned_loss=0.1099, over 19353.00 frames. ], tot_loss[loss=0.307, simple_loss=0.3605, pruned_loss=0.1268, over 3795926.03 frames. ], batch size: 66, lr: 2.07e-02, grad_scale: 8.0 2023-04-01 02:19:51,427 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=21690.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:20:14,901 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 2023-04-01 02:20:17,275 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1890, 0.9771, 1.3971, 1.1085, 2.4173, 3.4022, 3.2213, 3.6923], device='cuda:1'), covar=tensor([0.1682, 0.3926, 0.3725, 0.2250, 0.0547, 0.0134, 0.0293, 0.0101], device='cuda:1'), in_proj_covar=tensor([0.0266, 0.0279, 0.0323, 0.0262, 0.0193, 0.0108, 0.0201, 0.0121], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0001, 0.0002, 0.0001], device='cuda:1') 2023-04-01 02:20:35,528 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.245e+02 7.515e+02 9.038e+02 1.111e+03 2.454e+03, threshold=1.808e+03, percent-clipped=2.0 2023-04-01 02:20:45,434 INFO [train.py:903] (1/4) Epoch 4, batch 1250, loss[loss=0.3548, simple_loss=0.4005, pruned_loss=0.1546, over 19671.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.3608, pruned_loss=0.1271, over 3810920.16 frames. ], batch size: 58, lr: 2.06e-02, grad_scale: 8.0 2023-04-01 02:21:39,800 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=21781.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:21:44,769 INFO [train.py:903] (1/4) Epoch 4, batch 1300, loss[loss=0.2508, simple_loss=0.3091, pruned_loss=0.09627, over 19779.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.362, pruned_loss=0.1277, over 3808961.52 frames. ], batch size: 46, lr: 2.06e-02, grad_scale: 8.0 2023-04-01 02:21:45,186 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=21785.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:21:49,611 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=21789.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:22:08,435 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.46 vs. limit=2.0 2023-04-01 02:22:10,063 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=21805.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:22:15,769 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21810.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:22:34,650 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.842e+02 7.410e+02 9.034e+02 1.121e+03 1.849e+03, threshold=1.807e+03, percent-clipped=1.0 2023-04-01 02:22:45,117 INFO [train.py:903] (1/4) Epoch 4, batch 1350, loss[loss=0.3451, simple_loss=0.3898, pruned_loss=0.1502, over 19652.00 frames. ], tot_loss[loss=0.3068, simple_loss=0.3603, pruned_loss=0.1267, over 3819611.65 frames. ], batch size: 55, lr: 2.06e-02, grad_scale: 8.0 2023-04-01 02:23:04,109 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=21851.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:23:23,939 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=21868.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:23:33,958 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21876.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:23:45,406 INFO [train.py:903] (1/4) Epoch 4, batch 1400, loss[loss=0.3878, simple_loss=0.4197, pruned_loss=0.1779, over 19133.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3593, pruned_loss=0.1267, over 3820459.34 frames. ], batch size: 69, lr: 2.06e-02, grad_scale: 8.0 2023-04-01 02:24:35,700 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.358e+02 7.364e+02 9.517e+02 1.310e+03 2.254e+03, threshold=1.903e+03, percent-clipped=6.0 2023-04-01 02:24:42,745 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_rvb from training. Duration: 25.85 2023-04-01 02:24:46,194 INFO [train.py:903] (1/4) Epoch 4, batch 1450, loss[loss=0.2563, simple_loss=0.3118, pruned_loss=0.1004, over 19733.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.3589, pruned_loss=0.1265, over 3816271.78 frames. ], batch size: 46, lr: 2.05e-02, grad_scale: 8.0 2023-04-01 02:24:48,096 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 2023-04-01 02:24:54,260 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9986, 1.1632, 1.1907, 1.9001, 1.6285, 1.9452, 2.2953, 1.7150], device='cuda:1'), covar=tensor([0.0713, 0.1219, 0.1432, 0.1238, 0.1198, 0.0834, 0.0900, 0.0832], device='cuda:1'), in_proj_covar=tensor([0.0246, 0.0269, 0.0267, 0.0302, 0.0303, 0.0245, 0.0265, 0.0239], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:1') 2023-04-01 02:25:13,215 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=21957.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:25:34,253 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=21975.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:25:45,922 INFO [train.py:903] (1/4) Epoch 4, batch 1500, loss[loss=0.3497, simple_loss=0.3967, pruned_loss=0.1514, over 19125.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.3599, pruned_loss=0.1276, over 3823226.46 frames. ], batch size: 69, lr: 2.05e-02, grad_scale: 8.0 2023-04-01 02:25:49,637 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=21988.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:26:36,547 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.570e+02 6.805e+02 9.241e+02 1.170e+03 2.581e+03, threshold=1.848e+03, percent-clipped=2.0 2023-04-01 02:26:47,125 INFO [train.py:903] (1/4) Epoch 4, batch 1550, loss[loss=0.3019, simple_loss=0.3647, pruned_loss=0.1195, over 19617.00 frames. ], tot_loss[loss=0.3051, simple_loss=0.358, pruned_loss=0.1261, over 3822710.46 frames. ], batch size: 57, lr: 2.05e-02, grad_scale: 8.0 2023-04-01 02:27:01,643 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22045.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:27:08,632 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.66 vs. limit=5.0 2023-04-01 02:27:19,321 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22061.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:27:30,468 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22070.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:27:32,771 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22072.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:27:49,224 INFO [train.py:903] (1/4) Epoch 4, batch 1600, loss[loss=0.2579, simple_loss=0.3195, pruned_loss=0.09819, over 19611.00 frames. ], tot_loss[loss=0.3057, simple_loss=0.3582, pruned_loss=0.1266, over 3829663.80 frames. ], batch size: 50, lr: 2.05e-02, grad_scale: 8.0 2023-04-01 02:27:50,752 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22086.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:27:55,259 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22090.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:28:10,563 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9_rvb from training. Duration: 30.1555625 2023-04-01 02:28:37,641 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=22125.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:28:39,601 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.648e+02 8.565e+02 1.081e+03 1.346e+03 3.673e+03, threshold=2.162e+03, percent-clipped=6.0 2023-04-01 02:28:44,589 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5899, 1.2290, 1.3219, 1.8592, 1.5436, 1.7115, 2.0449, 1.6274], device='cuda:1'), covar=tensor([0.0789, 0.1180, 0.1220, 0.1000, 0.1039, 0.0801, 0.0812, 0.0718], device='cuda:1'), in_proj_covar=tensor([0.0243, 0.0269, 0.0265, 0.0301, 0.0300, 0.0245, 0.0265, 0.0238], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:1') 2023-04-01 02:28:48,843 INFO [train.py:903] (1/4) Epoch 4, batch 1650, loss[loss=0.3224, simple_loss=0.3782, pruned_loss=0.1333, over 19673.00 frames. ], tot_loss[loss=0.3057, simple_loss=0.3584, pruned_loss=0.1264, over 3827250.67 frames. ], batch size: 60, lr: 2.05e-02, grad_scale: 4.0 2023-04-01 02:29:04,258 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22148.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:29:12,053 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.9064, 2.3225, 2.2453, 2.9662, 2.7076, 2.5120, 2.5410, 2.9862], device='cuda:1'), covar=tensor([0.0690, 0.1647, 0.1177, 0.0710, 0.0993, 0.0413, 0.0775, 0.0459], device='cuda:1'), in_proj_covar=tensor([0.0254, 0.0368, 0.0280, 0.0245, 0.0301, 0.0251, 0.0260, 0.0234], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 02:29:18,652 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3019, 1.0720, 1.5088, 1.1148, 2.6752, 3.5302, 3.4097, 3.7436], device='cuda:1'), covar=tensor([0.1320, 0.2815, 0.2730, 0.1924, 0.0387, 0.0111, 0.0200, 0.0102], device='cuda:1'), in_proj_covar=tensor([0.0266, 0.0278, 0.0322, 0.0261, 0.0193, 0.0111, 0.0200, 0.0122], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0001, 0.0002, 0.0001], device='cuda:1') 2023-04-01 02:29:49,248 INFO [train.py:903] (1/4) Epoch 4, batch 1700, loss[loss=0.3227, simple_loss=0.3743, pruned_loss=0.1355, over 19278.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.3588, pruned_loss=0.1264, over 3828331.83 frames. ], batch size: 66, lr: 2.04e-02, grad_scale: 4.0 2023-04-01 02:30:00,383 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22194.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:30:23,322 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=22212.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:30:25,807 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.6544, 1.7970, 1.7821, 2.2439, 4.3072, 1.2300, 2.3385, 4.1463], device='cuda:1'), covar=tensor([0.0226, 0.1951, 0.1996, 0.1191, 0.0331, 0.1939, 0.1092, 0.0279], device='cuda:1'), in_proj_covar=tensor([0.0261, 0.0298, 0.0299, 0.0272, 0.0284, 0.0311, 0.0270, 0.0280], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-01 02:30:27,862 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 2023-04-01 02:30:40,332 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.920e+02 6.276e+02 7.753e+02 9.050e+02 1.909e+03, threshold=1.551e+03, percent-clipped=1.0 2023-04-01 02:30:48,105 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22233.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:30:49,927 INFO [train.py:903] (1/4) Epoch 4, batch 1750, loss[loss=0.2958, simple_loss=0.3376, pruned_loss=0.127, over 19852.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.3599, pruned_loss=0.1267, over 3815582.66 frames. ], batch size: 52, lr: 2.04e-02, grad_scale: 4.0 2023-04-01 02:30:57,753 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22240.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:31:15,616 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22255.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:31:50,782 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4792, 2.4243, 1.4573, 1.8084, 2.0931, 1.2023, 1.1729, 1.8827], device='cuda:1'), covar=tensor([0.0924, 0.0402, 0.0884, 0.0494, 0.0403, 0.0928, 0.0751, 0.0385], device='cuda:1'), in_proj_covar=tensor([0.0278, 0.0242, 0.0309, 0.0245, 0.0212, 0.0306, 0.0278, 0.0245], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 02:31:52,438 INFO [train.py:903] (1/4) Epoch 4, batch 1800, loss[loss=0.3218, simple_loss=0.3689, pruned_loss=0.1373, over 19652.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3598, pruned_loss=0.1266, over 3829422.64 frames. ], batch size: 53, lr: 2.04e-02, grad_scale: 4.0 2023-04-01 02:32:43,154 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.733e+02 7.313e+02 8.961e+02 1.128e+03 3.443e+03, threshold=1.792e+03, percent-clipped=8.0 2023-04-01 02:32:43,415 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22327.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:32:44,439 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22328.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:32:46,329 WARNING [train.py:1073] (1/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9_rvb from training. Duration: 25.0944375 2023-04-01 02:32:48,652 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=22332.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:32:52,003 INFO [train.py:903] (1/4) Epoch 4, batch 1850, loss[loss=0.257, simple_loss=0.3131, pruned_loss=0.1005, over 19737.00 frames. ], tot_loss[loss=0.306, simple_loss=0.3592, pruned_loss=0.1264, over 3830033.85 frames. ], batch size: 45, lr: 2.04e-02, grad_scale: 4.0 2023-04-01 02:33:00,794 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.85 vs. limit=2.0 2023-04-01 02:33:04,821 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22346.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:33:13,957 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22353.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:33:15,100 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.0827, 1.1683, 1.4604, 0.4993, 2.3218, 2.4203, 2.0987, 2.5600], device='cuda:1'), covar=tensor([0.1191, 0.2725, 0.2606, 0.2192, 0.0378, 0.0164, 0.0365, 0.0153], device='cuda:1'), in_proj_covar=tensor([0.0263, 0.0278, 0.0322, 0.0259, 0.0191, 0.0109, 0.0199, 0.0121], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0001, 0.0002, 0.0001], device='cuda:1') 2023-04-01 02:33:24,813 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9_rvb from training. Duration: 27.8166875 2023-04-01 02:33:36,225 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22371.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:33:37,407 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2273, 1.2045, 2.0366, 1.5154, 3.1789, 2.8345, 3.4496, 1.2886], device='cuda:1'), covar=tensor([0.1735, 0.2952, 0.1538, 0.1339, 0.1079, 0.1160, 0.1353, 0.2656], device='cuda:1'), in_proj_covar=tensor([0.0426, 0.0471, 0.0441, 0.0395, 0.0514, 0.0410, 0.0592, 0.0422], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 02:33:39,600 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.0288, 1.0353, 1.3347, 1.1058, 1.5987, 1.5588, 1.6875, 0.5275], device='cuda:1'), covar=tensor([0.1500, 0.2448, 0.1269, 0.1265, 0.0930, 0.1215, 0.0898, 0.2169], device='cuda:1'), in_proj_covar=tensor([0.0425, 0.0470, 0.0440, 0.0394, 0.0513, 0.0409, 0.0590, 0.0421], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 02:33:51,847 INFO [train.py:903] (1/4) Epoch 4, batch 1900, loss[loss=0.3385, simple_loss=0.3912, pruned_loss=0.1429, over 19788.00 frames. ], tot_loss[loss=0.3047, simple_loss=0.3583, pruned_loss=0.1256, over 3840243.56 frames. ], batch size: 56, lr: 2.03e-02, grad_scale: 4.0 2023-04-01 02:34:09,607 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 2023-04-01 02:34:14,802 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9_rvb from training. Duration: 27.02225 2023-04-01 02:34:15,074 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22403.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:34:36,874 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.53 vs. limit=5.0 2023-04-01 02:34:39,429 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 2023-04-01 02:34:42,867 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.044e+02 7.549e+02 9.520e+02 1.192e+03 3.384e+03, threshold=1.904e+03, percent-clipped=5.0 2023-04-01 02:34:51,945 INFO [train.py:903] (1/4) Epoch 4, batch 1950, loss[loss=0.2625, simple_loss=0.3139, pruned_loss=0.1056, over 19747.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3569, pruned_loss=0.1253, over 3830997.78 frames. ], batch size: 46, lr: 2.03e-02, grad_scale: 4.0 2023-04-01 02:34:56,976 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.20 vs. limit=5.0 2023-04-01 02:35:08,568 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22447.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:35:53,485 INFO [train.py:903] (1/4) Epoch 4, batch 2000, loss[loss=0.3741, simple_loss=0.4043, pruned_loss=0.1719, over 12982.00 frames. ], tot_loss[loss=0.3048, simple_loss=0.358, pruned_loss=0.1257, over 3823846.38 frames. ], batch size: 136, lr: 2.03e-02, grad_scale: 8.0 2023-04-01 02:36:01,813 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=22492.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:36:06,598 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22496.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:36:37,353 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22521.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:36:45,511 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.709e+02 6.735e+02 8.799e+02 1.102e+03 2.294e+03, threshold=1.760e+03, percent-clipped=1.0 2023-04-01 02:36:46,719 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_rvb from training. Duration: 26.8349375 2023-04-01 02:36:54,492 INFO [train.py:903] (1/4) Epoch 4, batch 2050, loss[loss=0.4219, simple_loss=0.4537, pruned_loss=0.1951, over 19344.00 frames. ], tot_loss[loss=0.3039, simple_loss=0.3577, pruned_loss=0.1251, over 3803486.53 frames. ], batch size: 66, lr: 2.03e-02, grad_scale: 8.0 2023-04-01 02:36:58,157 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=22538.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:37:04,877 WARNING [train.py:1073] (1/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 2023-04-01 02:37:06,032 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 2023-04-01 02:37:27,447 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 2023-04-01 02:37:40,490 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 2023-04-01 02:37:45,363 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=22577.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:37:52,332 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22583.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:37:54,190 INFO [train.py:903] (1/4) Epoch 4, batch 2100, loss[loss=0.2864, simple_loss=0.3454, pruned_loss=0.1137, over 19666.00 frames. ], tot_loss[loss=0.3045, simple_loss=0.3582, pruned_loss=0.1254, over 3817417.84 frames. ], batch size: 53, lr: 2.03e-02, grad_scale: 8.0 2023-04-01 02:38:10,699 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=22599.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:38:21,642 WARNING [train.py:1073] (1/4) Exclude cut with ID 453-131332-0000-131866_sp0.9_rvb from training. Duration: 25.3333125 2023-04-01 02:38:22,048 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22607.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:38:23,350 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22608.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:38:42,451 WARNING [train.py:1073] (1/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9_rvb from training. Duration: 26.6166875 2023-04-01 02:38:44,807 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.772e+02 6.954e+02 8.861e+02 1.126e+03 2.028e+03, threshold=1.772e+03, percent-clipped=3.0 2023-04-01 02:38:52,955 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.5602, 1.9299, 1.9701, 2.2994, 1.8014, 2.6730, 2.5280, 2.6919], device='cuda:1'), covar=tensor([0.0632, 0.0900, 0.0989, 0.1040, 0.1099, 0.0603, 0.0913, 0.0493], device='cuda:1'), in_proj_covar=tensor([0.0246, 0.0264, 0.0263, 0.0298, 0.0298, 0.0246, 0.0262, 0.0236], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:1') 2023-04-01 02:38:53,754 INFO [train.py:903] (1/4) Epoch 4, batch 2150, loss[loss=0.3033, simple_loss=0.3603, pruned_loss=0.1232, over 19342.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.3588, pruned_loss=0.1258, over 3821948.33 frames. ], batch size: 66, lr: 2.02e-02, grad_scale: 8.0 2023-04-01 02:39:17,733 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22653.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:39:22,203 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22657.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:39:56,340 INFO [train.py:903] (1/4) Epoch 4, batch 2200, loss[loss=0.3194, simple_loss=0.3812, pruned_loss=0.1288, over 19290.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3594, pruned_loss=0.1264, over 3792780.96 frames. ], batch size: 66, lr: 2.02e-02, grad_scale: 8.0 2023-04-01 02:40:05,420 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22692.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:40:18,048 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22703.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:40:30,198 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22714.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:40:46,719 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1823, 0.9555, 0.9831, 1.4526, 1.2555, 1.3520, 1.4949, 1.1956], device='cuda:1'), covar=tensor([0.0697, 0.0906, 0.0960, 0.0672, 0.0747, 0.0675, 0.0705, 0.0614], device='cuda:1'), in_proj_covar=tensor([0.0246, 0.0264, 0.0264, 0.0298, 0.0296, 0.0248, 0.0263, 0.0238], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:1') 2023-04-01 02:40:47,464 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 5.077e+02 6.866e+02 8.417e+02 1.098e+03 2.160e+03, threshold=1.683e+03, percent-clipped=4.0 2023-04-01 02:40:49,019 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22728.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:40:56,701 INFO [train.py:903] (1/4) Epoch 4, batch 2250, loss[loss=0.2831, simple_loss=0.3516, pruned_loss=0.1073, over 19564.00 frames. ], tot_loss[loss=0.3074, simple_loss=0.3604, pruned_loss=0.1272, over 3791224.42 frames. ], batch size: 61, lr: 2.02e-02, grad_scale: 8.0 2023-04-01 02:41:10,274 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=22747.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:41:57,019 INFO [train.py:903] (1/4) Epoch 4, batch 2300, loss[loss=0.3084, simple_loss=0.3618, pruned_loss=0.1275, over 19592.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3597, pruned_loss=0.1262, over 3804934.06 frames. ], batch size: 52, lr: 2.02e-02, grad_scale: 8.0 2023-04-01 02:42:09,275 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_rvb from training. Duration: 26.205 2023-04-01 02:42:31,170 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22812.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:42:47,875 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.632e+02 7.223e+02 9.387e+02 1.200e+03 1.860e+03, threshold=1.877e+03, percent-clipped=8.0 2023-04-01 02:42:56,885 INFO [train.py:903] (1/4) Epoch 4, batch 2350, loss[loss=0.2913, simple_loss=0.3377, pruned_loss=0.1224, over 19715.00 frames. ], tot_loss[loss=0.3066, simple_loss=0.3596, pruned_loss=0.1267, over 3806734.61 frames. ], batch size: 45, lr: 2.01e-02, grad_scale: 8.0 2023-04-01 02:43:06,745 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22843.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:43:30,995 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22862.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:43:32,262 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22863.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:43:38,744 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_rvb from training. Duration: 25.775 2023-04-01 02:43:54,379 WARNING [train.py:1073] (1/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9_rvb from training. Duration: 25.45 2023-04-01 02:43:57,664 INFO [train.py:903] (1/4) Epoch 4, batch 2400, loss[loss=0.2977, simple_loss=0.3515, pruned_loss=0.1219, over 19681.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3603, pruned_loss=0.1271, over 3803410.23 frames. ], batch size: 53, lr: 2.01e-02, grad_scale: 8.0 2023-04-01 02:44:03,297 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22888.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:44:27,690 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22909.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:44:49,610 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.703e+02 7.425e+02 9.466e+02 1.182e+03 3.064e+03, threshold=1.893e+03, percent-clipped=2.0 2023-04-01 02:44:56,186 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.45 vs. limit=2.0 2023-04-01 02:44:58,858 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22934.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:44:59,530 INFO [train.py:903] (1/4) Epoch 4, batch 2450, loss[loss=0.3978, simple_loss=0.415, pruned_loss=0.1903, over 13911.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.3605, pruned_loss=0.1272, over 3795895.85 frames. ], batch size: 137, lr: 2.01e-02, grad_scale: 8.0 2023-04-01 02:45:14,076 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22948.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:45:31,525 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1038, 1.1648, 1.4225, 1.2381, 1.7954, 1.7418, 1.9328, 0.5352], device='cuda:1'), covar=tensor([0.1409, 0.2338, 0.1306, 0.1179, 0.0857, 0.1141, 0.0815, 0.2189], device='cuda:1'), in_proj_covar=tensor([0.0425, 0.0477, 0.0449, 0.0398, 0.0516, 0.0416, 0.0599, 0.0423], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 02:45:40,908 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22970.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:45:44,372 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22973.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:45:57,747 INFO [train.py:903] (1/4) Epoch 4, batch 2500, loss[loss=0.3038, simple_loss=0.3567, pruned_loss=0.1254, over 19663.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3619, pruned_loss=0.1282, over 3809682.93 frames. ], batch size: 53, lr: 2.01e-02, grad_scale: 8.0 2023-04-01 02:46:09,322 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22995.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:46:16,262 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=23001.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:46:35,134 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3482, 1.1614, 1.4212, 1.1883, 2.6491, 3.3820, 3.3036, 3.6894], device='cuda:1'), covar=tensor([0.1366, 0.2793, 0.2893, 0.2015, 0.0458, 0.0150, 0.0192, 0.0102], device='cuda:1'), in_proj_covar=tensor([0.0266, 0.0282, 0.0323, 0.0260, 0.0194, 0.0109, 0.0200, 0.0123], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0002, 0.0001, 0.0002, 0.0001], device='cuda:1') 2023-04-01 02:46:48,542 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.107e+02 7.308e+02 8.619e+02 1.083e+03 2.930e+03, threshold=1.724e+03, percent-clipped=3.0 2023-04-01 02:46:57,735 INFO [train.py:903] (1/4) Epoch 4, batch 2550, loss[loss=0.3312, simple_loss=0.3852, pruned_loss=0.1386, over 19167.00 frames. ], tot_loss[loss=0.3074, simple_loss=0.3607, pruned_loss=0.127, over 3801071.99 frames. ], batch size: 69, lr: 2.01e-02, grad_scale: 8.0 2023-04-01 02:47:47,719 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6594, 1.4216, 1.9486, 1.7282, 2.8875, 4.5525, 4.4987, 4.9483], device='cuda:1'), covar=tensor([0.1287, 0.2598, 0.2472, 0.1769, 0.0443, 0.0091, 0.0124, 0.0061], device='cuda:1'), in_proj_covar=tensor([0.0265, 0.0279, 0.0319, 0.0261, 0.0193, 0.0107, 0.0200, 0.0124], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0001, 0.0002, 0.0001], device='cuda:1') 2023-04-01 02:47:49,710 WARNING [train.py:1073] (1/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 2023-04-01 02:47:57,382 INFO [train.py:903] (1/4) Epoch 4, batch 2600, loss[loss=0.3734, simple_loss=0.4045, pruned_loss=0.1712, over 12720.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3596, pruned_loss=0.1265, over 3807996.36 frames. ], batch size: 136, lr: 2.00e-02, grad_scale: 8.0 2023-04-01 02:48:33,813 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=23116.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:48:36,237 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=23118.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:48:46,620 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=23126.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:48:48,301 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.619e+02 6.946e+02 8.555e+02 1.074e+03 2.756e+03, threshold=1.711e+03, percent-clipped=6.0 2023-04-01 02:48:58,589 INFO [train.py:903] (1/4) Epoch 4, batch 2650, loss[loss=0.2799, simple_loss=0.3499, pruned_loss=0.105, over 19702.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3593, pruned_loss=0.1265, over 3797446.14 frames. ], batch size: 59, lr: 2.00e-02, grad_scale: 8.0 2023-04-01 02:49:08,268 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=23143.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:49:17,081 WARNING [train.py:1073] (1/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9_rvb from training. Duration: 27.25 2023-04-01 02:49:22,930 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=23156.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:49:58,046 INFO [train.py:903] (1/4) Epoch 4, batch 2700, loss[loss=0.3071, simple_loss=0.3593, pruned_loss=0.1275, over 17524.00 frames. ], tot_loss[loss=0.304, simple_loss=0.3578, pruned_loss=0.1251, over 3802509.33 frames. ], batch size: 101, lr: 2.00e-02, grad_scale: 8.0 2023-04-01 02:50:00,483 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=23187.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:50:48,019 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.628e+02 7.002e+02 8.947e+02 1.091e+03 2.361e+03, threshold=1.789e+03, percent-clipped=7.0 2023-04-01 02:50:57,145 INFO [train.py:903] (1/4) Epoch 4, batch 2750, loss[loss=0.2357, simple_loss=0.3004, pruned_loss=0.08547, over 19776.00 frames. ], tot_loss[loss=0.3035, simple_loss=0.357, pruned_loss=0.125, over 3803203.66 frames. ], batch size: 47, lr: 2.00e-02, grad_scale: 8.0 2023-04-01 02:50:57,417 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=23235.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:51:11,540 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=23247.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:51:41,024 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=23271.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:51:56,445 INFO [train.py:903] (1/4) Epoch 4, batch 2800, loss[loss=0.3462, simple_loss=0.3988, pruned_loss=0.1468, over 19614.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.3591, pruned_loss=0.1259, over 3798787.10 frames. ], batch size: 57, lr: 2.00e-02, grad_scale: 8.0 2023-04-01 02:52:17,019 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=23302.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:52:31,864 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.81 vs. limit=2.0 2023-04-01 02:52:45,199 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.435e+02 7.909e+02 1.044e+03 1.347e+03 2.323e+03, threshold=2.087e+03, percent-clipped=7.0 2023-04-01 02:52:56,802 INFO [train.py:903] (1/4) Epoch 4, batch 2850, loss[loss=0.3547, simple_loss=0.3958, pruned_loss=0.1568, over 19678.00 frames. ], tot_loss[loss=0.3048, simple_loss=0.3587, pruned_loss=0.1255, over 3802863.06 frames. ], batch size: 53, lr: 1.99e-02, grad_scale: 8.0 2023-04-01 02:52:57,426 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=4.14 vs. limit=5.0 2023-04-01 02:53:41,811 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=23372.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:53:56,270 INFO [train.py:903] (1/4) Epoch 4, batch 2900, loss[loss=0.2873, simple_loss=0.3487, pruned_loss=0.1129, over 19672.00 frames. ], tot_loss[loss=0.3032, simple_loss=0.3573, pruned_loss=0.1246, over 3805925.29 frames. ], batch size: 53, lr: 1.99e-02, grad_scale: 4.0 2023-04-01 02:53:56,290 WARNING [train.py:1073] (1/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 2023-04-01 02:54:10,043 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=23397.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:54:44,138 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 2023-04-01 02:54:45,629 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.212e+02 7.852e+02 1.023e+03 1.284e+03 2.319e+03, threshold=2.047e+03, percent-clipped=2.0 2023-04-01 02:54:53,633 INFO [train.py:903] (1/4) Epoch 4, batch 2950, loss[loss=0.3302, simple_loss=0.3864, pruned_loss=0.137, over 19516.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3577, pruned_loss=0.125, over 3813601.06 frames. ], batch size: 64, lr: 1.99e-02, grad_scale: 4.0 2023-04-01 02:55:03,494 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 2023-04-01 02:55:15,417 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 2023-04-01 02:55:21,784 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.1506, 2.6861, 2.1136, 2.1236, 1.8850, 2.3435, 0.7604, 2.0692], device='cuda:1'), covar=tensor([0.0301, 0.0213, 0.0201, 0.0324, 0.0448, 0.0353, 0.0506, 0.0433], device='cuda:1'), in_proj_covar=tensor([0.0272, 0.0261, 0.0261, 0.0288, 0.0350, 0.0276, 0.0266, 0.0283], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 02:55:35,111 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=23470.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:55:52,223 INFO [train.py:903] (1/4) Epoch 4, batch 3000, loss[loss=0.3131, simple_loss=0.3566, pruned_loss=0.1348, over 19844.00 frames. ], tot_loss[loss=0.3026, simple_loss=0.3571, pruned_loss=0.124, over 3821486.26 frames. ], batch size: 52, lr: 1.99e-02, grad_scale: 4.0 2023-04-01 02:55:52,223 INFO [train.py:928] (1/4) Computing validation loss 2023-04-01 02:56:05,134 INFO [train.py:937] (1/4) Epoch 4, validation: loss=0.2145, simple_loss=0.3118, pruned_loss=0.05862, over 944034.00 frames. 2023-04-01 02:56:05,135 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18649MB 2023-04-01 02:56:09,790 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_rvb from training. Duration: 29.735 2023-04-01 02:56:32,401 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=23506.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:56:56,155 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=23527.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:56:56,886 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.281e+02 6.234e+02 7.977e+02 1.046e+03 2.333e+03, threshold=1.595e+03, percent-clipped=2.0 2023-04-01 02:57:05,034 INFO [train.py:903] (1/4) Epoch 4, batch 3050, loss[loss=0.2866, simple_loss=0.3475, pruned_loss=0.1128, over 19620.00 frames. ], tot_loss[loss=0.3034, simple_loss=0.3578, pruned_loss=0.1245, over 3819668.45 frames. ], batch size: 50, lr: 1.99e-02, grad_scale: 4.0 2023-04-01 02:57:26,951 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=23552.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:57:28,441 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 2023-04-01 02:57:33,681 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=23558.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:57:53,409 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.92 vs. limit=5.0 2023-04-01 02:57:58,299 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=23579.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:58:04,150 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=23583.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:58:06,131 INFO [train.py:903] (1/4) Epoch 4, batch 3100, loss[loss=0.2497, simple_loss=0.3097, pruned_loss=0.09491, over 19751.00 frames. ], tot_loss[loss=0.3035, simple_loss=0.3579, pruned_loss=0.1246, over 3801279.16 frames. ], batch size: 45, lr: 1.98e-02, grad_scale: 4.0 2023-04-01 02:58:06,480 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=23585.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:58:13,285 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=23591.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 02:58:55,942 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.118e+02 6.915e+02 8.546e+02 1.092e+03 2.878e+03, threshold=1.709e+03, percent-clipped=7.0 2023-04-01 02:59:03,920 INFO [train.py:903] (1/4) Epoch 4, batch 3150, loss[loss=0.35, simple_loss=0.3949, pruned_loss=0.1526, over 18438.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.3586, pruned_loss=0.1253, over 3803190.89 frames. ], batch size: 84, lr: 1.98e-02, grad_scale: 4.0 2023-04-01 02:59:28,008 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 2023-04-01 03:00:02,534 INFO [train.py:903] (1/4) Epoch 4, batch 3200, loss[loss=0.3138, simple_loss=0.3679, pruned_loss=0.1298, over 19504.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3595, pruned_loss=0.1264, over 3804308.66 frames. ], batch size: 64, lr: 1.98e-02, grad_scale: 8.0 2023-04-01 03:00:13,168 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=23694.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 03:00:29,436 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=23706.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 03:00:53,663 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.799e+02 7.379e+02 9.197e+02 1.143e+03 1.957e+03, threshold=1.839e+03, percent-clipped=5.0 2023-04-01 03:01:02,115 INFO [train.py:903] (1/4) Epoch 4, batch 3250, loss[loss=0.2918, simple_loss=0.3567, pruned_loss=0.1135, over 19658.00 frames. ], tot_loss[loss=0.3044, simple_loss=0.3584, pruned_loss=0.1252, over 3807372.65 frames. ], batch size: 55, lr: 1.98e-02, grad_scale: 8.0 2023-04-01 03:02:01,903 INFO [train.py:903] (1/4) Epoch 4, batch 3300, loss[loss=0.2988, simple_loss=0.3507, pruned_loss=0.1234, over 19585.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3588, pruned_loss=0.1257, over 3809317.85 frames. ], batch size: 52, lr: 1.98e-02, grad_scale: 8.0 2023-04-01 03:02:04,053 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 2023-04-01 03:02:29,554 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.0332, 1.0138, 1.4531, 1.1468, 1.7727, 1.6868, 1.9699, 0.6329], device='cuda:1'), covar=tensor([0.1314, 0.2213, 0.1139, 0.1129, 0.0787, 0.1019, 0.0782, 0.1981], device='cuda:1'), in_proj_covar=tensor([0.0433, 0.0485, 0.0452, 0.0404, 0.0523, 0.0428, 0.0609, 0.0436], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 03:02:54,104 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.057e+02 7.772e+02 9.614e+02 1.210e+03 2.492e+03, threshold=1.923e+03, percent-clipped=5.0 2023-04-01 03:03:02,118 INFO [train.py:903] (1/4) Epoch 4, batch 3350, loss[loss=0.2466, simple_loss=0.3025, pruned_loss=0.09535, over 19799.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.359, pruned_loss=0.1257, over 3805672.52 frames. ], batch size: 48, lr: 1.97e-02, grad_scale: 8.0 2023-04-01 03:03:09,292 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=23841.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 03:03:19,201 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=23850.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 03:03:40,413 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=23866.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 03:03:59,993 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.2321, 2.9519, 1.9562, 2.8022, 1.0331, 2.7863, 2.6822, 2.7351], device='cuda:1'), covar=tensor([0.0927, 0.1246, 0.2021, 0.0877, 0.3555, 0.1086, 0.0857, 0.1250], device='cuda:1'), in_proj_covar=tensor([0.0329, 0.0295, 0.0345, 0.0279, 0.0345, 0.0295, 0.0258, 0.0294], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 03:04:01,842 INFO [train.py:903] (1/4) Epoch 4, batch 3400, loss[loss=0.3259, simple_loss=0.3821, pruned_loss=0.1349, over 19344.00 frames. ], tot_loss[loss=0.303, simple_loss=0.3571, pruned_loss=0.1244, over 3814094.99 frames. ], batch size: 70, lr: 1.97e-02, grad_scale: 8.0 2023-04-01 03:04:53,666 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.081e+02 7.350e+02 9.318e+02 1.202e+03 2.145e+03, threshold=1.864e+03, percent-clipped=3.0 2023-04-01 03:05:01,726 INFO [train.py:903] (1/4) Epoch 4, batch 3450, loss[loss=0.2649, simple_loss=0.3248, pruned_loss=0.1025, over 16045.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.3584, pruned_loss=0.1254, over 3796541.76 frames. ], batch size: 35, lr: 1.97e-02, grad_scale: 8.0 2023-04-01 03:05:01,750 WARNING [train.py:1073] (1/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9_rvb from training. Duration: 25.2444375 2023-04-01 03:05:16,206 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0 2023-04-01 03:05:22,649 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=23950.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 03:05:36,134 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=23962.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 03:05:39,402 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=23965.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 03:05:50,599 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=23975.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 03:06:04,980 INFO [train.py:903] (1/4) Epoch 4, batch 3500, loss[loss=0.2975, simple_loss=0.3496, pruned_loss=0.1226, over 19398.00 frames. ], tot_loss[loss=0.3057, simple_loss=0.3594, pruned_loss=0.126, over 3807929.05 frames. ], batch size: 48, lr: 1.97e-02, grad_scale: 8.0 2023-04-01 03:06:07,715 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=23987.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 03:06:14,718 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.8534, 1.4893, 1.3545, 1.9668, 1.3896, 2.0576, 2.1804, 1.8577], device='cuda:1'), covar=tensor([0.0742, 0.1039, 0.1246, 0.1020, 0.1144, 0.0735, 0.0952, 0.0650], device='cuda:1'), in_proj_covar=tensor([0.0245, 0.0264, 0.0260, 0.0293, 0.0296, 0.0249, 0.0256, 0.0238], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:1') 2023-04-01 03:06:58,144 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.373e+02 7.196e+02 8.612e+02 1.120e+03 2.630e+03, threshold=1.722e+03, percent-clipped=5.0 2023-04-01 03:07:06,280 INFO [train.py:903] (1/4) Epoch 4, batch 3550, loss[loss=0.2514, simple_loss=0.3226, pruned_loss=0.09011, over 19845.00 frames. ], tot_loss[loss=0.3043, simple_loss=0.358, pruned_loss=0.1253, over 3822159.89 frames. ], batch size: 52, lr: 1.97e-02, grad_scale: 8.0 2023-04-01 03:07:28,433 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=24055.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 03:08:05,263 INFO [train.py:903] (1/4) Epoch 4, batch 3600, loss[loss=0.3296, simple_loss=0.3887, pruned_loss=0.1353, over 18879.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.358, pruned_loss=0.1257, over 3819027.33 frames. ], batch size: 74, lr: 1.96e-02, grad_scale: 8.0 2023-04-01 03:08:56,914 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.624e+02 7.148e+02 8.733e+02 1.077e+03 2.339e+03, threshold=1.747e+03, percent-clipped=4.0 2023-04-01 03:09:04,848 INFO [train.py:903] (1/4) Epoch 4, batch 3650, loss[loss=0.2814, simple_loss=0.3301, pruned_loss=0.1164, over 19052.00 frames. ], tot_loss[loss=0.3025, simple_loss=0.3564, pruned_loss=0.1243, over 3824480.01 frames. ], batch size: 42, lr: 1.96e-02, grad_scale: 8.0 2023-04-01 03:10:05,501 INFO [train.py:903] (1/4) Epoch 4, batch 3700, loss[loss=0.3285, simple_loss=0.3776, pruned_loss=0.1397, over 19590.00 frames. ], tot_loss[loss=0.3019, simple_loss=0.3563, pruned_loss=0.1238, over 3842258.37 frames. ], batch size: 61, lr: 1.96e-02, grad_scale: 4.0 2023-04-01 03:10:48,438 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=24221.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 03:10:58,949 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.206e+02 7.207e+02 9.640e+02 1.134e+03 2.323e+03, threshold=1.928e+03, percent-clipped=6.0 2023-04-01 03:11:07,279 INFO [train.py:903] (1/4) Epoch 4, batch 3750, loss[loss=0.3307, simple_loss=0.3825, pruned_loss=0.1394, over 19728.00 frames. ], tot_loss[loss=0.3034, simple_loss=0.3576, pruned_loss=0.1246, over 3832197.68 frames. ], batch size: 63, lr: 1.96e-02, grad_scale: 4.0 2023-04-01 03:11:20,290 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=24246.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 03:12:07,331 INFO [train.py:903] (1/4) Epoch 4, batch 3800, loss[loss=0.2511, simple_loss=0.3169, pruned_loss=0.09268, over 19804.00 frames. ], tot_loss[loss=0.3034, simple_loss=0.3575, pruned_loss=0.1247, over 3815228.02 frames. ], batch size: 48, lr: 1.96e-02, grad_scale: 4.0 2023-04-01 03:12:38,479 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9_rvb from training. Duration: 29.1166875 2023-04-01 03:13:00,187 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 5.239e+02 7.507e+02 9.076e+02 1.248e+03 3.254e+03, threshold=1.815e+03, percent-clipped=3.0 2023-04-01 03:13:07,202 INFO [train.py:903] (1/4) Epoch 4, batch 3850, loss[loss=0.2678, simple_loss=0.3197, pruned_loss=0.108, over 19755.00 frames. ], tot_loss[loss=0.3031, simple_loss=0.3576, pruned_loss=0.1243, over 3817424.32 frames. ], batch size: 45, lr: 1.95e-02, grad_scale: 4.0 2023-04-01 03:13:38,775 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.6795, 3.0899, 3.1513, 3.1420, 1.1892, 2.8359, 2.6236, 2.8317], device='cuda:1'), covar=tensor([0.0977, 0.0646, 0.0734, 0.0611, 0.3603, 0.0488, 0.0608, 0.1389], device='cuda:1'), in_proj_covar=tensor([0.0445, 0.0392, 0.0532, 0.0423, 0.0535, 0.0307, 0.0347, 0.0511], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1') 2023-04-01 03:14:06,612 INFO [train.py:903] (1/4) Epoch 4, batch 3900, loss[loss=0.2631, simple_loss=0.3169, pruned_loss=0.1046, over 19406.00 frames. ], tot_loss[loss=0.3022, simple_loss=0.3568, pruned_loss=0.1238, over 3814376.07 frames. ], batch size: 48, lr: 1.95e-02, grad_scale: 4.0 2023-04-01 03:14:25,893 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=24399.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 03:14:28,916 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.66 vs. limit=5.0 2023-04-01 03:14:51,401 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.91 vs. limit=2.0 2023-04-01 03:15:00,804 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.463e+02 8.238e+02 9.729e+02 1.230e+03 4.971e+03, threshold=1.946e+03, percent-clipped=9.0 2023-04-01 03:15:09,420 INFO [train.py:903] (1/4) Epoch 4, batch 3950, loss[loss=0.3158, simple_loss=0.381, pruned_loss=0.1253, over 19688.00 frames. ], tot_loss[loss=0.3014, simple_loss=0.356, pruned_loss=0.1234, over 3815334.52 frames. ], batch size: 59, lr: 1.95e-02, grad_scale: 4.0 2023-04-01 03:15:17,176 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 2023-04-01 03:15:37,402 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.48 vs. limit=5.0 2023-04-01 03:16:10,710 INFO [train.py:903] (1/4) Epoch 4, batch 4000, loss[loss=0.3106, simple_loss=0.366, pruned_loss=0.1276, over 19770.00 frames. ], tot_loss[loss=0.3011, simple_loss=0.3557, pruned_loss=0.1232, over 3816740.60 frames. ], batch size: 54, lr: 1.95e-02, grad_scale: 8.0 2023-04-01 03:16:42,943 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1766, 1.1054, 1.5259, 0.8609, 2.4328, 2.8911, 2.6801, 3.0622], device='cuda:1'), covar=tensor([0.1340, 0.3043, 0.2904, 0.1984, 0.0420, 0.0133, 0.0291, 0.0139], device='cuda:1'), in_proj_covar=tensor([0.0264, 0.0284, 0.0327, 0.0259, 0.0193, 0.0108, 0.0206, 0.0127], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0002, 0.0001, 0.0002, 0.0001], device='cuda:1') 2023-04-01 03:16:45,808 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=24514.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 03:16:58,866 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1_rvb from training. Duration: 0.7545625 2023-04-01 03:17:03,420 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.288e+02 6.850e+02 8.525e+02 1.041e+03 2.187e+03, threshold=1.705e+03, percent-clipped=1.0 2023-04-01 03:17:09,917 INFO [train.py:903] (1/4) Epoch 4, batch 4050, loss[loss=0.3218, simple_loss=0.3737, pruned_loss=0.1349, over 19574.00 frames. ], tot_loss[loss=0.2999, simple_loss=0.3547, pruned_loss=0.1225, over 3827252.70 frames. ], batch size: 52, lr: 1.95e-02, grad_scale: 8.0 2023-04-01 03:17:47,506 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=24565.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 03:18:02,830 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 2023-04-01 03:18:10,047 INFO [train.py:903] (1/4) Epoch 4, batch 4100, loss[loss=0.2591, simple_loss=0.3248, pruned_loss=0.09671, over 19622.00 frames. ], tot_loss[loss=0.299, simple_loss=0.3541, pruned_loss=0.1219, over 3834943.85 frames. ], batch size: 50, lr: 1.94e-02, grad_scale: 8.0 2023-04-01 03:18:12,612 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=24587.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 03:18:19,369 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([0.8724, 1.2923, 0.9690, 1.0267, 1.1575, 0.8371, 0.6464, 1.2169], device='cuda:1'), covar=tensor([0.0425, 0.0458, 0.0798, 0.0357, 0.0407, 0.0873, 0.0581, 0.0348], device='cuda:1'), in_proj_covar=tensor([0.0271, 0.0249, 0.0313, 0.0239, 0.0218, 0.0306, 0.0278, 0.0251], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 03:18:49,633 WARNING [train.py:1073] (1/4) Exclude cut with ID 497-129325-0061-9566_sp1.1_rvb from training. Duration: 0.97725 2023-04-01 03:19:04,403 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.283e+02 6.813e+02 8.989e+02 1.047e+03 2.179e+03, threshold=1.798e+03, percent-clipped=2.0 2023-04-01 03:19:12,696 INFO [train.py:903] (1/4) Epoch 4, batch 4150, loss[loss=0.3121, simple_loss=0.3684, pruned_loss=0.1279, over 19412.00 frames. ], tot_loss[loss=0.2999, simple_loss=0.3547, pruned_loss=0.1225, over 3825261.68 frames. ], batch size: 70, lr: 1.94e-02, grad_scale: 8.0 2023-04-01 03:20:12,643 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.0819, 2.0080, 2.3078, 2.8026, 2.1151, 2.8836, 2.7192, 2.9197], device='cuda:1'), covar=tensor([0.0511, 0.0948, 0.0872, 0.0982, 0.1065, 0.0594, 0.0983, 0.0455], device='cuda:1'), in_proj_covar=tensor([0.0243, 0.0264, 0.0255, 0.0294, 0.0296, 0.0243, 0.0251, 0.0239], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:1') 2023-04-01 03:20:13,431 INFO [train.py:903] (1/4) Epoch 4, batch 4200, loss[loss=0.2457, simple_loss=0.3211, pruned_loss=0.08517, over 19595.00 frames. ], tot_loss[loss=0.2996, simple_loss=0.3547, pruned_loss=0.1222, over 3824500.62 frames. ], batch size: 52, lr: 1.94e-02, grad_scale: 8.0 2023-04-01 03:20:19,916 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 2023-04-01 03:21:05,880 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.764e+02 7.220e+02 8.850e+02 1.090e+03 2.101e+03, threshold=1.770e+03, percent-clipped=3.0 2023-04-01 03:21:12,813 INFO [train.py:903] (1/4) Epoch 4, batch 4250, loss[loss=0.3775, simple_loss=0.3989, pruned_loss=0.178, over 13450.00 frames. ], tot_loss[loss=0.2998, simple_loss=0.3548, pruned_loss=0.1224, over 3828460.99 frames. ], batch size: 136, lr: 1.94e-02, grad_scale: 8.0 2023-04-01 03:21:29,767 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9_rvb from training. Duration: 29.816625 2023-04-01 03:21:41,557 WARNING [train.py:1073] (1/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 2023-04-01 03:21:45,258 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.5941, 1.1622, 1.4641, 1.7637, 2.9011, 1.1177, 2.1774, 3.1556], device='cuda:1'), covar=tensor([0.0472, 0.2904, 0.2580, 0.1466, 0.0735, 0.2319, 0.1191, 0.0483], device='cuda:1'), in_proj_covar=tensor([0.0272, 0.0309, 0.0303, 0.0279, 0.0298, 0.0315, 0.0283, 0.0288], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 03:21:56,461 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=24770.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 03:22:10,841 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=24783.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 03:22:12,642 INFO [train.py:903] (1/4) Epoch 4, batch 4300, loss[loss=0.3026, simple_loss=0.3673, pruned_loss=0.119, over 19480.00 frames. ], tot_loss[loss=0.301, simple_loss=0.3556, pruned_loss=0.1232, over 3821928.35 frames. ], batch size: 64, lr: 1.94e-02, grad_scale: 8.0 2023-04-01 03:22:26,972 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=24795.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 03:23:06,661 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.009e+02 7.093e+02 8.869e+02 1.163e+03 2.104e+03, threshold=1.774e+03, percent-clipped=1.0 2023-04-01 03:23:08,962 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 2023-04-01 03:23:14,529 INFO [train.py:903] (1/4) Epoch 4, batch 4350, loss[loss=0.3306, simple_loss=0.3809, pruned_loss=0.1402, over 19130.00 frames. ], tot_loss[loss=0.301, simple_loss=0.3558, pruned_loss=0.1231, over 3831911.24 frames. ], batch size: 69, lr: 1.94e-02, grad_scale: 8.0 2023-04-01 03:23:52,214 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=24866.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 03:23:58,049 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1379, 1.2046, 1.8618, 1.2917, 2.4533, 2.0654, 2.6672, 1.0385], device='cuda:1'), covar=tensor([0.1832, 0.2894, 0.1529, 0.1692, 0.1247, 0.1462, 0.1462, 0.2708], device='cuda:1'), in_proj_covar=tensor([0.0437, 0.0485, 0.0458, 0.0409, 0.0532, 0.0427, 0.0608, 0.0433], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 03:24:15,080 INFO [train.py:903] (1/4) Epoch 4, batch 4400, loss[loss=0.2893, simple_loss=0.3583, pruned_loss=0.1101, over 18764.00 frames. ], tot_loss[loss=0.2994, simple_loss=0.3544, pruned_loss=0.1222, over 3828880.67 frames. ], batch size: 74, lr: 1.93e-02, grad_scale: 8.0 2023-04-01 03:24:40,905 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_rvb from training. Duration: 25.285 2023-04-01 03:24:44,308 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=24909.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 03:24:50,766 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 2023-04-01 03:25:09,651 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 5.331e+02 8.078e+02 9.889e+02 1.280e+03 3.768e+03, threshold=1.978e+03, percent-clipped=10.0 2023-04-01 03:25:12,015 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=24931.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 03:25:16,496 INFO [train.py:903] (1/4) Epoch 4, batch 4450, loss[loss=0.2829, simple_loss=0.3418, pruned_loss=0.112, over 19615.00 frames. ], tot_loss[loss=0.302, simple_loss=0.3562, pruned_loss=0.124, over 3817008.43 frames. ], batch size: 50, lr: 1.93e-02, grad_scale: 8.0 2023-04-01 03:25:49,039 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.2862, 2.2997, 2.0206, 3.4660, 2.2891, 3.6163, 3.4601, 2.2434], device='cuda:1'), covar=tensor([0.1610, 0.1284, 0.0688, 0.0811, 0.1626, 0.0391, 0.0949, 0.1019], device='cuda:1'), in_proj_covar=tensor([0.0569, 0.0550, 0.0515, 0.0714, 0.0611, 0.0461, 0.0620, 0.0531], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 03:26:17,087 INFO [train.py:903] (1/4) Epoch 4, batch 4500, loss[loss=0.327, simple_loss=0.3725, pruned_loss=0.1407, over 19675.00 frames. ], tot_loss[loss=0.3013, simple_loss=0.3554, pruned_loss=0.1236, over 3801740.25 frames. ], batch size: 53, lr: 1.93e-02, grad_scale: 8.0 2023-04-01 03:26:49,247 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1744, 1.2367, 1.5801, 1.2973, 2.2516, 2.0180, 2.4016, 0.8467], device='cuda:1'), covar=tensor([0.1552, 0.2510, 0.1406, 0.1294, 0.0920, 0.1196, 0.0971, 0.2364], device='cuda:1'), in_proj_covar=tensor([0.0428, 0.0477, 0.0452, 0.0397, 0.0521, 0.0418, 0.0596, 0.0430], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 03:26:52,463 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=25013.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 03:27:04,620 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=25024.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 03:27:11,028 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.968e+02 6.473e+02 7.865e+02 1.057e+03 2.211e+03, threshold=1.573e+03, percent-clipped=1.0 2023-04-01 03:27:16,564 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.2818, 1.2419, 1.2901, 1.6523, 2.8086, 1.2269, 1.9283, 2.8793], device='cuda:1'), covar=tensor([0.0410, 0.2647, 0.2605, 0.1477, 0.0615, 0.2282, 0.1227, 0.0475], device='cuda:1'), in_proj_covar=tensor([0.0277, 0.0313, 0.0312, 0.0284, 0.0305, 0.0321, 0.0287, 0.0299], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 03:27:18,545 INFO [train.py:903] (1/4) Epoch 4, batch 4550, loss[loss=0.2969, simple_loss=0.3648, pruned_loss=0.1145, over 18841.00 frames. ], tot_loss[loss=0.3014, simple_loss=0.3557, pruned_loss=0.1235, over 3811978.18 frames. ], batch size: 74, lr: 1.93e-02, grad_scale: 8.0 2023-04-01 03:27:27,123 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9_rvb from training. Duration: 28.72225 2023-04-01 03:27:31,937 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=25046.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 03:27:50,584 WARNING [train.py:1073] (1/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 2023-04-01 03:28:18,971 INFO [train.py:903] (1/4) Epoch 4, batch 4600, loss[loss=0.3498, simple_loss=0.3934, pruned_loss=0.1531, over 19623.00 frames. ], tot_loss[loss=0.3001, simple_loss=0.3549, pruned_loss=0.1227, over 3816298.03 frames. ], batch size: 57, lr: 1.93e-02, grad_scale: 8.0 2023-04-01 03:29:10,645 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=25127.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 03:29:12,893 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.682e+02 7.418e+02 9.211e+02 1.176e+03 2.853e+03, threshold=1.842e+03, percent-clipped=7.0 2023-04-01 03:29:20,333 INFO [train.py:903] (1/4) Epoch 4, batch 4650, loss[loss=0.4104, simple_loss=0.4184, pruned_loss=0.2012, over 13126.00 frames. ], tot_loss[loss=0.2992, simple_loss=0.3541, pruned_loss=0.1222, over 3810823.70 frames. ], batch size: 137, lr: 1.92e-02, grad_scale: 8.0 2023-04-01 03:29:37,186 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_rvb from training. Duration: 0.92 2023-04-01 03:29:47,537 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_rvb from training. Duration: 0.83 2023-04-01 03:30:19,351 INFO [train.py:903] (1/4) Epoch 4, batch 4700, loss[loss=0.2539, simple_loss=0.3127, pruned_loss=0.09749, over 19746.00 frames. ], tot_loss[loss=0.2989, simple_loss=0.3536, pruned_loss=0.1221, over 3817745.62 frames. ], batch size: 46, lr: 1.92e-02, grad_scale: 8.0 2023-04-01 03:30:42,833 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9_rvb from training. Duration: 0.92225 2023-04-01 03:30:50,629 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=25210.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 03:31:13,779 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.903e+02 7.587e+02 9.394e+02 1.259e+03 3.233e+03, threshold=1.879e+03, percent-clipped=11.0 2023-04-01 03:31:21,429 INFO [train.py:903] (1/4) Epoch 4, batch 4750, loss[loss=0.292, simple_loss=0.3488, pruned_loss=0.1176, over 19767.00 frames. ], tot_loss[loss=0.2993, simple_loss=0.3539, pruned_loss=0.1224, over 3833835.04 frames. ], batch size: 56, lr: 1.92e-02, grad_scale: 8.0 2023-04-01 03:31:30,698 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=25242.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 03:31:31,680 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.6320, 4.2007, 2.7194, 3.7179, 1.1560, 3.8170, 3.7836, 3.9932], device='cuda:1'), covar=tensor([0.0533, 0.1014, 0.1738, 0.0711, 0.3793, 0.0966, 0.0635, 0.0915], device='cuda:1'), in_proj_covar=tensor([0.0326, 0.0289, 0.0342, 0.0272, 0.0340, 0.0298, 0.0265, 0.0295], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 03:32:16,214 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=25280.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 03:32:21,434 INFO [train.py:903] (1/4) Epoch 4, batch 4800, loss[loss=0.3504, simple_loss=0.3875, pruned_loss=0.1566, over 13583.00 frames. ], tot_loss[loss=0.2997, simple_loss=0.3544, pruned_loss=0.1225, over 3834179.74 frames. ], batch size: 136, lr: 1.92e-02, grad_scale: 8.0 2023-04-01 03:32:41,277 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=25302.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 03:32:44,325 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=25305.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 03:33:09,547 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=25325.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 03:33:11,886 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=25327.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 03:33:13,790 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.433e+02 7.095e+02 8.715e+02 1.261e+03 2.828e+03, threshold=1.743e+03, percent-clipped=4.0 2023-04-01 03:33:21,567 INFO [train.py:903] (1/4) Epoch 4, batch 4850, loss[loss=0.302, simple_loss=0.3573, pruned_loss=0.1233, over 19531.00 frames. ], tot_loss[loss=0.299, simple_loss=0.3538, pruned_loss=0.1221, over 3840080.49 frames. ], batch size: 54, lr: 1.92e-02, grad_scale: 8.0 2023-04-01 03:33:22,981 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.7596, 1.8675, 1.7856, 2.8911, 2.4523, 2.3030, 2.4124, 2.5372], device='cuda:1'), covar=tensor([0.0641, 0.1518, 0.1229, 0.0687, 0.1033, 0.0416, 0.0687, 0.0524], device='cuda:1'), in_proj_covar=tensor([0.0251, 0.0366, 0.0281, 0.0243, 0.0304, 0.0255, 0.0268, 0.0232], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 03:33:45,814 WARNING [train.py:1073] (1/4) Exclude cut with ID 774-127930-0014-48411_sp1.1_rvb from training. Duration: 0.95 2023-04-01 03:33:48,285 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=25357.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 03:34:03,171 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 2023-04-01 03:34:04,841 WARNING [train.py:1073] (1/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1_rvb from training. Duration: 0.9409375 2023-04-01 03:34:11,328 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 2023-04-01 03:34:12,494 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 2023-04-01 03:34:21,580 INFO [train.py:903] (1/4) Epoch 4, batch 4900, loss[loss=0.3037, simple_loss=0.3657, pruned_loss=0.1208, over 18749.00 frames. ], tot_loss[loss=0.2995, simple_loss=0.3545, pruned_loss=0.1223, over 3824542.24 frames. ], batch size: 74, lr: 1.92e-02, grad_scale: 8.0 2023-04-01 03:34:21,606 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1_rvb from training. Duration: 25.3818125 2023-04-01 03:34:23,046 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.1740, 1.5998, 2.4727, 2.8579, 2.5000, 2.6645, 2.6841, 2.8941], device='cuda:1'), covar=tensor([0.0504, 0.1752, 0.0901, 0.0753, 0.1070, 0.0368, 0.0655, 0.0450], device='cuda:1'), in_proj_covar=tensor([0.0248, 0.0362, 0.0276, 0.0242, 0.0301, 0.0252, 0.0267, 0.0230], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 03:34:31,773 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.9645, 1.2474, 1.4477, 1.5146, 2.5763, 1.1768, 1.7966, 2.6360], device='cuda:1'), covar=tensor([0.0447, 0.2313, 0.2212, 0.1390, 0.0590, 0.1824, 0.0983, 0.0474], device='cuda:1'), in_proj_covar=tensor([0.0268, 0.0302, 0.0299, 0.0277, 0.0292, 0.0312, 0.0278, 0.0287], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 03:34:42,596 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 2023-04-01 03:35:16,119 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.171e+02 7.094e+02 8.660e+02 1.069e+03 1.655e+03, threshold=1.732e+03, percent-clipped=0.0 2023-04-01 03:35:23,610 INFO [train.py:903] (1/4) Epoch 4, batch 4950, loss[loss=0.2748, simple_loss=0.3418, pruned_loss=0.1039, over 19699.00 frames. ], tot_loss[loss=0.2977, simple_loss=0.3531, pruned_loss=0.1211, over 3823065.86 frames. ], batch size: 53, lr: 1.91e-02, grad_scale: 8.0 2023-04-01 03:35:37,194 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=25446.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 03:35:41,238 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 2023-04-01 03:36:04,435 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 2023-04-01 03:36:07,331 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=4.12 vs. limit=5.0 2023-04-01 03:36:08,203 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=25472.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 03:36:10,077 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.5355, 4.1160, 2.5327, 3.7010, 1.0629, 3.6111, 3.6012, 3.8972], device='cuda:1'), covar=tensor([0.0629, 0.1481, 0.1899, 0.0699, 0.4156, 0.0910, 0.0721, 0.0759], device='cuda:1'), in_proj_covar=tensor([0.0334, 0.0296, 0.0355, 0.0277, 0.0352, 0.0301, 0.0267, 0.0298], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 03:36:14,162 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.84 vs. limit=5.0 2023-04-01 03:36:24,197 INFO [train.py:903] (1/4) Epoch 4, batch 5000, loss[loss=0.2948, simple_loss=0.3563, pruned_loss=0.1167, over 19482.00 frames. ], tot_loss[loss=0.2976, simple_loss=0.3531, pruned_loss=0.121, over 3830304.62 frames. ], batch size: 64, lr: 1.91e-02, grad_scale: 8.0 2023-04-01 03:36:33,150 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_rvb from training. Duration: 27.14 2023-04-01 03:36:40,147 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=25498.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 03:36:44,425 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 2023-04-01 03:37:10,982 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=25523.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 03:37:17,568 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.118e+02 6.823e+02 8.824e+02 1.078e+03 2.588e+03, threshold=1.765e+03, percent-clipped=9.0 2023-04-01 03:37:24,345 INFO [train.py:903] (1/4) Epoch 4, batch 5050, loss[loss=0.328, simple_loss=0.3785, pruned_loss=0.1388, over 17278.00 frames. ], tot_loss[loss=0.2977, simple_loss=0.3534, pruned_loss=0.121, over 3827708.06 frames. ], batch size: 101, lr: 1.91e-02, grad_scale: 8.0 2023-04-01 03:37:36,450 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.42 vs. limit=5.0 2023-04-01 03:38:02,118 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9_rvb from training. Duration: 26.62775 2023-04-01 03:38:21,770 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=25581.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 03:38:25,842 INFO [train.py:903] (1/4) Epoch 4, batch 5100, loss[loss=0.2771, simple_loss=0.3276, pruned_loss=0.1133, over 19733.00 frames. ], tot_loss[loss=0.2962, simple_loss=0.3522, pruned_loss=0.1201, over 3831029.64 frames. ], batch size: 45, lr: 1.91e-02, grad_scale: 8.0 2023-04-01 03:38:36,809 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 2023-04-01 03:38:40,924 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_rvb from training. Duration: 25.035 2023-04-01 03:38:44,293 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_rvb from training. Duration: 27.92 2023-04-01 03:38:52,397 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=25606.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 03:39:19,437 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.903e+02 6.354e+02 8.688e+02 1.168e+03 2.387e+03, threshold=1.738e+03, percent-clipped=4.0 2023-04-01 03:39:27,222 INFO [train.py:903] (1/4) Epoch 4, batch 5150, loss[loss=0.3569, simple_loss=0.3932, pruned_loss=0.1603, over 13613.00 frames. ], tot_loss[loss=0.299, simple_loss=0.3548, pruned_loss=0.1216, over 3828512.63 frames. ], batch size: 136, lr: 1.91e-02, grad_scale: 8.0 2023-04-01 03:39:29,728 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.7463, 4.2194, 2.7204, 3.7871, 1.0082, 4.0812, 3.9266, 4.2582], device='cuda:1'), covar=tensor([0.0521, 0.0947, 0.1769, 0.0695, 0.3845, 0.0773, 0.0654, 0.0712], device='cuda:1'), in_proj_covar=tensor([0.0339, 0.0296, 0.0357, 0.0281, 0.0349, 0.0304, 0.0269, 0.0302], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 03:39:39,301 WARNING [train.py:1073] (1/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 2023-04-01 03:39:46,304 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2112, 1.3033, 0.9414, 0.9194, 1.0018, 1.0890, 0.0231, 0.3848], device='cuda:1'), covar=tensor([0.0262, 0.0250, 0.0183, 0.0205, 0.0516, 0.0193, 0.0474, 0.0403], device='cuda:1'), in_proj_covar=tensor([0.0274, 0.0276, 0.0267, 0.0297, 0.0356, 0.0285, 0.0277, 0.0282], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 03:39:58,757 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2626, 1.3116, 1.8974, 1.4471, 3.0463, 2.6576, 3.4250, 1.4437], device='cuda:1'), covar=tensor([0.1947, 0.3102, 0.1707, 0.1528, 0.1338, 0.1481, 0.1537, 0.2752], device='cuda:1'), in_proj_covar=tensor([0.0447, 0.0495, 0.0461, 0.0409, 0.0534, 0.0440, 0.0614, 0.0434], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 03:40:09,899 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.7742, 1.3759, 1.5449, 1.9016, 3.1558, 1.2357, 2.2481, 3.4098], device='cuda:1'), covar=tensor([0.0355, 0.2734, 0.2534, 0.1519, 0.0627, 0.2425, 0.1207, 0.0386], device='cuda:1'), in_proj_covar=tensor([0.0274, 0.0310, 0.0309, 0.0281, 0.0305, 0.0319, 0.0287, 0.0292], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 03:40:12,884 WARNING [train.py:1073] (1/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1_rvb from training. Duration: 0.9681875 2023-04-01 03:40:28,111 INFO [train.py:903] (1/4) Epoch 4, batch 5200, loss[loss=0.2406, simple_loss=0.3043, pruned_loss=0.08843, over 17370.00 frames. ], tot_loss[loss=0.2979, simple_loss=0.3534, pruned_loss=0.1212, over 3827310.52 frames. ], batch size: 38, lr: 1.90e-02, grad_scale: 8.0 2023-04-01 03:40:38,655 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.4392, 1.3691, 1.4947, 1.8221, 2.9692, 1.1846, 1.8666, 3.0188], device='cuda:1'), covar=tensor([0.0376, 0.2513, 0.2328, 0.1418, 0.0566, 0.2273, 0.1239, 0.0444], device='cuda:1'), in_proj_covar=tensor([0.0274, 0.0307, 0.0307, 0.0280, 0.0303, 0.0317, 0.0284, 0.0291], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 03:40:42,802 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 2023-04-01 03:41:21,050 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=25728.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 03:41:21,774 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.810e+02 7.304e+02 9.145e+02 1.165e+03 2.884e+03, threshold=1.829e+03, percent-clipped=6.0 2023-04-01 03:41:25,397 WARNING [train.py:1073] (1/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 2023-04-01 03:41:28,709 INFO [train.py:903] (1/4) Epoch 4, batch 5250, loss[loss=0.3319, simple_loss=0.3793, pruned_loss=0.1422, over 19473.00 frames. ], tot_loss[loss=0.2976, simple_loss=0.3534, pruned_loss=0.1209, over 3816587.74 frames. ], batch size: 49, lr: 1.90e-02, grad_scale: 8.0 2023-04-01 03:41:51,369 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=25753.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 03:42:30,031 INFO [train.py:903] (1/4) Epoch 4, batch 5300, loss[loss=0.3459, simple_loss=0.3774, pruned_loss=0.1572, over 19629.00 frames. ], tot_loss[loss=0.2952, simple_loss=0.3513, pruned_loss=0.1196, over 3825975.50 frames. ], batch size: 50, lr: 1.90e-02, grad_scale: 8.0 2023-04-01 03:42:35,924 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=25790.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 03:42:48,924 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 2023-04-01 03:42:49,426 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9748, 1.5869, 1.4518, 1.9889, 1.8259, 1.8682, 1.7007, 1.8857], device='cuda:1'), covar=tensor([0.0780, 0.1406, 0.1246, 0.0827, 0.1014, 0.0419, 0.0778, 0.0553], device='cuda:1'), in_proj_covar=tensor([0.0256, 0.0371, 0.0285, 0.0245, 0.0311, 0.0257, 0.0274, 0.0235], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 03:43:23,327 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.781e+02 7.857e+02 9.756e+02 1.201e+03 3.803e+03, threshold=1.951e+03, percent-clipped=8.0 2023-04-01 03:43:28,060 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0104, 2.0345, 1.8711, 3.0331, 2.0651, 3.0661, 2.7836, 1.8436], device='cuda:1'), covar=tensor([0.1657, 0.1316, 0.0733, 0.0774, 0.1546, 0.0423, 0.1155, 0.1216], device='cuda:1'), in_proj_covar=tensor([0.0575, 0.0561, 0.0524, 0.0721, 0.0624, 0.0479, 0.0630, 0.0537], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 03:43:31,864 INFO [train.py:903] (1/4) Epoch 4, batch 5350, loss[loss=0.3311, simple_loss=0.3837, pruned_loss=0.1392, over 18829.00 frames. ], tot_loss[loss=0.2956, simple_loss=0.3517, pruned_loss=0.1198, over 3811498.70 frames. ], batch size: 74, lr: 1.90e-02, grad_scale: 8.0 2023-04-01 03:43:45,891 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.91 vs. limit=2.0 2023-04-01 03:44:04,161 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9_rvb from training. Duration: 26.438875 2023-04-01 03:44:13,281 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3392, 1.2246, 1.8734, 1.4996, 2.9534, 2.3918, 3.0631, 1.3574], device='cuda:1'), covar=tensor([0.1871, 0.3127, 0.1763, 0.1519, 0.1179, 0.1469, 0.1478, 0.2719], device='cuda:1'), in_proj_covar=tensor([0.0446, 0.0494, 0.0464, 0.0406, 0.0538, 0.0443, 0.0620, 0.0438], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 03:44:21,895 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=25876.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 03:44:32,189 INFO [train.py:903] (1/4) Epoch 4, batch 5400, loss[loss=0.2317, simple_loss=0.3052, pruned_loss=0.07915, over 19718.00 frames. ], tot_loss[loss=0.2968, simple_loss=0.3529, pruned_loss=0.1204, over 3809885.83 frames. ], batch size: 51, lr: 1.90e-02, grad_scale: 8.0 2023-04-01 03:44:56,685 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=25905.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 03:45:16,356 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=2.01 vs. limit=2.0 2023-04-01 03:45:26,555 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.180e+02 6.939e+02 8.636e+02 1.056e+03 2.577e+03, threshold=1.727e+03, percent-clipped=2.0 2023-04-01 03:45:33,337 INFO [train.py:903] (1/4) Epoch 4, batch 5450, loss[loss=0.3198, simple_loss=0.3596, pruned_loss=0.14, over 19401.00 frames. ], tot_loss[loss=0.2968, simple_loss=0.3529, pruned_loss=0.1203, over 3808966.38 frames. ], batch size: 48, lr: 1.90e-02, grad_scale: 8.0 2023-04-01 03:46:34,625 INFO [train.py:903] (1/4) Epoch 4, batch 5500, loss[loss=0.2624, simple_loss=0.3124, pruned_loss=0.1062, over 15181.00 frames. ], tot_loss[loss=0.2977, simple_loss=0.3535, pruned_loss=0.1209, over 3796956.25 frames. ], batch size: 33, lr: 1.89e-02, grad_scale: 4.0 2023-04-01 03:46:42,600 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4705, 1.1752, 1.1837, 1.7969, 1.4739, 1.6543, 1.8784, 1.3315], device='cuda:1'), covar=tensor([0.0857, 0.1040, 0.1070, 0.0760, 0.0822, 0.0725, 0.0717, 0.0775], device='cuda:1'), in_proj_covar=tensor([0.0242, 0.0257, 0.0248, 0.0285, 0.0286, 0.0240, 0.0248, 0.0233], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:1') 2023-04-01 03:46:58,094 WARNING [train.py:1073] (1/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9_rvb from training. Duration: 27.47775 2023-04-01 03:47:31,757 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.407e+02 7.568e+02 9.013e+02 1.115e+03 1.816e+03, threshold=1.803e+03, percent-clipped=1.0 2023-04-01 03:47:37,482 INFO [train.py:903] (1/4) Epoch 4, batch 5550, loss[loss=0.2639, simple_loss=0.3254, pruned_loss=0.1012, over 19841.00 frames. ], tot_loss[loss=0.2963, simple_loss=0.3525, pruned_loss=0.1201, over 3816307.32 frames. ], batch size: 52, lr: 1.89e-02, grad_scale: 4.0 2023-04-01 03:47:45,435 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 2023-04-01 03:47:48,990 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.0836, 3.5386, 3.6771, 3.7035, 1.3440, 3.4005, 3.0380, 3.2241], device='cuda:1'), covar=tensor([0.1130, 0.0709, 0.0654, 0.0592, 0.3811, 0.0480, 0.0586, 0.1292], device='cuda:1'), in_proj_covar=tensor([0.0463, 0.0413, 0.0548, 0.0445, 0.0540, 0.0320, 0.0362, 0.0517], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1') 2023-04-01 03:48:08,130 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.0787, 1.8647, 1.3788, 1.3023, 1.7234, 1.1313, 0.9129, 1.7011], device='cuda:1'), covar=tensor([0.0603, 0.0488, 0.0880, 0.0432, 0.0375, 0.0911, 0.0606, 0.0282], device='cuda:1'), in_proj_covar=tensor([0.0280, 0.0258, 0.0315, 0.0243, 0.0218, 0.0312, 0.0282, 0.0247], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 03:48:33,398 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 2023-04-01 03:48:38,940 INFO [train.py:903] (1/4) Epoch 4, batch 5600, loss[loss=0.2707, simple_loss=0.3276, pruned_loss=0.1068, over 19060.00 frames. ], tot_loss[loss=0.2963, simple_loss=0.3528, pruned_loss=0.1199, over 3832672.08 frames. ], batch size: 42, lr: 1.89e-02, grad_scale: 8.0 2023-04-01 03:48:41,419 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9322, 1.5479, 2.0976, 1.7685, 2.7906, 4.7432, 4.6354, 4.9008], device='cuda:1'), covar=tensor([0.1062, 0.2472, 0.2339, 0.1560, 0.0423, 0.0082, 0.0112, 0.0070], device='cuda:1'), in_proj_covar=tensor([0.0257, 0.0279, 0.0317, 0.0259, 0.0196, 0.0116, 0.0203, 0.0127], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0002, 0.0001, 0.0002, 0.0001], device='cuda:1') 2023-04-01 03:49:22,515 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=26120.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 03:49:34,176 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.637e+02 7.669e+02 9.359e+02 1.114e+03 3.409e+03, threshold=1.872e+03, percent-clipped=3.0 2023-04-01 03:49:40,064 INFO [train.py:903] (1/4) Epoch 4, batch 5650, loss[loss=0.3163, simple_loss=0.3721, pruned_loss=0.1303, over 18161.00 frames. ], tot_loss[loss=0.2964, simple_loss=0.3529, pruned_loss=0.12, over 3831344.04 frames. ], batch size: 84, lr: 1.89e-02, grad_scale: 8.0 2023-04-01 03:50:12,574 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=26161.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 03:50:25,381 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9_rvb from training. Duration: 28.0944375 2023-04-01 03:50:38,345 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9048, 1.8610, 1.7967, 2.8435, 1.9932, 2.7596, 2.6269, 1.7822], device='cuda:1'), covar=tensor([0.1677, 0.1197, 0.0719, 0.0654, 0.1349, 0.0433, 0.1102, 0.1158], device='cuda:1'), in_proj_covar=tensor([0.0577, 0.0556, 0.0517, 0.0711, 0.0622, 0.0475, 0.0629, 0.0536], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 03:50:40,063 INFO [train.py:903] (1/4) Epoch 4, batch 5700, loss[loss=0.3192, simple_loss=0.3667, pruned_loss=0.1358, over 19752.00 frames. ], tot_loss[loss=0.2968, simple_loss=0.3528, pruned_loss=0.1203, over 3817626.30 frames. ], batch size: 54, lr: 1.89e-02, grad_scale: 8.0 2023-04-01 03:50:42,193 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=26186.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 03:50:42,303 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=26186.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 03:51:23,014 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=26220.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 03:51:35,339 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.235e+02 7.760e+02 9.507e+02 1.157e+03 2.773e+03, threshold=1.901e+03, percent-clipped=5.0 2023-04-01 03:51:39,942 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9_rvb from training. Duration: 33.038875 2023-04-01 03:51:40,952 INFO [train.py:903] (1/4) Epoch 4, batch 5750, loss[loss=0.2785, simple_loss=0.3429, pruned_loss=0.1071, over 19569.00 frames. ], tot_loss[loss=0.2975, simple_loss=0.3534, pruned_loss=0.1208, over 3813783.23 frames. ], batch size: 52, lr: 1.89e-02, grad_scale: 8.0 2023-04-01 03:51:48,678 WARNING [train.py:1073] (1/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 2023-04-01 03:51:52,101 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.9569, 4.9112, 5.7672, 5.7302, 1.7595, 5.3595, 4.7950, 5.2067], device='cuda:1'), covar=tensor([0.0692, 0.0504, 0.0355, 0.0277, 0.3619, 0.0197, 0.0353, 0.0765], device='cuda:1'), in_proj_covar=tensor([0.0457, 0.0410, 0.0536, 0.0437, 0.0529, 0.0315, 0.0357, 0.0508], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1') 2023-04-01 03:51:52,967 WARNING [train.py:1073] (1/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 2023-04-01 03:52:10,485 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.2430, 3.6302, 3.7870, 3.7536, 1.3646, 3.3521, 3.0831, 3.3959], device='cuda:1'), covar=tensor([0.0919, 0.0617, 0.0513, 0.0461, 0.3836, 0.0480, 0.0557, 0.1049], device='cuda:1'), in_proj_covar=tensor([0.0461, 0.0414, 0.0539, 0.0441, 0.0537, 0.0318, 0.0359, 0.0514], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1') 2023-04-01 03:52:20,348 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=26268.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 03:52:40,302 INFO [train.py:903] (1/4) Epoch 4, batch 5800, loss[loss=0.2843, simple_loss=0.3408, pruned_loss=0.1139, over 19867.00 frames. ], tot_loss[loss=0.2996, simple_loss=0.3552, pruned_loss=0.122, over 3817460.88 frames. ], batch size: 52, lr: 1.88e-02, grad_scale: 8.0 2023-04-01 03:53:35,866 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.8942, 1.8162, 1.6804, 2.3648, 4.3446, 1.1868, 2.2398, 4.3692], device='cuda:1'), covar=tensor([0.0229, 0.2275, 0.2398, 0.1264, 0.0423, 0.2231, 0.1228, 0.0305], device='cuda:1'), in_proj_covar=tensor([0.0278, 0.0310, 0.0307, 0.0283, 0.0304, 0.0313, 0.0287, 0.0293], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 03:53:36,644 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.499e+02 6.787e+02 8.542e+02 1.114e+03 2.576e+03, threshold=1.708e+03, percent-clipped=4.0 2023-04-01 03:53:39,459 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.8708, 1.9586, 1.8418, 2.9002, 2.0120, 2.9957, 2.6742, 1.7887], device='cuda:1'), covar=tensor([0.1782, 0.1309, 0.0733, 0.0783, 0.1532, 0.0438, 0.1331, 0.1232], device='cuda:1'), in_proj_covar=tensor([0.0578, 0.0556, 0.0519, 0.0722, 0.0618, 0.0479, 0.0627, 0.0538], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 03:53:41,253 INFO [train.py:903] (1/4) Epoch 4, batch 5850, loss[loss=0.2692, simple_loss=0.3425, pruned_loss=0.09796, over 19668.00 frames. ], tot_loss[loss=0.298, simple_loss=0.3535, pruned_loss=0.1213, over 3825573.05 frames. ], batch size: 55, lr: 1.88e-02, grad_scale: 4.0 2023-04-01 03:53:41,609 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=26335.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 03:54:30,690 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.4398, 4.0247, 2.3793, 3.6154, 1.1414, 3.6704, 3.6566, 3.9015], device='cuda:1'), covar=tensor([0.0664, 0.1012, 0.2111, 0.0745, 0.3743, 0.0907, 0.0739, 0.0849], device='cuda:1'), in_proj_covar=tensor([0.0335, 0.0298, 0.0352, 0.0278, 0.0343, 0.0294, 0.0265, 0.0291], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 03:54:40,807 INFO [train.py:903] (1/4) Epoch 4, batch 5900, loss[loss=0.3527, simple_loss=0.3987, pruned_loss=0.1534, over 19588.00 frames. ], tot_loss[loss=0.2992, simple_loss=0.3546, pruned_loss=0.1219, over 3826971.40 frames. ], batch size: 61, lr: 1.88e-02, grad_scale: 4.0 2023-04-01 03:54:41,869 WARNING [train.py:1073] (1/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 2023-04-01 03:55:03,874 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9_rvb from training. Duration: 27.511125 2023-04-01 03:55:37,697 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.672e+02 6.673e+02 8.574e+02 1.115e+03 3.080e+03, threshold=1.715e+03, percent-clipped=4.0 2023-04-01 03:55:39,778 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=26432.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 03:55:43,005 INFO [train.py:903] (1/4) Epoch 4, batch 5950, loss[loss=0.2996, simple_loss=0.369, pruned_loss=0.1151, over 19659.00 frames. ], tot_loss[loss=0.2994, simple_loss=0.3546, pruned_loss=0.1221, over 3813379.26 frames. ], batch size: 58, lr: 1.88e-02, grad_scale: 4.0 2023-04-01 03:56:17,507 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=26464.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 03:56:43,845 INFO [train.py:903] (1/4) Epoch 4, batch 6000, loss[loss=0.3184, simple_loss=0.3808, pruned_loss=0.128, over 19669.00 frames. ], tot_loss[loss=0.2973, simple_loss=0.3533, pruned_loss=0.1207, over 3820738.65 frames. ], batch size: 58, lr: 1.88e-02, grad_scale: 8.0 2023-04-01 03:56:43,846 INFO [train.py:928] (1/4) Computing validation loss 2023-04-01 03:56:57,349 INFO [train.py:937] (1/4) Epoch 4, validation: loss=0.2103, simple_loss=0.3081, pruned_loss=0.05622, over 944034.00 frames. 2023-04-01 03:56:57,350 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18649MB 2023-04-01 03:56:58,915 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.7033, 1.2549, 1.5577, 2.0547, 3.2535, 1.0357, 2.0561, 3.1830], device='cuda:1'), covar=tensor([0.0275, 0.2482, 0.2212, 0.1174, 0.0487, 0.2328, 0.1212, 0.0418], device='cuda:1'), in_proj_covar=tensor([0.0279, 0.0315, 0.0307, 0.0284, 0.0307, 0.0317, 0.0287, 0.0294], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 03:57:41,682 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0378, 1.7927, 1.6686, 2.0167, 1.8544, 1.9678, 1.6855, 1.9312], device='cuda:1'), covar=tensor([0.0757, 0.1291, 0.1150, 0.0809, 0.0968, 0.0444, 0.1000, 0.0594], device='cuda:1'), in_proj_covar=tensor([0.0250, 0.0359, 0.0278, 0.0241, 0.0301, 0.0252, 0.0268, 0.0236], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 03:57:51,992 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=26530.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 03:57:52,984 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.403e+02 7.085e+02 8.722e+02 1.145e+03 2.334e+03, threshold=1.744e+03, percent-clipped=4.0 2023-04-01 03:57:57,456 INFO [train.py:903] (1/4) Epoch 4, batch 6050, loss[loss=0.317, simple_loss=0.3718, pruned_loss=0.1311, over 19763.00 frames. ], tot_loss[loss=0.2971, simple_loss=0.353, pruned_loss=0.1206, over 3827525.80 frames. ], batch size: 63, lr: 1.87e-02, grad_scale: 8.0 2023-04-01 03:58:49,946 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=26579.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 03:58:57,302 INFO [train.py:903] (1/4) Epoch 4, batch 6100, loss[loss=0.2834, simple_loss=0.3434, pruned_loss=0.1117, over 19598.00 frames. ], tot_loss[loss=0.2964, simple_loss=0.3524, pruned_loss=0.1202, over 3830233.39 frames. ], batch size: 57, lr: 1.87e-02, grad_scale: 8.0 2023-04-01 03:59:04,729 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=26591.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 03:59:29,869 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=26612.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 03:59:30,170 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4961, 1.5280, 1.3441, 2.0362, 1.3776, 1.8619, 1.7529, 1.1542], device='cuda:1'), covar=tensor([0.2135, 0.1732, 0.1377, 0.0940, 0.1802, 0.0773, 0.2343, 0.2330], device='cuda:1'), in_proj_covar=tensor([0.0586, 0.0570, 0.0534, 0.0732, 0.0628, 0.0491, 0.0635, 0.0543], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 03:59:34,627 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=26616.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 03:59:51,597 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.600e+02 6.768e+02 7.963e+02 1.032e+03 2.370e+03, threshold=1.593e+03, percent-clipped=5.0 2023-04-01 03:59:56,247 INFO [train.py:903] (1/4) Epoch 4, batch 6150, loss[loss=0.2499, simple_loss=0.3103, pruned_loss=0.09479, over 19423.00 frames. ], tot_loss[loss=0.2951, simple_loss=0.3512, pruned_loss=0.1195, over 3830748.39 frames. ], batch size: 48, lr: 1.87e-02, grad_scale: 8.0 2023-04-01 04:00:08,578 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=26645.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 04:00:23,801 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9_rvb from training. Duration: 31.02225 2023-04-01 04:00:54,795 INFO [train.py:903] (1/4) Epoch 4, batch 6200, loss[loss=0.3043, simple_loss=0.3656, pruned_loss=0.1215, over 19675.00 frames. ], tot_loss[loss=0.2967, simple_loss=0.352, pruned_loss=0.1207, over 3829807.80 frames. ], batch size: 59, lr: 1.87e-02, grad_scale: 8.0 2023-04-01 04:01:39,826 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.1683, 2.7319, 1.7761, 1.9711, 2.0178, 2.2156, 0.4348, 2.1577], device='cuda:1'), covar=tensor([0.0262, 0.0254, 0.0311, 0.0431, 0.0475, 0.0364, 0.0609, 0.0407], device='cuda:1'), in_proj_covar=tensor([0.0273, 0.0274, 0.0268, 0.0298, 0.0353, 0.0281, 0.0273, 0.0280], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 04:01:45,243 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=26727.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 04:01:51,166 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.725e+02 7.340e+02 8.907e+02 1.200e+03 2.630e+03, threshold=1.781e+03, percent-clipped=5.0 2023-04-01 04:01:55,634 INFO [train.py:903] (1/4) Epoch 4, batch 6250, loss[loss=0.3023, simple_loss=0.3562, pruned_loss=0.1242, over 19580.00 frames. ], tot_loss[loss=0.297, simple_loss=0.3522, pruned_loss=0.1209, over 3827097.89 frames. ], batch size: 52, lr: 1.87e-02, grad_scale: 8.0 2023-04-01 04:02:24,771 WARNING [train.py:1073] (1/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9_rvb from training. Duration: 25.988875 2023-04-01 04:02:44,564 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=26776.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 04:02:55,314 INFO [train.py:903] (1/4) Epoch 4, batch 6300, loss[loss=0.2959, simple_loss=0.3404, pruned_loss=0.1257, over 19749.00 frames. ], tot_loss[loss=0.2949, simple_loss=0.3508, pruned_loss=0.1195, over 3831557.94 frames. ], batch size: 47, lr: 1.87e-02, grad_scale: 8.0 2023-04-01 04:03:49,878 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.898e+02 7.473e+02 9.358e+02 1.215e+03 3.413e+03, threshold=1.872e+03, percent-clipped=4.0 2023-04-01 04:03:54,555 INFO [train.py:903] (1/4) Epoch 4, batch 6350, loss[loss=0.2972, simple_loss=0.3459, pruned_loss=0.1242, over 19389.00 frames. ], tot_loss[loss=0.2956, simple_loss=0.351, pruned_loss=0.1201, over 3821179.31 frames. ], batch size: 48, lr: 1.86e-02, grad_scale: 8.0 2023-04-01 04:03:54,982 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=26835.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 04:04:25,999 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=26860.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 04:04:55,431 INFO [train.py:903] (1/4) Epoch 4, batch 6400, loss[loss=0.288, simple_loss=0.3571, pruned_loss=0.1095, over 19653.00 frames. ], tot_loss[loss=0.2936, simple_loss=0.3496, pruned_loss=0.1188, over 3821763.80 frames. ], batch size: 58, lr: 1.86e-02, grad_scale: 8.0 2023-04-01 04:05:05,491 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=26891.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 04:05:16,518 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=26901.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 04:05:28,815 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6394, 1.7257, 1.6763, 2.4917, 1.7235, 2.2349, 2.2400, 1.6109], device='cuda:1'), covar=tensor([0.1660, 0.1276, 0.0783, 0.0701, 0.1398, 0.0564, 0.1352, 0.1276], device='cuda:1'), in_proj_covar=tensor([0.0584, 0.0566, 0.0531, 0.0726, 0.0627, 0.0485, 0.0637, 0.0537], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 04:05:33,331 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2414, 1.1801, 1.9672, 1.4269, 2.8925, 2.5133, 3.1880, 1.2834], device='cuda:1'), covar=tensor([0.1883, 0.3071, 0.1570, 0.1533, 0.1310, 0.1369, 0.1458, 0.2852], device='cuda:1'), in_proj_covar=tensor([0.0445, 0.0495, 0.0459, 0.0413, 0.0538, 0.0435, 0.0611, 0.0435], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 04:05:41,179 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=26922.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 04:05:45,916 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=26926.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 04:05:52,840 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.488e+02 6.991e+02 8.678e+02 1.023e+03 2.915e+03, threshold=1.736e+03, percent-clipped=2.0 2023-04-01 04:05:58,077 INFO [train.py:903] (1/4) Epoch 4, batch 6450, loss[loss=0.3029, simple_loss=0.3516, pruned_loss=0.1271, over 19772.00 frames. ], tot_loss[loss=0.2934, simple_loss=0.3498, pruned_loss=0.1184, over 3824542.92 frames. ], batch size: 56, lr: 1.86e-02, grad_scale: 8.0 2023-04-01 04:06:05,252 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3168, 2.1271, 1.4938, 1.5530, 1.8918, 1.0150, 1.2243, 1.8279], device='cuda:1'), covar=tensor([0.0650, 0.0397, 0.0750, 0.0375, 0.0382, 0.0977, 0.0557, 0.0265], device='cuda:1'), in_proj_covar=tensor([0.0279, 0.0256, 0.0315, 0.0237, 0.0214, 0.0308, 0.0284, 0.0253], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 04:06:33,660 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.7193, 3.9846, 4.2945, 4.2955, 1.4753, 3.8808, 3.5025, 3.8338], device='cuda:1'), covar=tensor([0.0856, 0.0666, 0.0545, 0.0404, 0.4060, 0.0361, 0.0492, 0.1023], device='cuda:1'), in_proj_covar=tensor([0.0475, 0.0420, 0.0565, 0.0446, 0.0549, 0.0322, 0.0369, 0.0528], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1') 2023-04-01 04:06:36,883 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=26968.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 04:06:39,711 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9_rvb from training. Duration: 28.638875 2023-04-01 04:06:56,648 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=26983.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 04:06:58,600 INFO [train.py:903] (1/4) Epoch 4, batch 6500, loss[loss=0.2558, simple_loss=0.3099, pruned_loss=0.1009, over 19165.00 frames. ], tot_loss[loss=0.2934, simple_loss=0.3498, pruned_loss=0.1185, over 3821820.51 frames. ], batch size: 42, lr: 1.86e-02, grad_scale: 8.0 2023-04-01 04:07:03,003 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1_rvb from training. Duration: 0.836375 2023-04-01 04:07:25,326 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=27008.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 04:07:53,581 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.310e+02 7.867e+02 9.982e+02 1.245e+03 2.621e+03, threshold=1.996e+03, percent-clipped=6.0 2023-04-01 04:07:57,611 INFO [train.py:903] (1/4) Epoch 4, batch 6550, loss[loss=0.2866, simple_loss=0.3285, pruned_loss=0.1223, over 19759.00 frames. ], tot_loss[loss=0.295, simple_loss=0.3504, pruned_loss=0.1198, over 3833307.84 frames. ], batch size: 45, lr: 1.86e-02, grad_scale: 8.0 2023-04-01 04:08:54,149 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4520, 1.3130, 1.8716, 1.6002, 2.5059, 2.0838, 2.6094, 1.4519], device='cuda:1'), covar=tensor([0.1415, 0.2330, 0.1156, 0.1188, 0.0850, 0.1167, 0.0968, 0.1948], device='cuda:1'), in_proj_covar=tensor([0.0439, 0.0486, 0.0453, 0.0404, 0.0532, 0.0432, 0.0602, 0.0427], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 04:08:57,146 INFO [train.py:903] (1/4) Epoch 4, batch 6600, loss[loss=0.2527, simple_loss=0.3217, pruned_loss=0.09185, over 19599.00 frames. ], tot_loss[loss=0.2958, simple_loss=0.3512, pruned_loss=0.1202, over 3826156.54 frames. ], batch size: 52, lr: 1.86e-02, grad_scale: 8.0 2023-04-01 04:09:53,358 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.938e+02 7.635e+02 9.605e+02 1.185e+03 2.942e+03, threshold=1.921e+03, percent-clipped=6.0 2023-04-01 04:09:58,811 INFO [train.py:903] (1/4) Epoch 4, batch 6650, loss[loss=0.2601, simple_loss=0.3178, pruned_loss=0.1012, over 19398.00 frames. ], tot_loss[loss=0.2937, simple_loss=0.3494, pruned_loss=0.119, over 3828692.96 frames. ], batch size: 48, lr: 1.85e-02, grad_scale: 8.0 2023-04-01 04:10:13,160 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=27147.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 04:10:35,295 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=27166.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 04:10:42,762 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=27172.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 04:10:51,989 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.7085, 4.2027, 2.3956, 3.8191, 0.9278, 3.9231, 3.8631, 4.0283], device='cuda:1'), covar=tensor([0.0597, 0.1202, 0.2373, 0.0784, 0.4903, 0.0954, 0.0866, 0.1138], device='cuda:1'), in_proj_covar=tensor([0.0343, 0.0306, 0.0355, 0.0282, 0.0349, 0.0297, 0.0267, 0.0302], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 04:10:58,823 INFO [train.py:903] (1/4) Epoch 4, batch 6700, loss[loss=0.2389, simple_loss=0.2991, pruned_loss=0.08939, over 19774.00 frames. ], tot_loss[loss=0.2932, simple_loss=0.3489, pruned_loss=0.1188, over 3834862.19 frames. ], batch size: 48, lr: 1.85e-02, grad_scale: 8.0 2023-04-01 04:11:52,644 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.097e+02 7.210e+02 9.176e+02 1.266e+03 4.477e+03, threshold=1.835e+03, percent-clipped=7.0 2023-04-01 04:11:57,026 INFO [train.py:903] (1/4) Epoch 4, batch 6750, loss[loss=0.3001, simple_loss=0.3585, pruned_loss=0.1208, over 19543.00 frames. ], tot_loss[loss=0.2931, simple_loss=0.3494, pruned_loss=0.1184, over 3833743.80 frames. ], batch size: 56, lr: 1.85e-02, grad_scale: 8.0 2023-04-01 04:12:30,348 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.70 vs. limit=5.0 2023-04-01 04:12:31,960 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=27266.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 04:12:52,952 INFO [train.py:903] (1/4) Epoch 4, batch 6800, loss[loss=0.3078, simple_loss=0.3679, pruned_loss=0.1239, over 19529.00 frames. ], tot_loss[loss=0.2943, simple_loss=0.3503, pruned_loss=0.1192, over 3832696.85 frames. ], batch size: 56, lr: 1.85e-02, grad_scale: 8.0 2023-04-01 04:13:37,202 WARNING [train.py:1073] (1/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 2023-04-01 04:13:37,650 WARNING [train.py:1073] (1/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 2023-04-01 04:13:40,504 INFO [train.py:903] (1/4) Epoch 5, batch 0, loss[loss=0.3154, simple_loss=0.3692, pruned_loss=0.1308, over 19693.00 frames. ], tot_loss[loss=0.3154, simple_loss=0.3692, pruned_loss=0.1308, over 19693.00 frames. ], batch size: 60, lr: 1.72e-02, grad_scale: 8.0 2023-04-01 04:13:40,504 INFO [train.py:928] (1/4) Computing validation loss 2023-04-01 04:13:52,268 INFO [train.py:937] (1/4) Epoch 5, validation: loss=0.2121, simple_loss=0.3102, pruned_loss=0.05704, over 944034.00 frames. 2023-04-01 04:13:52,268 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18649MB 2023-04-01 04:13:52,406 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=27312.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 04:13:55,638 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([5.9600, 5.3028, 2.8228, 4.6499, 1.4931, 5.2751, 5.0847, 5.4132], device='cuda:1'), covar=tensor([0.0413, 0.0920, 0.1903, 0.0575, 0.3675, 0.0653, 0.0587, 0.0686], device='cuda:1'), in_proj_covar=tensor([0.0336, 0.0303, 0.0352, 0.0280, 0.0347, 0.0293, 0.0263, 0.0297], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 04:14:04,628 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 2023-04-01 04:14:16,054 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 5.218e+02 7.861e+02 9.859e+02 1.236e+03 2.711e+03, threshold=1.972e+03, percent-clipped=3.0 2023-04-01 04:14:37,962 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([5.0746, 1.9310, 1.9906, 2.2795, 4.5347, 1.3158, 2.5202, 4.6052], device='cuda:1'), covar=tensor([0.0242, 0.2182, 0.2071, 0.1382, 0.0425, 0.2191, 0.1028, 0.0287], device='cuda:1'), in_proj_covar=tensor([0.0274, 0.0305, 0.0305, 0.0282, 0.0296, 0.0311, 0.0278, 0.0291], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 04:14:52,470 INFO [train.py:903] (1/4) Epoch 5, batch 50, loss[loss=0.295, simple_loss=0.3596, pruned_loss=0.1152, over 17449.00 frames. ], tot_loss[loss=0.2979, simple_loss=0.3546, pruned_loss=0.1206, over 858507.72 frames. ], batch size: 101, lr: 1.72e-02, grad_scale: 8.0 2023-04-01 04:14:55,322 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 2023-04-01 04:15:15,084 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=27381.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 04:15:22,372 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 2023-04-01 04:15:26,080 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1_rvb from training. Duration: 27.0318125 2023-04-01 04:15:26,410 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.2954, 1.3249, 1.6951, 2.4851, 1.8491, 2.2557, 2.5420, 2.4081], device='cuda:1'), covar=tensor([0.0808, 0.1349, 0.1261, 0.1183, 0.1227, 0.0875, 0.1127, 0.0747], device='cuda:1'), in_proj_covar=tensor([0.0238, 0.0254, 0.0250, 0.0288, 0.0280, 0.0233, 0.0247, 0.0230], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:1') 2023-04-01 04:15:53,858 INFO [train.py:903] (1/4) Epoch 5, batch 100, loss[loss=0.2503, simple_loss=0.3199, pruned_loss=0.09039, over 19602.00 frames. ], tot_loss[loss=0.2957, simple_loss=0.3533, pruned_loss=0.1191, over 1522542.99 frames. ], batch size: 52, lr: 1.72e-02, grad_scale: 8.0 2023-04-01 04:16:05,290 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 2023-04-01 04:16:11,150 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=27427.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 04:16:15,198 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.977e+02 6.953e+02 8.679e+02 1.081e+03 2.199e+03, threshold=1.736e+03, percent-clipped=1.0 2023-04-01 04:16:53,730 INFO [train.py:903] (1/4) Epoch 5, batch 150, loss[loss=0.275, simple_loss=0.3368, pruned_loss=0.1065, over 19607.00 frames. ], tot_loss[loss=0.2945, simple_loss=0.3513, pruned_loss=0.1188, over 2010904.50 frames. ], batch size: 50, lr: 1.72e-02, grad_scale: 8.0 2023-04-01 04:17:52,391 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=27510.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 04:17:54,460 INFO [train.py:903] (1/4) Epoch 5, batch 200, loss[loss=0.2959, simple_loss=0.3624, pruned_loss=0.1147, over 19596.00 frames. ], tot_loss[loss=0.2928, simple_loss=0.3503, pruned_loss=0.1176, over 2419797.87 frames. ], batch size: 57, lr: 1.72e-02, grad_scale: 8.0 2023-04-01 04:17:54,474 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 2023-04-01 04:18:19,387 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.602e+02 6.870e+02 8.382e+02 1.064e+03 2.606e+03, threshold=1.676e+03, percent-clipped=3.0 2023-04-01 04:18:40,888 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.4004, 1.4365, 1.7023, 2.3560, 1.8724, 2.1830, 2.5777, 2.5405], device='cuda:1'), covar=tensor([0.0681, 0.1086, 0.1052, 0.1100, 0.1073, 0.0760, 0.0890, 0.0630], device='cuda:1'), in_proj_covar=tensor([0.0238, 0.0254, 0.0250, 0.0289, 0.0283, 0.0234, 0.0245, 0.0232], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:1') 2023-04-01 04:18:43,129 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2555, 2.0931, 1.5780, 1.4127, 1.9089, 1.0782, 1.0663, 1.5615], device='cuda:1'), covar=tensor([0.0615, 0.0464, 0.0889, 0.0469, 0.0333, 0.0968, 0.0588, 0.0359], device='cuda:1'), in_proj_covar=tensor([0.0277, 0.0260, 0.0322, 0.0241, 0.0221, 0.0312, 0.0283, 0.0257], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 04:18:56,502 INFO [train.py:903] (1/4) Epoch 5, batch 250, loss[loss=0.3085, simple_loss=0.3603, pruned_loss=0.1283, over 19828.00 frames. ], tot_loss[loss=0.2926, simple_loss=0.35, pruned_loss=0.1176, over 2721180.90 frames. ], batch size: 52, lr: 1.71e-02, grad_scale: 4.0 2023-04-01 04:19:51,630 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 2023-04-01 04:19:58,938 INFO [train.py:903] (1/4) Epoch 5, batch 300, loss[loss=0.2919, simple_loss=0.3478, pruned_loss=0.118, over 19799.00 frames. ], tot_loss[loss=0.293, simple_loss=0.3503, pruned_loss=0.1179, over 2973810.20 frames. ], batch size: 56, lr: 1.71e-02, grad_scale: 4.0 2023-04-01 04:20:15,324 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=27625.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 04:20:22,961 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.497e+02 6.581e+02 8.607e+02 1.103e+03 1.922e+03, threshold=1.721e+03, percent-clipped=6.0 2023-04-01 04:20:28,906 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=27637.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 04:21:01,019 INFO [train.py:903] (1/4) Epoch 5, batch 350, loss[loss=0.2842, simple_loss=0.3405, pruned_loss=0.114, over 19667.00 frames. ], tot_loss[loss=0.2941, simple_loss=0.3509, pruned_loss=0.1187, over 3162457.82 frames. ], batch size: 53, lr: 1.71e-02, grad_scale: 4.0 2023-04-01 04:21:01,361 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=27662.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 04:21:07,182 WARNING [train.py:1073] (1/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1_rvb from training. Duration: 0.9681875 2023-04-01 04:21:26,433 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=27683.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 04:21:27,596 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2470, 1.2813, 1.9314, 1.3560, 2.5789, 2.3785, 2.6869, 0.9778], device='cuda:1'), covar=tensor([0.1749, 0.2900, 0.1457, 0.1446, 0.1113, 0.1202, 0.1198, 0.2659], device='cuda:1'), in_proj_covar=tensor([0.0448, 0.0493, 0.0466, 0.0414, 0.0541, 0.0439, 0.0615, 0.0436], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 04:21:58,722 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=27708.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 04:22:02,932 INFO [train.py:903] (1/4) Epoch 5, batch 400, loss[loss=0.3034, simple_loss=0.3728, pruned_loss=0.117, over 19540.00 frames. ], tot_loss[loss=0.2925, simple_loss=0.3498, pruned_loss=0.1176, over 3320229.46 frames. ], batch size: 56, lr: 1.71e-02, grad_scale: 8.0 2023-04-01 04:22:08,108 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.87 vs. limit=5.0 2023-04-01 04:22:27,973 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.948e+02 7.305e+02 9.017e+02 1.065e+03 1.815e+03, threshold=1.803e+03, percent-clipped=3.0 2023-04-01 04:22:28,314 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=27732.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 04:22:58,635 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2667, 1.1461, 1.3580, 1.0969, 2.6627, 3.3801, 3.3273, 3.6966], device='cuda:1'), covar=tensor([0.1729, 0.4081, 0.4207, 0.2403, 0.0505, 0.0203, 0.0282, 0.0125], device='cuda:1'), in_proj_covar=tensor([0.0260, 0.0281, 0.0318, 0.0258, 0.0197, 0.0115, 0.0206, 0.0135], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-04-01 04:23:04,208 INFO [train.py:903] (1/4) Epoch 5, batch 450, loss[loss=0.2766, simple_loss=0.336, pruned_loss=0.1086, over 19583.00 frames. ], tot_loss[loss=0.2911, simple_loss=0.3489, pruned_loss=0.1167, over 3445098.33 frames. ], batch size: 52, lr: 1.71e-02, grad_scale: 8.0 2023-04-01 04:23:45,988 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 2023-04-01 04:23:47,109 WARNING [train.py:1073] (1/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 2023-04-01 04:24:07,111 INFO [train.py:903] (1/4) Epoch 5, batch 500, loss[loss=0.3351, simple_loss=0.3892, pruned_loss=0.1404, over 19605.00 frames. ], tot_loss[loss=0.2906, simple_loss=0.3486, pruned_loss=0.1163, over 3537152.68 frames. ], batch size: 57, lr: 1.71e-02, grad_scale: 8.0 2023-04-01 04:24:31,779 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.519e+02 6.150e+02 8.318e+02 1.057e+03 1.987e+03, threshold=1.664e+03, percent-clipped=1.0 2023-04-01 04:24:49,523 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3034, 1.2612, 1.5936, 0.9893, 2.5862, 3.2125, 3.0579, 3.4085], device='cuda:1'), covar=tensor([0.1323, 0.2775, 0.2606, 0.1992, 0.0377, 0.0139, 0.0204, 0.0116], device='cuda:1'), in_proj_covar=tensor([0.0258, 0.0279, 0.0315, 0.0255, 0.0194, 0.0114, 0.0203, 0.0134], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-04-01 04:25:11,308 INFO [train.py:903] (1/4) Epoch 5, batch 550, loss[loss=0.3276, simple_loss=0.3794, pruned_loss=0.138, over 19552.00 frames. ], tot_loss[loss=0.2905, simple_loss=0.3485, pruned_loss=0.1163, over 3603848.63 frames. ], batch size: 61, lr: 1.71e-02, grad_scale: 8.0 2023-04-01 04:25:35,631 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=27881.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 04:26:07,925 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=27906.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 04:26:14,567 INFO [train.py:903] (1/4) Epoch 5, batch 600, loss[loss=0.2775, simple_loss=0.3262, pruned_loss=0.1145, over 19736.00 frames. ], tot_loss[loss=0.29, simple_loss=0.3479, pruned_loss=0.116, over 3648616.85 frames. ], batch size: 45, lr: 1.70e-02, grad_scale: 8.0 2023-04-01 04:26:38,771 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.822e+02 6.595e+02 8.388e+02 1.023e+03 2.578e+03, threshold=1.678e+03, percent-clipped=3.0 2023-04-01 04:26:48,522 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.68 vs. limit=2.0 2023-04-01 04:27:02,819 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 2023-04-01 04:27:17,951 INFO [train.py:903] (1/4) Epoch 5, batch 650, loss[loss=0.2689, simple_loss=0.3334, pruned_loss=0.1022, over 19676.00 frames. ], tot_loss[loss=0.2897, simple_loss=0.3475, pruned_loss=0.1159, over 3677190.22 frames. ], batch size: 53, lr: 1.70e-02, grad_scale: 8.0 2023-04-01 04:28:20,085 INFO [train.py:903] (1/4) Epoch 5, batch 700, loss[loss=0.3719, simple_loss=0.4103, pruned_loss=0.1667, over 18810.00 frames. ], tot_loss[loss=0.2902, simple_loss=0.348, pruned_loss=0.1162, over 3711893.85 frames. ], batch size: 74, lr: 1.70e-02, grad_scale: 8.0 2023-04-01 04:28:47,055 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.733e+02 7.486e+02 9.333e+02 1.140e+03 2.488e+03, threshold=1.867e+03, percent-clipped=5.0 2023-04-01 04:28:48,612 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.8712, 4.2136, 4.5650, 4.4925, 1.3534, 4.1292, 3.6166, 4.1400], device='cuda:1'), covar=tensor([0.0925, 0.0533, 0.0454, 0.0396, 0.4191, 0.0308, 0.0515, 0.0969], device='cuda:1'), in_proj_covar=tensor([0.0477, 0.0426, 0.0567, 0.0450, 0.0552, 0.0323, 0.0367, 0.0525], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1') 2023-04-01 04:29:25,954 INFO [train.py:903] (1/4) Epoch 5, batch 750, loss[loss=0.2999, simple_loss=0.3574, pruned_loss=0.1212, over 19775.00 frames. ], tot_loss[loss=0.2895, simple_loss=0.3474, pruned_loss=0.1158, over 3734520.67 frames. ], batch size: 54, lr: 1.70e-02, grad_scale: 8.0 2023-04-01 04:29:43,607 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=28076.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 04:30:27,876 INFO [train.py:903] (1/4) Epoch 5, batch 800, loss[loss=0.281, simple_loss=0.3437, pruned_loss=0.1092, over 19689.00 frames. ], tot_loss[loss=0.2882, simple_loss=0.3467, pruned_loss=0.1149, over 3757854.80 frames. ], batch size: 59, lr: 1.70e-02, grad_scale: 8.0 2023-04-01 04:30:48,156 WARNING [train.py:1073] (1/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 2023-04-01 04:30:52,980 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.190e+02 6.427e+02 8.567e+02 1.032e+03 2.729e+03, threshold=1.713e+03, percent-clipped=3.0 2023-04-01 04:31:32,197 INFO [train.py:903] (1/4) Epoch 5, batch 850, loss[loss=0.295, simple_loss=0.3568, pruned_loss=0.1166, over 19611.00 frames. ], tot_loss[loss=0.2884, simple_loss=0.3472, pruned_loss=0.1148, over 3777197.79 frames. ], batch size: 57, lr: 1.70e-02, grad_scale: 8.0 2023-04-01 04:32:08,722 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=28191.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 04:32:29,187 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9_rvb from training. Duration: 25.061125 2023-04-01 04:32:33,621 INFO [train.py:903] (1/4) Epoch 5, batch 900, loss[loss=0.285, simple_loss=0.3468, pruned_loss=0.1116, over 17302.00 frames. ], tot_loss[loss=0.2882, simple_loss=0.3465, pruned_loss=0.1149, over 3790717.22 frames. ], batch size: 101, lr: 1.69e-02, grad_scale: 8.0 2023-04-01 04:32:59,334 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.597e+02 6.938e+02 8.196e+02 1.125e+03 2.658e+03, threshold=1.639e+03, percent-clipped=4.0 2023-04-01 04:33:17,414 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=28247.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 04:33:22,451 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.17 vs. limit=5.0 2023-04-01 04:33:36,683 INFO [train.py:903] (1/4) Epoch 5, batch 950, loss[loss=0.3249, simple_loss=0.3745, pruned_loss=0.1376, over 19536.00 frames. ], tot_loss[loss=0.289, simple_loss=0.3474, pruned_loss=0.1153, over 3793106.16 frames. ], batch size: 54, lr: 1.69e-02, grad_scale: 8.0 2023-04-01 04:33:42,370 WARNING [train.py:1073] (1/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9_rvb from training. Duration: 26.32775 2023-04-01 04:33:49,427 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=28273.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 04:34:28,459 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=28305.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 04:34:36,191 INFO [train.py:903] (1/4) Epoch 5, batch 1000, loss[loss=0.2592, simple_loss=0.3084, pruned_loss=0.105, over 19372.00 frames. ], tot_loss[loss=0.2895, simple_loss=0.3477, pruned_loss=0.1156, over 3806577.74 frames. ], batch size: 47, lr: 1.69e-02, grad_scale: 8.0 2023-04-01 04:34:45,493 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1097, 1.8769, 1.3901, 1.1998, 1.7067, 0.9272, 0.8312, 1.5346], device='cuda:1'), covar=tensor([0.0549, 0.0432, 0.0881, 0.0499, 0.0379, 0.0997, 0.0582, 0.0331], device='cuda:1'), in_proj_covar=tensor([0.0273, 0.0259, 0.0311, 0.0236, 0.0218, 0.0308, 0.0277, 0.0254], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 04:34:59,300 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.517e+02 7.143e+02 8.888e+02 1.138e+03 2.880e+03, threshold=1.778e+03, percent-clipped=9.0 2023-04-01 04:35:30,267 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 2023-04-01 04:35:36,615 INFO [train.py:903] (1/4) Epoch 5, batch 1050, loss[loss=0.2849, simple_loss=0.3377, pruned_loss=0.116, over 19723.00 frames. ], tot_loss[loss=0.2904, simple_loss=0.348, pruned_loss=0.1164, over 3812819.55 frames. ], batch size: 51, lr: 1.69e-02, grad_scale: 8.0 2023-04-01 04:35:44,075 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2583, 1.3183, 1.2210, 1.0071, 0.9857, 1.2213, 0.0259, 0.4736], device='cuda:1'), covar=tensor([0.0262, 0.0237, 0.0159, 0.0200, 0.0551, 0.0183, 0.0404, 0.0389], device='cuda:1'), in_proj_covar=tensor([0.0280, 0.0273, 0.0269, 0.0294, 0.0355, 0.0279, 0.0267, 0.0279], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 04:36:09,492 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 2023-04-01 04:36:36,346 INFO [train.py:903] (1/4) Epoch 5, batch 1100, loss[loss=0.3699, simple_loss=0.4151, pruned_loss=0.1624, over 19512.00 frames. ], tot_loss[loss=0.2913, simple_loss=0.3488, pruned_loss=0.1169, over 3814180.45 frames. ], batch size: 64, lr: 1.69e-02, grad_scale: 8.0 2023-04-01 04:37:01,574 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.535e+02 6.917e+02 8.996e+02 1.190e+03 3.192e+03, threshold=1.799e+03, percent-clipped=6.0 2023-04-01 04:37:19,648 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=28447.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 04:37:37,664 INFO [train.py:903] (1/4) Epoch 5, batch 1150, loss[loss=0.3378, simple_loss=0.3755, pruned_loss=0.15, over 13435.00 frames. ], tot_loss[loss=0.2913, simple_loss=0.349, pruned_loss=0.1169, over 3804322.14 frames. ], batch size: 137, lr: 1.69e-02, grad_scale: 8.0 2023-04-01 04:37:50,298 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=28472.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 04:38:11,028 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.9895, 1.8781, 1.9457, 2.0983, 4.4295, 1.1855, 2.4475, 4.6577], device='cuda:1'), covar=tensor([0.0228, 0.2196, 0.2083, 0.1256, 0.0440, 0.2262, 0.1100, 0.0228], device='cuda:1'), in_proj_covar=tensor([0.0275, 0.0305, 0.0306, 0.0283, 0.0299, 0.0313, 0.0282, 0.0296], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 04:38:29,147 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=28504.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 04:38:36,075 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=28510.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 04:38:38,083 INFO [train.py:903] (1/4) Epoch 5, batch 1200, loss[loss=0.268, simple_loss=0.3424, pruned_loss=0.09681, over 19322.00 frames. ], tot_loss[loss=0.2909, simple_loss=0.3488, pruned_loss=0.1165, over 3818610.91 frames. ], batch size: 70, lr: 1.69e-02, grad_scale: 8.0 2023-04-01 04:39:01,758 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.620e+02 6.666e+02 7.999e+02 1.012e+03 1.920e+03, threshold=1.600e+03, percent-clipped=0.0 2023-04-01 04:39:12,213 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 2023-04-01 04:39:37,127 INFO [train.py:903] (1/4) Epoch 5, batch 1250, loss[loss=0.2571, simple_loss=0.32, pruned_loss=0.09706, over 19391.00 frames. ], tot_loss[loss=0.291, simple_loss=0.3487, pruned_loss=0.1166, over 3803631.64 frames. ], batch size: 47, lr: 1.68e-02, grad_scale: 8.0 2023-04-01 04:40:13,035 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=28591.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 04:40:37,857 INFO [train.py:903] (1/4) Epoch 5, batch 1300, loss[loss=0.2649, simple_loss=0.3375, pruned_loss=0.09617, over 19668.00 frames. ], tot_loss[loss=0.2895, simple_loss=0.3478, pruned_loss=0.1156, over 3820466.84 frames. ], batch size: 58, lr: 1.68e-02, grad_scale: 8.0 2023-04-01 04:40:43,785 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=28617.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 04:41:03,192 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.211e+02 7.072e+02 8.596e+02 1.219e+03 1.879e+03, threshold=1.719e+03, percent-clipped=8.0 2023-04-01 04:41:23,362 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=28649.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 04:41:37,882 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.4487, 1.3005, 1.3222, 1.7310, 3.0049, 1.2577, 2.0346, 3.2204], device='cuda:1'), covar=tensor([0.0333, 0.2386, 0.2537, 0.1379, 0.0524, 0.2033, 0.1144, 0.0353], device='cuda:1'), in_proj_covar=tensor([0.0276, 0.0306, 0.0307, 0.0284, 0.0298, 0.0313, 0.0281, 0.0295], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 04:41:39,854 INFO [train.py:903] (1/4) Epoch 5, batch 1350, loss[loss=0.3382, simple_loss=0.3836, pruned_loss=0.1464, over 19514.00 frames. ], tot_loss[loss=0.2877, simple_loss=0.3464, pruned_loss=0.1145, over 3831258.31 frames. ], batch size: 54, lr: 1.68e-02, grad_scale: 8.0 2023-04-01 04:42:14,813 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.2854, 2.3213, 2.0687, 3.4263, 2.2452, 3.2893, 2.9532, 2.0154], device='cuda:1'), covar=tensor([0.1792, 0.1342, 0.0726, 0.0819, 0.1692, 0.0517, 0.1218, 0.1221], device='cuda:1'), in_proj_covar=tensor([0.0600, 0.0591, 0.0540, 0.0745, 0.0642, 0.0508, 0.0653, 0.0557], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 04:42:34,081 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=28706.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 04:42:37,528 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3273, 2.2477, 1.5494, 1.5421, 2.0549, 1.0367, 1.2007, 1.7233], device='cuda:1'), covar=tensor([0.0718, 0.0433, 0.0819, 0.0508, 0.0382, 0.0964, 0.0602, 0.0375], device='cuda:1'), in_proj_covar=tensor([0.0274, 0.0258, 0.0309, 0.0238, 0.0219, 0.0308, 0.0277, 0.0255], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 04:42:40,551 INFO [train.py:903] (1/4) Epoch 5, batch 1400, loss[loss=0.303, simple_loss=0.3628, pruned_loss=0.1216, over 19663.00 frames. ], tot_loss[loss=0.2874, simple_loss=0.3463, pruned_loss=0.1142, over 3829248.32 frames. ], batch size: 55, lr: 1.68e-02, grad_scale: 8.0 2023-04-01 04:43:04,141 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.747e+02 6.792e+02 9.169e+02 1.129e+03 1.829e+03, threshold=1.834e+03, percent-clipped=1.0 2023-04-01 04:43:04,462 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=28732.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 04:43:40,834 INFO [train.py:903] (1/4) Epoch 5, batch 1450, loss[loss=0.2693, simple_loss=0.3401, pruned_loss=0.09927, over 19666.00 frames. ], tot_loss[loss=0.2879, simple_loss=0.3469, pruned_loss=0.1145, over 3836984.35 frames. ], batch size: 59, lr: 1.68e-02, grad_scale: 8.0 2023-04-01 04:43:43,179 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_rvb from training. Duration: 25.85 2023-04-01 04:43:43,512 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=28764.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 04:44:37,599 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.2132, 1.6179, 1.7366, 2.6002, 1.7416, 2.3610, 2.6017, 2.2270], device='cuda:1'), covar=tensor([0.0726, 0.1052, 0.1101, 0.1005, 0.1208, 0.0743, 0.0932, 0.0738], device='cuda:1'), in_proj_covar=tensor([0.0238, 0.0261, 0.0253, 0.0289, 0.0286, 0.0241, 0.0247, 0.0238], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:1') 2023-04-01 04:44:41,294 INFO [train.py:903] (1/4) Epoch 5, batch 1500, loss[loss=0.2876, simple_loss=0.3244, pruned_loss=0.1254, over 19034.00 frames. ], tot_loss[loss=0.2866, simple_loss=0.3456, pruned_loss=0.1138, over 3847760.21 frames. ], batch size: 42, lr: 1.68e-02, grad_scale: 8.0 2023-04-01 04:45:06,145 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.730e+02 6.912e+02 8.562e+02 1.022e+03 2.509e+03, threshold=1.712e+03, percent-clipped=1.0 2023-04-01 04:45:24,298 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=28848.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 04:45:31,721 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=28854.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 04:45:42,352 INFO [train.py:903] (1/4) Epoch 5, batch 1550, loss[loss=0.2877, simple_loss=0.3574, pruned_loss=0.109, over 19665.00 frames. ], tot_loss[loss=0.289, simple_loss=0.3479, pruned_loss=0.115, over 3843760.28 frames. ], batch size: 60, lr: 1.68e-02, grad_scale: 8.0 2023-04-01 04:45:56,766 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4437, 1.2578, 1.7762, 1.2420, 2.8291, 3.5981, 3.5311, 3.7696], device='cuda:1'), covar=tensor([0.1333, 0.2767, 0.2527, 0.1832, 0.0398, 0.0118, 0.0178, 0.0120], device='cuda:1'), in_proj_covar=tensor([0.0261, 0.0282, 0.0320, 0.0257, 0.0199, 0.0117, 0.0206, 0.0139], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-04-01 04:46:02,322 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=28879.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 04:46:04,681 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.6230, 4.1252, 4.2819, 4.2226, 1.3979, 3.9255, 3.5462, 3.8233], device='cuda:1'), covar=tensor([0.1029, 0.0598, 0.0579, 0.0508, 0.3809, 0.0357, 0.0515, 0.1194], device='cuda:1'), in_proj_covar=tensor([0.0504, 0.0451, 0.0600, 0.0479, 0.0572, 0.0342, 0.0387, 0.0553], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1') 2023-04-01 04:46:41,466 INFO [train.py:903] (1/4) Epoch 5, batch 1600, loss[loss=0.2603, simple_loss=0.3039, pruned_loss=0.1084, over 19734.00 frames. ], tot_loss[loss=0.2896, simple_loss=0.3483, pruned_loss=0.1154, over 3845247.23 frames. ], batch size: 47, lr: 1.67e-02, grad_scale: 8.0 2023-04-01 04:47:04,538 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.940e+02 6.984e+02 8.420e+02 1.102e+03 2.946e+03, threshold=1.684e+03, percent-clipped=4.0 2023-04-01 04:47:04,569 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9_rvb from training. Duration: 30.1555625 2023-04-01 04:47:28,988 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.44 vs. limit=2.0 2023-04-01 04:47:41,581 INFO [train.py:903] (1/4) Epoch 5, batch 1650, loss[loss=0.2384, simple_loss=0.3007, pruned_loss=0.08805, over 19776.00 frames. ], tot_loss[loss=0.2887, simple_loss=0.3477, pruned_loss=0.1148, over 3839559.95 frames. ], batch size: 45, lr: 1.67e-02, grad_scale: 4.0 2023-04-01 04:47:41,971 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=28962.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 04:47:43,165 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=28963.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 04:47:49,859 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=28969.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 04:48:00,800 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1192, 1.2643, 1.9569, 1.3799, 2.6712, 2.1868, 2.8165, 1.0692], device='cuda:1'), covar=tensor([0.1854, 0.3015, 0.1487, 0.1553, 0.1201, 0.1456, 0.1346, 0.2657], device='cuda:1'), in_proj_covar=tensor([0.0435, 0.0498, 0.0469, 0.0402, 0.0541, 0.0437, 0.0620, 0.0432], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 04:48:12,488 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=28987.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 04:48:14,414 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=28988.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 04:48:30,715 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9449, 1.4089, 1.5206, 2.0732, 1.6877, 1.6673, 1.8295, 1.8512], device='cuda:1'), covar=tensor([0.0800, 0.1656, 0.1291, 0.0851, 0.1147, 0.0520, 0.0819, 0.0677], device='cuda:1'), in_proj_covar=tensor([0.0249, 0.0354, 0.0277, 0.0238, 0.0304, 0.0248, 0.0270, 0.0231], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 04:48:42,351 INFO [train.py:903] (1/4) Epoch 5, batch 1700, loss[loss=0.3554, simple_loss=0.3899, pruned_loss=0.1604, over 18167.00 frames. ], tot_loss[loss=0.2893, simple_loss=0.3479, pruned_loss=0.1153, over 3833082.23 frames. ], batch size: 83, lr: 1.67e-02, grad_scale: 4.0 2023-04-01 04:48:43,820 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=29013.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 04:48:51,931 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=29020.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 04:48:52,063 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=29020.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 04:49:04,396 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.1811, 1.4859, 1.6443, 2.1107, 1.6698, 1.9076, 1.8965, 2.0265], device='cuda:1'), covar=tensor([0.0738, 0.1698, 0.1291, 0.0854, 0.1249, 0.0430, 0.0924, 0.0576], device='cuda:1'), in_proj_covar=tensor([0.0250, 0.0360, 0.0281, 0.0239, 0.0306, 0.0249, 0.0274, 0.0233], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 04:49:08,763 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.635e+02 6.699e+02 8.227e+02 1.083e+03 2.721e+03, threshold=1.645e+03, percent-clipped=4.0 2023-04-01 04:49:21,047 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 2023-04-01 04:49:22,490 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=29045.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 04:49:42,474 INFO [train.py:903] (1/4) Epoch 5, batch 1750, loss[loss=0.3077, simple_loss=0.3638, pruned_loss=0.1258, over 19536.00 frames. ], tot_loss[loss=0.2878, simple_loss=0.3468, pruned_loss=0.1145, over 3829874.83 frames. ], batch size: 54, lr: 1.67e-02, grad_scale: 4.0 2023-04-01 04:50:43,926 INFO [train.py:903] (1/4) Epoch 5, batch 1800, loss[loss=0.3417, simple_loss=0.3904, pruned_loss=0.1465, over 17398.00 frames. ], tot_loss[loss=0.2879, simple_loss=0.3469, pruned_loss=0.1144, over 3819222.44 frames. ], batch size: 101, lr: 1.67e-02, grad_scale: 4.0 2023-04-01 04:50:54,482 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.5087, 1.2487, 1.3464, 1.7350, 2.9777, 0.9793, 2.0622, 3.1326], device='cuda:1'), covar=tensor([0.0347, 0.2532, 0.2510, 0.1445, 0.0567, 0.2491, 0.1173, 0.0422], device='cuda:1'), in_proj_covar=tensor([0.0280, 0.0307, 0.0310, 0.0287, 0.0301, 0.0314, 0.0281, 0.0300], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 04:51:07,717 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.740e+02 6.050e+02 7.849e+02 1.011e+03 2.328e+03, threshold=1.570e+03, percent-clipped=7.0 2023-04-01 04:51:27,789 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 2023-04-01 04:51:39,751 WARNING [train.py:1073] (1/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9_rvb from training. Duration: 25.0944375 2023-04-01 04:51:44,192 INFO [train.py:903] (1/4) Epoch 5, batch 1850, loss[loss=0.2672, simple_loss=0.3251, pruned_loss=0.1047, over 19748.00 frames. ], tot_loss[loss=0.2856, simple_loss=0.3451, pruned_loss=0.1131, over 3820821.99 frames. ], batch size: 46, lr: 1.67e-02, grad_scale: 4.0 2023-04-01 04:52:17,655 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9_rvb from training. Duration: 27.8166875 2023-04-01 04:52:43,727 INFO [train.py:903] (1/4) Epoch 5, batch 1900, loss[loss=0.2529, simple_loss=0.316, pruned_loss=0.09496, over 19618.00 frames. ], tot_loss[loss=0.2864, simple_loss=0.3456, pruned_loss=0.1136, over 3825564.53 frames. ], batch size: 50, lr: 1.67e-02, grad_scale: 4.0 2023-04-01 04:52:53,173 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=29219.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 04:52:58,362 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=29223.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 04:53:01,718 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=29225.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 04:53:03,249 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 2023-04-01 04:53:07,800 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9_rvb from training. Duration: 27.02225 2023-04-01 04:53:11,315 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.676e+02 6.988e+02 9.073e+02 1.156e+03 1.890e+03, threshold=1.815e+03, percent-clipped=5.0 2023-04-01 04:53:12,554 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6539, 1.3516, 1.2533, 1.7152, 1.4273, 1.5515, 1.2952, 1.5985], device='cuda:1'), covar=tensor([0.0803, 0.1246, 0.1303, 0.0690, 0.0949, 0.0465, 0.0954, 0.0671], device='cuda:1'), in_proj_covar=tensor([0.0252, 0.0368, 0.0287, 0.0239, 0.0305, 0.0247, 0.0274, 0.0234], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 04:53:23,913 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=29244.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 04:53:28,888 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 2023-04-01 04:53:30,419 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=29250.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 04:53:35,748 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=29254.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 04:53:44,960 INFO [train.py:903] (1/4) Epoch 5, batch 1950, loss[loss=0.2852, simple_loss=0.3502, pruned_loss=0.1101, over 19515.00 frames. ], tot_loss[loss=0.2874, simple_loss=0.3464, pruned_loss=0.1142, over 3824482.57 frames. ], batch size: 64, lr: 1.67e-02, grad_scale: 4.0 2023-04-01 04:54:46,875 INFO [train.py:903] (1/4) Epoch 5, batch 2000, loss[loss=0.2338, simple_loss=0.293, pruned_loss=0.08728, over 19725.00 frames. ], tot_loss[loss=0.2868, simple_loss=0.3461, pruned_loss=0.1138, over 3834566.55 frames. ], batch size: 46, lr: 1.66e-02, grad_scale: 8.0 2023-04-01 04:55:10,279 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.142e+02 7.102e+02 9.141e+02 1.135e+03 3.050e+03, threshold=1.828e+03, percent-clipped=2.0 2023-04-01 04:55:16,876 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=29338.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 04:55:21,655 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.81 vs. limit=2.0 2023-04-01 04:55:36,606 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3650, 1.1904, 1.6893, 1.1420, 2.7896, 3.3812, 3.2777, 3.6307], device='cuda:1'), covar=tensor([0.1280, 0.2865, 0.2644, 0.1915, 0.0388, 0.0131, 0.0183, 0.0115], device='cuda:1'), in_proj_covar=tensor([0.0255, 0.0280, 0.0314, 0.0256, 0.0196, 0.0115, 0.0204, 0.0135], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-04-01 04:55:43,159 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_rvb from training. Duration: 26.8349375 2023-04-01 04:55:46,565 INFO [train.py:903] (1/4) Epoch 5, batch 2050, loss[loss=0.2817, simple_loss=0.3454, pruned_loss=0.1089, over 19625.00 frames. ], tot_loss[loss=0.2874, simple_loss=0.3467, pruned_loss=0.114, over 3816962.25 frames. ], batch size: 57, lr: 1.66e-02, grad_scale: 8.0 2023-04-01 04:55:49,013 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=29364.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 04:55:59,868 WARNING [train.py:1073] (1/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 2023-04-01 04:56:00,828 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 2023-04-01 04:56:23,069 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 2023-04-01 04:56:37,156 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3950, 1.5359, 2.0722, 1.5028, 2.9429, 2.9010, 3.3930, 1.3411], device='cuda:1'), covar=tensor([0.1716, 0.2840, 0.1634, 0.1418, 0.1323, 0.1224, 0.1401, 0.2762], device='cuda:1'), in_proj_covar=tensor([0.0440, 0.0502, 0.0472, 0.0405, 0.0538, 0.0434, 0.0620, 0.0436], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 04:56:45,827 INFO [train.py:903] (1/4) Epoch 5, batch 2100, loss[loss=0.2436, simple_loss=0.3123, pruned_loss=0.0875, over 19758.00 frames. ], tot_loss[loss=0.2875, simple_loss=0.3463, pruned_loss=0.1143, over 3799425.16 frames. ], batch size: 51, lr: 1.66e-02, grad_scale: 8.0 2023-04-01 04:57:07,794 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.47 vs. limit=2.0 2023-04-01 04:57:12,304 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.101e+02 7.145e+02 9.347e+02 1.270e+03 4.921e+03, threshold=1.869e+03, percent-clipped=10.0 2023-04-01 04:57:13,527 WARNING [train.py:1073] (1/4) Exclude cut with ID 453-131332-0000-131866_sp0.9_rvb from training. Duration: 25.3333125 2023-04-01 04:57:20,943 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1343, 2.1227, 1.6162, 1.2105, 1.9387, 0.8447, 0.9226, 1.7657], device='cuda:1'), covar=tensor([0.0672, 0.0364, 0.0685, 0.0560, 0.0278, 0.1050, 0.0595, 0.0252], device='cuda:1'), in_proj_covar=tensor([0.0270, 0.0260, 0.0306, 0.0236, 0.0212, 0.0306, 0.0280, 0.0252], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 04:57:34,264 WARNING [train.py:1073] (1/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9_rvb from training. Duration: 26.6166875 2023-04-01 04:57:46,682 INFO [train.py:903] (1/4) Epoch 5, batch 2150, loss[loss=0.2545, simple_loss=0.3195, pruned_loss=0.09468, over 19741.00 frames. ], tot_loss[loss=0.289, simple_loss=0.3476, pruned_loss=0.1152, over 3804339.81 frames. ], batch size: 51, lr: 1.66e-02, grad_scale: 8.0 2023-04-01 04:58:08,975 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=29479.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 04:58:48,386 INFO [train.py:903] (1/4) Epoch 5, batch 2200, loss[loss=0.283, simple_loss=0.3322, pruned_loss=0.1169, over 19811.00 frames. ], tot_loss[loss=0.29, simple_loss=0.3479, pruned_loss=0.116, over 3810981.61 frames. ], batch size: 49, lr: 1.66e-02, grad_scale: 8.0 2023-04-01 04:59:11,934 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.753e+02 6.760e+02 8.285e+02 1.117e+03 1.782e+03, threshold=1.657e+03, percent-clipped=0.0 2023-04-01 04:59:47,483 INFO [train.py:903] (1/4) Epoch 5, batch 2250, loss[loss=0.2944, simple_loss=0.3588, pruned_loss=0.1151, over 19688.00 frames. ], tot_loss[loss=0.2899, simple_loss=0.3482, pruned_loss=0.1158, over 3822122.10 frames. ], batch size: 60, lr: 1.66e-02, grad_scale: 8.0 2023-04-01 04:59:55,512 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=29569.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:00:28,206 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=29594.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:00:31,572 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6466, 1.1515, 1.2017, 1.8542, 1.4371, 1.8777, 1.9929, 1.5965], device='cuda:1'), covar=tensor([0.0832, 0.1149, 0.1163, 0.1049, 0.1012, 0.0740, 0.0970, 0.0802], device='cuda:1'), in_proj_covar=tensor([0.0234, 0.0253, 0.0245, 0.0283, 0.0277, 0.0230, 0.0240, 0.0229], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:1') 2023-04-01 05:00:32,406 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=29598.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:00:48,023 INFO [train.py:903] (1/4) Epoch 5, batch 2300, loss[loss=0.2453, simple_loss=0.3187, pruned_loss=0.08594, over 19675.00 frames. ], tot_loss[loss=0.29, simple_loss=0.3483, pruned_loss=0.1158, over 3813478.50 frames. ], batch size: 60, lr: 1.66e-02, grad_scale: 8.0 2023-04-01 05:00:56,341 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=29619.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:01:03,671 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_rvb from training. Duration: 26.205 2023-04-01 05:01:15,151 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.694e+02 6.380e+02 7.654e+02 9.572e+02 1.859e+03, threshold=1.531e+03, percent-clipped=2.0 2023-04-01 05:01:48,834 INFO [train.py:903] (1/4) Epoch 5, batch 2350, loss[loss=0.2873, simple_loss=0.3531, pruned_loss=0.1108, over 17387.00 frames. ], tot_loss[loss=0.2873, simple_loss=0.3465, pruned_loss=0.114, over 3827715.62 frames. ], batch size: 101, lr: 1.65e-02, grad_scale: 8.0 2023-04-01 05:02:11,372 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=29680.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:02:28,857 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_rvb from training. Duration: 25.775 2023-04-01 05:02:45,294 WARNING [train.py:1073] (1/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9_rvb from training. Duration: 25.45 2023-04-01 05:02:49,425 INFO [train.py:903] (1/4) Epoch 5, batch 2400, loss[loss=0.2738, simple_loss=0.3429, pruned_loss=0.1023, over 19741.00 frames. ], tot_loss[loss=0.2869, simple_loss=0.3463, pruned_loss=0.1138, over 3828292.31 frames. ], batch size: 63, lr: 1.65e-02, grad_scale: 8.0 2023-04-01 05:02:50,615 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=29713.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:03:12,593 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.376e+02 6.843e+02 7.889e+02 1.103e+03 3.246e+03, threshold=1.578e+03, percent-clipped=5.0 2023-04-01 05:03:15,322 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=29735.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:03:46,786 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=29760.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:03:48,491 INFO [train.py:903] (1/4) Epoch 5, batch 2450, loss[loss=0.3223, simple_loss=0.3578, pruned_loss=0.1434, over 19373.00 frames. ], tot_loss[loss=0.2865, simple_loss=0.3458, pruned_loss=0.1136, over 3839501.91 frames. ], batch size: 47, lr: 1.65e-02, grad_scale: 8.0 2023-04-01 05:04:48,603 INFO [train.py:903] (1/4) Epoch 5, batch 2500, loss[loss=0.2088, simple_loss=0.2767, pruned_loss=0.07041, over 19793.00 frames. ], tot_loss[loss=0.2862, simple_loss=0.345, pruned_loss=0.1137, over 3838057.88 frames. ], batch size: 48, lr: 1.65e-02, grad_scale: 8.0 2023-04-01 05:04:57,742 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=29820.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:05:14,590 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.471e+02 6.790e+02 8.500e+02 1.086e+03 2.138e+03, threshold=1.700e+03, percent-clipped=3.0 2023-04-01 05:05:48,274 INFO [train.py:903] (1/4) Epoch 5, batch 2550, loss[loss=0.3296, simple_loss=0.384, pruned_loss=0.1376, over 18009.00 frames. ], tot_loss[loss=0.2867, simple_loss=0.3454, pruned_loss=0.114, over 3811281.60 frames. ], batch size: 83, lr: 1.65e-02, grad_scale: 8.0 2023-04-01 05:06:40,399 WARNING [train.py:1073] (1/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 2023-04-01 05:06:48,587 INFO [train.py:903] (1/4) Epoch 5, batch 2600, loss[loss=0.2647, simple_loss=0.3264, pruned_loss=0.1015, over 19621.00 frames. ], tot_loss[loss=0.2889, simple_loss=0.3469, pruned_loss=0.1155, over 3794034.89 frames. ], batch size: 50, lr: 1.65e-02, grad_scale: 8.0 2023-04-01 05:06:50,692 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=29913.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:07:05,031 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9177, 1.8343, 1.9169, 2.9363, 1.8737, 2.7580, 2.6507, 1.8719], device='cuda:1'), covar=tensor([0.1848, 0.1494, 0.0780, 0.0781, 0.1693, 0.0558, 0.1345, 0.1352], device='cuda:1'), in_proj_covar=tensor([0.0607, 0.0597, 0.0541, 0.0755, 0.0646, 0.0516, 0.0654, 0.0563], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 05:07:13,514 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.137e+02 6.882e+02 9.175e+02 1.239e+03 1.828e+03, threshold=1.835e+03, percent-clipped=5.0 2023-04-01 05:07:50,211 INFO [train.py:903] (1/4) Epoch 5, batch 2650, loss[loss=0.2665, simple_loss=0.3375, pruned_loss=0.09771, over 19751.00 frames. ], tot_loss[loss=0.2879, simple_loss=0.3462, pruned_loss=0.1148, over 3802353.59 frames. ], batch size: 54, lr: 1.65e-02, grad_scale: 8.0 2023-04-01 05:07:58,249 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=29969.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:08:08,349 WARNING [train.py:1073] (1/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9_rvb from training. Duration: 27.25 2023-04-01 05:08:29,864 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=29994.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:08:48,857 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([0.9846, 1.8568, 1.3606, 1.2241, 1.7326, 0.9310, 0.9484, 1.6699], device='cuda:1'), covar=tensor([0.0641, 0.0417, 0.0787, 0.0459, 0.0305, 0.0941, 0.0565, 0.0273], device='cuda:1'), in_proj_covar=tensor([0.0279, 0.0266, 0.0313, 0.0237, 0.0221, 0.0310, 0.0288, 0.0255], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 05:08:50,700 INFO [train.py:903] (1/4) Epoch 5, batch 2700, loss[loss=0.2702, simple_loss=0.3323, pruned_loss=0.1041, over 19659.00 frames. ], tot_loss[loss=0.2871, simple_loss=0.3455, pruned_loss=0.1143, over 3794218.12 frames. ], batch size: 53, lr: 1.64e-02, grad_scale: 8.0 2023-04-01 05:09:04,702 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=30024.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:09:10,326 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=30028.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:09:17,304 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.255e+02 6.674e+02 8.375e+02 9.964e+02 1.932e+03, threshold=1.675e+03, percent-clipped=1.0 2023-04-01 05:09:31,506 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.1550, 2.1085, 1.9930, 3.2136, 2.2633, 3.1744, 2.8532, 1.9549], device='cuda:1'), covar=tensor([0.1972, 0.1520, 0.0828, 0.0920, 0.1822, 0.0583, 0.1462, 0.1387], device='cuda:1'), in_proj_covar=tensor([0.0604, 0.0596, 0.0543, 0.0754, 0.0651, 0.0518, 0.0662, 0.0564], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 05:09:49,948 INFO [train.py:903] (1/4) Epoch 5, batch 2750, loss[loss=0.2695, simple_loss=0.329, pruned_loss=0.105, over 19394.00 frames. ], tot_loss[loss=0.2864, simple_loss=0.3452, pruned_loss=0.1139, over 3804493.16 frames. ], batch size: 48, lr: 1.64e-02, grad_scale: 4.0 2023-04-01 05:10:10,883 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.9725, 4.4120, 4.6710, 4.5895, 1.5700, 4.2663, 3.8076, 4.2014], device='cuda:1'), covar=tensor([0.0879, 0.0504, 0.0422, 0.0369, 0.4058, 0.0323, 0.0439, 0.0968], device='cuda:1'), in_proj_covar=tensor([0.0503, 0.0448, 0.0598, 0.0487, 0.0575, 0.0353, 0.0381, 0.0552], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1') 2023-04-01 05:10:42,536 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.4739, 1.2728, 1.3406, 1.6068, 3.0739, 1.0603, 2.0284, 2.9815], device='cuda:1'), covar=tensor([0.0356, 0.2308, 0.2397, 0.1458, 0.0537, 0.2135, 0.1244, 0.0435], device='cuda:1'), in_proj_covar=tensor([0.0284, 0.0313, 0.0312, 0.0287, 0.0304, 0.0311, 0.0288, 0.0303], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 05:10:50,795 INFO [train.py:903] (1/4) Epoch 5, batch 2800, loss[loss=0.2826, simple_loss=0.3434, pruned_loss=0.1109, over 19529.00 frames. ], tot_loss[loss=0.2874, simple_loss=0.3462, pruned_loss=0.1143, over 3813633.43 frames. ], batch size: 54, lr: 1.64e-02, grad_scale: 8.0 2023-04-01 05:11:17,048 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.279e+02 7.041e+02 9.072e+02 1.191e+03 2.188e+03, threshold=1.814e+03, percent-clipped=6.0 2023-04-01 05:11:22,780 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=30139.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:11:27,357 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=30143.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:11:51,983 INFO [train.py:903] (1/4) Epoch 5, batch 2850, loss[loss=0.2446, simple_loss=0.3103, pruned_loss=0.08949, over 19783.00 frames. ], tot_loss[loss=0.2879, simple_loss=0.3468, pruned_loss=0.1145, over 3813802.66 frames. ], batch size: 49, lr: 1.64e-02, grad_scale: 4.0 2023-04-01 05:11:54,295 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=30164.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:12:15,906 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.8398, 1.4571, 1.4446, 2.0025, 1.5593, 2.0418, 2.1481, 2.0074], device='cuda:1'), covar=tensor([0.0746, 0.1022, 0.1064, 0.0965, 0.1013, 0.0769, 0.0832, 0.0635], device='cuda:1'), in_proj_covar=tensor([0.0237, 0.0259, 0.0249, 0.0288, 0.0282, 0.0239, 0.0242, 0.0235], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:1') 2023-04-01 05:12:46,630 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2784, 2.1908, 1.6740, 1.4904, 1.9865, 1.1710, 1.1313, 1.8856], device='cuda:1'), covar=tensor([0.0560, 0.0401, 0.0783, 0.0483, 0.0342, 0.0961, 0.0563, 0.0266], device='cuda:1'), in_proj_covar=tensor([0.0279, 0.0265, 0.0310, 0.0235, 0.0219, 0.0310, 0.0286, 0.0254], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 05:12:51,657 INFO [train.py:903] (1/4) Epoch 5, batch 2900, loss[loss=0.3045, simple_loss=0.3684, pruned_loss=0.1203, over 19577.00 frames. ], tot_loss[loss=0.2886, simple_loss=0.3475, pruned_loss=0.1148, over 3817162.37 frames. ], batch size: 61, lr: 1.64e-02, grad_scale: 4.0 2023-04-01 05:12:51,674 WARNING [train.py:1073] (1/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 2023-04-01 05:13:09,285 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=30227.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:13:20,142 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.023e+02 7.237e+02 9.091e+02 1.153e+03 2.755e+03, threshold=1.818e+03, percent-clipped=7.0 2023-04-01 05:13:28,894 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5474, 1.2447, 1.2139, 2.0751, 1.6016, 1.7925, 2.1892, 1.8013], device='cuda:1'), covar=tensor([0.0873, 0.1124, 0.1186, 0.0835, 0.1006, 0.0823, 0.0781, 0.0689], device='cuda:1'), in_proj_covar=tensor([0.0238, 0.0259, 0.0248, 0.0286, 0.0280, 0.0237, 0.0242, 0.0233], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:1') 2023-04-01 05:13:51,429 INFO [train.py:903] (1/4) Epoch 5, batch 2950, loss[loss=0.2413, simple_loss=0.2915, pruned_loss=0.09557, over 18969.00 frames. ], tot_loss[loss=0.2887, simple_loss=0.3476, pruned_loss=0.1149, over 3807402.84 frames. ], batch size: 42, lr: 1.64e-02, grad_scale: 4.0 2023-04-01 05:14:12,794 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=30279.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:14:18,160 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=30284.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:14:29,249 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9402, 1.6051, 1.6578, 2.2990, 1.8883, 2.1281, 2.2458, 2.0417], device='cuda:1'), covar=tensor([0.0670, 0.0934, 0.0865, 0.0667, 0.0844, 0.0626, 0.0679, 0.0552], device='cuda:1'), in_proj_covar=tensor([0.0236, 0.0258, 0.0247, 0.0282, 0.0279, 0.0235, 0.0240, 0.0232], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:1') 2023-04-01 05:14:46,949 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=30309.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:14:50,603 INFO [train.py:903] (1/4) Epoch 5, batch 3000, loss[loss=0.3269, simple_loss=0.3783, pruned_loss=0.1378, over 18077.00 frames. ], tot_loss[loss=0.2901, simple_loss=0.3486, pruned_loss=0.1158, over 3806522.43 frames. ], batch size: 83, lr: 1.64e-02, grad_scale: 4.0 2023-04-01 05:14:50,603 INFO [train.py:928] (1/4) Computing validation loss 2023-04-01 05:15:03,126 INFO [train.py:937] (1/4) Epoch 5, validation: loss=0.2047, simple_loss=0.3034, pruned_loss=0.05296, over 944034.00 frames. 2023-04-01 05:15:03,127 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18649MB 2023-04-01 05:15:05,714 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_rvb from training. Duration: 29.735 2023-04-01 05:15:08,541 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1497, 1.1952, 1.6838, 1.3680, 2.5733, 2.2016, 2.7471, 0.9618], device='cuda:1'), covar=tensor([0.1793, 0.2982, 0.1643, 0.1464, 0.1178, 0.1423, 0.1288, 0.2763], device='cuda:1'), in_proj_covar=tensor([0.0444, 0.0507, 0.0478, 0.0408, 0.0550, 0.0439, 0.0626, 0.0441], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 05:15:33,552 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.690e+02 7.155e+02 8.736e+02 1.085e+03 2.346e+03, threshold=1.747e+03, percent-clipped=4.0 2023-04-01 05:16:06,358 INFO [train.py:903] (1/4) Epoch 5, batch 3050, loss[loss=0.2684, simple_loss=0.3433, pruned_loss=0.09676, over 19643.00 frames. ], tot_loss[loss=0.2878, simple_loss=0.3469, pruned_loss=0.1143, over 3812191.99 frames. ], batch size: 55, lr: 1.64e-02, grad_scale: 4.0 2023-04-01 05:16:19,957 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.3691, 2.1835, 1.8293, 1.8055, 1.5357, 1.8157, 0.4906, 1.2346], device='cuda:1'), covar=tensor([0.0238, 0.0233, 0.0183, 0.0269, 0.0495, 0.0301, 0.0481, 0.0428], device='cuda:1'), in_proj_covar=tensor([0.0283, 0.0280, 0.0282, 0.0299, 0.0365, 0.0287, 0.0275, 0.0291], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 05:16:22,185 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.9107, 1.7540, 1.7587, 2.0365, 4.2815, 0.9554, 2.2603, 4.2092], device='cuda:1'), covar=tensor([0.0249, 0.2221, 0.2180, 0.1332, 0.0457, 0.2346, 0.1185, 0.0319], device='cuda:1'), in_proj_covar=tensor([0.0286, 0.0311, 0.0311, 0.0286, 0.0307, 0.0317, 0.0289, 0.0304], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 05:16:26,382 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=30378.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:16:45,447 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=30395.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:17:07,569 INFO [train.py:903] (1/4) Epoch 5, batch 3100, loss[loss=0.2781, simple_loss=0.3387, pruned_loss=0.1088, over 19734.00 frames. ], tot_loss[loss=0.2876, simple_loss=0.3467, pruned_loss=0.1142, over 3808600.73 frames. ], batch size: 51, lr: 1.63e-02, grad_scale: 4.0 2023-04-01 05:17:17,149 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=30420.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:17:33,697 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.705e+02 6.847e+02 8.274e+02 1.001e+03 3.134e+03, threshold=1.655e+03, percent-clipped=2.0 2023-04-01 05:17:41,071 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.62 vs. limit=2.0 2023-04-01 05:18:06,710 INFO [train.py:903] (1/4) Epoch 5, batch 3150, loss[loss=0.3162, simple_loss=0.3674, pruned_loss=0.1325, over 18781.00 frames. ], tot_loss[loss=0.2881, simple_loss=0.3469, pruned_loss=0.1147, over 3818768.97 frames. ], batch size: 74, lr: 1.63e-02, grad_scale: 4.0 2023-04-01 05:18:34,306 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 2023-04-01 05:18:37,227 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=30487.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:19:06,281 INFO [train.py:903] (1/4) Epoch 5, batch 3200, loss[loss=0.2923, simple_loss=0.3592, pruned_loss=0.1127, over 19691.00 frames. ], tot_loss[loss=0.2868, simple_loss=0.346, pruned_loss=0.1138, over 3818907.67 frames. ], batch size: 59, lr: 1.63e-02, grad_scale: 8.0 2023-04-01 05:19:23,695 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.4626, 1.3493, 1.4713, 1.7548, 3.1612, 1.0469, 2.0244, 3.2698], device='cuda:1'), covar=tensor([0.0354, 0.2120, 0.2204, 0.1354, 0.0438, 0.2169, 0.1139, 0.0336], device='cuda:1'), in_proj_covar=tensor([0.0282, 0.0306, 0.0306, 0.0285, 0.0300, 0.0312, 0.0284, 0.0302], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 05:19:26,759 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.9934, 4.3166, 4.6533, 4.6481, 1.6642, 4.2735, 3.8006, 4.2208], device='cuda:1'), covar=tensor([0.0931, 0.0628, 0.0438, 0.0403, 0.3929, 0.0367, 0.0473, 0.0939], device='cuda:1'), in_proj_covar=tensor([0.0498, 0.0437, 0.0589, 0.0479, 0.0566, 0.0352, 0.0380, 0.0554], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1') 2023-04-01 05:19:35,545 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.065e+02 7.005e+02 8.588e+02 1.128e+03 3.335e+03, threshold=1.718e+03, percent-clipped=13.0 2023-04-01 05:19:35,989 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=30535.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:20:04,478 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=30560.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:20:06,322 INFO [train.py:903] (1/4) Epoch 5, batch 3250, loss[loss=0.3064, simple_loss=0.3632, pruned_loss=0.1248, over 18193.00 frames. ], tot_loss[loss=0.2852, simple_loss=0.3449, pruned_loss=0.1128, over 3826838.22 frames. ], batch size: 83, lr: 1.63e-02, grad_scale: 8.0 2023-04-01 05:20:19,560 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=30571.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:20:55,340 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=30602.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:21:08,677 INFO [train.py:903] (1/4) Epoch 5, batch 3300, loss[loss=0.416, simple_loss=0.4445, pruned_loss=0.1937, over 13845.00 frames. ], tot_loss[loss=0.2865, simple_loss=0.3461, pruned_loss=0.1134, over 3828759.11 frames. ], batch size: 136, lr: 1.63e-02, grad_scale: 8.0 2023-04-01 05:21:16,483 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 2023-04-01 05:21:35,516 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.601e+02 6.699e+02 7.755e+02 9.198e+02 2.155e+03, threshold=1.551e+03, percent-clipped=1.0 2023-04-01 05:22:09,096 INFO [train.py:903] (1/4) Epoch 5, batch 3350, loss[loss=0.2753, simple_loss=0.3492, pruned_loss=0.1007, over 19687.00 frames. ], tot_loss[loss=0.2851, simple_loss=0.345, pruned_loss=0.1126, over 3844121.09 frames. ], batch size: 59, lr: 1.63e-02, grad_scale: 8.0 2023-04-01 05:22:21,636 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=30673.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:22:38,249 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=30686.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:23:09,773 INFO [train.py:903] (1/4) Epoch 5, batch 3400, loss[loss=0.2852, simple_loss=0.355, pruned_loss=0.1078, over 19474.00 frames. ], tot_loss[loss=0.2849, simple_loss=0.3451, pruned_loss=0.1124, over 3845809.46 frames. ], batch size: 64, lr: 1.63e-02, grad_scale: 8.0 2023-04-01 05:23:22,329 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=30722.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:23:39,933 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.254e+02 6.893e+02 8.724e+02 1.073e+03 2.213e+03, threshold=1.745e+03, percent-clipped=7.0 2023-04-01 05:24:11,363 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 2023-04-01 05:24:11,823 INFO [train.py:903] (1/4) Epoch 5, batch 3450, loss[loss=0.2581, simple_loss=0.3099, pruned_loss=0.1031, over 19705.00 frames. ], tot_loss[loss=0.2855, simple_loss=0.3459, pruned_loss=0.1126, over 3835743.78 frames. ], batch size: 45, lr: 1.63e-02, grad_scale: 8.0 2023-04-01 05:24:15,170 WARNING [train.py:1073] (1/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9_rvb from training. Duration: 25.2444375 2023-04-01 05:25:13,747 INFO [train.py:903] (1/4) Epoch 5, batch 3500, loss[loss=0.3141, simple_loss=0.3751, pruned_loss=0.1266, over 19609.00 frames. ], tot_loss[loss=0.2847, simple_loss=0.3452, pruned_loss=0.1121, over 3832294.99 frames. ], batch size: 57, lr: 1.62e-02, grad_scale: 8.0 2023-04-01 05:25:39,211 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.495e+02 7.015e+02 8.195e+02 1.097e+03 2.546e+03, threshold=1.639e+03, percent-clipped=5.0 2023-04-01 05:25:41,856 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=30837.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:26:09,300 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=30858.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:26:13,373 INFO [train.py:903] (1/4) Epoch 5, batch 3550, loss[loss=0.3158, simple_loss=0.3708, pruned_loss=0.1304, over 19320.00 frames. ], tot_loss[loss=0.2839, simple_loss=0.344, pruned_loss=0.1119, over 3826918.97 frames. ], batch size: 66, lr: 1.62e-02, grad_scale: 8.0 2023-04-01 05:26:38,722 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=30883.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:27:13,893 INFO [train.py:903] (1/4) Epoch 5, batch 3600, loss[loss=0.2764, simple_loss=0.3281, pruned_loss=0.1123, over 19320.00 frames. ], tot_loss[loss=0.283, simple_loss=0.3431, pruned_loss=0.1114, over 3828887.20 frames. ], batch size: 44, lr: 1.62e-02, grad_scale: 8.0 2023-04-01 05:27:24,193 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=30921.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:27:43,239 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.778e+02 6.769e+02 8.289e+02 1.066e+03 2.218e+03, threshold=1.658e+03, percent-clipped=4.0 2023-04-01 05:27:51,229 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=30942.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:27:52,184 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=30943.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:27:57,956 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1917, 1.1589, 1.6157, 1.3917, 2.4485, 2.0641, 2.5623, 0.9344], device='cuda:1'), covar=tensor([0.1867, 0.3137, 0.1743, 0.1511, 0.1068, 0.1493, 0.1175, 0.2875], device='cuda:1'), in_proj_covar=tensor([0.0445, 0.0507, 0.0481, 0.0409, 0.0556, 0.0446, 0.0629, 0.0448], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 05:28:02,219 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=30952.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:28:09,088 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([5.9922, 5.3653, 2.8573, 4.7589, 1.1654, 5.1026, 5.2262, 5.3676], device='cuda:1'), covar=tensor([0.0392, 0.0801, 0.1772, 0.0517, 0.3928, 0.0608, 0.0496, 0.0729], device='cuda:1'), in_proj_covar=tensor([0.0348, 0.0308, 0.0363, 0.0286, 0.0355, 0.0301, 0.0284, 0.0313], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 05:28:13,483 INFO [train.py:903] (1/4) Epoch 5, batch 3650, loss[loss=0.252, simple_loss=0.3124, pruned_loss=0.09578, over 19786.00 frames. ], tot_loss[loss=0.2824, simple_loss=0.3425, pruned_loss=0.1112, over 3837517.97 frames. ], batch size: 48, lr: 1.62e-02, grad_scale: 8.0 2023-04-01 05:28:20,082 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=30967.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:28:45,408 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9891, 2.0659, 1.6094, 1.5276, 1.4634, 1.6238, 0.3065, 0.8777], device='cuda:1'), covar=tensor([0.0277, 0.0258, 0.0175, 0.0235, 0.0511, 0.0273, 0.0498, 0.0447], device='cuda:1'), in_proj_covar=tensor([0.0286, 0.0284, 0.0287, 0.0309, 0.0376, 0.0298, 0.0280, 0.0293], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 05:29:02,306 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([0.8916, 1.3307, 1.0097, 1.0305, 1.2570, 0.9223, 0.8977, 1.2208], device='cuda:1'), covar=tensor([0.0511, 0.0627, 0.1004, 0.0459, 0.0415, 0.1027, 0.0504, 0.0376], device='cuda:1'), in_proj_covar=tensor([0.0292, 0.0274, 0.0317, 0.0243, 0.0225, 0.0314, 0.0286, 0.0256], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 05:29:14,331 INFO [train.py:903] (1/4) Epoch 5, batch 3700, loss[loss=0.3187, simple_loss=0.3614, pruned_loss=0.138, over 14015.00 frames. ], tot_loss[loss=0.2834, simple_loss=0.3431, pruned_loss=0.1118, over 3824155.65 frames. ], batch size: 136, lr: 1.62e-02, grad_scale: 8.0 2023-04-01 05:29:21,021 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=31017.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:29:34,006 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31029.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 05:29:40,640 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.405e+02 7.099e+02 8.962e+02 1.140e+03 3.223e+03, threshold=1.792e+03, percent-clipped=9.0 2023-04-01 05:30:15,304 INFO [train.py:903] (1/4) Epoch 5, batch 3750, loss[loss=0.259, simple_loss=0.3315, pruned_loss=0.09323, over 19481.00 frames. ], tot_loss[loss=0.2843, simple_loss=0.3437, pruned_loss=0.1124, over 3824243.58 frames. ], batch size: 64, lr: 1.62e-02, grad_scale: 8.0 2023-04-01 05:30:53,645 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=31093.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:31:15,740 INFO [train.py:903] (1/4) Epoch 5, batch 3800, loss[loss=0.2781, simple_loss=0.3461, pruned_loss=0.1051, over 19487.00 frames. ], tot_loss[loss=0.2838, simple_loss=0.3438, pruned_loss=0.112, over 3830300.43 frames. ], batch size: 64, lr: 1.62e-02, grad_scale: 8.0 2023-04-01 05:31:22,804 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31118.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:31:26,008 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.1527, 1.5492, 1.7031, 2.0270, 1.8334, 1.8223, 1.5917, 1.9579], device='cuda:1'), covar=tensor([0.0655, 0.1347, 0.1130, 0.0746, 0.1040, 0.0418, 0.0937, 0.0506], device='cuda:1'), in_proj_covar=tensor([0.0248, 0.0359, 0.0279, 0.0231, 0.0299, 0.0240, 0.0265, 0.0225], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 05:31:40,727 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=31132.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:31:44,604 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.073e+02 5.856e+02 8.467e+02 1.115e+03 2.554e+03, threshold=1.693e+03, percent-clipped=5.0 2023-04-01 05:31:49,118 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9_rvb from training. Duration: 29.1166875 2023-04-01 05:32:15,378 INFO [train.py:903] (1/4) Epoch 5, batch 3850, loss[loss=0.2675, simple_loss=0.3292, pruned_loss=0.1029, over 19707.00 frames. ], tot_loss[loss=0.2828, simple_loss=0.343, pruned_loss=0.1113, over 3837415.77 frames. ], batch size: 59, lr: 1.61e-02, grad_scale: 8.0 2023-04-01 05:32:37,689 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.71 vs. limit=2.0 2023-04-01 05:33:19,167 INFO [train.py:903] (1/4) Epoch 5, batch 3900, loss[loss=0.318, simple_loss=0.3745, pruned_loss=0.1307, over 19343.00 frames. ], tot_loss[loss=0.2825, simple_loss=0.3428, pruned_loss=0.1112, over 3835802.06 frames. ], batch size: 66, lr: 1.61e-02, grad_scale: 8.0 2023-04-01 05:33:45,288 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.653e+02 6.709e+02 8.069e+02 1.055e+03 2.198e+03, threshold=1.614e+03, percent-clipped=3.0 2023-04-01 05:34:09,845 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3119, 2.0351, 1.5634, 1.3477, 1.8873, 1.0946, 1.1862, 1.6755], device='cuda:1'), covar=tensor([0.0689, 0.0524, 0.0865, 0.0488, 0.0396, 0.0918, 0.0498, 0.0349], device='cuda:1'), in_proj_covar=tensor([0.0278, 0.0270, 0.0313, 0.0233, 0.0219, 0.0306, 0.0279, 0.0248], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 05:34:18,967 INFO [train.py:903] (1/4) Epoch 5, batch 3950, loss[loss=0.3412, simple_loss=0.385, pruned_loss=0.1487, over 19668.00 frames. ], tot_loss[loss=0.2829, simple_loss=0.3431, pruned_loss=0.1113, over 3841880.06 frames. ], batch size: 60, lr: 1.61e-02, grad_scale: 4.0 2023-04-01 05:34:22,383 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=31265.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:34:24,549 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 2023-04-01 05:34:47,790 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=31287.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:35:00,476 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=31296.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:35:18,261 INFO [train.py:903] (1/4) Epoch 5, batch 4000, loss[loss=0.3015, simple_loss=0.3619, pruned_loss=0.1205, over 19525.00 frames. ], tot_loss[loss=0.2849, simple_loss=0.344, pruned_loss=0.1129, over 3820360.50 frames. ], batch size: 54, lr: 1.61e-02, grad_scale: 8.0 2023-04-01 05:35:28,433 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.7264, 1.3891, 1.3581, 2.0486, 1.5469, 2.2170, 2.1004, 1.9287], device='cuda:1'), covar=tensor([0.0765, 0.1045, 0.1132, 0.0870, 0.0919, 0.0591, 0.0834, 0.0592], device='cuda:1'), in_proj_covar=tensor([0.0230, 0.0252, 0.0246, 0.0280, 0.0273, 0.0233, 0.0231, 0.0223], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1') 2023-04-01 05:35:29,583 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.0553, 1.0059, 1.2715, 1.5700, 2.6058, 1.0939, 1.8466, 2.6531], device='cuda:1'), covar=tensor([0.0389, 0.2484, 0.2385, 0.1304, 0.0605, 0.2031, 0.1078, 0.0472], device='cuda:1'), in_proj_covar=tensor([0.0280, 0.0307, 0.0307, 0.0279, 0.0294, 0.0306, 0.0283, 0.0295], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 05:35:48,957 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.459e+02 7.061e+02 8.767e+02 1.081e+03 2.366e+03, threshold=1.753e+03, percent-clipped=7.0 2023-04-01 05:36:06,113 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1_rvb from training. Duration: 0.7545625 2023-04-01 05:36:11,766 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2983, 1.1931, 1.6375, 1.0564, 2.7092, 3.4254, 3.3489, 3.6680], device='cuda:1'), covar=tensor([0.1351, 0.2944, 0.2739, 0.1979, 0.0408, 0.0123, 0.0184, 0.0108], device='cuda:1'), in_proj_covar=tensor([0.0253, 0.0280, 0.0317, 0.0249, 0.0197, 0.0120, 0.0205, 0.0136], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-04-01 05:36:18,326 INFO [train.py:903] (1/4) Epoch 5, batch 4050, loss[loss=0.2495, simple_loss=0.3272, pruned_loss=0.0859, over 19712.00 frames. ], tot_loss[loss=0.2854, simple_loss=0.3444, pruned_loss=0.1132, over 3821296.00 frames. ], batch size: 59, lr: 1.61e-02, grad_scale: 4.0 2023-04-01 05:36:20,389 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.46 vs. limit=2.0 2023-04-01 05:36:33,360 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=31373.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 05:36:42,200 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=31380.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:36:51,310 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=31388.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:37:07,163 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=31402.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:37:18,596 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=31411.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:37:19,399 INFO [train.py:903] (1/4) Epoch 5, batch 4100, loss[loss=0.271, simple_loss=0.3498, pruned_loss=0.09612, over 19308.00 frames. ], tot_loss[loss=0.2857, simple_loss=0.3449, pruned_loss=0.1133, over 3814242.19 frames. ], batch size: 66, lr: 1.61e-02, grad_scale: 4.0 2023-04-01 05:37:20,940 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31413.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:37:37,721 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31427.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:37:48,916 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.057e+02 5.964e+02 7.397e+02 9.605e+02 3.908e+03, threshold=1.479e+03, percent-clipped=5.0 2023-04-01 05:37:49,641 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 2023-04-01 05:37:53,722 WARNING [train.py:1073] (1/4) Exclude cut with ID 497-129325-0061-9566_sp1.1_rvb from training. Duration: 0.97725 2023-04-01 05:38:20,924 INFO [train.py:903] (1/4) Epoch 5, batch 4150, loss[loss=0.2785, simple_loss=0.3458, pruned_loss=0.1056, over 19602.00 frames. ], tot_loss[loss=0.2843, simple_loss=0.3437, pruned_loss=0.1125, over 3819451.56 frames. ], batch size: 57, lr: 1.61e-02, grad_scale: 4.0 2023-04-01 05:38:50,999 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=31488.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 05:39:16,822 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31509.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 05:39:20,063 INFO [train.py:903] (1/4) Epoch 5, batch 4200, loss[loss=0.2685, simple_loss=0.3362, pruned_loss=0.1004, over 18761.00 frames. ], tot_loss[loss=0.2838, simple_loss=0.3437, pruned_loss=0.112, over 3823562.34 frames. ], batch size: 74, lr: 1.61e-02, grad_scale: 4.0 2023-04-01 05:39:23,664 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 2023-04-01 05:39:50,934 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.017e+02 6.856e+02 8.101e+02 1.045e+03 3.023e+03, threshold=1.620e+03, percent-clipped=6.0 2023-04-01 05:40:19,233 INFO [train.py:903] (1/4) Epoch 5, batch 4250, loss[loss=0.2764, simple_loss=0.3482, pruned_loss=0.1023, over 19737.00 frames. ], tot_loss[loss=0.2838, simple_loss=0.3439, pruned_loss=0.1118, over 3832204.85 frames. ], batch size: 63, lr: 1.60e-02, grad_scale: 4.0 2023-04-01 05:40:21,977 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6957, 1.2420, 1.2411, 1.8629, 1.4560, 2.0257, 2.0820, 1.9149], device='cuda:1'), covar=tensor([0.0834, 0.1187, 0.1275, 0.1099, 0.1130, 0.0745, 0.1005, 0.0641], device='cuda:1'), in_proj_covar=tensor([0.0231, 0.0253, 0.0245, 0.0280, 0.0274, 0.0234, 0.0237, 0.0223], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:1') 2023-04-01 05:40:35,451 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9_rvb from training. Duration: 29.816625 2023-04-01 05:40:46,369 WARNING [train.py:1073] (1/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 2023-04-01 05:41:20,811 INFO [train.py:903] (1/4) Epoch 5, batch 4300, loss[loss=0.2616, simple_loss=0.324, pruned_loss=0.09956, over 19738.00 frames. ], tot_loss[loss=0.2816, simple_loss=0.3422, pruned_loss=0.1105, over 3843629.11 frames. ], batch size: 51, lr: 1.60e-02, grad_scale: 4.0 2023-04-01 05:41:41,978 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4080, 1.0033, 1.0175, 1.7114, 1.3956, 1.3577, 1.6671, 1.3469], device='cuda:1'), covar=tensor([0.0828, 0.1241, 0.1271, 0.0790, 0.0953, 0.0992, 0.0914, 0.0830], device='cuda:1'), in_proj_covar=tensor([0.0229, 0.0251, 0.0245, 0.0277, 0.0271, 0.0232, 0.0233, 0.0221], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1') 2023-04-01 05:41:51,069 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=31636.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:41:51,761 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.826e+02 7.144e+02 8.425e+02 1.091e+03 2.021e+03, threshold=1.685e+03, percent-clipped=5.0 2023-04-01 05:42:13,882 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 2023-04-01 05:42:18,613 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=31658.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:42:21,719 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31661.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:42:22,334 INFO [train.py:903] (1/4) Epoch 5, batch 4350, loss[loss=0.25, simple_loss=0.306, pruned_loss=0.09701, over 19111.00 frames. ], tot_loss[loss=0.2827, simple_loss=0.343, pruned_loss=0.1112, over 3849576.28 frames. ], batch size: 42, lr: 1.60e-02, grad_scale: 4.0 2023-04-01 05:42:28,428 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=31667.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:42:45,942 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31683.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:42:57,800 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31692.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:43:04,358 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.3741, 1.2996, 1.4550, 1.5332, 2.8201, 0.9723, 1.9686, 3.1094], device='cuda:1'), covar=tensor([0.0457, 0.2557, 0.2473, 0.1718, 0.0694, 0.2483, 0.1355, 0.0442], device='cuda:1'), in_proj_covar=tensor([0.0277, 0.0310, 0.0307, 0.0280, 0.0297, 0.0308, 0.0282, 0.0296], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 05:43:22,083 INFO [train.py:903] (1/4) Epoch 5, batch 4400, loss[loss=0.2578, simple_loss=0.3355, pruned_loss=0.0901, over 19526.00 frames. ], tot_loss[loss=0.283, simple_loss=0.3434, pruned_loss=0.1113, over 3857111.31 frames. ], batch size: 56, lr: 1.60e-02, grad_scale: 8.0 2023-04-01 05:43:34,003 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31722.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:43:44,919 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_rvb from training. Duration: 25.285 2023-04-01 05:43:53,319 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.183e+02 7.191e+02 8.879e+02 1.082e+03 1.961e+03, threshold=1.776e+03, percent-clipped=1.0 2023-04-01 05:43:55,410 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 2023-04-01 05:43:58,384 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.79 vs. limit=5.0 2023-04-01 05:44:02,752 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=31744.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 05:44:16,187 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([5.0767, 5.0774, 5.9197, 5.7817, 1.5655, 5.4966, 4.6806, 5.3494], device='cuda:1'), covar=tensor([0.0836, 0.0507, 0.0401, 0.0344, 0.4644, 0.0299, 0.0462, 0.0985], device='cuda:1'), in_proj_covar=tensor([0.0503, 0.0440, 0.0583, 0.0483, 0.0571, 0.0356, 0.0384, 0.0557], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1') 2023-04-01 05:44:22,697 INFO [train.py:903] (1/4) Epoch 5, batch 4450, loss[loss=0.2432, simple_loss=0.3048, pruned_loss=0.0908, over 19776.00 frames. ], tot_loss[loss=0.2823, simple_loss=0.3429, pruned_loss=0.1108, over 3852436.21 frames. ], batch size: 48, lr: 1.60e-02, grad_scale: 8.0 2023-04-01 05:44:30,906 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31769.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 05:44:32,868 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=31771.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:45:16,682 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31806.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:45:20,581 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 2023-04-01 05:45:23,177 INFO [train.py:903] (1/4) Epoch 5, batch 4500, loss[loss=0.3083, simple_loss=0.3683, pruned_loss=0.1241, over 19073.00 frames. ], tot_loss[loss=0.2828, simple_loss=0.343, pruned_loss=0.1113, over 3840099.48 frames. ], batch size: 69, lr: 1.60e-02, grad_scale: 8.0 2023-04-01 05:45:53,689 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.229e+02 6.666e+02 8.776e+02 1.149e+03 2.550e+03, threshold=1.755e+03, percent-clipped=7.0 2023-04-01 05:45:58,564 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31841.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:46:12,834 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=31853.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 05:46:24,485 INFO [train.py:903] (1/4) Epoch 5, batch 4550, loss[loss=0.244, simple_loss=0.3074, pruned_loss=0.09034, over 19028.00 frames. ], tot_loss[loss=0.2835, simple_loss=0.3437, pruned_loss=0.1116, over 3833734.29 frames. ], batch size: 42, lr: 1.60e-02, grad_scale: 8.0 2023-04-01 05:46:30,265 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9_rvb from training. Duration: 28.72225 2023-04-01 05:46:51,968 WARNING [train.py:1073] (1/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 2023-04-01 05:46:52,273 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=31886.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:47:05,630 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31896.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:47:25,654 INFO [train.py:903] (1/4) Epoch 5, batch 4600, loss[loss=0.256, simple_loss=0.3139, pruned_loss=0.09907, over 19776.00 frames. ], tot_loss[loss=0.2835, simple_loss=0.3438, pruned_loss=0.1116, over 3833173.42 frames. ], batch size: 47, lr: 1.60e-02, grad_scale: 8.0 2023-04-01 05:47:50,433 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31933.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:47:55,743 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.582e+02 6.775e+02 7.918e+02 9.900e+02 3.222e+03, threshold=1.584e+03, percent-clipped=3.0 2023-04-01 05:48:10,574 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.8854, 1.4896, 1.4444, 2.1462, 1.8018, 2.1296, 2.3163, 1.9852], device='cuda:1'), covar=tensor([0.0695, 0.1044, 0.1096, 0.0900, 0.0881, 0.0721, 0.0848, 0.0612], device='cuda:1'), in_proj_covar=tensor([0.0229, 0.0255, 0.0246, 0.0280, 0.0269, 0.0233, 0.0234, 0.0224], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:1') 2023-04-01 05:48:25,663 INFO [train.py:903] (1/4) Epoch 5, batch 4650, loss[loss=0.3255, simple_loss=0.3796, pruned_loss=0.1357, over 19684.00 frames. ], tot_loss[loss=0.2826, simple_loss=0.3434, pruned_loss=0.1109, over 3838725.67 frames. ], batch size: 53, lr: 1.60e-02, grad_scale: 8.0 2023-04-01 05:48:32,723 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31968.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:48:32,788 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=31968.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 05:48:41,062 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_rvb from training. Duration: 0.92 2023-04-01 05:48:52,751 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_rvb from training. Duration: 0.83 2023-04-01 05:49:25,340 INFO [train.py:903] (1/4) Epoch 5, batch 4700, loss[loss=0.2454, simple_loss=0.3095, pruned_loss=0.09061, over 19749.00 frames. ], tot_loss[loss=0.2823, simple_loss=0.3428, pruned_loss=0.1109, over 3835093.51 frames. ], batch size: 47, lr: 1.59e-02, grad_scale: 8.0 2023-04-01 05:49:48,439 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9_rvb from training. Duration: 0.92225 2023-04-01 05:49:56,507 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.386e+02 6.540e+02 8.637e+02 1.061e+03 2.519e+03, threshold=1.727e+03, percent-clipped=5.0 2023-04-01 05:50:27,324 INFO [train.py:903] (1/4) Epoch 5, batch 4750, loss[loss=0.2918, simple_loss=0.362, pruned_loss=0.1108, over 19467.00 frames. ], tot_loss[loss=0.2829, simple_loss=0.3435, pruned_loss=0.1112, over 3838462.10 frames. ], batch size: 64, lr: 1.59e-02, grad_scale: 8.0 2023-04-01 05:50:32,824 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=32066.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:50:34,215 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1982, 2.1510, 1.6271, 1.5718, 1.9729, 1.1762, 1.0958, 1.7288], device='cuda:1'), covar=tensor([0.0687, 0.0414, 0.0796, 0.0452, 0.0288, 0.0926, 0.0549, 0.0329], device='cuda:1'), in_proj_covar=tensor([0.0279, 0.0264, 0.0312, 0.0234, 0.0215, 0.0309, 0.0281, 0.0252], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 05:51:28,012 INFO [train.py:903] (1/4) Epoch 5, batch 4800, loss[loss=0.2775, simple_loss=0.346, pruned_loss=0.1045, over 19742.00 frames. ], tot_loss[loss=0.2839, simple_loss=0.3439, pruned_loss=0.112, over 3832965.57 frames. ], batch size: 63, lr: 1.59e-02, grad_scale: 8.0 2023-04-01 05:51:57,641 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.741e+02 7.051e+02 8.763e+02 1.042e+03 3.094e+03, threshold=1.753e+03, percent-clipped=4.0 2023-04-01 05:52:01,634 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3438, 1.3381, 1.8547, 1.5216, 3.1303, 4.4990, 4.5009, 4.9330], device='cuda:1'), covar=tensor([0.1470, 0.2897, 0.2828, 0.1773, 0.0408, 0.0130, 0.0148, 0.0076], device='cuda:1'), in_proj_covar=tensor([0.0255, 0.0280, 0.0317, 0.0251, 0.0197, 0.0117, 0.0204, 0.0138], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-04-01 05:52:04,785 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=32142.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:52:14,225 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=32150.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:52:27,937 INFO [train.py:903] (1/4) Epoch 5, batch 4850, loss[loss=0.2964, simple_loss=0.3553, pruned_loss=0.1187, over 19533.00 frames. ], tot_loss[loss=0.2839, simple_loss=0.3439, pruned_loss=0.112, over 3832472.15 frames. ], batch size: 64, lr: 1.59e-02, grad_scale: 8.0 2023-04-01 05:52:34,137 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=32167.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:52:51,451 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=32181.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:52:54,112 WARNING [train.py:1073] (1/4) Exclude cut with ID 774-127930-0014-48411_sp1.1_rvb from training. Duration: 0.95 2023-04-01 05:52:56,539 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=32185.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:53:10,262 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.14 vs. limit=2.0 2023-04-01 05:53:12,907 WARNING [train.py:1073] (1/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1_rvb from training. Duration: 0.9409375 2023-04-01 05:53:17,713 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 2023-04-01 05:53:18,683 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 2023-04-01 05:53:26,876 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1_rvb from training. Duration: 25.3818125 2023-04-01 05:53:27,947 INFO [train.py:903] (1/4) Epoch 5, batch 4900, loss[loss=0.288, simple_loss=0.3552, pruned_loss=0.1104, over 19491.00 frames. ], tot_loss[loss=0.2836, simple_loss=0.3433, pruned_loss=0.112, over 3820348.08 frames. ], batch size: 64, lr: 1.59e-02, grad_scale: 8.0 2023-04-01 05:53:44,404 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=32224.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 05:53:48,952 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 2023-04-01 05:53:59,194 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.307e+02 6.585e+02 7.912e+02 1.039e+03 2.328e+03, threshold=1.582e+03, percent-clipped=1.0 2023-04-01 05:54:02,587 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=32240.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:54:13,257 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=32249.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 05:54:29,205 INFO [train.py:903] (1/4) Epoch 5, batch 4950, loss[loss=0.2421, simple_loss=0.3148, pruned_loss=0.08473, over 19659.00 frames. ], tot_loss[loss=0.2846, simple_loss=0.3442, pruned_loss=0.1125, over 3817737.76 frames. ], batch size: 55, lr: 1.59e-02, grad_scale: 8.0 2023-04-01 05:54:33,930 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=32265.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:54:48,150 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=32277.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:54:49,216 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 2023-04-01 05:55:12,815 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 2023-04-01 05:55:15,562 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=32300.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:55:30,485 INFO [train.py:903] (1/4) Epoch 5, batch 5000, loss[loss=0.2775, simple_loss=0.3283, pruned_loss=0.1134, over 19357.00 frames. ], tot_loss[loss=0.2846, simple_loss=0.3438, pruned_loss=0.1127, over 3830016.41 frames. ], batch size: 47, lr: 1.59e-02, grad_scale: 8.0 2023-04-01 05:55:30,635 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=32312.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:55:40,412 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_rvb from training. Duration: 27.14 2023-04-01 05:55:50,536 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 2023-04-01 05:55:58,320 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.834e+02 7.030e+02 8.789e+02 1.165e+03 2.289e+03, threshold=1.758e+03, percent-clipped=7.0 2023-04-01 05:56:21,716 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=32355.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:56:29,552 INFO [train.py:903] (1/4) Epoch 5, batch 5050, loss[loss=0.2393, simple_loss=0.3042, pruned_loss=0.08717, over 19322.00 frames. ], tot_loss[loss=0.2851, simple_loss=0.3443, pruned_loss=0.1129, over 3806394.18 frames. ], batch size: 44, lr: 1.59e-02, grad_scale: 8.0 2023-04-01 05:57:05,108 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9_rvb from training. Duration: 26.62775 2023-04-01 05:57:07,774 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=32392.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:57:29,764 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 2023-04-01 05:57:30,257 INFO [train.py:903] (1/4) Epoch 5, batch 5100, loss[loss=0.2332, simple_loss=0.3008, pruned_loss=0.08276, over 19393.00 frames. ], tot_loss[loss=0.2839, simple_loss=0.3437, pruned_loss=0.1121, over 3815460.10 frames. ], batch size: 47, lr: 1.58e-02, grad_scale: 8.0 2023-04-01 05:57:33,095 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.6248, 1.8233, 2.4275, 2.5945, 2.4161, 2.2621, 2.5205, 2.7726], device='cuda:1'), covar=tensor([0.0780, 0.1785, 0.1137, 0.0897, 0.1175, 0.0463, 0.0864, 0.0482], device='cuda:1'), in_proj_covar=tensor([0.0253, 0.0359, 0.0282, 0.0240, 0.0304, 0.0242, 0.0268, 0.0229], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 05:57:40,586 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 2023-04-01 05:57:44,911 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_rvb from training. Duration: 25.035 2023-04-01 05:57:51,242 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_rvb from training. Duration: 27.92 2023-04-01 05:57:51,527 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=32427.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:58:02,216 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.811e+02 6.879e+02 8.267e+02 1.044e+03 2.791e+03, threshold=1.653e+03, percent-clipped=3.0 2023-04-01 05:58:02,672 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=32437.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:58:32,000 INFO [train.py:903] (1/4) Epoch 5, batch 5150, loss[loss=0.2678, simple_loss=0.3407, pruned_loss=0.09738, over 19666.00 frames. ], tot_loss[loss=0.2804, simple_loss=0.3412, pruned_loss=0.1098, over 3821797.67 frames. ], batch size: 58, lr: 1.58e-02, grad_scale: 8.0 2023-04-01 05:58:32,465 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=32462.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:58:45,282 WARNING [train.py:1073] (1/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 2023-04-01 05:59:19,204 WARNING [train.py:1073] (1/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1_rvb from training. Duration: 0.9681875 2023-04-01 05:59:33,454 INFO [train.py:903] (1/4) Epoch 5, batch 5200, loss[loss=0.2849, simple_loss=0.3567, pruned_loss=0.1065, over 19641.00 frames. ], tot_loss[loss=0.281, simple_loss=0.3416, pruned_loss=0.1102, over 3822159.71 frames. ], batch size: 60, lr: 1.58e-02, grad_scale: 8.0 2023-04-01 05:59:42,703 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.5557, 1.7314, 2.1502, 2.6196, 2.2544, 2.2518, 2.3923, 2.6332], device='cuda:1'), covar=tensor([0.0731, 0.1906, 0.1238, 0.0839, 0.1221, 0.0423, 0.0823, 0.0496], device='cuda:1'), in_proj_covar=tensor([0.0251, 0.0365, 0.0283, 0.0238, 0.0305, 0.0239, 0.0268, 0.0228], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 05:59:43,923 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=32521.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 05:59:45,961 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 2023-04-01 06:00:02,761 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.318e+02 6.175e+02 7.903e+02 1.065e+03 1.799e+03, threshold=1.581e+03, percent-clipped=1.0 2023-04-01 06:00:15,191 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=32546.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 06:00:26,231 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([5.1446, 5.1053, 5.9989, 5.9112, 1.8870, 5.5756, 4.7388, 5.4550], device='cuda:1'), covar=tensor([0.0833, 0.0478, 0.0433, 0.0362, 0.4194, 0.0268, 0.0414, 0.0967], device='cuda:1'), in_proj_covar=tensor([0.0509, 0.0447, 0.0589, 0.0492, 0.0571, 0.0359, 0.0388, 0.0559], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1') 2023-04-01 06:00:26,391 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=32556.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 06:00:29,357 WARNING [train.py:1073] (1/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 2023-04-01 06:00:32,701 INFO [train.py:903] (1/4) Epoch 5, batch 5250, loss[loss=0.3319, simple_loss=0.3871, pruned_loss=0.1384, over 17222.00 frames. ], tot_loss[loss=0.28, simple_loss=0.3409, pruned_loss=0.1096, over 3823781.40 frames. ], batch size: 101, lr: 1.58e-02, grad_scale: 8.0 2023-04-01 06:00:37,577 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.6924, 4.1493, 4.3370, 4.3307, 1.4468, 3.9835, 3.4893, 3.9635], device='cuda:1'), covar=tensor([0.0841, 0.0495, 0.0466, 0.0404, 0.3979, 0.0344, 0.0506, 0.0952], device='cuda:1'), in_proj_covar=tensor([0.0509, 0.0448, 0.0590, 0.0493, 0.0573, 0.0360, 0.0389, 0.0559], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1') 2023-04-01 06:00:55,260 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=32581.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 06:01:32,157 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=32611.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 06:01:32,913 INFO [train.py:903] (1/4) Epoch 5, batch 5300, loss[loss=0.2812, simple_loss=0.3472, pruned_loss=0.1076, over 19660.00 frames. ], tot_loss[loss=0.2805, simple_loss=0.3412, pruned_loss=0.1098, over 3824383.87 frames. ], batch size: 55, lr: 1.58e-02, grad_scale: 8.0 2023-04-01 06:01:51,389 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 2023-04-01 06:02:04,557 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=32636.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 06:02:05,307 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.788e+02 6.583e+02 8.041e+02 1.027e+03 2.106e+03, threshold=1.608e+03, percent-clipped=4.0 2023-04-01 06:02:17,829 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=32648.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 06:02:34,322 INFO [train.py:903] (1/4) Epoch 5, batch 5350, loss[loss=0.261, simple_loss=0.3312, pruned_loss=0.09542, over 19624.00 frames. ], tot_loss[loss=0.2813, simple_loss=0.342, pruned_loss=0.1103, over 3836865.39 frames. ], batch size: 57, lr: 1.58e-02, grad_scale: 8.0 2023-04-01 06:02:48,920 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=32673.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 06:03:00,295 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=32683.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 06:03:06,670 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9_rvb from training. Duration: 26.438875 2023-04-01 06:03:27,094 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.69 vs. limit=2.0 2023-04-01 06:03:31,105 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=32708.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 06:03:35,120 INFO [train.py:903] (1/4) Epoch 5, batch 5400, loss[loss=0.2517, simple_loss=0.3265, pruned_loss=0.08848, over 19685.00 frames. ], tot_loss[loss=0.2822, simple_loss=0.3426, pruned_loss=0.1109, over 3832724.13 frames. ], batch size: 53, lr: 1.58e-02, grad_scale: 8.0 2023-04-01 06:04:03,208 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.192e+02 6.673e+02 8.431e+02 1.064e+03 2.658e+03, threshold=1.686e+03, percent-clipped=8.0 2023-04-01 06:04:04,938 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0 2023-04-01 06:04:11,107 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3273, 2.4259, 1.7199, 1.3684, 2.2869, 1.1370, 1.1225, 1.7778], device='cuda:1'), covar=tensor([0.0745, 0.0393, 0.0700, 0.0556, 0.0299, 0.0885, 0.0598, 0.0376], device='cuda:1'), in_proj_covar=tensor([0.0278, 0.0264, 0.0316, 0.0238, 0.0216, 0.0304, 0.0282, 0.0250], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 06:04:33,649 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.7107, 4.0518, 4.3144, 4.3354, 1.5206, 3.9866, 3.5488, 3.8920], device='cuda:1'), covar=tensor([0.0941, 0.0666, 0.0499, 0.0398, 0.4094, 0.0401, 0.0491, 0.1040], device='cuda:1'), in_proj_covar=tensor([0.0517, 0.0456, 0.0603, 0.0500, 0.0578, 0.0366, 0.0392, 0.0568], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1') 2023-04-01 06:04:34,472 INFO [train.py:903] (1/4) Epoch 5, batch 5450, loss[loss=0.2608, simple_loss=0.3297, pruned_loss=0.09593, over 19668.00 frames. ], tot_loss[loss=0.2808, simple_loss=0.3411, pruned_loss=0.1102, over 3840596.33 frames. ], batch size: 55, lr: 1.58e-02, grad_scale: 8.0 2023-04-01 06:05:33,890 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0155, 1.4920, 1.7786, 2.2851, 2.0324, 1.8831, 2.3250, 1.9418], device='cuda:1'), covar=tensor([0.0879, 0.1433, 0.1135, 0.1054, 0.0999, 0.1151, 0.0926, 0.0854], device='cuda:1'), in_proj_covar=tensor([0.0230, 0.0246, 0.0241, 0.0269, 0.0263, 0.0232, 0.0228, 0.0220], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1') 2023-04-01 06:05:34,678 INFO [train.py:903] (1/4) Epoch 5, batch 5500, loss[loss=0.3308, simple_loss=0.3774, pruned_loss=0.1421, over 13342.00 frames. ], tot_loss[loss=0.2811, simple_loss=0.3415, pruned_loss=0.1103, over 3834985.15 frames. ], batch size: 136, lr: 1.57e-02, grad_scale: 8.0 2023-04-01 06:05:56,690 WARNING [train.py:1073] (1/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9_rvb from training. Duration: 27.47775 2023-04-01 06:06:05,437 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.564e+02 6.283e+02 7.782e+02 1.000e+03 2.107e+03, threshold=1.556e+03, percent-clipped=4.0 2023-04-01 06:06:34,471 INFO [train.py:903] (1/4) Epoch 5, batch 5550, loss[loss=0.2757, simple_loss=0.3451, pruned_loss=0.1032, over 19687.00 frames. ], tot_loss[loss=0.2816, simple_loss=0.342, pruned_loss=0.1106, over 3835201.93 frames. ], batch size: 60, lr: 1.57e-02, grad_scale: 8.0 2023-04-01 06:06:34,912 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.1600, 1.9766, 1.4783, 1.4929, 1.3146, 1.6048, 0.2694, 0.9377], device='cuda:1'), covar=tensor([0.0282, 0.0267, 0.0230, 0.0320, 0.0595, 0.0340, 0.0548, 0.0483], device='cuda:1'), in_proj_covar=tensor([0.0285, 0.0290, 0.0285, 0.0306, 0.0381, 0.0299, 0.0280, 0.0299], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 06:06:40,854 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 2023-04-01 06:07:29,977 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 2023-04-01 06:07:36,764 INFO [train.py:903] (1/4) Epoch 5, batch 5600, loss[loss=0.2737, simple_loss=0.3209, pruned_loss=0.1133, over 19356.00 frames. ], tot_loss[loss=0.2824, simple_loss=0.3427, pruned_loss=0.1111, over 3820559.53 frames. ], batch size: 47, lr: 1.57e-02, grad_scale: 8.0 2023-04-01 06:08:06,605 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.479e+02 7.188e+02 9.159e+02 1.163e+03 2.158e+03, threshold=1.832e+03, percent-clipped=9.0 2023-04-01 06:08:23,137 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1330, 1.2633, 1.8868, 1.3427, 2.7869, 2.1774, 2.8805, 1.1163], device='cuda:1'), covar=tensor([0.1741, 0.2884, 0.1509, 0.1424, 0.1071, 0.1388, 0.1164, 0.2671], device='cuda:1'), in_proj_covar=tensor([0.0443, 0.0500, 0.0479, 0.0403, 0.0550, 0.0444, 0.0622, 0.0439], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 06:08:38,295 INFO [train.py:903] (1/4) Epoch 5, batch 5650, loss[loss=0.2507, simple_loss=0.3268, pruned_loss=0.08731, over 19723.00 frames. ], tot_loss[loss=0.281, simple_loss=0.3418, pruned_loss=0.1101, over 3815837.81 frames. ], batch size: 63, lr: 1.57e-02, grad_scale: 8.0 2023-04-01 06:08:54,132 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.65 vs. limit=2.0 2023-04-01 06:09:04,596 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=32984.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 06:09:24,870 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9_rvb from training. Duration: 28.0944375 2023-04-01 06:09:38,279 INFO [train.py:903] (1/4) Epoch 5, batch 5700, loss[loss=0.3061, simple_loss=0.3659, pruned_loss=0.1232, over 18272.00 frames. ], tot_loss[loss=0.2816, simple_loss=0.342, pruned_loss=0.1106, over 3803363.30 frames. ], batch size: 84, lr: 1.57e-02, grad_scale: 8.0 2023-04-01 06:09:42,492 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.12 vs. limit=2.0 2023-04-01 06:10:09,258 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.491e+02 7.117e+02 8.783e+02 1.086e+03 2.576e+03, threshold=1.757e+03, percent-clipped=4.0 2023-04-01 06:10:38,546 INFO [train.py:903] (1/4) Epoch 5, batch 5750, loss[loss=0.2943, simple_loss=0.3565, pruned_loss=0.1161, over 19729.00 frames. ], tot_loss[loss=0.2826, simple_loss=0.3431, pruned_loss=0.111, over 3797009.19 frames. ], batch size: 63, lr: 1.57e-02, grad_scale: 8.0 2023-04-01 06:10:39,675 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9_rvb from training. Duration: 33.038875 2023-04-01 06:10:47,554 WARNING [train.py:1073] (1/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 2023-04-01 06:10:52,568 WARNING [train.py:1073] (1/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 2023-04-01 06:11:01,652 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1731, 1.2326, 1.6182, 1.3881, 2.1550, 2.0610, 2.3820, 0.8097], device='cuda:1'), covar=tensor([0.1889, 0.3128, 0.1662, 0.1484, 0.1177, 0.1447, 0.1153, 0.2824], device='cuda:1'), in_proj_covar=tensor([0.0442, 0.0496, 0.0476, 0.0402, 0.0549, 0.0437, 0.0617, 0.0437], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 06:11:09,077 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=33087.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 06:11:40,192 INFO [train.py:903] (1/4) Epoch 5, batch 5800, loss[loss=0.3001, simple_loss=0.3534, pruned_loss=0.1234, over 19501.00 frames. ], tot_loss[loss=0.2805, simple_loss=0.3413, pruned_loss=0.1099, over 3800127.53 frames. ], batch size: 49, lr: 1.57e-02, grad_scale: 8.0 2023-04-01 06:12:08,911 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.601e+02 7.070e+02 8.520e+02 1.129e+03 2.712e+03, threshold=1.704e+03, percent-clipped=8.0 2023-04-01 06:12:26,073 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6918, 1.2974, 1.3939, 2.0519, 1.7068, 1.9980, 2.2510, 1.9296], device='cuda:1'), covar=tensor([0.0772, 0.1056, 0.0969, 0.0819, 0.0874, 0.0689, 0.0748, 0.0597], device='cuda:1'), in_proj_covar=tensor([0.0233, 0.0249, 0.0245, 0.0274, 0.0267, 0.0232, 0.0232, 0.0223], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:1') 2023-04-01 06:12:40,679 INFO [train.py:903] (1/4) Epoch 5, batch 5850, loss[loss=0.2456, simple_loss=0.3156, pruned_loss=0.08784, over 19582.00 frames. ], tot_loss[loss=0.2812, simple_loss=0.3418, pruned_loss=0.1104, over 3804872.87 frames. ], batch size: 52, lr: 1.57e-02, grad_scale: 8.0 2023-04-01 06:12:53,495 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=33173.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 06:13:40,939 INFO [train.py:903] (1/4) Epoch 5, batch 5900, loss[loss=0.2513, simple_loss=0.3287, pruned_loss=0.08697, over 19520.00 frames. ], tot_loss[loss=0.2807, simple_loss=0.3409, pruned_loss=0.1103, over 3814802.52 frames. ], batch size: 56, lr: 1.57e-02, grad_scale: 8.0 2023-04-01 06:13:43,338 WARNING [train.py:1073] (1/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 2023-04-01 06:14:04,498 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9_rvb from training. Duration: 27.511125 2023-04-01 06:14:11,989 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.345e+02 7.066e+02 8.762e+02 1.130e+03 2.300e+03, threshold=1.752e+03, percent-clipped=6.0 2023-04-01 06:14:41,682 INFO [train.py:903] (1/4) Epoch 5, batch 5950, loss[loss=0.2733, simple_loss=0.3393, pruned_loss=0.1037, over 19654.00 frames. ], tot_loss[loss=0.2808, simple_loss=0.3412, pruned_loss=0.1102, over 3809379.95 frames. ], batch size: 58, lr: 1.56e-02, grad_scale: 8.0 2023-04-01 06:15:23,367 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.1935, 2.1798, 2.0985, 3.3103, 2.3426, 3.6783, 3.0511, 1.9628], device='cuda:1'), covar=tensor([0.2468, 0.1793, 0.0878, 0.1163, 0.2073, 0.0534, 0.1762, 0.1619], device='cuda:1'), in_proj_covar=tensor([0.0646, 0.0634, 0.0560, 0.0797, 0.0677, 0.0558, 0.0697, 0.0594], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 06:15:43,947 INFO [train.py:903] (1/4) Epoch 5, batch 6000, loss[loss=0.2654, simple_loss=0.3238, pruned_loss=0.1035, over 19481.00 frames. ], tot_loss[loss=0.2802, simple_loss=0.3409, pruned_loss=0.1098, over 3818262.06 frames. ], batch size: 49, lr: 1.56e-02, grad_scale: 8.0 2023-04-01 06:15:43,947 INFO [train.py:928] (1/4) Computing validation loss 2023-04-01 06:15:56,873 INFO [train.py:937] (1/4) Epoch 5, validation: loss=0.203, simple_loss=0.3017, pruned_loss=0.05213, over 944034.00 frames. 2023-04-01 06:15:56,874 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18649MB 2023-04-01 06:16:18,147 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=33328.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 06:16:28,852 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.050e+02 6.893e+02 8.503e+02 1.056e+03 1.945e+03, threshold=1.701e+03, percent-clipped=4.0 2023-04-01 06:16:59,412 INFO [train.py:903] (1/4) Epoch 5, batch 6050, loss[loss=0.2811, simple_loss=0.3328, pruned_loss=0.1147, over 19629.00 frames. ], tot_loss[loss=0.2797, simple_loss=0.3403, pruned_loss=0.1096, over 3819465.06 frames. ], batch size: 50, lr: 1.56e-02, grad_scale: 16.0 2023-04-01 06:17:34,354 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.8751, 1.3839, 1.4150, 2.0809, 1.6409, 1.9698, 2.0484, 1.8068], device='cuda:1'), covar=tensor([0.0714, 0.1110, 0.1113, 0.0892, 0.0978, 0.0778, 0.1015, 0.0688], device='cuda:1'), in_proj_covar=tensor([0.0225, 0.0243, 0.0240, 0.0269, 0.0258, 0.0227, 0.0225, 0.0217], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1') 2023-04-01 06:18:00,559 INFO [train.py:903] (1/4) Epoch 5, batch 6100, loss[loss=0.2965, simple_loss=0.3655, pruned_loss=0.1137, over 19569.00 frames. ], tot_loss[loss=0.2795, simple_loss=0.3406, pruned_loss=0.1092, over 3830616.24 frames. ], batch size: 61, lr: 1.56e-02, grad_scale: 8.0 2023-04-01 06:18:23,031 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=33431.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 06:18:32,606 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.702e+02 6.262e+02 7.464e+02 9.853e+02 2.581e+03, threshold=1.493e+03, percent-clipped=2.0 2023-04-01 06:18:39,223 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=33443.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 06:19:00,961 INFO [train.py:903] (1/4) Epoch 5, batch 6150, loss[loss=0.2994, simple_loss=0.3552, pruned_loss=0.1218, over 19382.00 frames. ], tot_loss[loss=0.2807, simple_loss=0.3417, pruned_loss=0.1098, over 3830604.80 frames. ], batch size: 70, lr: 1.56e-02, grad_scale: 8.0 2023-04-01 06:19:29,657 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9_rvb from training. Duration: 31.02225 2023-04-01 06:19:35,772 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=33490.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 06:20:01,183 INFO [train.py:903] (1/4) Epoch 5, batch 6200, loss[loss=0.2924, simple_loss=0.3568, pruned_loss=0.114, over 19605.00 frames. ], tot_loss[loss=0.2812, simple_loss=0.3417, pruned_loss=0.1103, over 3832349.79 frames. ], batch size: 61, lr: 1.56e-02, grad_scale: 8.0 2023-04-01 06:20:08,945 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=33517.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 06:20:09,240 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.3584, 2.9468, 2.0076, 2.2488, 1.9854, 2.2463, 0.8526, 2.1745], device='cuda:1'), covar=tensor([0.0234, 0.0253, 0.0313, 0.0373, 0.0546, 0.0452, 0.0574, 0.0450], device='cuda:1'), in_proj_covar=tensor([0.0278, 0.0278, 0.0281, 0.0297, 0.0372, 0.0290, 0.0271, 0.0290], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 06:20:34,185 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.213e+02 6.728e+02 8.677e+02 1.165e+03 2.777e+03, threshold=1.735e+03, percent-clipped=13.0 2023-04-01 06:20:43,559 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=33546.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 06:21:03,395 INFO [train.py:903] (1/4) Epoch 5, batch 6250, loss[loss=0.2319, simple_loss=0.2971, pruned_loss=0.08337, over 19731.00 frames. ], tot_loss[loss=0.281, simple_loss=0.3418, pruned_loss=0.1101, over 3833940.58 frames. ], batch size: 46, lr: 1.56e-02, grad_scale: 8.0 2023-04-01 06:21:31,242 WARNING [train.py:1073] (1/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9_rvb from training. Duration: 25.988875 2023-04-01 06:22:03,933 INFO [train.py:903] (1/4) Epoch 5, batch 6300, loss[loss=0.2256, simple_loss=0.2903, pruned_loss=0.08047, over 19380.00 frames. ], tot_loss[loss=0.2799, simple_loss=0.3404, pruned_loss=0.1097, over 3831685.89 frames. ], batch size: 47, lr: 1.56e-02, grad_scale: 8.0 2023-04-01 06:22:28,442 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=33632.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 06:22:35,563 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.303e+02 6.435e+02 8.030e+02 9.840e+02 2.632e+03, threshold=1.606e+03, percent-clipped=3.0 2023-04-01 06:23:04,055 INFO [train.py:903] (1/4) Epoch 5, batch 6350, loss[loss=0.1983, simple_loss=0.2723, pruned_loss=0.0622, over 19764.00 frames. ], tot_loss[loss=0.279, simple_loss=0.3398, pruned_loss=0.1091, over 3814568.43 frames. ], batch size: 47, lr: 1.56e-02, grad_scale: 8.0 2023-04-01 06:23:45,924 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=33695.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 06:23:50,552 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=33699.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 06:24:05,037 INFO [train.py:903] (1/4) Epoch 5, batch 6400, loss[loss=0.2612, simple_loss=0.3163, pruned_loss=0.1031, over 19794.00 frames. ], tot_loss[loss=0.2787, simple_loss=0.3398, pruned_loss=0.1088, over 3813682.14 frames. ], batch size: 48, lr: 1.55e-02, grad_scale: 8.0 2023-04-01 06:24:20,305 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=33724.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 06:24:37,365 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.704e+02 6.874e+02 8.420e+02 1.031e+03 3.616e+03, threshold=1.684e+03, percent-clipped=3.0 2023-04-01 06:25:05,852 INFO [train.py:903] (1/4) Epoch 5, batch 6450, loss[loss=0.3145, simple_loss=0.3633, pruned_loss=0.1328, over 19580.00 frames. ], tot_loss[loss=0.2799, simple_loss=0.3407, pruned_loss=0.1095, over 3819192.51 frames. ], batch size: 52, lr: 1.55e-02, grad_scale: 8.0 2023-04-01 06:25:18,529 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3577, 1.3298, 1.5449, 1.3932, 3.0092, 4.4501, 4.4186, 4.8449], device='cuda:1'), covar=tensor([0.1465, 0.2878, 0.3072, 0.1849, 0.0413, 0.0122, 0.0136, 0.0065], device='cuda:1'), in_proj_covar=tensor([0.0254, 0.0282, 0.0316, 0.0250, 0.0198, 0.0116, 0.0203, 0.0143], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-04-01 06:25:49,027 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9_rvb from training. Duration: 28.638875 2023-04-01 06:25:54,052 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=33802.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 06:26:06,349 INFO [train.py:903] (1/4) Epoch 5, batch 6500, loss[loss=0.2713, simple_loss=0.3329, pruned_loss=0.1049, over 17407.00 frames. ], tot_loss[loss=0.2808, simple_loss=0.3415, pruned_loss=0.11, over 3823825.52 frames. ], batch size: 101, lr: 1.55e-02, grad_scale: 8.0 2023-04-01 06:26:12,165 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1_rvb from training. Duration: 0.836375 2023-04-01 06:26:24,735 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=33827.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 06:26:32,483 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=33834.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 06:26:36,708 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.072e+02 6.680e+02 8.171e+02 1.132e+03 2.519e+03, threshold=1.634e+03, percent-clipped=6.0 2023-04-01 06:27:07,199 INFO [train.py:903] (1/4) Epoch 5, batch 6550, loss[loss=0.336, simple_loss=0.3826, pruned_loss=0.1447, over 13213.00 frames. ], tot_loss[loss=0.281, simple_loss=0.3419, pruned_loss=0.1101, over 3823479.85 frames. ], batch size: 135, lr: 1.55e-02, grad_scale: 8.0 2023-04-01 06:27:38,528 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=33888.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 06:27:59,331 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5115, 2.4858, 1.6492, 1.6417, 2.3970, 1.1842, 1.1328, 1.7551], device='cuda:1'), covar=tensor([0.0737, 0.0440, 0.0954, 0.0493, 0.0347, 0.0991, 0.0635, 0.0411], device='cuda:1'), in_proj_covar=tensor([0.0276, 0.0269, 0.0312, 0.0237, 0.0222, 0.0303, 0.0285, 0.0247], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 06:28:06,986 INFO [train.py:903] (1/4) Epoch 5, batch 6600, loss[loss=0.2244, simple_loss=0.2879, pruned_loss=0.08046, over 19749.00 frames. ], tot_loss[loss=0.2803, simple_loss=0.341, pruned_loss=0.1098, over 3826036.74 frames. ], batch size: 46, lr: 1.55e-02, grad_scale: 4.0 2023-04-01 06:28:08,496 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=33913.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 06:28:40,230 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.407e+02 7.502e+02 9.137e+02 1.060e+03 2.817e+03, threshold=1.827e+03, percent-clipped=6.0 2023-04-01 06:28:52,863 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=33949.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 06:29:01,842 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=33957.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 06:29:09,037 INFO [train.py:903] (1/4) Epoch 5, batch 6650, loss[loss=0.2604, simple_loss=0.3228, pruned_loss=0.09906, over 19730.00 frames. ], tot_loss[loss=0.2798, simple_loss=0.3404, pruned_loss=0.1096, over 3812011.43 frames. ], batch size: 51, lr: 1.55e-02, grad_scale: 4.0 2023-04-01 06:30:10,060 INFO [train.py:903] (1/4) Epoch 5, batch 6700, loss[loss=0.2957, simple_loss=0.3592, pruned_loss=0.1162, over 19676.00 frames. ], tot_loss[loss=0.2814, simple_loss=0.3417, pruned_loss=0.1106, over 3811278.83 frames. ], batch size: 60, lr: 1.55e-02, grad_scale: 4.0 2023-04-01 06:30:40,325 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.004e+02 7.257e+02 9.131e+02 1.100e+03 2.314e+03, threshold=1.826e+03, percent-clipped=7.0 2023-04-01 06:30:40,465 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=34039.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 06:31:06,264 INFO [train.py:903] (1/4) Epoch 5, batch 6750, loss[loss=0.2089, simple_loss=0.2847, pruned_loss=0.06655, over 19476.00 frames. ], tot_loss[loss=0.2823, simple_loss=0.3423, pruned_loss=0.1112, over 3810106.99 frames. ], batch size: 49, lr: 1.55e-02, grad_scale: 4.0 2023-04-01 06:31:06,529 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=34062.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 06:31:22,403 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3538, 1.1924, 1.2689, 1.7100, 1.4085, 1.5611, 1.6449, 1.4313], device='cuda:1'), covar=tensor([0.0861, 0.1071, 0.1146, 0.0895, 0.0938, 0.0821, 0.0908, 0.0751], device='cuda:1'), in_proj_covar=tensor([0.0232, 0.0251, 0.0249, 0.0278, 0.0268, 0.0233, 0.0231, 0.0224], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:1') 2023-04-01 06:32:02,682 INFO [train.py:903] (1/4) Epoch 5, batch 6800, loss[loss=0.3041, simple_loss=0.3447, pruned_loss=0.1318, over 19777.00 frames. ], tot_loss[loss=0.2824, simple_loss=0.3426, pruned_loss=0.1111, over 3803958.52 frames. ], batch size: 49, lr: 1.55e-02, grad_scale: 8.0 2023-04-01 06:32:30,647 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.271e+02 6.317e+02 7.808e+02 9.230e+02 1.582e+03, threshold=1.562e+03, percent-clipped=0.0 2023-04-01 06:32:47,618 WARNING [train.py:1073] (1/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 2023-04-01 06:32:48,644 WARNING [train.py:1073] (1/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 2023-04-01 06:32:50,744 INFO [train.py:903] (1/4) Epoch 6, batch 0, loss[loss=0.3173, simple_loss=0.3712, pruned_loss=0.1318, over 19592.00 frames. ], tot_loss[loss=0.3173, simple_loss=0.3712, pruned_loss=0.1318, over 19592.00 frames. ], batch size: 61, lr: 1.44e-02, grad_scale: 8.0 2023-04-01 06:32:50,744 INFO [train.py:928] (1/4) Computing validation loss 2023-04-01 06:33:02,090 INFO [train.py:937] (1/4) Epoch 6, validation: loss=0.2022, simple_loss=0.3015, pruned_loss=0.05149, over 944034.00 frames. 2023-04-01 06:33:02,091 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18649MB 2023-04-01 06:33:15,311 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 2023-04-01 06:33:20,309 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=34154.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 06:33:30,289 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=34163.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 06:34:03,526 INFO [train.py:903] (1/4) Epoch 6, batch 50, loss[loss=0.2657, simple_loss=0.339, pruned_loss=0.09617, over 19486.00 frames. ], tot_loss[loss=0.284, simple_loss=0.3458, pruned_loss=0.1111, over 857011.86 frames. ], batch size: 64, lr: 1.44e-02, grad_scale: 8.0 2023-04-01 06:34:22,167 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=34205.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 06:34:40,722 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1_rvb from training. Duration: 27.0318125 2023-04-01 06:34:53,304 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=34230.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 06:35:05,223 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.721e+02 5.756e+02 7.149e+02 1.025e+03 3.166e+03, threshold=1.430e+03, percent-clipped=7.0 2023-04-01 06:35:06,296 INFO [train.py:903] (1/4) Epoch 6, batch 100, loss[loss=0.2567, simple_loss=0.3206, pruned_loss=0.09646, over 19599.00 frames. ], tot_loss[loss=0.2764, simple_loss=0.3394, pruned_loss=0.1067, over 1521808.10 frames. ], batch size: 52, lr: 1.44e-02, grad_scale: 8.0 2023-04-01 06:35:18,598 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 2023-04-01 06:35:23,498 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.4034, 1.1442, 1.2016, 1.2036, 2.0755, 0.9424, 1.8547, 2.1914], device='cuda:1'), covar=tensor([0.0558, 0.2327, 0.2303, 0.1394, 0.0704, 0.1889, 0.0860, 0.0516], device='cuda:1'), in_proj_covar=tensor([0.0285, 0.0311, 0.0313, 0.0286, 0.0306, 0.0316, 0.0283, 0.0301], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 06:35:26,672 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.2726, 3.9044, 2.4370, 3.5711, 0.9742, 3.5608, 3.5883, 3.7796], device='cuda:1'), covar=tensor([0.0660, 0.0962, 0.1916, 0.0657, 0.3966, 0.0729, 0.0748, 0.0757], device='cuda:1'), in_proj_covar=tensor([0.0349, 0.0313, 0.0371, 0.0287, 0.0353, 0.0304, 0.0287, 0.0315], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 06:36:06,448 INFO [train.py:903] (1/4) Epoch 6, batch 150, loss[loss=0.271, simple_loss=0.3362, pruned_loss=0.1029, over 19697.00 frames. ], tot_loss[loss=0.2782, simple_loss=0.3398, pruned_loss=0.1082, over 2017729.07 frames. ], batch size: 59, lr: 1.44e-02, grad_scale: 4.0 2023-04-01 06:36:19,471 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=34301.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 06:37:08,895 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.164e+02 6.478e+02 8.283e+02 9.993e+02 1.951e+03, threshold=1.657e+03, percent-clipped=7.0 2023-04-01 06:37:08,922 INFO [train.py:903] (1/4) Epoch 6, batch 200, loss[loss=0.2563, simple_loss=0.3285, pruned_loss=0.09205, over 19667.00 frames. ], tot_loss[loss=0.2736, simple_loss=0.3362, pruned_loss=0.1055, over 2426769.22 frames. ], batch size: 55, lr: 1.44e-02, grad_scale: 4.0 2023-04-01 06:37:08,940 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 2023-04-01 06:37:44,178 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3128, 1.4676, 2.2352, 1.6608, 3.2310, 2.7291, 3.4456, 1.6357], device='cuda:1'), covar=tensor([0.1907, 0.3082, 0.1716, 0.1464, 0.1216, 0.1436, 0.1472, 0.2831], device='cuda:1'), in_proj_covar=tensor([0.0450, 0.0503, 0.0485, 0.0411, 0.0553, 0.0448, 0.0627, 0.0444], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 06:38:12,099 INFO [train.py:903] (1/4) Epoch 6, batch 250, loss[loss=0.2876, simple_loss=0.3541, pruned_loss=0.1106, over 19681.00 frames. ], tot_loss[loss=0.2734, simple_loss=0.3361, pruned_loss=0.1054, over 2742035.77 frames. ], batch size: 59, lr: 1.44e-02, grad_scale: 4.0 2023-04-01 06:38:33,065 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=34406.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 06:38:38,006 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=34410.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 06:38:38,440 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.74 vs. limit=5.0 2023-04-01 06:38:44,824 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.9177, 3.5551, 2.3161, 3.2883, 1.0889, 3.1754, 3.2439, 3.3749], device='cuda:1'), covar=tensor([0.0734, 0.1304, 0.2034, 0.0842, 0.3835, 0.1092, 0.0884, 0.0972], device='cuda:1'), in_proj_covar=tensor([0.0350, 0.0310, 0.0369, 0.0286, 0.0355, 0.0308, 0.0286, 0.0315], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 06:38:44,990 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=34416.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 06:39:08,854 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=34435.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 06:39:14,123 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.775e+02 6.949e+02 8.663e+02 1.115e+03 2.860e+03, threshold=1.733e+03, percent-clipped=3.0 2023-04-01 06:39:14,141 INFO [train.py:903] (1/4) Epoch 6, batch 300, loss[loss=0.3138, simple_loss=0.3545, pruned_loss=0.1366, over 13087.00 frames. ], tot_loss[loss=0.2743, simple_loss=0.3369, pruned_loss=0.1059, over 2969694.92 frames. ], batch size: 135, lr: 1.44e-02, grad_scale: 4.0 2023-04-01 06:39:15,668 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.3179, 1.2904, 1.3960, 1.5575, 2.8440, 1.0337, 1.9761, 2.9823], device='cuda:1'), covar=tensor([0.0389, 0.2489, 0.2478, 0.1526, 0.0644, 0.2407, 0.1202, 0.0443], device='cuda:1'), in_proj_covar=tensor([0.0290, 0.0316, 0.0320, 0.0288, 0.0311, 0.0318, 0.0289, 0.0307], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 06:40:17,202 INFO [train.py:903] (1/4) Epoch 6, batch 350, loss[loss=0.2258, simple_loss=0.2851, pruned_loss=0.08325, over 19716.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3365, pruned_loss=0.1059, over 3159717.33 frames. ], batch size: 46, lr: 1.43e-02, grad_scale: 4.0 2023-04-01 06:40:22,949 WARNING [train.py:1073] (1/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1_rvb from training. Duration: 0.9681875 2023-04-01 06:40:37,841 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=34507.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 06:40:55,506 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=34521.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 06:41:18,578 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.691e+02 6.755e+02 8.258e+02 9.979e+02 1.871e+03, threshold=1.652e+03, percent-clipped=1.0 2023-04-01 06:41:18,597 INFO [train.py:903] (1/4) Epoch 6, batch 400, loss[loss=0.2378, simple_loss=0.3116, pruned_loss=0.08201, over 19481.00 frames. ], tot_loss[loss=0.2768, simple_loss=0.3387, pruned_loss=0.1075, over 3298169.31 frames. ], batch size: 49, lr: 1.43e-02, grad_scale: 8.0 2023-04-01 06:42:20,647 INFO [train.py:903] (1/4) Epoch 6, batch 450, loss[loss=0.2671, simple_loss=0.3299, pruned_loss=0.1021, over 19766.00 frames. ], tot_loss[loss=0.2767, simple_loss=0.3384, pruned_loss=0.1075, over 3428216.93 frames. ], batch size: 54, lr: 1.43e-02, grad_scale: 8.0 2023-04-01 06:42:48,995 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=34611.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 06:42:54,482 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 2023-04-01 06:42:55,448 WARNING [train.py:1073] (1/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 2023-04-01 06:43:01,475 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=34622.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 06:43:23,824 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.432e+02 6.970e+02 8.269e+02 1.071e+03 2.551e+03, threshold=1.654e+03, percent-clipped=6.0 2023-04-01 06:43:23,841 INFO [train.py:903] (1/4) Epoch 6, batch 500, loss[loss=0.2986, simple_loss=0.3649, pruned_loss=0.1161, over 19611.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.3379, pruned_loss=0.1068, over 3528132.34 frames. ], batch size: 57, lr: 1.43e-02, grad_scale: 8.0 2023-04-01 06:43:24,075 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=34640.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 06:43:29,887 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=34644.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 06:44:03,467 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=34672.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 06:44:27,115 INFO [train.py:903] (1/4) Epoch 6, batch 550, loss[loss=0.2956, simple_loss=0.356, pruned_loss=0.1176, over 19675.00 frames. ], tot_loss[loss=0.2752, simple_loss=0.3369, pruned_loss=0.1068, over 3592438.65 frames. ], batch size: 55, lr: 1.43e-02, grad_scale: 8.0 2023-04-01 06:44:37,112 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=34697.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 06:45:17,117 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=34728.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 06:45:31,835 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.620e+02 6.191e+02 8.093e+02 9.820e+02 1.880e+03, threshold=1.619e+03, percent-clipped=2.0 2023-04-01 06:45:31,854 INFO [train.py:903] (1/4) Epoch 6, batch 600, loss[loss=0.224, simple_loss=0.2967, pruned_loss=0.07562, over 19484.00 frames. ], tot_loss[loss=0.2751, simple_loss=0.3372, pruned_loss=0.1065, over 3648813.71 frames. ], batch size: 49, lr: 1.43e-02, grad_scale: 8.0 2023-04-01 06:46:13,298 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 2023-04-01 06:46:15,092 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 2023-04-01 06:46:20,223 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=34777.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 06:46:35,482 INFO [train.py:903] (1/4) Epoch 6, batch 650, loss[loss=0.217, simple_loss=0.2868, pruned_loss=0.07361, over 19062.00 frames. ], tot_loss[loss=0.2775, simple_loss=0.3394, pruned_loss=0.1078, over 3678724.06 frames. ], batch size: 42, lr: 1.43e-02, grad_scale: 8.0 2023-04-01 06:46:51,841 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=34802.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 06:47:38,641 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.675e+02 6.280e+02 8.587e+02 1.153e+03 3.497e+03, threshold=1.717e+03, percent-clipped=9.0 2023-04-01 06:47:38,660 INFO [train.py:903] (1/4) Epoch 6, batch 700, loss[loss=0.2277, simple_loss=0.2968, pruned_loss=0.07935, over 19482.00 frames. ], tot_loss[loss=0.2778, simple_loss=0.3393, pruned_loss=0.1082, over 3716257.93 frames. ], batch size: 49, lr: 1.43e-02, grad_scale: 8.0 2023-04-01 06:48:27,752 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=34878.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 06:48:43,528 INFO [train.py:903] (1/4) Epoch 6, batch 750, loss[loss=0.2561, simple_loss=0.3291, pruned_loss=0.09148, over 19538.00 frames. ], tot_loss[loss=0.2772, simple_loss=0.3389, pruned_loss=0.1077, over 3731303.03 frames. ], batch size: 56, lr: 1.43e-02, grad_scale: 8.0 2023-04-01 06:49:00,050 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=34903.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 06:49:06,790 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2884, 1.1483, 1.5758, 1.0657, 2.5250, 3.4107, 3.1588, 3.4778], device='cuda:1'), covar=tensor([0.1483, 0.3082, 0.2770, 0.1985, 0.0453, 0.0188, 0.0205, 0.0137], device='cuda:1'), in_proj_covar=tensor([0.0253, 0.0281, 0.0311, 0.0246, 0.0198, 0.0118, 0.0201, 0.0145], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-04-01 06:49:45,087 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.784e+02 6.254e+02 7.861e+02 1.094e+03 2.828e+03, threshold=1.572e+03, percent-clipped=5.0 2023-04-01 06:49:45,105 INFO [train.py:903] (1/4) Epoch 6, batch 800, loss[loss=0.2793, simple_loss=0.3422, pruned_loss=0.1083, over 19782.00 frames. ], tot_loss[loss=0.2763, simple_loss=0.3382, pruned_loss=0.1072, over 3768305.00 frames. ], batch size: 56, lr: 1.43e-02, grad_scale: 8.0 2023-04-01 06:50:02,472 WARNING [train.py:1073] (1/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 2023-04-01 06:50:03,746 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=34955.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 06:50:21,725 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 2023-04-01 06:50:41,616 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=34984.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 06:50:46,219 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=34988.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 06:50:48,467 INFO [train.py:903] (1/4) Epoch 6, batch 850, loss[loss=0.2566, simple_loss=0.3112, pruned_loss=0.101, over 18675.00 frames. ], tot_loss[loss=0.2763, simple_loss=0.3381, pruned_loss=0.1073, over 3774191.04 frames. ], batch size: 41, lr: 1.42e-02, grad_scale: 8.0 2023-04-01 06:51:32,728 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2096, 1.2238, 1.7223, 1.4775, 2.3252, 2.0696, 2.4994, 0.8664], device='cuda:1'), covar=tensor([0.1860, 0.3110, 0.1654, 0.1425, 0.1169, 0.1513, 0.1285, 0.2842], device='cuda:1'), in_proj_covar=tensor([0.0448, 0.0511, 0.0488, 0.0412, 0.0556, 0.0449, 0.0625, 0.0448], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 06:51:42,518 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9_rvb from training. Duration: 25.061125 2023-04-01 06:51:49,532 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.027e+02 6.223e+02 7.935e+02 9.772e+02 2.166e+03, threshold=1.587e+03, percent-clipped=2.0 2023-04-01 06:51:49,550 INFO [train.py:903] (1/4) Epoch 6, batch 900, loss[loss=0.3149, simple_loss=0.3797, pruned_loss=0.125, over 19574.00 frames. ], tot_loss[loss=0.2764, simple_loss=0.338, pruned_loss=0.1074, over 3795659.78 frames. ], batch size: 61, lr: 1.42e-02, grad_scale: 8.0 2023-04-01 06:51:51,218 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3843, 1.0326, 1.4262, 1.2079, 2.5123, 3.3454, 3.2799, 3.6852], device='cuda:1'), covar=tensor([0.1442, 0.3835, 0.3621, 0.1939, 0.0504, 0.0194, 0.0243, 0.0127], device='cuda:1'), in_proj_covar=tensor([0.0253, 0.0280, 0.0311, 0.0248, 0.0198, 0.0119, 0.0201, 0.0145], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-04-01 06:52:28,231 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=35070.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 06:52:30,297 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=35072.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 06:52:51,733 INFO [train.py:903] (1/4) Epoch 6, batch 950, loss[loss=0.2703, simple_loss=0.3276, pruned_loss=0.1065, over 19736.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.3376, pruned_loss=0.1069, over 3811881.41 frames. ], batch size: 46, lr: 1.42e-02, grad_scale: 8.0 2023-04-01 06:52:57,534 WARNING [train.py:1073] (1/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9_rvb from training. Duration: 26.32775 2023-04-01 06:53:04,328 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=35099.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 06:53:09,190 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=35103.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 06:53:18,081 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=35110.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 06:53:55,217 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.390e+02 7.194e+02 8.589e+02 1.083e+03 2.096e+03, threshold=1.718e+03, percent-clipped=5.0 2023-04-01 06:53:55,236 INFO [train.py:903] (1/4) Epoch 6, batch 1000, loss[loss=0.2946, simple_loss=0.3428, pruned_loss=0.1232, over 19463.00 frames. ], tot_loss[loss=0.2756, simple_loss=0.3376, pruned_loss=0.1068, over 3818344.30 frames. ], batch size: 49, lr: 1.42e-02, grad_scale: 8.0 2023-04-01 06:54:48,376 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 2023-04-01 06:54:52,231 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=35187.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 06:54:55,176 INFO [train.py:903] (1/4) Epoch 6, batch 1050, loss[loss=0.3119, simple_loss=0.3632, pruned_loss=0.1303, over 17334.00 frames. ], tot_loss[loss=0.2782, simple_loss=0.3397, pruned_loss=0.1083, over 3817162.63 frames. ], batch size: 101, lr: 1.42e-02, grad_scale: 8.0 2023-04-01 06:55:30,964 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 2023-04-01 06:55:49,752 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 2023-04-01 06:55:57,166 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.102e+02 7.066e+02 8.619e+02 1.238e+03 3.302e+03, threshold=1.724e+03, percent-clipped=8.0 2023-04-01 06:55:57,185 INFO [train.py:903] (1/4) Epoch 6, batch 1100, loss[loss=0.3319, simple_loss=0.375, pruned_loss=0.1443, over 13313.00 frames. ], tot_loss[loss=0.278, simple_loss=0.3393, pruned_loss=0.1083, over 3813077.03 frames. ], batch size: 136, lr: 1.42e-02, grad_scale: 8.0 2023-04-01 06:56:09,913 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9913, 1.9457, 1.6323, 1.3756, 1.4141, 1.6189, 0.2745, 0.7614], device='cuda:1'), covar=tensor([0.0253, 0.0288, 0.0172, 0.0294, 0.0547, 0.0311, 0.0545, 0.0500], device='cuda:1'), in_proj_covar=tensor([0.0285, 0.0290, 0.0289, 0.0308, 0.0374, 0.0294, 0.0285, 0.0298], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 06:56:48,020 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.78 vs. limit=5.0 2023-04-01 06:56:59,592 INFO [train.py:903] (1/4) Epoch 6, batch 1150, loss[loss=0.2932, simple_loss=0.3594, pruned_loss=0.1135, over 19681.00 frames. ], tot_loss[loss=0.2782, simple_loss=0.3397, pruned_loss=0.1083, over 3813845.91 frames. ], batch size: 59, lr: 1.42e-02, grad_scale: 8.0 2023-04-01 06:57:45,412 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=35326.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 06:58:04,197 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.615e+02 6.124e+02 7.469e+02 9.050e+02 1.884e+03, threshold=1.494e+03, percent-clipped=1.0 2023-04-01 06:58:04,215 INFO [train.py:903] (1/4) Epoch 6, batch 1200, loss[loss=0.288, simple_loss=0.3606, pruned_loss=0.1077, over 19786.00 frames. ], tot_loss[loss=0.2784, simple_loss=0.3401, pruned_loss=0.1084, over 3810100.20 frames. ], batch size: 56, lr: 1.42e-02, grad_scale: 8.0 2023-04-01 06:58:17,322 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=35351.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 06:58:23,058 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=35355.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 06:58:27,680 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=35359.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 06:58:32,172 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=35363.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 06:58:36,592 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 2023-04-01 06:58:55,127 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=35380.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 06:58:56,267 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2503, 1.2730, 1.8219, 1.4437, 2.6599, 2.2850, 2.8140, 0.9393], device='cuda:1'), covar=tensor([0.1798, 0.2997, 0.1545, 0.1496, 0.1032, 0.1376, 0.1071, 0.2812], device='cuda:1'), in_proj_covar=tensor([0.0447, 0.0508, 0.0494, 0.0413, 0.0562, 0.0451, 0.0626, 0.0452], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 06:58:59,580 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=35384.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 06:59:05,730 INFO [train.py:903] (1/4) Epoch 6, batch 1250, loss[loss=0.2941, simple_loss=0.3513, pruned_loss=0.1184, over 18701.00 frames. ], tot_loss[loss=0.2774, simple_loss=0.3393, pruned_loss=0.1078, over 3820550.83 frames. ], batch size: 74, lr: 1.42e-02, grad_scale: 8.0 2023-04-01 06:59:09,373 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=35392.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 06:59:42,219 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6449, 1.7083, 1.4574, 2.4580, 1.7129, 2.2770, 1.8879, 1.2993], device='cuda:1'), covar=tensor([0.2812, 0.2283, 0.1585, 0.1253, 0.2331, 0.0970, 0.2975, 0.2873], device='cuda:1'), in_proj_covar=tensor([0.0655, 0.0640, 0.0570, 0.0793, 0.0677, 0.0565, 0.0697, 0.0605], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 07:00:08,109 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.399e+02 6.685e+02 8.523e+02 1.077e+03 2.432e+03, threshold=1.705e+03, percent-clipped=5.0 2023-04-01 07:00:08,127 INFO [train.py:903] (1/4) Epoch 6, batch 1300, loss[loss=0.2844, simple_loss=0.347, pruned_loss=0.1109, over 19758.00 frames. ], tot_loss[loss=0.2793, simple_loss=0.3409, pruned_loss=0.1089, over 3802719.30 frames. ], batch size: 54, lr: 1.42e-02, grad_scale: 8.0 2023-04-01 07:00:12,109 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=35443.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 07:00:25,337 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0409, 3.7624, 2.1839, 2.4373, 3.3258, 1.8691, 1.1390, 1.9374], device='cuda:1'), covar=tensor([0.1026, 0.0295, 0.0776, 0.0546, 0.0398, 0.0881, 0.0893, 0.0609], device='cuda:1'), in_proj_covar=tensor([0.0281, 0.0277, 0.0318, 0.0236, 0.0231, 0.0309, 0.0283, 0.0253], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 07:00:26,224 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=35454.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 07:00:29,030 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.79 vs. limit=2.0 2023-04-01 07:00:37,007 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.0986, 1.0569, 1.0565, 1.3218, 1.0363, 1.3041, 1.2938, 1.2219], device='cuda:1'), covar=tensor([0.0966, 0.1150, 0.1121, 0.0741, 0.0950, 0.0871, 0.0875, 0.0780], device='cuda:1'), in_proj_covar=tensor([0.0230, 0.0250, 0.0239, 0.0274, 0.0265, 0.0232, 0.0226, 0.0221], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:1') 2023-04-01 07:00:43,921 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=35468.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 07:00:57,890 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([6.2788, 5.5816, 2.7857, 4.9461, 1.2970, 5.5519, 5.4861, 5.6913], device='cuda:1'), covar=tensor([0.0297, 0.0699, 0.1753, 0.0496, 0.3457, 0.0437, 0.0421, 0.0510], device='cuda:1'), in_proj_covar=tensor([0.0352, 0.0312, 0.0371, 0.0289, 0.0350, 0.0310, 0.0289, 0.0321], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 07:01:08,277 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 2023-04-01 07:01:11,030 INFO [train.py:903] (1/4) Epoch 6, batch 1350, loss[loss=0.2705, simple_loss=0.3359, pruned_loss=0.1025, over 19402.00 frames. ], tot_loss[loss=0.2792, simple_loss=0.341, pruned_loss=0.1087, over 3804283.14 frames. ], batch size: 48, lr: 1.41e-02, grad_scale: 8.0 2023-04-01 07:01:40,808 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3943, 1.5643, 2.0567, 1.6734, 2.8417, 4.6134, 4.5197, 4.8847], device='cuda:1'), covar=tensor([0.1408, 0.2776, 0.2732, 0.1696, 0.0487, 0.0113, 0.0163, 0.0075], device='cuda:1'), in_proj_covar=tensor([0.0254, 0.0285, 0.0312, 0.0249, 0.0201, 0.0122, 0.0202, 0.0146], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-04-01 07:02:13,027 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.489e+02 6.783e+02 8.495e+02 1.071e+03 2.340e+03, threshold=1.699e+03, percent-clipped=3.0 2023-04-01 07:02:13,045 INFO [train.py:903] (1/4) Epoch 6, batch 1400, loss[loss=0.3267, simple_loss=0.3683, pruned_loss=0.1426, over 19833.00 frames. ], tot_loss[loss=0.2778, simple_loss=0.3397, pruned_loss=0.1079, over 3821484.85 frames. ], batch size: 52, lr: 1.41e-02, grad_scale: 8.0 2023-04-01 07:02:22,733 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=35548.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 07:02:47,168 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=35569.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 07:03:13,361 INFO [train.py:903] (1/4) Epoch 6, batch 1450, loss[loss=0.3472, simple_loss=0.3926, pruned_loss=0.1509, over 19267.00 frames. ], tot_loss[loss=0.279, simple_loss=0.3403, pruned_loss=0.1088, over 3825741.68 frames. ], batch size: 66, lr: 1.41e-02, grad_scale: 8.0 2023-04-01 07:03:13,400 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_rvb from training. Duration: 25.85 2023-04-01 07:03:52,451 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.3375, 1.4382, 1.4895, 1.5180, 2.8101, 1.0636, 1.9248, 2.9132], device='cuda:1'), covar=tensor([0.0394, 0.2375, 0.2399, 0.1526, 0.0714, 0.2302, 0.1182, 0.0447], device='cuda:1'), in_proj_covar=tensor([0.0295, 0.0313, 0.0316, 0.0286, 0.0313, 0.0312, 0.0286, 0.0305], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 07:04:15,922 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.374e+02 6.609e+02 8.520e+02 1.101e+03 2.891e+03, threshold=1.704e+03, percent-clipped=3.0 2023-04-01 07:04:15,941 INFO [train.py:903] (1/4) Epoch 6, batch 1500, loss[loss=0.2888, simple_loss=0.3387, pruned_loss=0.1195, over 19829.00 frames. ], tot_loss[loss=0.2786, simple_loss=0.34, pruned_loss=0.1086, over 3812842.26 frames. ], batch size: 52, lr: 1.41e-02, grad_scale: 8.0 2023-04-01 07:05:17,132 INFO [train.py:903] (1/4) Epoch 6, batch 1550, loss[loss=0.311, simple_loss=0.3522, pruned_loss=0.1349, over 19486.00 frames. ], tot_loss[loss=0.2794, simple_loss=0.3404, pruned_loss=0.1092, over 3812008.23 frames. ], batch size: 49, lr: 1.41e-02, grad_scale: 8.0 2023-04-01 07:05:39,874 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=35707.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 07:05:57,553 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6112, 1.3085, 1.3424, 2.0136, 1.5129, 2.0677, 2.0737, 1.9005], device='cuda:1'), covar=tensor([0.0828, 0.1010, 0.1111, 0.0885, 0.1032, 0.0675, 0.0843, 0.0591], device='cuda:1'), in_proj_covar=tensor([0.0228, 0.0248, 0.0239, 0.0270, 0.0263, 0.0231, 0.0223, 0.0218], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1') 2023-04-01 07:06:05,680 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([0.8856, 1.3160, 1.0406, 1.0224, 1.2145, 0.8510, 0.8607, 1.2400], device='cuda:1'), covar=tensor([0.0420, 0.0545, 0.0837, 0.0439, 0.0374, 0.0913, 0.0504, 0.0332], device='cuda:1'), in_proj_covar=tensor([0.0280, 0.0274, 0.0313, 0.0238, 0.0228, 0.0305, 0.0284, 0.0253], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 07:06:16,821 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=35736.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 07:06:22,433 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.104e+02 6.450e+02 9.026e+02 1.093e+03 2.835e+03, threshold=1.805e+03, percent-clipped=5.0 2023-04-01 07:06:22,452 INFO [train.py:903] (1/4) Epoch 6, batch 1600, loss[loss=0.2228, simple_loss=0.2955, pruned_loss=0.07509, over 19845.00 frames. ], tot_loss[loss=0.2772, simple_loss=0.3388, pruned_loss=0.1078, over 3826990.97 frames. ], batch size: 52, lr: 1.41e-02, grad_scale: 8.0 2023-04-01 07:06:44,232 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9_rvb from training. Duration: 30.1555625 2023-04-01 07:06:53,853 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.0573, 1.1352, 1.4246, 1.6005, 2.6714, 1.0166, 1.9283, 2.6779], device='cuda:1'), covar=tensor([0.0403, 0.2475, 0.2374, 0.1312, 0.0599, 0.2043, 0.0969, 0.0471], device='cuda:1'), in_proj_covar=tensor([0.0294, 0.0310, 0.0313, 0.0286, 0.0308, 0.0309, 0.0283, 0.0302], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 07:07:23,499 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=35789.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 07:07:24,376 INFO [train.py:903] (1/4) Epoch 6, batch 1650, loss[loss=0.3481, simple_loss=0.3959, pruned_loss=0.1502, over 18902.00 frames. ], tot_loss[loss=0.2773, simple_loss=0.3391, pruned_loss=0.1078, over 3830837.62 frames. ], batch size: 74, lr: 1.41e-02, grad_scale: 8.0 2023-04-01 07:07:35,329 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0344, 2.0643, 1.9541, 3.1698, 1.9959, 3.3329, 2.8405, 1.8678], device='cuda:1'), covar=tensor([0.2776, 0.2207, 0.1059, 0.1390, 0.2718, 0.0740, 0.2058, 0.1964], device='cuda:1'), in_proj_covar=tensor([0.0659, 0.0649, 0.0575, 0.0807, 0.0686, 0.0568, 0.0701, 0.0612], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 07:07:50,596 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 2023-04-01 07:07:51,784 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.13 vs. limit=2.0 2023-04-01 07:08:05,691 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=35822.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 07:08:09,298 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=35825.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 07:08:13,893 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=35829.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 07:08:27,427 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.226e+02 6.593e+02 7.720e+02 9.793e+02 2.227e+03, threshold=1.544e+03, percent-clipped=1.0 2023-04-01 07:08:27,445 INFO [train.py:903] (1/4) Epoch 6, batch 1700, loss[loss=0.254, simple_loss=0.3209, pruned_loss=0.09354, over 19843.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.3378, pruned_loss=0.1068, over 3836352.70 frames. ], batch size: 52, lr: 1.41e-02, grad_scale: 8.0 2023-04-01 07:08:38,865 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=35850.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 07:08:40,021 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=35851.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 07:09:06,037 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 2023-04-01 07:09:29,408 INFO [train.py:903] (1/4) Epoch 6, batch 1750, loss[loss=0.3245, simple_loss=0.3849, pruned_loss=0.1321, over 19699.00 frames. ], tot_loss[loss=0.2749, simple_loss=0.3374, pruned_loss=0.1062, over 3829208.31 frames. ], batch size: 58, lr: 1.41e-02, grad_scale: 8.0 2023-04-01 07:09:31,965 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=35892.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 07:10:33,848 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.141e+02 6.695e+02 8.372e+02 1.116e+03 2.634e+03, threshold=1.674e+03, percent-clipped=7.0 2023-04-01 07:10:33,866 INFO [train.py:903] (1/4) Epoch 6, batch 1800, loss[loss=0.3091, simple_loss=0.3798, pruned_loss=0.1192, over 19490.00 frames. ], tot_loss[loss=0.2756, simple_loss=0.3379, pruned_loss=0.1067, over 3820361.73 frames. ], batch size: 64, lr: 1.41e-02, grad_scale: 8.0 2023-04-01 07:10:57,795 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1135, 1.2253, 1.5319, 0.9579, 2.5693, 2.8692, 2.6973, 3.0808], device='cuda:1'), covar=tensor([0.1392, 0.2754, 0.2680, 0.1909, 0.0420, 0.0313, 0.0254, 0.0158], device='cuda:1'), in_proj_covar=tensor([0.0254, 0.0281, 0.0315, 0.0247, 0.0201, 0.0123, 0.0201, 0.0147], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-04-01 07:11:31,904 WARNING [train.py:1073] (1/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9_rvb from training. Duration: 25.0944375 2023-04-01 07:11:36,740 INFO [train.py:903] (1/4) Epoch 6, batch 1850, loss[loss=0.2918, simple_loss=0.3561, pruned_loss=0.1138, over 18889.00 frames. ], tot_loss[loss=0.2769, simple_loss=0.3387, pruned_loss=0.1076, over 3820382.31 frames. ], batch size: 74, lr: 1.40e-02, grad_scale: 8.0 2023-04-01 07:11:59,155 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=36007.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 07:12:11,437 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9_rvb from training. Duration: 27.8166875 2023-04-01 07:12:40,884 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.024e+02 7.330e+02 9.053e+02 1.086e+03 1.723e+03, threshold=1.811e+03, percent-clipped=2.0 2023-04-01 07:12:40,902 INFO [train.py:903] (1/4) Epoch 6, batch 1900, loss[loss=0.1961, simple_loss=0.2716, pruned_loss=0.06024, over 19774.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.3377, pruned_loss=0.1067, over 3831862.81 frames. ], batch size: 46, lr: 1.40e-02, grad_scale: 8.0 2023-04-01 07:12:57,328 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 2023-04-01 07:13:04,077 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9_rvb from training. Duration: 27.02225 2023-04-01 07:13:27,622 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 2023-04-01 07:13:29,155 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=36078.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 07:13:36,009 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.5477, 1.2545, 1.3823, 1.5741, 3.1017, 1.0630, 1.9524, 3.2052], device='cuda:1'), covar=tensor([0.0351, 0.2454, 0.2547, 0.1560, 0.0596, 0.2367, 0.1301, 0.0383], device='cuda:1'), in_proj_covar=tensor([0.0299, 0.0317, 0.0321, 0.0295, 0.0311, 0.0320, 0.0291, 0.0310], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 07:13:42,467 INFO [train.py:903] (1/4) Epoch 6, batch 1950, loss[loss=0.2725, simple_loss=0.3404, pruned_loss=0.1023, over 19438.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.3382, pruned_loss=0.1066, over 3833781.13 frames. ], batch size: 70, lr: 1.40e-02, grad_scale: 8.0 2023-04-01 07:14:00,154 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=36103.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 07:14:05,015 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=36107.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 07:14:35,254 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=36132.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 07:14:36,138 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=36133.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 07:14:45,004 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.894e+02 6.352e+02 7.868e+02 9.806e+02 1.510e+03, threshold=1.574e+03, percent-clipped=0.0 2023-04-01 07:14:45,021 INFO [train.py:903] (1/4) Epoch 6, batch 2000, loss[loss=0.3132, simple_loss=0.3615, pruned_loss=0.1324, over 13214.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.3385, pruned_loss=0.1066, over 3818733.81 frames. ], batch size: 136, lr: 1.40e-02, grad_scale: 8.0 2023-04-01 07:15:24,945 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=36173.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 07:15:42,582 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_rvb from training. Duration: 26.8349375 2023-04-01 07:15:46,069 INFO [train.py:903] (1/4) Epoch 6, batch 2050, loss[loss=0.2096, simple_loss=0.2772, pruned_loss=0.07093, over 18638.00 frames. ], tot_loss[loss=0.2768, simple_loss=0.3388, pruned_loss=0.1074, over 3805086.10 frames. ], batch size: 41, lr: 1.40e-02, grad_scale: 8.0 2023-04-01 07:15:57,096 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=36199.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 07:16:00,513 WARNING [train.py:1073] (1/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 2023-04-01 07:16:03,090 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 2023-04-01 07:16:22,950 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 2023-04-01 07:16:23,186 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([6.2213, 5.5153, 3.0399, 4.8265, 1.2662, 5.3755, 5.2719, 5.5620], device='cuda:1'), covar=tensor([0.0379, 0.0923, 0.1880, 0.0609, 0.4032, 0.0601, 0.0543, 0.0590], device='cuda:1'), in_proj_covar=tensor([0.0368, 0.0321, 0.0377, 0.0296, 0.0360, 0.0317, 0.0297, 0.0331], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-01 07:16:37,667 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4135, 2.3661, 1.6117, 1.5781, 2.2249, 1.3149, 1.1465, 1.8515], device='cuda:1'), covar=tensor([0.0779, 0.0419, 0.0807, 0.0510, 0.0312, 0.0917, 0.0661, 0.0418], device='cuda:1'), in_proj_covar=tensor([0.0280, 0.0276, 0.0310, 0.0240, 0.0228, 0.0309, 0.0287, 0.0254], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 07:16:47,772 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.109e+02 6.690e+02 8.798e+02 1.173e+03 2.442e+03, threshold=1.760e+03, percent-clipped=12.0 2023-04-01 07:16:47,790 INFO [train.py:903] (1/4) Epoch 6, batch 2100, loss[loss=0.2246, simple_loss=0.2967, pruned_loss=0.07621, over 19858.00 frames. ], tot_loss[loss=0.2758, simple_loss=0.338, pruned_loss=0.1068, over 3798855.18 frames. ], batch size: 52, lr: 1.40e-02, grad_scale: 8.0 2023-04-01 07:16:57,873 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=36248.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 07:17:17,845 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=36263.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 07:17:18,576 WARNING [train.py:1073] (1/4) Exclude cut with ID 453-131332-0000-131866_sp0.9_rvb from training. Duration: 25.3333125 2023-04-01 07:17:39,999 WARNING [train.py:1073] (1/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9_rvb from training. Duration: 26.6166875 2023-04-01 07:17:49,930 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=36288.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 07:17:49,972 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=36288.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 07:17:51,846 INFO [train.py:903] (1/4) Epoch 6, batch 2150, loss[loss=0.3227, simple_loss=0.3654, pruned_loss=0.14, over 19845.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.3376, pruned_loss=0.1068, over 3792713.25 frames. ], batch size: 52, lr: 1.40e-02, grad_scale: 16.0 2023-04-01 07:18:48,679 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.7453, 4.2210, 4.4512, 4.3969, 1.5095, 4.0191, 3.6396, 4.0623], device='cuda:1'), covar=tensor([0.1017, 0.0470, 0.0364, 0.0387, 0.3736, 0.0436, 0.0450, 0.0777], device='cuda:1'), in_proj_covar=tensor([0.0539, 0.0470, 0.0627, 0.0522, 0.0599, 0.0391, 0.0399, 0.0592], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1') 2023-04-01 07:18:53,904 INFO [train.py:903] (1/4) Epoch 6, batch 2200, loss[loss=0.2162, simple_loss=0.2916, pruned_loss=0.07042, over 19618.00 frames. ], tot_loss[loss=0.2762, simple_loss=0.3375, pruned_loss=0.1075, over 3785209.43 frames. ], batch size: 50, lr: 1.40e-02, grad_scale: 8.0 2023-04-01 07:18:55,064 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.621e+02 6.300e+02 8.031e+02 1.073e+03 2.013e+03, threshold=1.606e+03, percent-clipped=1.0 2023-04-01 07:19:06,902 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0 2023-04-01 07:19:54,195 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.81 vs. limit=2.0 2023-04-01 07:19:57,016 INFO [train.py:903] (1/4) Epoch 6, batch 2250, loss[loss=0.262, simple_loss=0.316, pruned_loss=0.104, over 16006.00 frames. ], tot_loss[loss=0.2766, simple_loss=0.3382, pruned_loss=0.1075, over 3797392.30 frames. ], batch size: 35, lr: 1.40e-02, grad_scale: 8.0 2023-04-01 07:20:58,357 INFO [train.py:903] (1/4) Epoch 6, batch 2300, loss[loss=0.2633, simple_loss=0.327, pruned_loss=0.09976, over 19476.00 frames. ], tot_loss[loss=0.2732, simple_loss=0.3359, pruned_loss=0.1053, over 3807173.18 frames. ], batch size: 49, lr: 1.40e-02, grad_scale: 8.0 2023-04-01 07:20:59,562 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.730e+02 6.689e+02 7.482e+02 9.822e+02 1.768e+03, threshold=1.496e+03, percent-clipped=2.0 2023-04-01 07:21:11,047 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.5630, 4.0764, 2.4177, 3.6616, 1.0287, 3.5535, 3.7568, 3.8102], device='cuda:1'), covar=tensor([0.0602, 0.1111, 0.2167, 0.0670, 0.3964, 0.0911, 0.0699, 0.1010], device='cuda:1'), in_proj_covar=tensor([0.0362, 0.0321, 0.0374, 0.0294, 0.0358, 0.0319, 0.0294, 0.0329], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-01 07:21:14,318 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_rvb from training. Duration: 26.205 2023-04-01 07:22:00,467 INFO [train.py:903] (1/4) Epoch 6, batch 2350, loss[loss=0.3212, simple_loss=0.372, pruned_loss=0.1352, over 18022.00 frames. ], tot_loss[loss=0.2735, simple_loss=0.3361, pruned_loss=0.1054, over 3817402.48 frames. ], batch size: 83, lr: 1.40e-02, grad_scale: 8.0 2023-04-01 07:22:19,306 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=36504.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 07:22:43,438 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_rvb from training. Duration: 25.775 2023-04-01 07:22:49,441 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=36529.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 07:23:00,562 WARNING [train.py:1073] (1/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9_rvb from training. Duration: 25.45 2023-04-01 07:23:02,881 INFO [train.py:903] (1/4) Epoch 6, batch 2400, loss[loss=0.289, simple_loss=0.3553, pruned_loss=0.1113, over 17384.00 frames. ], tot_loss[loss=0.2725, simple_loss=0.3357, pruned_loss=0.1047, over 3821483.07 frames. ], batch size: 101, lr: 1.39e-02, grad_scale: 8.0 2023-04-01 07:23:04,016 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.742e+02 5.696e+02 7.249e+02 9.157e+02 1.479e+03, threshold=1.450e+03, percent-clipped=0.0 2023-04-01 07:23:07,400 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=36543.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 07:23:08,778 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=36544.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 07:23:26,795 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.49 vs. limit=5.0 2023-04-01 07:23:39,016 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=36569.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 07:24:06,992 INFO [train.py:903] (1/4) Epoch 6, batch 2450, loss[loss=0.3296, simple_loss=0.37, pruned_loss=0.1446, over 13150.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.3375, pruned_loss=0.1057, over 3815412.89 frames. ], batch size: 136, lr: 1.39e-02, grad_scale: 8.0 2023-04-01 07:24:46,721 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=36623.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 07:24:51,017 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=36626.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 07:24:56,997 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=36630.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 07:25:07,900 INFO [train.py:903] (1/4) Epoch 6, batch 2500, loss[loss=0.2698, simple_loss=0.3408, pruned_loss=0.09942, over 19547.00 frames. ], tot_loss[loss=0.274, simple_loss=0.3371, pruned_loss=0.1055, over 3805406.12 frames. ], batch size: 56, lr: 1.39e-02, grad_scale: 8.0 2023-04-01 07:25:09,076 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.995e+02 6.265e+02 8.006e+02 9.274e+02 1.564e+03, threshold=1.601e+03, percent-clipped=1.0 2023-04-01 07:25:30,329 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=36658.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 07:26:09,567 INFO [train.py:903] (1/4) Epoch 6, batch 2550, loss[loss=0.3398, simple_loss=0.3788, pruned_loss=0.1504, over 13241.00 frames. ], tot_loss[loss=0.2728, simple_loss=0.3359, pruned_loss=0.1049, over 3800848.90 frames. ], batch size: 136, lr: 1.39e-02, grad_scale: 8.0 2023-04-01 07:26:44,789 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=36718.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 07:26:46,949 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([0.8142, 1.0295, 1.3499, 0.5885, 2.1195, 2.1990, 1.9580, 2.2875], device='cuda:1'), covar=tensor([0.1423, 0.2960, 0.2816, 0.2133, 0.0450, 0.0313, 0.0357, 0.0222], device='cuda:1'), in_proj_covar=tensor([0.0253, 0.0282, 0.0312, 0.0247, 0.0200, 0.0122, 0.0200, 0.0148], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-04-01 07:26:53,871 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([5.9033, 5.3068, 2.8920, 4.7663, 1.5378, 4.9841, 5.1490, 5.3066], device='cuda:1'), covar=tensor([0.0445, 0.0996, 0.1987, 0.0510, 0.3711, 0.0713, 0.0568, 0.0670], device='cuda:1'), in_proj_covar=tensor([0.0364, 0.0314, 0.0372, 0.0291, 0.0353, 0.0313, 0.0294, 0.0327], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-01 07:27:05,088 WARNING [train.py:1073] (1/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 2023-04-01 07:27:10,904 INFO [train.py:903] (1/4) Epoch 6, batch 2600, loss[loss=0.2865, simple_loss=0.3449, pruned_loss=0.114, over 19543.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3366, pruned_loss=0.1059, over 3811300.82 frames. ], batch size: 54, lr: 1.39e-02, grad_scale: 8.0 2023-04-01 07:27:12,685 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.869e+02 6.492e+02 8.253e+02 1.085e+03 2.742e+03, threshold=1.651e+03, percent-clipped=10.0 2023-04-01 07:28:14,513 INFO [train.py:903] (1/4) Epoch 6, batch 2650, loss[loss=0.2518, simple_loss=0.3261, pruned_loss=0.0888, over 19535.00 frames. ], tot_loss[loss=0.273, simple_loss=0.3359, pruned_loss=0.1051, over 3813499.48 frames. ], batch size: 54, lr: 1.39e-02, grad_scale: 8.0 2023-04-01 07:28:37,554 WARNING [train.py:1073] (1/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9_rvb from training. Duration: 27.25 2023-04-01 07:28:55,452 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=36823.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 07:29:16,467 INFO [train.py:903] (1/4) Epoch 6, batch 2700, loss[loss=0.3096, simple_loss=0.3586, pruned_loss=0.1303, over 19481.00 frames. ], tot_loss[loss=0.2732, simple_loss=0.3358, pruned_loss=0.1053, over 3822964.59 frames. ], batch size: 49, lr: 1.39e-02, grad_scale: 8.0 2023-04-01 07:29:17,596 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.977e+02 6.342e+02 7.427e+02 9.812e+02 2.890e+03, threshold=1.485e+03, percent-clipped=2.0 2023-04-01 07:30:05,316 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 2023-04-01 07:30:18,801 INFO [train.py:903] (1/4) Epoch 6, batch 2750, loss[loss=0.221, simple_loss=0.2901, pruned_loss=0.07597, over 19828.00 frames. ], tot_loss[loss=0.2734, simple_loss=0.3359, pruned_loss=0.1055, over 3832374.73 frames. ], batch size: 49, lr: 1.39e-02, grad_scale: 8.0 2023-04-01 07:30:49,706 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 2023-04-01 07:30:51,757 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=36914.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 07:31:09,562 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=36929.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 07:31:22,192 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=36939.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 07:31:23,024 INFO [train.py:903] (1/4) Epoch 6, batch 2800, loss[loss=0.2677, simple_loss=0.3292, pruned_loss=0.1031, over 19849.00 frames. ], tot_loss[loss=0.2717, simple_loss=0.3349, pruned_loss=0.1043, over 3838408.07 frames. ], batch size: 52, lr: 1.39e-02, grad_scale: 8.0 2023-04-01 07:31:24,217 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.898e+02 7.168e+02 8.701e+02 1.243e+03 3.330e+03, threshold=1.740e+03, percent-clipped=17.0 2023-04-01 07:31:47,718 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.9663, 4.2742, 4.6341, 4.5657, 1.6629, 4.1765, 3.7006, 4.2503], device='cuda:1'), covar=tensor([0.0884, 0.0631, 0.0468, 0.0384, 0.3955, 0.0378, 0.0537, 0.0875], device='cuda:1'), in_proj_covar=tensor([0.0554, 0.0477, 0.0649, 0.0532, 0.0609, 0.0399, 0.0413, 0.0606], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1') 2023-04-01 07:31:56,922 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=36967.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 07:32:00,392 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=36970.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 07:32:06,250 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=36974.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 07:32:08,492 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=36976.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 07:32:23,712 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0837, 2.0365, 1.6463, 1.5212, 1.5269, 1.6722, 0.4184, 0.8759], device='cuda:1'), covar=tensor([0.0250, 0.0253, 0.0186, 0.0263, 0.0538, 0.0286, 0.0501, 0.0465], device='cuda:1'), in_proj_covar=tensor([0.0297, 0.0298, 0.0296, 0.0310, 0.0385, 0.0309, 0.0286, 0.0306], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 07:32:26,410 INFO [train.py:903] (1/4) Epoch 6, batch 2850, loss[loss=0.2647, simple_loss=0.3433, pruned_loss=0.09304, over 19694.00 frames. ], tot_loss[loss=0.2716, simple_loss=0.3347, pruned_loss=0.1042, over 3834878.31 frames. ], batch size: 59, lr: 1.39e-02, grad_scale: 8.0 2023-04-01 07:33:05,604 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=37022.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 07:33:28,652 INFO [train.py:903] (1/4) Epoch 6, batch 2900, loss[loss=0.2809, simple_loss=0.3446, pruned_loss=0.1087, over 19671.00 frames. ], tot_loss[loss=0.2711, simple_loss=0.3343, pruned_loss=0.1039, over 3833429.69 frames. ], batch size: 60, lr: 1.38e-02, grad_scale: 8.0 2023-04-01 07:33:28,666 WARNING [train.py:1073] (1/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 2023-04-01 07:33:29,872 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.185e+02 6.105e+02 7.947e+02 1.025e+03 2.308e+03, threshold=1.589e+03, percent-clipped=2.0 2023-04-01 07:33:55,107 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=37062.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 07:34:20,473 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37082.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 07:34:24,058 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37085.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 07:34:28,759 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37089.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 07:34:29,562 INFO [train.py:903] (1/4) Epoch 6, batch 2950, loss[loss=0.2431, simple_loss=0.3027, pruned_loss=0.09174, over 19318.00 frames. ], tot_loss[loss=0.2706, simple_loss=0.3338, pruned_loss=0.1037, over 3835936.88 frames. ], batch size: 44, lr: 1.38e-02, grad_scale: 8.0 2023-04-01 07:35:31,156 INFO [train.py:903] (1/4) Epoch 6, batch 3000, loss[loss=0.2458, simple_loss=0.3034, pruned_loss=0.09414, over 19428.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.3332, pruned_loss=0.1033, over 3846311.35 frames. ], batch size: 48, lr: 1.38e-02, grad_scale: 8.0 2023-04-01 07:35:31,157 INFO [train.py:928] (1/4) Computing validation loss 2023-04-01 07:35:43,639 INFO [train.py:937] (1/4) Epoch 6, validation: loss=0.1968, simple_loss=0.2962, pruned_loss=0.04867, over 944034.00 frames. 2023-04-01 07:35:43,640 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18649MB 2023-04-01 07:35:44,844 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.497e+02 6.001e+02 7.289e+02 9.626e+02 1.809e+03, threshold=1.458e+03, percent-clipped=5.0 2023-04-01 07:35:48,602 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_rvb from training. Duration: 29.735 2023-04-01 07:36:18,756 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=37167.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 07:36:30,287 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37177.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 07:36:45,870 INFO [train.py:903] (1/4) Epoch 6, batch 3050, loss[loss=0.2809, simple_loss=0.3507, pruned_loss=0.1056, over 19668.00 frames. ], tot_loss[loss=0.2723, simple_loss=0.335, pruned_loss=0.1047, over 3839336.13 frames. ], batch size: 55, lr: 1.38e-02, grad_scale: 8.0 2023-04-01 07:37:48,502 INFO [train.py:903] (1/4) Epoch 6, batch 3100, loss[loss=0.2696, simple_loss=0.3351, pruned_loss=0.1021, over 19664.00 frames. ], tot_loss[loss=0.2726, simple_loss=0.335, pruned_loss=0.1051, over 3825858.36 frames. ], batch size: 53, lr: 1.38e-02, grad_scale: 8.0 2023-04-01 07:37:49,788 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.764e+02 6.699e+02 8.375e+02 1.038e+03 2.239e+03, threshold=1.675e+03, percent-clipped=7.0 2023-04-01 07:38:28,940 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=37273.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 07:38:41,380 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37282.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 07:38:46,546 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=37286.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 07:38:50,729 INFO [train.py:903] (1/4) Epoch 6, batch 3150, loss[loss=0.2094, simple_loss=0.2796, pruned_loss=0.06966, over 19568.00 frames. ], tot_loss[loss=0.2739, simple_loss=0.3359, pruned_loss=0.106, over 3808173.31 frames. ], batch size: 52, lr: 1.38e-02, grad_scale: 8.0 2023-04-01 07:38:57,015 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2237, 1.3127, 1.8601, 1.3952, 2.4919, 2.0873, 2.6931, 1.1232], device='cuda:1'), covar=tensor([0.2083, 0.3536, 0.1882, 0.1691, 0.1364, 0.1755, 0.1408, 0.3019], device='cuda:1'), in_proj_covar=tensor([0.0443, 0.0507, 0.0493, 0.0407, 0.0559, 0.0445, 0.0627, 0.0451], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 07:39:13,498 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 2023-04-01 07:39:27,512 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=37320.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 07:39:49,088 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=37338.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 07:39:51,047 INFO [train.py:903] (1/4) Epoch 6, batch 3200, loss[loss=0.246, simple_loss=0.3045, pruned_loss=0.09375, over 19422.00 frames. ], tot_loss[loss=0.2751, simple_loss=0.337, pruned_loss=0.1065, over 3820315.72 frames. ], batch size: 48, lr: 1.38e-02, grad_scale: 8.0 2023-04-01 07:39:52,144 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.326e+02 6.253e+02 8.171e+02 9.916e+02 1.975e+03, threshold=1.634e+03, percent-clipped=4.0 2023-04-01 07:39:52,618 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=37341.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 07:39:58,084 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=37345.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 07:39:59,734 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0 2023-04-01 07:40:20,389 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=37363.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 07:40:24,432 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=37366.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 07:40:24,647 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=37366.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 07:40:29,187 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=37370.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 07:40:42,562 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=37382.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 07:40:50,628 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37388.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 07:40:52,610 INFO [train.py:903] (1/4) Epoch 6, batch 3250, loss[loss=0.2204, simple_loss=0.2844, pruned_loss=0.07813, over 19766.00 frames. ], tot_loss[loss=0.2752, simple_loss=0.3372, pruned_loss=0.1066, over 3827407.64 frames. ], batch size: 45, lr: 1.38e-02, grad_scale: 4.0 2023-04-01 07:41:30,771 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2694, 2.1191, 1.5254, 1.3621, 1.8790, 1.0919, 1.0702, 1.7251], device='cuda:1'), covar=tensor([0.0630, 0.0399, 0.0836, 0.0498, 0.0349, 0.0922, 0.0572, 0.0301], device='cuda:1'), in_proj_covar=tensor([0.0276, 0.0276, 0.0309, 0.0241, 0.0221, 0.0310, 0.0280, 0.0242], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 07:41:45,872 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=37433.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 07:41:48,044 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37435.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 07:41:53,523 INFO [train.py:903] (1/4) Epoch 6, batch 3300, loss[loss=0.247, simple_loss=0.3162, pruned_loss=0.08893, over 19572.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.338, pruned_loss=0.1065, over 3834858.63 frames. ], batch size: 52, lr: 1.38e-02, grad_scale: 4.0 2023-04-01 07:41:57,418 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.613e+02 6.108e+02 8.102e+02 1.002e+03 3.053e+03, threshold=1.620e+03, percent-clipped=3.0 2023-04-01 07:42:01,115 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 2023-04-01 07:42:17,188 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=37458.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 07:42:44,820 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37481.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 07:42:56,298 INFO [train.py:903] (1/4) Epoch 6, batch 3350, loss[loss=0.2733, simple_loss=0.3284, pruned_loss=0.1091, over 19733.00 frames. ], tot_loss[loss=0.2738, simple_loss=0.3368, pruned_loss=0.1054, over 3831436.72 frames. ], batch size: 47, lr: 1.38e-02, grad_scale: 4.0 2023-04-01 07:43:56,155 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=37538.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 07:43:57,993 INFO [train.py:903] (1/4) Epoch 6, batch 3400, loss[loss=0.2775, simple_loss=0.3245, pruned_loss=0.1153, over 19765.00 frames. ], tot_loss[loss=0.273, simple_loss=0.336, pruned_loss=0.1051, over 3822868.62 frames. ], batch size: 46, lr: 1.38e-02, grad_scale: 4.0 2023-04-01 07:44:00,242 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.822e+02 6.377e+02 8.364e+02 1.096e+03 2.128e+03, threshold=1.673e+03, percent-clipped=5.0 2023-04-01 07:44:26,237 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=37563.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 07:44:59,269 INFO [train.py:903] (1/4) Epoch 6, batch 3450, loss[loss=0.2851, simple_loss=0.3513, pruned_loss=0.1095, over 19093.00 frames. ], tot_loss[loss=0.2707, simple_loss=0.3345, pruned_loss=0.1035, over 3834725.46 frames. ], batch size: 69, lr: 1.37e-02, grad_scale: 4.0 2023-04-01 07:45:07,183 WARNING [train.py:1073] (1/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9_rvb from training. Duration: 25.2444375 2023-04-01 07:45:24,714 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.49 vs. limit=5.0 2023-04-01 07:45:39,234 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.8204, 4.3516, 2.5146, 3.9421, 0.9507, 3.9908, 4.0796, 4.2430], device='cuda:1'), covar=tensor([0.0516, 0.0963, 0.1996, 0.0617, 0.3899, 0.0811, 0.0720, 0.0788], device='cuda:1'), in_proj_covar=tensor([0.0366, 0.0324, 0.0384, 0.0294, 0.0360, 0.0319, 0.0297, 0.0332], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-01 07:45:49,547 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=37630.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 07:46:02,064 INFO [train.py:903] (1/4) Epoch 6, batch 3500, loss[loss=0.2831, simple_loss=0.3518, pruned_loss=0.1072, over 19549.00 frames. ], tot_loss[loss=0.2713, simple_loss=0.3349, pruned_loss=0.1038, over 3830081.60 frames. ], batch size: 54, lr: 1.37e-02, grad_scale: 4.0 2023-04-01 07:46:04,578 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.712e+02 6.340e+02 8.060e+02 1.060e+03 3.220e+03, threshold=1.612e+03, percent-clipped=3.0 2023-04-01 07:46:07,415 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=37644.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 07:46:23,280 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.7118, 1.3580, 1.2192, 1.8190, 1.5164, 1.8222, 1.8954, 1.5857], device='cuda:1'), covar=tensor([0.0807, 0.1082, 0.1201, 0.0948, 0.0947, 0.0794, 0.0944, 0.0789], device='cuda:1'), in_proj_covar=tensor([0.0226, 0.0244, 0.0237, 0.0274, 0.0262, 0.0230, 0.0226, 0.0219], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:1') 2023-04-01 07:46:38,321 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=37669.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 07:46:54,394 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([5.0661, 1.7493, 1.9040, 2.0134, 4.4271, 0.9478, 2.3173, 4.5704], device='cuda:1'), covar=tensor([0.0228, 0.2374, 0.2234, 0.1433, 0.0494, 0.2548, 0.1277, 0.0266], device='cuda:1'), in_proj_covar=tensor([0.0298, 0.0316, 0.0311, 0.0287, 0.0311, 0.0312, 0.0288, 0.0304], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 07:47:04,849 INFO [train.py:903] (1/4) Epoch 6, batch 3550, loss[loss=0.2918, simple_loss=0.358, pruned_loss=0.1128, over 19366.00 frames. ], tot_loss[loss=0.2724, simple_loss=0.3358, pruned_loss=0.1045, over 3825832.65 frames. ], batch size: 66, lr: 1.37e-02, grad_scale: 4.0 2023-04-01 07:47:07,135 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=37691.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 07:47:12,748 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.8497, 1.2047, 1.4432, 1.8625, 3.3410, 1.1004, 2.0985, 3.3803], device='cuda:1'), covar=tensor([0.0299, 0.2543, 0.2405, 0.1292, 0.0526, 0.2222, 0.1265, 0.0343], device='cuda:1'), in_proj_covar=tensor([0.0299, 0.0318, 0.0315, 0.0288, 0.0312, 0.0313, 0.0288, 0.0305], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 07:47:34,902 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=37716.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 07:47:49,116 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=37726.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 07:48:02,995 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=37737.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 07:48:05,907 INFO [train.py:903] (1/4) Epoch 6, batch 3600, loss[loss=0.3084, simple_loss=0.3539, pruned_loss=0.1315, over 19751.00 frames. ], tot_loss[loss=0.2722, simple_loss=0.3356, pruned_loss=0.1044, over 3832702.06 frames. ], batch size: 51, lr: 1.37e-02, grad_scale: 8.0 2023-04-01 07:48:08,229 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.206e+02 7.040e+02 8.572e+02 1.227e+03 4.209e+03, threshold=1.714e+03, percent-clipped=12.0 2023-04-01 07:48:12,114 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37745.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 07:48:33,238 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=37762.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 07:49:08,248 INFO [train.py:903] (1/4) Epoch 6, batch 3650, loss[loss=0.3571, simple_loss=0.3824, pruned_loss=0.1658, over 13711.00 frames. ], tot_loss[loss=0.2728, simple_loss=0.3357, pruned_loss=0.1049, over 3820609.25 frames. ], batch size: 136, lr: 1.37e-02, grad_scale: 8.0 2023-04-01 07:50:10,246 INFO [train.py:903] (1/4) Epoch 6, batch 3700, loss[loss=0.2356, simple_loss=0.2991, pruned_loss=0.0861, over 19482.00 frames. ], tot_loss[loss=0.2723, simple_loss=0.3352, pruned_loss=0.1047, over 3820466.18 frames. ], batch size: 49, lr: 1.37e-02, grad_scale: 8.0 2023-04-01 07:50:11,834 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37841.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 07:50:12,525 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.880e+02 5.969e+02 7.272e+02 1.002e+03 1.787e+03, threshold=1.454e+03, percent-clipped=1.0 2023-04-01 07:51:00,331 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.6906, 1.7856, 1.9526, 2.6569, 2.4730, 2.1825, 2.1931, 2.4978], device='cuda:1'), covar=tensor([0.0693, 0.1862, 0.1281, 0.0886, 0.1226, 0.0432, 0.0891, 0.0529], device='cuda:1'), in_proj_covar=tensor([0.0249, 0.0360, 0.0283, 0.0237, 0.0304, 0.0240, 0.0265, 0.0224], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 07:51:13,475 INFO [train.py:903] (1/4) Epoch 6, batch 3750, loss[loss=0.2999, simple_loss=0.3573, pruned_loss=0.1212, over 18129.00 frames. ], tot_loss[loss=0.272, simple_loss=0.3351, pruned_loss=0.1044, over 3825163.84 frames. ], batch size: 83, lr: 1.37e-02, grad_scale: 8.0 2023-04-01 07:52:16,179 INFO [train.py:903] (1/4) Epoch 6, batch 3800, loss[loss=0.2305, simple_loss=0.292, pruned_loss=0.0845, over 15049.00 frames. ], tot_loss[loss=0.2705, simple_loss=0.3336, pruned_loss=0.1037, over 3821933.37 frames. ], batch size: 33, lr: 1.37e-02, grad_scale: 8.0 2023-04-01 07:52:18,420 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.239e+02 5.762e+02 7.572e+02 9.178e+02 2.007e+03, threshold=1.514e+03, percent-clipped=4.0 2023-04-01 07:52:47,800 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9_rvb from training. Duration: 29.1166875 2023-04-01 07:53:17,772 INFO [train.py:903] (1/4) Epoch 6, batch 3850, loss[loss=0.2597, simple_loss=0.3215, pruned_loss=0.09898, over 19480.00 frames. ], tot_loss[loss=0.2704, simple_loss=0.3336, pruned_loss=0.1036, over 3808707.88 frames. ], batch size: 49, lr: 1.37e-02, grad_scale: 8.0 2023-04-01 07:53:25,801 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=37997.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 07:53:32,695 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=38001.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 07:54:00,114 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.52 vs. limit=2.0 2023-04-01 07:54:03,030 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=38026.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 07:54:20,061 INFO [train.py:903] (1/4) Epoch 6, batch 3900, loss[loss=0.2075, simple_loss=0.278, pruned_loss=0.06848, over 19769.00 frames. ], tot_loss[loss=0.2728, simple_loss=0.3355, pruned_loss=0.105, over 3806573.91 frames. ], batch size: 46, lr: 1.37e-02, grad_scale: 8.0 2023-04-01 07:54:22,364 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.032e+02 6.773e+02 7.927e+02 9.711e+02 2.220e+03, threshold=1.585e+03, percent-clipped=5.0 2023-04-01 07:54:32,143 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.61 vs. limit=2.0 2023-04-01 07:55:24,002 INFO [train.py:903] (1/4) Epoch 6, batch 3950, loss[loss=0.2646, simple_loss=0.3176, pruned_loss=0.1058, over 19421.00 frames. ], tot_loss[loss=0.2722, simple_loss=0.3352, pruned_loss=0.1046, over 3816076.80 frames. ], batch size: 48, lr: 1.37e-02, grad_scale: 8.0 2023-04-01 07:55:27,714 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 2023-04-01 07:55:33,048 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=38097.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 07:56:03,031 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=38122.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 07:56:25,971 INFO [train.py:903] (1/4) Epoch 6, batch 4000, loss[loss=0.2821, simple_loss=0.3472, pruned_loss=0.1085, over 19706.00 frames. ], tot_loss[loss=0.271, simple_loss=0.3345, pruned_loss=0.1038, over 3821640.68 frames. ], batch size: 59, lr: 1.37e-02, grad_scale: 8.0 2023-04-01 07:56:28,273 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.559e+02 6.005e+02 7.523e+02 9.814e+02 1.567e+03, threshold=1.505e+03, percent-clipped=0.0 2023-04-01 07:57:10,232 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1_rvb from training. Duration: 0.7545625 2023-04-01 07:57:25,997 INFO [train.py:903] (1/4) Epoch 6, batch 4050, loss[loss=0.3004, simple_loss=0.3606, pruned_loss=0.1201, over 19612.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.3338, pruned_loss=0.1034, over 3814766.00 frames. ], batch size: 57, lr: 1.36e-02, grad_scale: 8.0 2023-04-01 07:58:27,528 INFO [train.py:903] (1/4) Epoch 6, batch 4100, loss[loss=0.3566, simple_loss=0.3928, pruned_loss=0.1602, over 17259.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.3334, pruned_loss=0.1031, over 3807409.62 frames. ], batch size: 101, lr: 1.36e-02, grad_scale: 8.0 2023-04-01 07:58:30,599 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.078e+02 6.486e+02 7.817e+02 9.790e+02 2.532e+03, threshold=1.563e+03, percent-clipped=8.0 2023-04-01 07:59:04,831 WARNING [train.py:1073] (1/4) Exclude cut with ID 497-129325-0061-9566_sp1.1_rvb from training. Duration: 0.97725 2023-04-01 07:59:30,597 INFO [train.py:903] (1/4) Epoch 6, batch 4150, loss[loss=0.2546, simple_loss=0.335, pruned_loss=0.08713, over 18741.00 frames. ], tot_loss[loss=0.2699, simple_loss=0.3337, pruned_loss=0.1031, over 3814161.30 frames. ], batch size: 74, lr: 1.36e-02, grad_scale: 8.0 2023-04-01 08:00:34,879 INFO [train.py:903] (1/4) Epoch 6, batch 4200, loss[loss=0.2658, simple_loss=0.3351, pruned_loss=0.09824, over 18707.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.3343, pruned_loss=0.1038, over 3813185.91 frames. ], batch size: 74, lr: 1.36e-02, grad_scale: 8.0 2023-04-01 08:00:36,323 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=38341.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 08:00:37,310 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.522e+02 6.104e+02 7.462e+02 9.650e+02 2.123e+03, threshold=1.492e+03, percent-clipped=5.0 2023-04-01 08:00:37,727 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1132, 1.1214, 1.3233, 1.1779, 1.7638, 1.6365, 1.8071, 0.5172], device='cuda:1'), covar=tensor([0.1644, 0.2744, 0.1473, 0.1382, 0.0960, 0.1497, 0.0927, 0.2640], device='cuda:1'), in_proj_covar=tensor([0.0448, 0.0515, 0.0497, 0.0409, 0.0561, 0.0452, 0.0633, 0.0450], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 08:00:39,345 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 2023-04-01 08:01:21,315 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=38378.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 08:01:34,727 INFO [train.py:903] (1/4) Epoch 6, batch 4250, loss[loss=0.2666, simple_loss=0.3186, pruned_loss=0.1073, over 19738.00 frames. ], tot_loss[loss=0.2711, simple_loss=0.3345, pruned_loss=0.1039, over 3822563.71 frames. ], batch size: 51, lr: 1.36e-02, grad_scale: 8.0 2023-04-01 08:01:48,234 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9_rvb from training. Duration: 29.816625 2023-04-01 08:02:01,562 WARNING [train.py:1073] (1/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 2023-04-01 08:02:34,846 INFO [train.py:903] (1/4) Epoch 6, batch 4300, loss[loss=0.2463, simple_loss=0.3147, pruned_loss=0.089, over 19676.00 frames. ], tot_loss[loss=0.2728, simple_loss=0.3359, pruned_loss=0.1048, over 3799480.96 frames. ], batch size: 55, lr: 1.36e-02, grad_scale: 8.0 2023-04-01 08:02:37,124 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.280e+02 6.582e+02 8.580e+02 1.078e+03 2.349e+03, threshold=1.716e+03, percent-clipped=8.0 2023-04-01 08:02:56,015 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=38456.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 08:02:57,124 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.1887, 1.5920, 1.5709, 2.0009, 1.7653, 1.6785, 1.5560, 2.0592], device='cuda:1'), covar=tensor([0.0906, 0.1891, 0.1647, 0.1093, 0.1592, 0.1015, 0.1431, 0.0827], device='cuda:1'), in_proj_covar=tensor([0.0249, 0.0355, 0.0281, 0.0236, 0.0299, 0.0243, 0.0266, 0.0226], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 08:03:06,290 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4302, 2.4140, 1.6822, 1.4595, 2.2085, 1.2148, 1.3025, 1.7891], device='cuda:1'), covar=tensor([0.0746, 0.0453, 0.0869, 0.0624, 0.0417, 0.1044, 0.0666, 0.0406], device='cuda:1'), in_proj_covar=tensor([0.0284, 0.0285, 0.0320, 0.0246, 0.0229, 0.0319, 0.0289, 0.0259], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 08:03:27,653 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 2023-04-01 08:03:35,913 INFO [train.py:903] (1/4) Epoch 6, batch 4350, loss[loss=0.2301, simple_loss=0.2925, pruned_loss=0.08385, over 19748.00 frames. ], tot_loss[loss=0.2728, simple_loss=0.3358, pruned_loss=0.1049, over 3821844.05 frames. ], batch size: 45, lr: 1.36e-02, grad_scale: 8.0 2023-04-01 08:04:40,410 INFO [train.py:903] (1/4) Epoch 6, batch 4400, loss[loss=0.3235, simple_loss=0.3814, pruned_loss=0.1328, over 19084.00 frames. ], tot_loss[loss=0.273, simple_loss=0.3361, pruned_loss=0.105, over 3815072.17 frames. ], batch size: 69, lr: 1.36e-02, grad_scale: 8.0 2023-04-01 08:04:42,534 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.411e+02 6.588e+02 8.122e+02 1.160e+03 2.348e+03, threshold=1.624e+03, percent-clipped=4.0 2023-04-01 08:05:05,902 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_rvb from training. Duration: 25.285 2023-04-01 08:05:13,671 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 2023-04-01 08:05:40,416 INFO [train.py:903] (1/4) Epoch 6, batch 4450, loss[loss=0.2848, simple_loss=0.3469, pruned_loss=0.1113, over 19699.00 frames. ], tot_loss[loss=0.2728, simple_loss=0.3358, pruned_loss=0.1049, over 3806492.69 frames. ], batch size: 59, lr: 1.36e-02, grad_scale: 8.0 2023-04-01 08:06:07,953 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 2023-04-01 08:06:42,157 INFO [train.py:903] (1/4) Epoch 6, batch 4500, loss[loss=0.2503, simple_loss=0.3214, pruned_loss=0.08961, over 19486.00 frames. ], tot_loss[loss=0.271, simple_loss=0.3342, pruned_loss=0.1039, over 3819775.01 frames. ], batch size: 49, lr: 1.36e-02, grad_scale: 8.0 2023-04-01 08:06:44,523 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.035e+02 6.139e+02 7.598e+02 9.442e+02 2.713e+03, threshold=1.520e+03, percent-clipped=3.0 2023-04-01 08:06:55,375 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.49 vs. limit=2.0 2023-04-01 08:07:42,586 INFO [train.py:903] (1/4) Epoch 6, batch 4550, loss[loss=0.2205, simple_loss=0.2947, pruned_loss=0.07313, over 19473.00 frames. ], tot_loss[loss=0.2705, simple_loss=0.3341, pruned_loss=0.1035, over 3817545.46 frames. ], batch size: 49, lr: 1.36e-02, grad_scale: 8.0 2023-04-01 08:07:52,243 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.5141, 3.9389, 4.1920, 4.1687, 1.4006, 3.8441, 3.4970, 3.7845], device='cuda:1'), covar=tensor([0.1007, 0.0604, 0.0496, 0.0433, 0.4092, 0.0435, 0.0473, 0.0983], device='cuda:1'), in_proj_covar=tensor([0.0561, 0.0475, 0.0663, 0.0534, 0.0614, 0.0403, 0.0413, 0.0606], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1') 2023-04-01 08:07:53,984 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9_rvb from training. Duration: 28.72225 2023-04-01 08:08:10,858 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=38712.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 08:08:17,330 WARNING [train.py:1073] (1/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 2023-04-01 08:08:22,021 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=38722.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 08:08:23,445 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.6969, 1.8498, 2.3602, 2.9919, 2.4537, 2.7174, 2.1985, 2.9992], device='cuda:1'), covar=tensor([0.0653, 0.1593, 0.1135, 0.0713, 0.1167, 0.0354, 0.0959, 0.0427], device='cuda:1'), in_proj_covar=tensor([0.0249, 0.0354, 0.0282, 0.0233, 0.0298, 0.0239, 0.0267, 0.0222], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 08:08:41,577 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=38737.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 08:08:45,375 INFO [train.py:903] (1/4) Epoch 6, batch 4600, loss[loss=0.2501, simple_loss=0.3153, pruned_loss=0.09239, over 19668.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.3347, pruned_loss=0.1035, over 3823926.22 frames. ], batch size: 53, lr: 1.35e-02, grad_scale: 8.0 2023-04-01 08:08:47,714 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.208e+02 6.588e+02 8.068e+02 1.040e+03 1.807e+03, threshold=1.614e+03, percent-clipped=3.0 2023-04-01 08:09:02,393 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3311, 1.4709, 2.1122, 1.5967, 3.2669, 2.6349, 3.5600, 1.5970], device='cuda:1'), covar=tensor([0.1851, 0.3301, 0.1777, 0.1420, 0.1274, 0.1471, 0.1483, 0.2840], device='cuda:1'), in_proj_covar=tensor([0.0452, 0.0522, 0.0505, 0.0413, 0.0565, 0.0456, 0.0639, 0.0455], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 08:09:45,874 INFO [train.py:903] (1/4) Epoch 6, batch 4650, loss[loss=0.2531, simple_loss=0.3062, pruned_loss=0.1, over 19757.00 frames. ], tot_loss[loss=0.2691, simple_loss=0.333, pruned_loss=0.1026, over 3821651.62 frames. ], batch size: 47, lr: 1.35e-02, grad_scale: 8.0 2023-04-01 08:10:01,545 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_rvb from training. Duration: 0.92 2023-04-01 08:10:11,697 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_rvb from training. Duration: 0.83 2023-04-01 08:10:43,218 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=38837.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 08:10:46,233 INFO [train.py:903] (1/4) Epoch 6, batch 4700, loss[loss=0.3668, simple_loss=0.3958, pruned_loss=0.1689, over 13568.00 frames. ], tot_loss[loss=0.2697, simple_loss=0.3332, pruned_loss=0.103, over 3814765.49 frames. ], batch size: 136, lr: 1.35e-02, grad_scale: 8.0 2023-04-01 08:10:47,736 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=38841.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 08:10:48,619 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.118e+02 6.701e+02 8.528e+02 1.085e+03 2.106e+03, threshold=1.706e+03, percent-clipped=3.0 2023-04-01 08:11:07,238 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9_rvb from training. Duration: 0.92225 2023-04-01 08:11:46,443 INFO [train.py:903] (1/4) Epoch 6, batch 4750, loss[loss=0.2831, simple_loss=0.3521, pruned_loss=0.107, over 19739.00 frames. ], tot_loss[loss=0.269, simple_loss=0.3329, pruned_loss=0.1026, over 3814421.48 frames. ], batch size: 63, lr: 1.35e-02, grad_scale: 8.0 2023-04-01 08:12:47,942 INFO [train.py:903] (1/4) Epoch 6, batch 4800, loss[loss=0.2476, simple_loss=0.3048, pruned_loss=0.09517, over 19734.00 frames. ], tot_loss[loss=0.2717, simple_loss=0.3351, pruned_loss=0.1041, over 3821600.85 frames. ], batch size: 46, lr: 1.35e-02, grad_scale: 8.0 2023-04-01 08:12:52,284 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.945e+02 6.641e+02 7.688e+02 1.063e+03 2.770e+03, threshold=1.538e+03, percent-clipped=4.0 2023-04-01 08:13:49,881 INFO [train.py:903] (1/4) Epoch 6, batch 4850, loss[loss=0.2584, simple_loss=0.3332, pruned_loss=0.09179, over 19611.00 frames. ], tot_loss[loss=0.2702, simple_loss=0.3339, pruned_loss=0.1033, over 3817736.60 frames. ], batch size: 57, lr: 1.35e-02, grad_scale: 8.0 2023-04-01 08:14:14,774 WARNING [train.py:1073] (1/4) Exclude cut with ID 774-127930-0014-48411_sp1.1_rvb from training. Duration: 0.95 2023-04-01 08:14:33,511 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.8954, 4.2948, 4.5940, 4.5456, 1.4509, 4.2320, 3.8152, 4.1990], device='cuda:1'), covar=tensor([0.1118, 0.0612, 0.0493, 0.0422, 0.4509, 0.0400, 0.0523, 0.0915], device='cuda:1'), in_proj_covar=tensor([0.0566, 0.0480, 0.0664, 0.0543, 0.0618, 0.0403, 0.0414, 0.0607], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1') 2023-04-01 08:14:36,521 WARNING [train.py:1073] (1/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1_rvb from training. Duration: 0.9409375 2023-04-01 08:14:42,285 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 2023-04-01 08:14:42,313 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 2023-04-01 08:14:51,528 INFO [train.py:903] (1/4) Epoch 6, batch 4900, loss[loss=0.3889, simple_loss=0.4233, pruned_loss=0.1772, over 13086.00 frames. ], tot_loss[loss=0.269, simple_loss=0.333, pruned_loss=0.1025, over 3829809.96 frames. ], batch size: 136, lr: 1.35e-02, grad_scale: 8.0 2023-04-01 08:14:51,568 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1_rvb from training. Duration: 25.3818125 2023-04-01 08:14:55,127 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.255e+02 6.820e+02 8.455e+02 1.082e+03 3.554e+03, threshold=1.691e+03, percent-clipped=3.0 2023-04-01 08:15:04,828 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([5.6765, 5.1346, 2.8390, 4.4423, 1.6095, 4.5370, 4.8032, 5.0591], device='cuda:1'), covar=tensor([0.0333, 0.0682, 0.1740, 0.0551, 0.3389, 0.0689, 0.0605, 0.0663], device='cuda:1'), in_proj_covar=tensor([0.0365, 0.0316, 0.0378, 0.0289, 0.0354, 0.0315, 0.0296, 0.0329], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-01 08:15:12,443 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 2023-04-01 08:15:15,321 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.1803, 3.8275, 2.2639, 2.3202, 3.3469, 1.9565, 1.4120, 1.9869], device='cuda:1'), covar=tensor([0.0910, 0.0332, 0.0779, 0.0544, 0.0419, 0.0879, 0.0824, 0.0633], device='cuda:1'), in_proj_covar=tensor([0.0284, 0.0286, 0.0315, 0.0239, 0.0228, 0.0313, 0.0287, 0.0255], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 08:15:52,286 INFO [train.py:903] (1/4) Epoch 6, batch 4950, loss[loss=0.2621, simple_loss=0.3281, pruned_loss=0.098, over 19481.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.3324, pruned_loss=0.1024, over 3821574.59 frames. ], batch size: 49, lr: 1.35e-02, grad_scale: 8.0 2023-04-01 08:15:56,133 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=39093.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 08:16:11,559 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 2023-04-01 08:16:27,918 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=39118.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 08:16:34,736 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 2023-04-01 08:16:54,838 INFO [train.py:903] (1/4) Epoch 6, batch 5000, loss[loss=0.2988, simple_loss=0.352, pruned_loss=0.1228, over 18087.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.3323, pruned_loss=0.1027, over 3813948.47 frames. ], batch size: 83, lr: 1.35e-02, grad_scale: 8.0 2023-04-01 08:16:59,187 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.473e+02 6.699e+02 7.802e+02 1.019e+03 2.317e+03, threshold=1.560e+03, percent-clipped=4.0 2023-04-01 08:17:05,608 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_rvb from training. Duration: 27.14 2023-04-01 08:17:10,518 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0313, 1.9956, 1.9662, 3.0735, 2.0403, 2.7366, 2.5760, 1.8847], device='cuda:1'), covar=tensor([0.2315, 0.1834, 0.0987, 0.1042, 0.2166, 0.0797, 0.1860, 0.1739], device='cuda:1'), in_proj_covar=tensor([0.0677, 0.0676, 0.0583, 0.0828, 0.0705, 0.0590, 0.0713, 0.0620], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 08:17:14,762 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 2023-04-01 08:17:31,083 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.4985, 1.8699, 2.0973, 2.6385, 2.5222, 2.3304, 2.1382, 2.7192], device='cuda:1'), covar=tensor([0.0719, 0.1444, 0.1101, 0.0797, 0.1009, 0.0396, 0.0890, 0.0484], device='cuda:1'), in_proj_covar=tensor([0.0250, 0.0359, 0.0285, 0.0236, 0.0303, 0.0245, 0.0272, 0.0230], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 08:17:51,514 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=39185.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 08:17:57,111 INFO [train.py:903] (1/4) Epoch 6, batch 5050, loss[loss=0.3249, simple_loss=0.3676, pruned_loss=0.1412, over 13476.00 frames. ], tot_loss[loss=0.2685, simple_loss=0.3321, pruned_loss=0.1024, over 3820727.62 frames. ], batch size: 136, lr: 1.35e-02, grad_scale: 8.0 2023-04-01 08:18:12,185 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=39203.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 08:18:29,777 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9_rvb from training. Duration: 26.62775 2023-04-01 08:18:46,960 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 2023-04-01 08:18:57,184 INFO [train.py:903] (1/4) Epoch 6, batch 5100, loss[loss=0.2637, simple_loss=0.325, pruned_loss=0.1012, over 19790.00 frames. ], tot_loss[loss=0.2696, simple_loss=0.3332, pruned_loss=0.103, over 3824222.73 frames. ], batch size: 48, lr: 1.35e-02, grad_scale: 8.0 2023-04-01 08:19:00,438 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.152e+02 6.811e+02 8.395e+02 1.062e+03 1.934e+03, threshold=1.679e+03, percent-clipped=3.0 2023-04-01 08:19:05,477 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 2023-04-01 08:19:08,940 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_rvb from training. Duration: 25.035 2023-04-01 08:19:13,350 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_rvb from training. Duration: 27.92 2023-04-01 08:19:31,957 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5723, 1.6300, 1.7503, 2.1521, 1.4242, 1.7403, 2.0505, 1.6305], device='cuda:1'), covar=tensor([0.2411, 0.1920, 0.1058, 0.1037, 0.2049, 0.0982, 0.2130, 0.1882], device='cuda:1'), in_proj_covar=tensor([0.0683, 0.0681, 0.0585, 0.0829, 0.0711, 0.0591, 0.0722, 0.0629], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 08:19:57,722 INFO [train.py:903] (1/4) Epoch 6, batch 5150, loss[loss=0.2281, simple_loss=0.3075, pruned_loss=0.07436, over 19782.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.3338, pruned_loss=0.1034, over 3831857.65 frames. ], batch size: 54, lr: 1.35e-02, grad_scale: 8.0 2023-04-01 08:20:08,410 WARNING [train.py:1073] (1/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 2023-04-01 08:20:11,736 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=39300.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 08:20:13,880 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=39302.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 08:20:43,360 WARNING [train.py:1073] (1/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1_rvb from training. Duration: 0.9681875 2023-04-01 08:21:00,827 INFO [train.py:903] (1/4) Epoch 6, batch 5200, loss[loss=0.2146, simple_loss=0.2762, pruned_loss=0.07653, over 19766.00 frames. ], tot_loss[loss=0.2719, simple_loss=0.335, pruned_loss=0.1045, over 3819290.88 frames. ], batch size: 45, lr: 1.34e-02, grad_scale: 8.0 2023-04-01 08:21:04,235 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.495e+02 6.191e+02 7.598e+02 1.014e+03 2.218e+03, threshold=1.520e+03, percent-clipped=2.0 2023-04-01 08:21:14,273 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 2023-04-01 08:21:36,148 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.95 vs. limit=5.0 2023-04-01 08:21:57,369 WARNING [train.py:1073] (1/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 2023-04-01 08:22:03,016 INFO [train.py:903] (1/4) Epoch 6, batch 5250, loss[loss=0.2931, simple_loss=0.3522, pruned_loss=0.117, over 19663.00 frames. ], tot_loss[loss=0.2725, simple_loss=0.3355, pruned_loss=0.1048, over 3823999.46 frames. ], batch size: 53, lr: 1.34e-02, grad_scale: 8.0 2023-04-01 08:22:29,345 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 2023-04-01 08:23:05,650 INFO [train.py:903] (1/4) Epoch 6, batch 5300, loss[loss=0.2519, simple_loss=0.3264, pruned_loss=0.08872, over 19520.00 frames. ], tot_loss[loss=0.2691, simple_loss=0.3329, pruned_loss=0.1026, over 3830971.67 frames. ], batch size: 56, lr: 1.34e-02, grad_scale: 8.0 2023-04-01 08:23:10,375 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.313e+02 6.384e+02 7.596e+02 9.799e+02 2.153e+03, threshold=1.519e+03, percent-clipped=7.0 2023-04-01 08:23:22,423 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 2023-04-01 08:23:56,520 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=39481.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 08:24:06,724 INFO [train.py:903] (1/4) Epoch 6, batch 5350, loss[loss=0.2663, simple_loss=0.3274, pruned_loss=0.1026, over 19761.00 frames. ], tot_loss[loss=0.27, simple_loss=0.3336, pruned_loss=0.1032, over 3825363.22 frames. ], batch size: 51, lr: 1.34e-02, grad_scale: 4.0 2023-04-01 08:24:13,821 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.8152, 4.3612, 2.6313, 3.9153, 1.0413, 4.0242, 4.1058, 4.2847], device='cuda:1'), covar=tensor([0.0521, 0.0920, 0.1884, 0.0603, 0.3967, 0.0797, 0.0632, 0.0697], device='cuda:1'), in_proj_covar=tensor([0.0371, 0.0325, 0.0384, 0.0291, 0.0360, 0.0317, 0.0300, 0.0332], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-01 08:24:41,834 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9_rvb from training. Duration: 26.438875 2023-04-01 08:25:07,430 INFO [train.py:903] (1/4) Epoch 6, batch 5400, loss[loss=0.3172, simple_loss=0.3752, pruned_loss=0.1296, over 19142.00 frames. ], tot_loss[loss=0.2706, simple_loss=0.3342, pruned_loss=0.1035, over 3807934.90 frames. ], batch size: 69, lr: 1.34e-02, grad_scale: 4.0 2023-04-01 08:25:15,197 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.754e+02 6.102e+02 7.285e+02 1.015e+03 2.320e+03, threshold=1.457e+03, percent-clipped=6.0 2023-04-01 08:25:19,789 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=39547.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 08:25:30,311 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=39556.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 08:25:39,825 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 2023-04-01 08:25:52,467 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 2023-04-01 08:26:00,359 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=39581.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 08:26:13,205 INFO [train.py:903] (1/4) Epoch 6, batch 5450, loss[loss=0.3129, simple_loss=0.3594, pruned_loss=0.1332, over 19533.00 frames. ], tot_loss[loss=0.2691, simple_loss=0.333, pruned_loss=0.1026, over 3813329.94 frames. ], batch size: 54, lr: 1.34e-02, grad_scale: 4.0 2023-04-01 08:27:13,643 INFO [train.py:903] (1/4) Epoch 6, batch 5500, loss[loss=0.3252, simple_loss=0.3778, pruned_loss=0.1363, over 19647.00 frames. ], tot_loss[loss=0.2726, simple_loss=0.3357, pruned_loss=0.1047, over 3819240.65 frames. ], batch size: 58, lr: 1.34e-02, grad_scale: 4.0 2023-04-01 08:27:18,180 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.238e+02 6.796e+02 8.775e+02 1.086e+03 2.226e+03, threshold=1.755e+03, percent-clipped=13.0 2023-04-01 08:27:20,742 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=39646.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 08:27:33,868 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=39657.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 08:27:36,847 WARNING [train.py:1073] (1/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9_rvb from training. Duration: 27.47775 2023-04-01 08:27:40,448 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=39662.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 08:28:14,676 INFO [train.py:903] (1/4) Epoch 6, batch 5550, loss[loss=0.2868, simple_loss=0.3422, pruned_loss=0.1157, over 19629.00 frames. ], tot_loss[loss=0.2719, simple_loss=0.3353, pruned_loss=0.1043, over 3810202.68 frames. ], batch size: 57, lr: 1.34e-02, grad_scale: 4.0 2023-04-01 08:28:21,811 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 2023-04-01 08:28:43,665 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.7216, 3.1270, 3.2049, 3.2087, 1.2125, 2.9954, 2.6494, 2.9315], device='cuda:1'), covar=tensor([0.1246, 0.0710, 0.0731, 0.0674, 0.4085, 0.0583, 0.0721, 0.1239], device='cuda:1'), in_proj_covar=tensor([0.0562, 0.0481, 0.0656, 0.0540, 0.0613, 0.0406, 0.0414, 0.0606], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1') 2023-04-01 08:29:11,944 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 2023-04-01 08:29:15,360 INFO [train.py:903] (1/4) Epoch 6, batch 5600, loss[loss=0.295, simple_loss=0.3586, pruned_loss=0.1157, over 19612.00 frames. ], tot_loss[loss=0.2717, simple_loss=0.335, pruned_loss=0.1042, over 3824406.63 frames. ], batch size: 57, lr: 1.34e-02, grad_scale: 8.0 2023-04-01 08:29:20,710 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.863e+02 6.221e+02 7.656e+02 9.358e+02 1.388e+03, threshold=1.531e+03, percent-clipped=0.0 2023-04-01 08:29:43,084 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=39761.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 08:30:11,515 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=39785.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 08:30:19,676 INFO [train.py:903] (1/4) Epoch 6, batch 5650, loss[loss=0.2215, simple_loss=0.285, pruned_loss=0.07897, over 19341.00 frames. ], tot_loss[loss=0.2717, simple_loss=0.3349, pruned_loss=0.1042, over 3824309.96 frames. ], batch size: 47, lr: 1.34e-02, grad_scale: 8.0 2023-04-01 08:30:47,697 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1753, 1.2127, 1.6823, 1.3601, 2.4716, 2.0804, 2.6244, 0.9156], device='cuda:1'), covar=tensor([0.1922, 0.3340, 0.1821, 0.1601, 0.1180, 0.1610, 0.1234, 0.2978], device='cuda:1'), in_proj_covar=tensor([0.0444, 0.0519, 0.0506, 0.0409, 0.0565, 0.0449, 0.0630, 0.0454], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 08:31:01,127 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=39825.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 08:31:08,824 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9_rvb from training. Duration: 28.0944375 2023-04-01 08:31:13,797 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=39833.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 08:31:21,777 INFO [train.py:903] (1/4) Epoch 6, batch 5700, loss[loss=0.2746, simple_loss=0.3358, pruned_loss=0.1068, over 19578.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.335, pruned_loss=0.1043, over 3809348.98 frames. ], batch size: 61, lr: 1.34e-02, grad_scale: 8.0 2023-04-01 08:31:22,132 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.0852, 1.3162, 1.3641, 1.3648, 2.6563, 0.9089, 1.8850, 2.6894], device='cuda:1'), covar=tensor([0.0427, 0.2285, 0.2354, 0.1496, 0.0677, 0.2259, 0.1066, 0.0444], device='cuda:1'), in_proj_covar=tensor([0.0304, 0.0317, 0.0319, 0.0300, 0.0320, 0.0319, 0.0295, 0.0312], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 08:31:26,538 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.066e+02 6.807e+02 8.639e+02 1.032e+03 2.369e+03, threshold=1.728e+03, percent-clipped=2.0 2023-04-01 08:31:57,155 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9530, 1.4011, 1.5683, 1.9951, 2.1039, 1.6326, 1.6137, 1.8070], device='cuda:1'), covar=tensor([0.0854, 0.1685, 0.1329, 0.0915, 0.0991, 0.0538, 0.1069, 0.0682], device='cuda:1'), in_proj_covar=tensor([0.0246, 0.0354, 0.0282, 0.0237, 0.0301, 0.0243, 0.0273, 0.0226], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 08:32:12,192 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.59 vs. limit=5.0 2023-04-01 08:32:17,452 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5396, 2.0757, 1.6028, 1.5799, 1.9605, 1.3000, 1.4178, 1.6050], device='cuda:1'), covar=tensor([0.0553, 0.0412, 0.0599, 0.0445, 0.0339, 0.0722, 0.0491, 0.0321], device='cuda:1'), in_proj_covar=tensor([0.0271, 0.0275, 0.0311, 0.0236, 0.0222, 0.0307, 0.0283, 0.0251], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 08:32:21,376 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.79 vs. limit=5.0 2023-04-01 08:32:21,660 INFO [train.py:903] (1/4) Epoch 6, batch 5750, loss[loss=0.2838, simple_loss=0.3532, pruned_loss=0.1071, over 18788.00 frames. ], tot_loss[loss=0.2712, simple_loss=0.3346, pruned_loss=0.104, over 3808558.06 frames. ], batch size: 74, lr: 1.34e-02, grad_scale: 8.0 2023-04-01 08:32:23,981 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9_rvb from training. Duration: 33.038875 2023-04-01 08:32:30,895 WARNING [train.py:1073] (1/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 2023-04-01 08:32:35,613 WARNING [train.py:1073] (1/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 2023-04-01 08:32:52,593 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.8450, 1.7659, 2.1540, 2.8152, 2.4328, 2.3641, 2.3703, 2.7398], device='cuda:1'), covar=tensor([0.0715, 0.1779, 0.1173, 0.0978, 0.1255, 0.0412, 0.0920, 0.0487], device='cuda:1'), in_proj_covar=tensor([0.0247, 0.0355, 0.0283, 0.0238, 0.0303, 0.0243, 0.0273, 0.0229], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 08:32:58,446 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=39918.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 08:33:23,258 INFO [train.py:903] (1/4) Epoch 6, batch 5800, loss[loss=0.2428, simple_loss=0.3145, pruned_loss=0.08557, over 19770.00 frames. ], tot_loss[loss=0.271, simple_loss=0.3342, pruned_loss=0.1039, over 3800671.71 frames. ], batch size: 54, lr: 1.33e-02, grad_scale: 8.0 2023-04-01 08:33:23,690 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=39940.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 08:33:28,189 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=39943.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 08:33:29,042 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.561e+02 6.519e+02 7.840e+02 9.394e+02 2.455e+03, threshold=1.568e+03, percent-clipped=4.0 2023-04-01 08:34:01,397 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([0.9902, 1.0839, 1.5406, 0.5905, 2.1856, 2.4584, 2.0995, 2.5856], device='cuda:1'), covar=tensor([0.1361, 0.3122, 0.2782, 0.2096, 0.0402, 0.0231, 0.0325, 0.0199], device='cuda:1'), in_proj_covar=tensor([0.0249, 0.0284, 0.0314, 0.0248, 0.0201, 0.0130, 0.0200, 0.0153], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-04-01 08:34:26,992 INFO [train.py:903] (1/4) Epoch 6, batch 5850, loss[loss=0.2583, simple_loss=0.332, pruned_loss=0.09232, over 19792.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.3328, pruned_loss=0.103, over 3809056.18 frames. ], batch size: 56, lr: 1.33e-02, grad_scale: 8.0 2023-04-01 08:34:28,639 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.2167, 2.1467, 1.7614, 1.7458, 1.5445, 1.7303, 0.3285, 1.0066], device='cuda:1'), covar=tensor([0.0284, 0.0297, 0.0224, 0.0309, 0.0665, 0.0376, 0.0647, 0.0525], device='cuda:1'), in_proj_covar=tensor([0.0296, 0.0298, 0.0299, 0.0314, 0.0386, 0.0312, 0.0289, 0.0303], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 08:34:40,714 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=40001.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 08:34:59,479 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=40017.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 08:35:28,999 INFO [train.py:903] (1/4) Epoch 6, batch 5900, loss[loss=0.2674, simple_loss=0.3205, pruned_loss=0.1071, over 19708.00 frames. ], tot_loss[loss=0.2702, simple_loss=0.3333, pruned_loss=0.1035, over 3809605.16 frames. ], batch size: 45, lr: 1.33e-02, grad_scale: 8.0 2023-04-01 08:35:31,444 WARNING [train.py:1073] (1/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 2023-04-01 08:35:31,819 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40042.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 08:35:33,716 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.177e+02 6.123e+02 7.695e+02 9.772e+02 2.844e+03, threshold=1.539e+03, percent-clipped=4.0 2023-04-01 08:35:53,058 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9_rvb from training. Duration: 27.511125 2023-04-01 08:36:10,183 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 2023-04-01 08:36:30,103 INFO [train.py:903] (1/4) Epoch 6, batch 5950, loss[loss=0.2699, simple_loss=0.3435, pruned_loss=0.09814, over 19665.00 frames. ], tot_loss[loss=0.2684, simple_loss=0.332, pruned_loss=0.1024, over 3824213.97 frames. ], batch size: 55, lr: 1.33e-02, grad_scale: 8.0 2023-04-01 08:37:03,209 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40116.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 08:37:18,121 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=40129.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 08:37:30,831 INFO [train.py:903] (1/4) Epoch 6, batch 6000, loss[loss=0.2744, simple_loss=0.3414, pruned_loss=0.1037, over 19301.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.3323, pruned_loss=0.1022, over 3839194.04 frames. ], batch size: 66, lr: 1.33e-02, grad_scale: 8.0 2023-04-01 08:37:30,831 INFO [train.py:928] (1/4) Computing validation loss 2023-04-01 08:37:43,219 INFO [train.py:937] (1/4) Epoch 6, validation: loss=0.1955, simple_loss=0.2951, pruned_loss=0.04789, over 944034.00 frames. 2023-04-01 08:37:43,220 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18649MB 2023-04-01 08:37:47,771 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.228e+02 6.298e+02 7.514e+02 9.544e+02 1.960e+03, threshold=1.503e+03, percent-clipped=1.0 2023-04-01 08:38:29,640 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=40177.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 08:38:44,845 INFO [train.py:903] (1/4) Epoch 6, batch 6050, loss[loss=0.2948, simple_loss=0.346, pruned_loss=0.1218, over 13118.00 frames. ], tot_loss[loss=0.2662, simple_loss=0.3306, pruned_loss=0.1009, over 3840977.66 frames. ], batch size: 136, lr: 1.33e-02, grad_scale: 8.0 2023-04-01 08:38:53,728 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=40196.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 08:39:14,913 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=40213.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 08:39:24,383 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40221.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 08:39:43,576 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=40236.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 08:39:48,074 INFO [train.py:903] (1/4) Epoch 6, batch 6100, loss[loss=0.2098, simple_loss=0.2818, pruned_loss=0.06893, over 19728.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3309, pruned_loss=0.1011, over 3822322.88 frames. ], batch size: 46, lr: 1.33e-02, grad_scale: 8.0 2023-04-01 08:39:54,266 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.428e+02 6.367e+02 7.728e+02 1.144e+03 2.582e+03, threshold=1.546e+03, percent-clipped=10.0 2023-04-01 08:39:54,673 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40244.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 08:40:15,215 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.73 vs. limit=5.0 2023-04-01 08:40:21,494 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.9417, 4.9307, 5.8204, 5.7217, 1.7222, 5.4739, 4.6333, 5.3029], device='cuda:1'), covar=tensor([0.1005, 0.0710, 0.0457, 0.0376, 0.4445, 0.0278, 0.0435, 0.1002], device='cuda:1'), in_proj_covar=tensor([0.0569, 0.0490, 0.0667, 0.0550, 0.0622, 0.0415, 0.0424, 0.0611], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1') 2023-04-01 08:40:51,266 INFO [train.py:903] (1/4) Epoch 6, batch 6150, loss[loss=0.2694, simple_loss=0.3305, pruned_loss=0.1041, over 19477.00 frames. ], tot_loss[loss=0.268, simple_loss=0.3322, pruned_loss=0.1019, over 3825189.26 frames. ], batch size: 49, lr: 1.33e-02, grad_scale: 8.0 2023-04-01 08:40:51,597 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=40290.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 08:40:54,065 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40292.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 08:41:19,441 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9_rvb from training. Duration: 31.02225 2023-04-01 08:41:52,300 INFO [train.py:903] (1/4) Epoch 6, batch 6200, loss[loss=0.2784, simple_loss=0.3424, pruned_loss=0.1072, over 19767.00 frames. ], tot_loss[loss=0.2684, simple_loss=0.3327, pruned_loss=0.1021, over 3836024.32 frames. ], batch size: 54, lr: 1.33e-02, grad_scale: 8.0 2023-04-01 08:41:57,219 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.101e+02 6.810e+02 8.218e+02 1.008e+03 2.334e+03, threshold=1.644e+03, percent-clipped=5.0 2023-04-01 08:42:01,834 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=40348.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 08:42:33,816 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=40372.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 08:42:54,271 INFO [train.py:903] (1/4) Epoch 6, batch 6250, loss[loss=0.3408, simple_loss=0.3848, pruned_loss=0.1484, over 19486.00 frames. ], tot_loss[loss=0.2685, simple_loss=0.3325, pruned_loss=0.1023, over 3832112.82 frames. ], batch size: 64, lr: 1.33e-02, grad_scale: 8.0 2023-04-01 08:43:04,601 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40397.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 08:43:14,690 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.4756, 1.9570, 2.0555, 2.6738, 2.3584, 2.0928, 1.8874, 2.6739], device='cuda:1'), covar=tensor([0.0747, 0.1501, 0.1132, 0.0735, 0.1058, 0.0453, 0.1013, 0.0469], device='cuda:1'), in_proj_covar=tensor([0.0249, 0.0362, 0.0287, 0.0239, 0.0306, 0.0248, 0.0277, 0.0232], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 08:43:27,274 WARNING [train.py:1073] (1/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9_rvb from training. Duration: 25.988875 2023-04-01 08:43:58,136 INFO [train.py:903] (1/4) Epoch 6, batch 6300, loss[loss=0.2362, simple_loss=0.305, pruned_loss=0.08367, over 19572.00 frames. ], tot_loss[loss=0.2676, simple_loss=0.3317, pruned_loss=0.1017, over 3821355.88 frames. ], batch size: 52, lr: 1.33e-02, grad_scale: 8.0 2023-04-01 08:44:03,767 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.716e+02 6.389e+02 8.265e+02 1.061e+03 2.633e+03, threshold=1.653e+03, percent-clipped=7.0 2023-04-01 08:44:46,061 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 2023-04-01 08:44:52,526 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 2023-04-01 08:45:00,749 INFO [train.py:903] (1/4) Epoch 6, batch 6350, loss[loss=0.2408, simple_loss=0.3185, pruned_loss=0.0815, over 19601.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.3331, pruned_loss=0.1029, over 3835157.73 frames. ], batch size: 57, lr: 1.33e-02, grad_scale: 8.0 2023-04-01 08:45:12,801 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=40500.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 08:45:24,327 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=40510.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 08:45:44,956 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40525.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 08:46:02,597 INFO [train.py:903] (1/4) Epoch 6, batch 6400, loss[loss=0.2713, simple_loss=0.3481, pruned_loss=0.09723, over 19271.00 frames. ], tot_loss[loss=0.268, simple_loss=0.332, pruned_loss=0.102, over 3845180.21 frames. ], batch size: 66, lr: 1.32e-02, grad_scale: 8.0 2023-04-01 08:46:07,241 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.716e+02 6.528e+02 8.039e+02 1.077e+03 1.980e+03, threshold=1.608e+03, percent-clipped=3.0 2023-04-01 08:46:12,175 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=40548.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 08:46:23,186 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=40557.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 08:46:44,134 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 2023-04-01 08:46:44,883 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40573.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 08:46:52,950 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=40580.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 08:47:04,378 INFO [train.py:903] (1/4) Epoch 6, batch 6450, loss[loss=0.2614, simple_loss=0.3225, pruned_loss=0.1002, over 19611.00 frames. ], tot_loss[loss=0.2677, simple_loss=0.3322, pruned_loss=0.1016, over 3840595.94 frames. ], batch size: 50, lr: 1.32e-02, grad_scale: 8.0 2023-04-01 08:47:20,546 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 2023-04-01 08:47:46,786 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([0.9585, 1.0961, 1.3699, 0.5701, 2.0774, 2.1987, 1.9479, 2.3015], device='cuda:1'), covar=tensor([0.1373, 0.2983, 0.2809, 0.2266, 0.0485, 0.0394, 0.0357, 0.0254], device='cuda:1'), in_proj_covar=tensor([0.0253, 0.0282, 0.0312, 0.0246, 0.0204, 0.0129, 0.0200, 0.0155], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-04-01 08:47:50,053 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9_rvb from training. Duration: 28.638875 2023-04-01 08:47:59,673 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=40634.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 08:48:07,603 INFO [train.py:903] (1/4) Epoch 6, batch 6500, loss[loss=0.244, simple_loss=0.3033, pruned_loss=0.09234, over 19384.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.3329, pruned_loss=0.1021, over 3819251.25 frames. ], batch size: 48, lr: 1.32e-02, grad_scale: 8.0 2023-04-01 08:48:13,017 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.640e+02 6.039e+02 7.833e+02 1.001e+03 2.233e+03, threshold=1.567e+03, percent-clipped=5.0 2023-04-01 08:48:15,440 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1_rvb from training. Duration: 0.836375 2023-04-01 08:48:44,319 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=40669.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 08:48:47,960 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40672.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 08:49:12,715 INFO [train.py:903] (1/4) Epoch 6, batch 6550, loss[loss=0.2705, simple_loss=0.3161, pruned_loss=0.1124, over 18603.00 frames. ], tot_loss[loss=0.2691, simple_loss=0.3332, pruned_loss=0.1025, over 3821470.46 frames. ], batch size: 41, lr: 1.32e-02, grad_scale: 8.0 2023-04-01 08:49:15,166 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=40692.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 08:49:18,789 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40695.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 08:50:15,125 INFO [train.py:903] (1/4) Epoch 6, batch 6600, loss[loss=0.3896, simple_loss=0.4184, pruned_loss=0.1804, over 17339.00 frames. ], tot_loss[loss=0.2699, simple_loss=0.3339, pruned_loss=0.1029, over 3827651.26 frames. ], batch size: 101, lr: 1.32e-02, grad_scale: 8.0 2023-04-01 08:50:19,763 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.418e+02 6.030e+02 7.716e+02 9.913e+02 2.888e+03, threshold=1.543e+03, percent-clipped=3.0 2023-04-01 08:50:23,601 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.3475, 3.9090, 2.4664, 3.5385, 1.2404, 3.4692, 3.4811, 3.6218], device='cuda:1'), covar=tensor([0.0661, 0.0997, 0.2009, 0.0766, 0.3521, 0.0904, 0.0824, 0.0999], device='cuda:1'), in_proj_covar=tensor([0.0372, 0.0327, 0.0382, 0.0297, 0.0361, 0.0316, 0.0307, 0.0334], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-01 08:50:26,126 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40749.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 08:51:17,591 INFO [train.py:903] (1/4) Epoch 6, batch 6650, loss[loss=0.2736, simple_loss=0.321, pruned_loss=0.1131, over 19776.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.3342, pruned_loss=0.1032, over 3814407.86 frames. ], batch size: 47, lr: 1.32e-02, grad_scale: 8.0 2023-04-01 08:51:40,494 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40807.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 08:52:19,368 INFO [train.py:903] (1/4) Epoch 6, batch 6700, loss[loss=0.215, simple_loss=0.2837, pruned_loss=0.07311, over 19781.00 frames. ], tot_loss[loss=0.2697, simple_loss=0.3334, pruned_loss=0.103, over 3827245.97 frames. ], batch size: 47, lr: 1.32e-02, grad_scale: 8.0 2023-04-01 08:52:24,133 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.981e+02 6.315e+02 8.385e+02 9.952e+02 2.559e+03, threshold=1.677e+03, percent-clipped=5.0 2023-04-01 08:52:39,409 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=40854.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 08:53:10,596 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.71 vs. limit=5.0 2023-04-01 08:53:13,429 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=40883.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 08:53:19,358 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.6263, 1.2493, 1.3519, 1.7387, 3.1777, 0.9653, 2.1070, 3.3698], device='cuda:1'), covar=tensor([0.0388, 0.2569, 0.2626, 0.1416, 0.0661, 0.2436, 0.1285, 0.0344], device='cuda:1'), in_proj_covar=tensor([0.0301, 0.0310, 0.0315, 0.0292, 0.0315, 0.0311, 0.0290, 0.0310], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 08:53:21,368 INFO [train.py:903] (1/4) Epoch 6, batch 6750, loss[loss=0.2381, simple_loss=0.2978, pruned_loss=0.08924, over 19367.00 frames. ], tot_loss[loss=0.2685, simple_loss=0.3324, pruned_loss=0.1023, over 3817370.63 frames. ], batch size: 47, lr: 1.32e-02, grad_scale: 8.0 2023-04-01 08:53:43,437 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.2071, 2.1594, 1.7133, 1.7010, 1.5746, 1.6538, 0.1923, 0.9598], device='cuda:1'), covar=tensor([0.0315, 0.0292, 0.0240, 0.0354, 0.0654, 0.0393, 0.0666, 0.0533], device='cuda:1'), in_proj_covar=tensor([0.0297, 0.0304, 0.0300, 0.0322, 0.0394, 0.0318, 0.0292, 0.0308], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 08:53:53,186 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=40918.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 08:54:04,516 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=40928.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 08:54:17,517 INFO [train.py:903] (1/4) Epoch 6, batch 6800, loss[loss=0.3068, simple_loss=0.3502, pruned_loss=0.1317, over 13384.00 frames. ], tot_loss[loss=0.2696, simple_loss=0.3332, pruned_loss=0.103, over 3810475.37 frames. ], batch size: 136, lr: 1.32e-02, grad_scale: 8.0 2023-04-01 08:54:23,019 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.567e+02 5.879e+02 7.609e+02 1.019e+03 2.150e+03, threshold=1.522e+03, percent-clipped=4.0 2023-04-01 08:54:31,421 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=40951.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 08:54:33,650 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40953.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 08:55:04,569 WARNING [train.py:1073] (1/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 2023-04-01 08:55:05,006 WARNING [train.py:1073] (1/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 2023-04-01 08:55:08,785 INFO [train.py:903] (1/4) Epoch 7, batch 0, loss[loss=0.3174, simple_loss=0.3633, pruned_loss=0.1357, over 19699.00 frames. ], tot_loss[loss=0.3174, simple_loss=0.3633, pruned_loss=0.1357, over 19699.00 frames. ], batch size: 53, lr: 1.24e-02, grad_scale: 8.0 2023-04-01 08:55:08,786 INFO [train.py:928] (1/4) Computing validation loss 2023-04-01 08:55:20,407 INFO [train.py:937] (1/4) Epoch 7, validation: loss=0.1957, simple_loss=0.2957, pruned_loss=0.04779, over 944034.00 frames. 2023-04-01 08:55:20,408 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18649MB 2023-04-01 08:55:21,965 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40969.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 08:55:30,897 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40976.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 08:55:32,922 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 2023-04-01 08:56:02,422 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3244, 1.5061, 2.0617, 1.6517, 3.0596, 2.5802, 3.1711, 1.4159], device='cuda:1'), covar=tensor([0.1905, 0.3242, 0.1882, 0.1517, 0.1478, 0.1635, 0.1628, 0.3082], device='cuda:1'), in_proj_covar=tensor([0.0452, 0.0528, 0.0509, 0.0409, 0.0569, 0.0456, 0.0638, 0.0456], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 08:56:04,723 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=41005.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 08:56:15,986 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=41013.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 08:56:19,115 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=41015.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 08:56:22,013 INFO [train.py:903] (1/4) Epoch 7, batch 50, loss[loss=0.2503, simple_loss=0.3224, pruned_loss=0.08909, over 19376.00 frames. ], tot_loss[loss=0.2645, simple_loss=0.3311, pruned_loss=0.09892, over 879878.00 frames. ], batch size: 47, lr: 1.23e-02, grad_scale: 8.0 2023-04-01 08:56:36,152 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=41030.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 08:56:51,483 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 2023-04-01 08:56:51,946 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.786e+02 6.089e+02 7.435e+02 1.027e+03 3.072e+03, threshold=1.487e+03, percent-clipped=7.0 2023-04-01 08:56:56,504 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1_rvb from training. Duration: 27.0318125 2023-04-01 08:57:17,924 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=41063.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 08:57:18,061 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=41063.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 08:57:23,453 INFO [train.py:903] (1/4) Epoch 7, batch 100, loss[loss=0.365, simple_loss=0.396, pruned_loss=0.167, over 13818.00 frames. ], tot_loss[loss=0.2655, simple_loss=0.3311, pruned_loss=0.09998, over 1534307.46 frames. ], batch size: 135, lr: 1.23e-02, grad_scale: 8.0 2023-04-01 08:57:34,768 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 2023-04-01 08:57:46,214 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=41088.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 08:58:24,482 INFO [train.py:903] (1/4) Epoch 7, batch 150, loss[loss=0.1994, simple_loss=0.2759, pruned_loss=0.06142, over 14749.00 frames. ], tot_loss[loss=0.2634, simple_loss=0.3288, pruned_loss=0.09899, over 2031629.28 frames. ], batch size: 32, lr: 1.23e-02, grad_scale: 8.0 2023-04-01 08:58:26,857 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=41120.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 08:58:36,364 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=41128.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 08:58:56,746 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.181e+02 6.219e+02 8.190e+02 1.094e+03 2.901e+03, threshold=1.638e+03, percent-clipped=4.0 2023-04-01 08:59:22,643 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 2023-04-01 08:59:23,716 INFO [train.py:903] (1/4) Epoch 7, batch 200, loss[loss=0.2802, simple_loss=0.344, pruned_loss=0.1081, over 19319.00 frames. ], tot_loss[loss=0.2661, simple_loss=0.3307, pruned_loss=0.1008, over 2433232.17 frames. ], batch size: 70, lr: 1.23e-02, grad_scale: 8.0 2023-04-01 09:00:06,211 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.3818, 0.9658, 1.2210, 1.1829, 1.9681, 0.7998, 1.8132, 2.0577], device='cuda:1'), covar=tensor([0.0818, 0.2832, 0.2704, 0.1585, 0.1104, 0.2204, 0.1098, 0.0701], device='cuda:1'), in_proj_covar=tensor([0.0305, 0.0318, 0.0321, 0.0299, 0.0322, 0.0316, 0.0295, 0.0313], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 09:00:27,760 INFO [train.py:903] (1/4) Epoch 7, batch 250, loss[loss=0.2487, simple_loss=0.3167, pruned_loss=0.09031, over 19837.00 frames. ], tot_loss[loss=0.2677, simple_loss=0.3322, pruned_loss=0.1016, over 2755062.50 frames. ], batch size: 52, lr: 1.23e-02, grad_scale: 8.0 2023-04-01 09:00:38,020 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=41225.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 09:00:40,036 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=41227.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 09:00:59,432 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.194e+02 6.866e+02 8.894e+02 1.080e+03 3.290e+03, threshold=1.779e+03, percent-clipped=6.0 2023-04-01 09:01:06,563 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=41250.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 09:01:23,879 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=41262.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 09:01:27,396 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.7253, 4.3080, 2.4666, 3.7830, 1.2913, 3.9090, 3.9059, 4.0360], device='cuda:1'), covar=tensor([0.0553, 0.0973, 0.1852, 0.0760, 0.3516, 0.0728, 0.0682, 0.0985], device='cuda:1'), in_proj_covar=tensor([0.0371, 0.0330, 0.0376, 0.0296, 0.0358, 0.0311, 0.0300, 0.0333], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-01 09:01:30,757 INFO [train.py:903] (1/4) Epoch 7, batch 300, loss[loss=0.2372, simple_loss=0.3047, pruned_loss=0.08483, over 19582.00 frames. ], tot_loss[loss=0.2649, simple_loss=0.3295, pruned_loss=0.1002, over 3005745.78 frames. ], batch size: 52, lr: 1.23e-02, grad_scale: 8.0 2023-04-01 09:02:31,485 INFO [train.py:903] (1/4) Epoch 7, batch 350, loss[loss=0.2193, simple_loss=0.2838, pruned_loss=0.07735, over 19473.00 frames. ], tot_loss[loss=0.2664, simple_loss=0.3314, pruned_loss=0.1008, over 3186394.97 frames. ], batch size: 49, lr: 1.23e-02, grad_scale: 8.0 2023-04-01 09:02:33,981 WARNING [train.py:1073] (1/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1_rvb from training. Duration: 0.9681875 2023-04-01 09:03:02,044 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=41342.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 09:03:04,940 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.733e+02 5.797e+02 7.460e+02 9.435e+02 2.818e+03, threshold=1.492e+03, percent-clipped=3.0 2023-04-01 09:03:22,750 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=41359.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 09:03:32,953 INFO [train.py:903] (1/4) Epoch 7, batch 400, loss[loss=0.2972, simple_loss=0.3589, pruned_loss=0.1178, over 19481.00 frames. ], tot_loss[loss=0.2642, simple_loss=0.3297, pruned_loss=0.0994, over 3332707.29 frames. ], batch size: 64, lr: 1.23e-02, grad_scale: 8.0 2023-04-01 09:03:43,461 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=41377.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 09:03:54,211 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=41384.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 09:04:07,919 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.3338, 2.1970, 1.9576, 1.8347, 1.5510, 1.8750, 0.4083, 1.1482], device='cuda:1'), covar=tensor([0.0274, 0.0292, 0.0211, 0.0276, 0.0608, 0.0357, 0.0608, 0.0484], device='cuda:1'), in_proj_covar=tensor([0.0299, 0.0305, 0.0301, 0.0323, 0.0392, 0.0320, 0.0290, 0.0306], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 09:04:20,760 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=41407.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 09:04:23,219 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=41409.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 09:04:34,710 INFO [train.py:903] (1/4) Epoch 7, batch 450, loss[loss=0.2657, simple_loss=0.3221, pruned_loss=0.1047, over 19744.00 frames. ], tot_loss[loss=0.2647, simple_loss=0.3298, pruned_loss=0.09983, over 3418137.86 frames. ], batch size: 51, lr: 1.23e-02, grad_scale: 8.0 2023-04-01 09:05:02,681 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 2023-04-01 09:05:03,838 WARNING [train.py:1073] (1/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 2023-04-01 09:05:06,099 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.663e+02 5.824e+02 7.719e+02 9.807e+02 3.448e+03, threshold=1.544e+03, percent-clipped=7.0 2023-04-01 09:05:31,583 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=41464.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 09:05:37,399 INFO [train.py:903] (1/4) Epoch 7, batch 500, loss[loss=0.2332, simple_loss=0.2972, pruned_loss=0.08461, over 19362.00 frames. ], tot_loss[loss=0.2667, simple_loss=0.3316, pruned_loss=0.1009, over 3513693.20 frames. ], batch size: 47, lr: 1.23e-02, grad_scale: 16.0 2023-04-01 09:05:44,885 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=41474.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 09:06:38,625 INFO [train.py:903] (1/4) Epoch 7, batch 550, loss[loss=0.259, simple_loss=0.3298, pruned_loss=0.09407, over 19694.00 frames. ], tot_loss[loss=0.2668, simple_loss=0.3313, pruned_loss=0.1011, over 3573133.04 frames. ], batch size: 59, lr: 1.23e-02, grad_scale: 8.0 2023-04-01 09:06:43,705 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=41522.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 09:07:09,850 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.859e+02 6.032e+02 7.411e+02 9.275e+02 1.625e+03, threshold=1.482e+03, percent-clipped=1.0 2023-04-01 09:07:37,911 INFO [train.py:903] (1/4) Epoch 7, batch 600, loss[loss=0.2445, simple_loss=0.3086, pruned_loss=0.09022, over 19816.00 frames. ], tot_loss[loss=0.2681, simple_loss=0.3323, pruned_loss=0.1019, over 3627029.55 frames. ], batch size: 48, lr: 1.23e-02, grad_scale: 8.0 2023-04-01 09:07:44,495 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.47 vs. limit=5.0 2023-04-01 09:07:50,844 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=41579.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 09:08:13,424 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 2023-04-01 09:08:16,202 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=41598.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 09:08:39,311 INFO [train.py:903] (1/4) Epoch 7, batch 650, loss[loss=0.2685, simple_loss=0.3369, pruned_loss=0.1001, over 19460.00 frames. ], tot_loss[loss=0.2662, simple_loss=0.331, pruned_loss=0.1007, over 3677798.97 frames. ], batch size: 64, lr: 1.23e-02, grad_scale: 8.0 2023-04-01 09:08:45,289 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=41623.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 09:09:00,389 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=41633.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 09:09:13,511 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.335e+02 5.719e+02 7.418e+02 1.065e+03 4.334e+03, threshold=1.484e+03, percent-clipped=7.0 2023-04-01 09:09:14,982 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.4763, 1.9914, 1.9892, 2.6730, 1.8202, 2.7661, 2.9226, 2.6382], device='cuda:1'), covar=tensor([0.0701, 0.0927, 0.0989, 0.0904, 0.1065, 0.0629, 0.0819, 0.0577], device='cuda:1'), in_proj_covar=tensor([0.0220, 0.0234, 0.0234, 0.0263, 0.0255, 0.0218, 0.0218, 0.0209], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1') 2023-04-01 09:09:28,654 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=41658.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 09:09:40,760 INFO [train.py:903] (1/4) Epoch 7, batch 700, loss[loss=0.2885, simple_loss=0.3577, pruned_loss=0.1096, over 19098.00 frames. ], tot_loss[loss=0.2658, simple_loss=0.331, pruned_loss=0.1003, over 3723612.88 frames. ], batch size: 69, lr: 1.23e-02, grad_scale: 8.0 2023-04-01 09:09:48,541 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=41672.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 09:10:02,987 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 2023-04-01 09:10:45,012 INFO [train.py:903] (1/4) Epoch 7, batch 750, loss[loss=0.2623, simple_loss=0.3421, pruned_loss=0.09118, over 19272.00 frames. ], tot_loss[loss=0.2645, simple_loss=0.3299, pruned_loss=0.09953, over 3749822.20 frames. ], batch size: 66, lr: 1.22e-02, grad_scale: 8.0 2023-04-01 09:10:59,406 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=41730.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 09:11:15,966 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.723e+02 5.751e+02 6.886e+02 8.721e+02 1.519e+03, threshold=1.377e+03, percent-clipped=2.0 2023-04-01 09:11:26,559 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.2142, 2.1566, 2.3594, 3.4886, 2.2068, 3.4734, 3.2542, 2.1429], device='cuda:1'), covar=tensor([0.2923, 0.2473, 0.1020, 0.1382, 0.2945, 0.0872, 0.1998, 0.1943], device='cuda:1'), in_proj_covar=tensor([0.0700, 0.0694, 0.0596, 0.0840, 0.0711, 0.0606, 0.0731, 0.0641], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 09:11:29,953 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 2023-04-01 09:11:31,818 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=41755.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 09:11:46,751 INFO [train.py:903] (1/4) Epoch 7, batch 800, loss[loss=0.2845, simple_loss=0.3479, pruned_loss=0.1105, over 19769.00 frames. ], tot_loss[loss=0.2639, simple_loss=0.3297, pruned_loss=0.09907, over 3772214.93 frames. ], batch size: 54, lr: 1.22e-02, grad_scale: 8.0 2023-04-01 09:11:56,200 WARNING [train.py:1073] (1/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 2023-04-01 09:11:58,775 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=41778.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 09:12:27,398 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.7674, 3.1848, 3.2373, 3.2808, 1.0992, 3.0677, 2.7253, 2.9371], device='cuda:1'), covar=tensor([0.1207, 0.0792, 0.0716, 0.0635, 0.4279, 0.0635, 0.0691, 0.1268], device='cuda:1'), in_proj_covar=tensor([0.0570, 0.0495, 0.0666, 0.0553, 0.0621, 0.0422, 0.0424, 0.0619], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1') 2023-04-01 09:12:31,075 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=41803.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 09:12:48,110 INFO [train.py:903] (1/4) Epoch 7, batch 850, loss[loss=0.248, simple_loss=0.3075, pruned_loss=0.09431, over 19765.00 frames. ], tot_loss[loss=0.264, simple_loss=0.3293, pruned_loss=0.09938, over 3775714.12 frames. ], batch size: 47, lr: 1.22e-02, grad_scale: 8.0 2023-04-01 09:13:10,282 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=41835.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 09:13:12,180 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.8652, 4.2304, 4.4977, 4.5543, 1.4908, 4.1376, 3.6368, 4.0748], device='cuda:1'), covar=tensor([0.1110, 0.0669, 0.0588, 0.0466, 0.4863, 0.0487, 0.0606, 0.1145], device='cuda:1'), in_proj_covar=tensor([0.0571, 0.0493, 0.0665, 0.0551, 0.0622, 0.0424, 0.0425, 0.0618], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1') 2023-04-01 09:13:23,186 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.000e+02 6.514e+02 8.083e+02 9.646e+02 1.896e+03, threshold=1.617e+03, percent-clipped=5.0 2023-04-01 09:13:38,259 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9_rvb from training. Duration: 25.061125 2023-04-01 09:13:41,028 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=41860.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 09:13:50,188 INFO [train.py:903] (1/4) Epoch 7, batch 900, loss[loss=0.2561, simple_loss=0.3239, pruned_loss=0.09418, over 19595.00 frames. ], tot_loss[loss=0.2646, simple_loss=0.3295, pruned_loss=0.09989, over 3792091.80 frames. ], batch size: 52, lr: 1.22e-02, grad_scale: 8.0 2023-04-01 09:14:05,378 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.4911, 4.0452, 2.2434, 3.6245, 0.9295, 3.6815, 3.7498, 3.8755], device='cuda:1'), covar=tensor([0.0621, 0.1101, 0.2295, 0.0738, 0.4089, 0.0827, 0.0757, 0.1033], device='cuda:1'), in_proj_covar=tensor([0.0379, 0.0332, 0.0383, 0.0300, 0.0357, 0.0313, 0.0307, 0.0339], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-01 09:14:51,584 INFO [train.py:903] (1/4) Epoch 7, batch 950, loss[loss=0.2825, simple_loss=0.3352, pruned_loss=0.1149, over 19757.00 frames. ], tot_loss[loss=0.2638, simple_loss=0.3291, pruned_loss=0.09923, over 3811253.63 frames. ], batch size: 47, lr: 1.22e-02, grad_scale: 4.0 2023-04-01 09:14:55,108 WARNING [train.py:1073] (1/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9_rvb from training. Duration: 26.32775 2023-04-01 09:15:01,406 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=41924.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 09:15:26,240 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.318e+02 6.505e+02 7.519e+02 9.487e+02 1.757e+03, threshold=1.504e+03, percent-clipped=1.0 2023-04-01 09:15:53,978 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=41966.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 09:15:55,890 INFO [train.py:903] (1/4) Epoch 7, batch 1000, loss[loss=0.2724, simple_loss=0.3418, pruned_loss=0.1015, over 19462.00 frames. ], tot_loss[loss=0.2657, simple_loss=0.3308, pruned_loss=0.1003, over 3799998.54 frames. ], batch size: 64, lr: 1.22e-02, grad_scale: 4.0 2023-04-01 09:16:45,430 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.7933, 1.5030, 1.3940, 1.8072, 1.7544, 1.6426, 1.6192, 1.8441], device='cuda:1'), covar=tensor([0.0875, 0.1518, 0.1284, 0.0874, 0.1026, 0.0459, 0.0937, 0.0588], device='cuda:1'), in_proj_covar=tensor([0.0246, 0.0361, 0.0289, 0.0238, 0.0303, 0.0248, 0.0274, 0.0231], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 09:16:48,566 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 2023-04-01 09:16:56,956 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=42016.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 09:16:59,084 INFO [train.py:903] (1/4) Epoch 7, batch 1050, loss[loss=0.2943, simple_loss=0.3426, pruned_loss=0.123, over 19714.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3316, pruned_loss=0.1007, over 3814300.61 frames. ], batch size: 51, lr: 1.22e-02, grad_scale: 4.0 2023-04-01 09:17:14,992 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6700, 1.3147, 1.3585, 1.8443, 1.4638, 1.9275, 1.7731, 1.6677], device='cuda:1'), covar=tensor([0.0813, 0.1081, 0.1158, 0.0893, 0.0976, 0.0703, 0.0967, 0.0731], device='cuda:1'), in_proj_covar=tensor([0.0219, 0.0234, 0.0234, 0.0265, 0.0255, 0.0217, 0.0218, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1') 2023-04-01 09:17:19,836 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.12 vs. limit=2.0 2023-04-01 09:17:29,383 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 2023-04-01 09:17:33,747 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.193e+02 5.894e+02 7.129e+02 9.584e+02 2.561e+03, threshold=1.426e+03, percent-clipped=5.0 2023-04-01 09:18:00,561 INFO [train.py:903] (1/4) Epoch 7, batch 1100, loss[loss=0.2717, simple_loss=0.3157, pruned_loss=0.1139, over 19760.00 frames. ], tot_loss[loss=0.2645, simple_loss=0.3296, pruned_loss=0.09971, over 3815834.47 frames. ], batch size: 47, lr: 1.22e-02, grad_scale: 4.0 2023-04-01 09:19:03,370 INFO [train.py:903] (1/4) Epoch 7, batch 1150, loss[loss=0.2717, simple_loss=0.3413, pruned_loss=0.1011, over 19050.00 frames. ], tot_loss[loss=0.2651, simple_loss=0.3299, pruned_loss=0.1001, over 3794565.72 frames. ], batch size: 69, lr: 1.22e-02, grad_scale: 4.0 2023-04-01 09:19:21,137 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=42131.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 09:19:37,737 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.517e+02 5.892e+02 7.369e+02 1.011e+03 1.805e+03, threshold=1.474e+03, percent-clipped=4.0 2023-04-01 09:20:05,775 INFO [train.py:903] (1/4) Epoch 7, batch 1200, loss[loss=0.2814, simple_loss=0.3508, pruned_loss=0.106, over 19303.00 frames. ], tot_loss[loss=0.2656, simple_loss=0.3305, pruned_loss=0.1003, over 3793234.29 frames. ], batch size: 66, lr: 1.22e-02, grad_scale: 8.0 2023-04-01 09:20:30,610 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 2023-04-01 09:21:08,028 INFO [train.py:903] (1/4) Epoch 7, batch 1250, loss[loss=0.2351, simple_loss=0.32, pruned_loss=0.07515, over 19529.00 frames. ], tot_loss[loss=0.2647, simple_loss=0.3295, pruned_loss=0.1, over 3799697.97 frames. ], batch size: 56, lr: 1.22e-02, grad_scale: 8.0 2023-04-01 09:21:43,466 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.288e+02 6.237e+02 7.824e+02 1.034e+03 2.254e+03, threshold=1.565e+03, percent-clipped=6.0 2023-04-01 09:22:09,629 INFO [train.py:903] (1/4) Epoch 7, batch 1300, loss[loss=0.2909, simple_loss=0.3529, pruned_loss=0.1144, over 19526.00 frames. ], tot_loss[loss=0.2655, simple_loss=0.3304, pruned_loss=0.1003, over 3800033.44 frames. ], batch size: 54, lr: 1.22e-02, grad_scale: 8.0 2023-04-01 09:22:09,809 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=42268.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 09:23:02,499 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=42310.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 09:23:12,314 INFO [train.py:903] (1/4) Epoch 7, batch 1350, loss[loss=0.3393, simple_loss=0.3802, pruned_loss=0.1492, over 13371.00 frames. ], tot_loss[loss=0.2646, simple_loss=0.3294, pruned_loss=0.09992, over 3807643.42 frames. ], batch size: 136, lr: 1.22e-02, grad_scale: 8.0 2023-04-01 09:23:29,490 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6788, 2.0990, 1.7041, 1.6749, 1.9995, 1.4440, 1.4382, 1.6978], device='cuda:1'), covar=tensor([0.0612, 0.0538, 0.0598, 0.0455, 0.0365, 0.0704, 0.0576, 0.0387], device='cuda:1'), in_proj_covar=tensor([0.0282, 0.0283, 0.0315, 0.0242, 0.0229, 0.0311, 0.0291, 0.0255], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 09:23:47,361 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.606e+02 6.578e+02 7.819e+02 1.024e+03 2.032e+03, threshold=1.564e+03, percent-clipped=3.0 2023-04-01 09:24:07,204 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2432, 1.2714, 1.6464, 1.3878, 2.1252, 1.9601, 2.3437, 0.7228], device='cuda:1'), covar=tensor([0.1850, 0.3367, 0.1867, 0.1551, 0.1230, 0.1602, 0.1157, 0.3074], device='cuda:1'), in_proj_covar=tensor([0.0458, 0.0526, 0.0518, 0.0413, 0.0565, 0.0457, 0.0631, 0.0457], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 09:24:15,698 INFO [train.py:903] (1/4) Epoch 7, batch 1400, loss[loss=0.232, simple_loss=0.3054, pruned_loss=0.07933, over 19747.00 frames. ], tot_loss[loss=0.263, simple_loss=0.3282, pruned_loss=0.09886, over 3805983.75 frames. ], batch size: 54, lr: 1.22e-02, grad_scale: 8.0 2023-04-01 09:24:34,411 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=42383.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 09:24:39,183 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=42387.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 09:24:50,665 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=42397.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 09:25:10,408 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=42412.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 09:25:12,454 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_rvb from training. Duration: 25.85 2023-04-01 09:25:17,935 INFO [train.py:903] (1/4) Epoch 7, batch 1450, loss[loss=0.2182, simple_loss=0.2885, pruned_loss=0.07395, over 19402.00 frames. ], tot_loss[loss=0.2613, simple_loss=0.3266, pruned_loss=0.09803, over 3803176.27 frames. ], batch size: 48, lr: 1.21e-02, grad_scale: 8.0 2023-04-01 09:25:26,426 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=42425.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 09:25:53,241 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.957e+02 5.854e+02 7.466e+02 9.791e+02 2.115e+03, threshold=1.493e+03, percent-clipped=4.0 2023-04-01 09:26:19,590 INFO [train.py:903] (1/4) Epoch 7, batch 1500, loss[loss=0.2207, simple_loss=0.2944, pruned_loss=0.07348, over 19751.00 frames. ], tot_loss[loss=0.2616, simple_loss=0.3268, pruned_loss=0.09822, over 3805491.11 frames. ], batch size: 51, lr: 1.21e-02, grad_scale: 8.0 2023-04-01 09:27:07,129 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.66 vs. limit=5.0 2023-04-01 09:27:20,486 INFO [train.py:903] (1/4) Epoch 7, batch 1550, loss[loss=0.29, simple_loss=0.356, pruned_loss=0.112, over 19308.00 frames. ], tot_loss[loss=0.2641, simple_loss=0.329, pruned_loss=0.09959, over 3802277.47 frames. ], batch size: 66, lr: 1.21e-02, grad_scale: 8.0 2023-04-01 09:27:26,957 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9812, 1.6480, 1.4538, 1.9103, 1.8270, 1.7207, 1.4350, 1.7945], device='cuda:1'), covar=tensor([0.0691, 0.1223, 0.1262, 0.0765, 0.0876, 0.0447, 0.1010, 0.0589], device='cuda:1'), in_proj_covar=tensor([0.0242, 0.0356, 0.0285, 0.0236, 0.0302, 0.0245, 0.0266, 0.0232], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 09:27:29,126 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=42524.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 09:27:55,714 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.300e+02 6.427e+02 8.071e+02 9.919e+02 2.182e+03, threshold=1.614e+03, percent-clipped=7.0 2023-04-01 09:28:23,008 INFO [train.py:903] (1/4) Epoch 7, batch 1600, loss[loss=0.2253, simple_loss=0.2893, pruned_loss=0.08067, over 19327.00 frames. ], tot_loss[loss=0.2638, simple_loss=0.3288, pruned_loss=0.09945, over 3811394.72 frames. ], batch size: 44, lr: 1.21e-02, grad_scale: 8.0 2023-04-01 09:28:41,307 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9_rvb from training. Duration: 30.1555625 2023-04-01 09:29:04,602 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.5367, 1.4826, 1.6541, 1.8007, 3.9685, 1.0171, 2.1735, 3.8508], device='cuda:1'), covar=tensor([0.0316, 0.2570, 0.2605, 0.1649, 0.0637, 0.2614, 0.1379, 0.0387], device='cuda:1'), in_proj_covar=tensor([0.0303, 0.0319, 0.0326, 0.0294, 0.0323, 0.0316, 0.0296, 0.0318], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 09:29:24,654 INFO [train.py:903] (1/4) Epoch 7, batch 1650, loss[loss=0.2223, simple_loss=0.2996, pruned_loss=0.07254, over 19478.00 frames. ], tot_loss[loss=0.2621, simple_loss=0.3269, pruned_loss=0.0986, over 3814103.51 frames. ], batch size: 49, lr: 1.21e-02, grad_scale: 8.0 2023-04-01 09:29:50,150 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=42639.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 09:29:59,296 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.52 vs. limit=2.0 2023-04-01 09:29:59,742 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.059e+02 6.175e+02 7.945e+02 9.824e+02 2.630e+03, threshold=1.589e+03, percent-clipped=4.0 2023-04-01 09:30:23,021 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=42664.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 09:30:26,333 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=42667.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 09:30:27,192 INFO [train.py:903] (1/4) Epoch 7, batch 1700, loss[loss=0.2791, simple_loss=0.3431, pruned_loss=0.1076, over 19493.00 frames. ], tot_loss[loss=0.2619, simple_loss=0.3271, pruned_loss=0.09832, over 3818271.52 frames. ], batch size: 64, lr: 1.21e-02, grad_scale: 8.0 2023-04-01 09:30:43,953 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=42681.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 09:31:02,728 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 2023-04-01 09:31:15,460 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=42706.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 09:31:29,071 INFO [train.py:903] (1/4) Epoch 7, batch 1750, loss[loss=0.2688, simple_loss=0.3363, pruned_loss=0.1006, over 19562.00 frames. ], tot_loss[loss=0.2604, simple_loss=0.326, pruned_loss=0.09746, over 3828816.52 frames. ], batch size: 61, lr: 1.21e-02, grad_scale: 8.0 2023-04-01 09:31:59,015 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.50 vs. limit=2.0 2023-04-01 09:31:59,542 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=42741.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 09:32:05,277 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.512e+02 5.682e+02 7.177e+02 9.179e+02 1.731e+03, threshold=1.435e+03, percent-clipped=1.0 2023-04-01 09:32:33,267 INFO [train.py:903] (1/4) Epoch 7, batch 1800, loss[loss=0.2749, simple_loss=0.3466, pruned_loss=0.1016, over 19552.00 frames. ], tot_loss[loss=0.26, simple_loss=0.326, pruned_loss=0.09701, over 3825564.11 frames. ], batch size: 54, lr: 1.21e-02, grad_scale: 8.0 2023-04-01 09:33:27,626 WARNING [train.py:1073] (1/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9_rvb from training. Duration: 25.0944375 2023-04-01 09:33:35,066 INFO [train.py:903] (1/4) Epoch 7, batch 1850, loss[loss=0.2702, simple_loss=0.3216, pruned_loss=0.1094, over 19348.00 frames. ], tot_loss[loss=0.2605, simple_loss=0.3264, pruned_loss=0.09734, over 3829102.58 frames. ], batch size: 47, lr: 1.21e-02, grad_scale: 8.0 2023-04-01 09:34:04,793 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9_rvb from training. Duration: 27.8166875 2023-04-01 09:34:09,013 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.550e+02 6.596e+02 7.909e+02 1.066e+03 2.536e+03, threshold=1.582e+03, percent-clipped=10.0 2023-04-01 09:34:22,466 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=42856.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 09:34:37,123 INFO [train.py:903] (1/4) Epoch 7, batch 1900, loss[loss=0.2742, simple_loss=0.3436, pruned_loss=0.1024, over 19449.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.3269, pruned_loss=0.09742, over 3820232.60 frames. ], batch size: 64, lr: 1.21e-02, grad_scale: 8.0 2023-04-01 09:34:37,305 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=42868.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 09:34:48,050 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=42877.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 09:34:51,101 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 2023-04-01 09:34:51,276 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([5.8338, 5.1507, 2.8513, 4.5320, 1.2746, 4.8482, 5.0276, 5.3010], device='cuda:1'), covar=tensor([0.0441, 0.0952, 0.1950, 0.0648, 0.4017, 0.0666, 0.0653, 0.0795], device='cuda:1'), in_proj_covar=tensor([0.0381, 0.0337, 0.0389, 0.0300, 0.0363, 0.0320, 0.0306, 0.0339], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-01 09:34:57,083 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9_rvb from training. Duration: 27.02225 2023-04-01 09:35:14,092 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=42897.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 09:35:21,941 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 2023-04-01 09:35:32,110 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.54 vs. limit=5.0 2023-04-01 09:35:38,556 INFO [train.py:903] (1/4) Epoch 7, batch 1950, loss[loss=0.2204, simple_loss=0.285, pruned_loss=0.07789, over 19715.00 frames. ], tot_loss[loss=0.2605, simple_loss=0.3266, pruned_loss=0.09717, over 3825024.32 frames. ], batch size: 45, lr: 1.21e-02, grad_scale: 8.0 2023-04-01 09:36:15,248 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.499e+02 6.682e+02 8.244e+02 9.665e+02 2.689e+03, threshold=1.649e+03, percent-clipped=3.0 2023-04-01 09:36:41,125 INFO [train.py:903] (1/4) Epoch 7, batch 2000, loss[loss=0.302, simple_loss=0.3617, pruned_loss=0.1211, over 19524.00 frames. ], tot_loss[loss=0.2624, simple_loss=0.3282, pruned_loss=0.09829, over 3833010.08 frames. ], batch size: 64, lr: 1.21e-02, grad_scale: 8.0 2023-04-01 09:37:00,296 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=42983.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 09:37:35,126 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=43011.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 09:37:36,083 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_rvb from training. Duration: 26.8349375 2023-04-01 09:37:43,660 INFO [train.py:903] (1/4) Epoch 7, batch 2050, loss[loss=0.2876, simple_loss=0.3493, pruned_loss=0.1129, over 19662.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.3285, pruned_loss=0.09823, over 3834318.81 frames. ], batch size: 55, lr: 1.21e-02, grad_scale: 8.0 2023-04-01 09:37:56,190 WARNING [train.py:1073] (1/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 2023-04-01 09:37:57,389 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 2023-04-01 09:38:08,484 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6541, 1.2666, 1.4027, 1.8660, 1.4471, 1.9541, 2.0133, 1.6960], device='cuda:1'), covar=tensor([0.0820, 0.1053, 0.1095, 0.1007, 0.0990, 0.0719, 0.0813, 0.0671], device='cuda:1'), in_proj_covar=tensor([0.0224, 0.0236, 0.0232, 0.0264, 0.0254, 0.0221, 0.0216, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1') 2023-04-01 09:38:17,409 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.956e+02 6.149e+02 7.669e+02 9.586e+02 2.177e+03, threshold=1.534e+03, percent-clipped=1.0 2023-04-01 09:38:19,526 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 2023-04-01 09:38:46,695 INFO [train.py:903] (1/4) Epoch 7, batch 2100, loss[loss=0.3089, simple_loss=0.3587, pruned_loss=0.1295, over 13879.00 frames. ], tot_loss[loss=0.2608, simple_loss=0.327, pruned_loss=0.09724, over 3837181.25 frames. ], batch size: 136, lr: 1.21e-02, grad_scale: 8.0 2023-04-01 09:39:13,039 WARNING [train.py:1073] (1/4) Exclude cut with ID 453-131332-0000-131866_sp0.9_rvb from training. Duration: 25.3333125 2023-04-01 09:39:16,693 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.5365, 1.1060, 1.2998, 1.2750, 2.1523, 0.9243, 1.8597, 2.2249], device='cuda:1'), covar=tensor([0.0533, 0.2344, 0.2323, 0.1348, 0.0795, 0.1891, 0.0910, 0.0553], device='cuda:1'), in_proj_covar=tensor([0.0307, 0.0316, 0.0326, 0.0295, 0.0322, 0.0318, 0.0298, 0.0320], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 09:39:35,491 WARNING [train.py:1073] (1/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9_rvb from training. Duration: 26.6166875 2023-04-01 09:39:41,467 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=43112.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 09:39:48,065 INFO [train.py:903] (1/4) Epoch 7, batch 2150, loss[loss=0.2807, simple_loss=0.3448, pruned_loss=0.1083, over 19515.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.3271, pruned_loss=0.09733, over 3843806.63 frames. ], batch size: 64, lr: 1.20e-02, grad_scale: 8.0 2023-04-01 09:39:57,658 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=43126.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 09:40:12,363 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=43137.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 09:40:23,149 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.978e+02 6.527e+02 7.673e+02 9.951e+02 2.226e+03, threshold=1.535e+03, percent-clipped=3.0 2023-04-01 09:40:40,238 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 2023-04-01 09:40:49,524 INFO [train.py:903] (1/4) Epoch 7, batch 2200, loss[loss=0.2531, simple_loss=0.3243, pruned_loss=0.09089, over 17562.00 frames. ], tot_loss[loss=0.2601, simple_loss=0.3266, pruned_loss=0.09687, over 3833149.08 frames. ], batch size: 101, lr: 1.20e-02, grad_scale: 8.0 2023-04-01 09:41:16,779 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3556, 1.1724, 1.6830, 1.2310, 2.9197, 3.7121, 3.6026, 3.9352], device='cuda:1'), covar=tensor([0.1380, 0.3056, 0.2801, 0.1817, 0.0391, 0.0141, 0.0183, 0.0129], device='cuda:1'), in_proj_covar=tensor([0.0250, 0.0284, 0.0314, 0.0247, 0.0203, 0.0134, 0.0204, 0.0161], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 09:41:53,418 INFO [train.py:903] (1/4) Epoch 7, batch 2250, loss[loss=0.2682, simple_loss=0.3392, pruned_loss=0.09865, over 18387.00 frames. ], tot_loss[loss=0.261, simple_loss=0.3271, pruned_loss=0.09746, over 3825940.09 frames. ], batch size: 84, lr: 1.20e-02, grad_scale: 8.0 2023-04-01 09:41:57,901 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=43221.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 09:42:19,734 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=43239.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 09:42:21,846 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=43241.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 09:42:27,686 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.233e+02 6.206e+02 7.577e+02 9.251e+02 2.641e+03, threshold=1.515e+03, percent-clipped=5.0 2023-04-01 09:42:52,555 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=43264.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 09:42:56,673 INFO [train.py:903] (1/4) Epoch 7, batch 2300, loss[loss=0.2352, simple_loss=0.3007, pruned_loss=0.08482, over 19803.00 frames. ], tot_loss[loss=0.261, simple_loss=0.3272, pruned_loss=0.09745, over 3829691.93 frames. ], batch size: 48, lr: 1.20e-02, grad_scale: 8.0 2023-04-01 09:43:10,466 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_rvb from training. Duration: 26.205 2023-04-01 09:43:50,078 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 2023-04-01 09:43:59,263 INFO [train.py:903] (1/4) Epoch 7, batch 2350, loss[loss=0.2329, simple_loss=0.2949, pruned_loss=0.08542, over 18674.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.326, pruned_loss=0.09642, over 3831148.99 frames. ], batch size: 41, lr: 1.20e-02, grad_scale: 8.0 2023-04-01 09:44:20,947 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=43336.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 09:44:34,240 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.645e+02 6.151e+02 7.487e+02 9.533e+02 1.563e+03, threshold=1.497e+03, percent-clipped=2.0 2023-04-01 09:44:43,149 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_rvb from training. Duration: 25.775 2023-04-01 09:44:46,800 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=43356.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 09:44:55,073 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.8915, 4.3203, 4.6386, 4.5970, 1.6636, 4.2354, 3.7034, 4.2617], device='cuda:1'), covar=tensor([0.1151, 0.0771, 0.0506, 0.0453, 0.4470, 0.0498, 0.0620, 0.1008], device='cuda:1'), in_proj_covar=tensor([0.0579, 0.0503, 0.0673, 0.0556, 0.0633, 0.0424, 0.0430, 0.0627], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1') 2023-04-01 09:44:59,570 WARNING [train.py:1073] (1/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9_rvb from training. Duration: 25.45 2023-04-01 09:45:00,595 INFO [train.py:903] (1/4) Epoch 7, batch 2400, loss[loss=0.3087, simple_loss=0.3696, pruned_loss=0.1239, over 19664.00 frames. ], tot_loss[loss=0.2606, simple_loss=0.3269, pruned_loss=0.09712, over 3830554.88 frames. ], batch size: 58, lr: 1.20e-02, grad_scale: 8.0 2023-04-01 09:45:20,074 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=43382.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 09:45:49,725 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=43407.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 09:46:04,429 INFO [train.py:903] (1/4) Epoch 7, batch 2450, loss[loss=0.2767, simple_loss=0.3423, pruned_loss=0.1056, over 18806.00 frames. ], tot_loss[loss=0.2602, simple_loss=0.3265, pruned_loss=0.09694, over 3837572.75 frames. ], batch size: 74, lr: 1.20e-02, grad_scale: 8.0 2023-04-01 09:46:38,155 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.412e+02 5.802e+02 7.669e+02 8.855e+02 2.284e+03, threshold=1.534e+03, percent-clipped=5.0 2023-04-01 09:46:43,553 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.62 vs. limit=2.0 2023-04-01 09:47:02,416 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.0156, 1.2504, 1.4197, 0.5707, 2.2586, 2.3996, 2.1483, 2.5765], device='cuda:1'), covar=tensor([0.1343, 0.3016, 0.2973, 0.2307, 0.0417, 0.0233, 0.0343, 0.0211], device='cuda:1'), in_proj_covar=tensor([0.0253, 0.0286, 0.0316, 0.0250, 0.0205, 0.0135, 0.0205, 0.0161], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 09:47:06,559 INFO [train.py:903] (1/4) Epoch 7, batch 2500, loss[loss=0.2598, simple_loss=0.319, pruned_loss=0.1003, over 19774.00 frames. ], tot_loss[loss=0.2602, simple_loss=0.3266, pruned_loss=0.09694, over 3838641.32 frames. ], batch size: 47, lr: 1.20e-02, grad_scale: 8.0 2023-04-01 09:47:38,455 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.79 vs. limit=2.0 2023-04-01 09:48:09,514 INFO [train.py:903] (1/4) Epoch 7, batch 2550, loss[loss=0.2351, simple_loss=0.3145, pruned_loss=0.07783, over 19669.00 frames. ], tot_loss[loss=0.2588, simple_loss=0.3257, pruned_loss=0.09597, over 3834409.34 frames. ], batch size: 55, lr: 1.20e-02, grad_scale: 8.0 2023-04-01 09:48:25,706 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=43532.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 09:48:44,536 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.143e+02 6.070e+02 7.288e+02 8.830e+02 1.707e+03, threshold=1.458e+03, percent-clipped=2.0 2023-04-01 09:49:01,582 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.0137, 1.1193, 1.2564, 1.2115, 2.6327, 0.8851, 1.7719, 2.7674], device='cuda:1'), covar=tensor([0.0438, 0.2470, 0.2573, 0.1666, 0.0697, 0.2307, 0.1214, 0.0422], device='cuda:1'), in_proj_covar=tensor([0.0307, 0.0315, 0.0323, 0.0295, 0.0321, 0.0319, 0.0300, 0.0316], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 09:49:05,678 WARNING [train.py:1073] (1/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 2023-04-01 09:49:10,228 INFO [train.py:903] (1/4) Epoch 7, batch 2600, loss[loss=0.3258, simple_loss=0.3723, pruned_loss=0.1396, over 12989.00 frames. ], tot_loss[loss=0.2593, simple_loss=0.326, pruned_loss=0.09625, over 3832317.85 frames. ], batch size: 135, lr: 1.20e-02, grad_scale: 8.0 2023-04-01 09:49:41,330 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=43592.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 09:49:59,872 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=4.12 vs. limit=5.0 2023-04-01 09:50:06,453 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=43612.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 09:50:12,269 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=43617.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 09:50:12,985 INFO [train.py:903] (1/4) Epoch 7, batch 2650, loss[loss=0.275, simple_loss=0.3433, pruned_loss=0.1033, over 19667.00 frames. ], tot_loss[loss=0.26, simple_loss=0.3266, pruned_loss=0.09669, over 3827084.89 frames. ], batch size: 58, lr: 1.20e-02, grad_scale: 8.0 2023-04-01 09:50:28,063 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.1849, 2.0770, 1.8042, 1.6805, 1.6178, 1.8282, 0.2594, 0.9773], device='cuda:1'), covar=tensor([0.0280, 0.0292, 0.0215, 0.0344, 0.0559, 0.0375, 0.0638, 0.0512], device='cuda:1'), in_proj_covar=tensor([0.0305, 0.0310, 0.0303, 0.0329, 0.0396, 0.0322, 0.0291, 0.0306], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 09:50:35,409 WARNING [train.py:1073] (1/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9_rvb from training. Duration: 27.25 2023-04-01 09:50:38,368 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=43637.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 09:50:49,494 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.166e+02 6.243e+02 7.316e+02 9.618e+02 1.411e+03, threshold=1.463e+03, percent-clipped=0.0 2023-04-01 09:51:16,493 INFO [train.py:903] (1/4) Epoch 7, batch 2700, loss[loss=0.2154, simple_loss=0.285, pruned_loss=0.07295, over 19732.00 frames. ], tot_loss[loss=0.2614, simple_loss=0.3277, pruned_loss=0.09755, over 3825160.74 frames. ], batch size: 45, lr: 1.20e-02, grad_scale: 4.0 2023-04-01 09:51:19,201 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6784, 1.5547, 1.3648, 1.9673, 1.5106, 2.1041, 2.1032, 2.0444], device='cuda:1'), covar=tensor([0.0816, 0.0928, 0.1060, 0.0888, 0.0917, 0.0641, 0.0791, 0.0558], device='cuda:1'), in_proj_covar=tensor([0.0229, 0.0241, 0.0236, 0.0271, 0.0258, 0.0223, 0.0219, 0.0216], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1') 2023-04-01 09:51:25,739 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5507, 1.5725, 1.3533, 1.8596, 1.4468, 2.0657, 1.9348, 1.8977], device='cuda:1'), covar=tensor([0.0775, 0.0853, 0.0976, 0.0805, 0.0896, 0.0588, 0.0801, 0.0564], device='cuda:1'), in_proj_covar=tensor([0.0228, 0.0241, 0.0236, 0.0271, 0.0258, 0.0222, 0.0219, 0.0216], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1') 2023-04-01 09:52:19,230 INFO [train.py:903] (1/4) Epoch 7, batch 2750, loss[loss=0.275, simple_loss=0.3398, pruned_loss=0.1051, over 19613.00 frames. ], tot_loss[loss=0.2621, simple_loss=0.3284, pruned_loss=0.09787, over 3823960.16 frames. ], batch size: 57, lr: 1.20e-02, grad_scale: 4.0 2023-04-01 09:52:43,660 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.3569, 1.4337, 1.5006, 1.5114, 2.9426, 0.8859, 1.9727, 3.1602], device='cuda:1'), covar=tensor([0.0377, 0.2269, 0.2320, 0.1504, 0.0628, 0.2347, 0.1121, 0.0344], device='cuda:1'), in_proj_covar=tensor([0.0305, 0.0313, 0.0322, 0.0291, 0.0319, 0.0312, 0.0297, 0.0314], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 09:52:55,466 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.581e+02 6.869e+02 8.299e+02 1.091e+03 2.331e+03, threshold=1.660e+03, percent-clipped=8.0 2023-04-01 09:53:20,548 INFO [train.py:903] (1/4) Epoch 7, batch 2800, loss[loss=0.2201, simple_loss=0.2792, pruned_loss=0.08047, over 19733.00 frames. ], tot_loss[loss=0.261, simple_loss=0.3275, pruned_loss=0.09723, over 3837511.08 frames. ], batch size: 46, lr: 1.20e-02, grad_scale: 8.0 2023-04-01 09:53:35,923 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5276, 1.1486, 1.5438, 1.2976, 2.5910, 3.7579, 3.5781, 3.9979], device='cuda:1'), covar=tensor([0.1265, 0.2939, 0.2721, 0.1916, 0.0502, 0.0148, 0.0176, 0.0118], device='cuda:1'), in_proj_covar=tensor([0.0251, 0.0283, 0.0311, 0.0245, 0.0200, 0.0132, 0.0203, 0.0160], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-04-01 09:54:07,570 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=43805.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 09:54:22,953 INFO [train.py:903] (1/4) Epoch 7, batch 2850, loss[loss=0.3048, simple_loss=0.3626, pruned_loss=0.1236, over 19631.00 frames. ], tot_loss[loss=0.2614, simple_loss=0.3279, pruned_loss=0.09742, over 3844823.56 frames. ], batch size: 57, lr: 1.20e-02, grad_scale: 8.0 2023-04-01 09:54:50,790 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.73 vs. limit=2.0 2023-04-01 09:54:59,106 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.801e+02 6.316e+02 8.520e+02 1.005e+03 1.613e+03, threshold=1.704e+03, percent-clipped=0.0 2023-04-01 09:55:25,670 WARNING [train.py:1073] (1/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 2023-04-01 09:55:26,751 INFO [train.py:903] (1/4) Epoch 7, batch 2900, loss[loss=0.2613, simple_loss=0.3236, pruned_loss=0.09956, over 17520.00 frames. ], tot_loss[loss=0.261, simple_loss=0.3276, pruned_loss=0.0972, over 3840280.70 frames. ], batch size: 101, lr: 1.19e-02, grad_scale: 8.0 2023-04-01 09:55:36,422 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=43876.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 09:56:28,083 INFO [train.py:903] (1/4) Epoch 7, batch 2950, loss[loss=0.3107, simple_loss=0.3704, pruned_loss=0.1254, over 19457.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.329, pruned_loss=0.09764, over 3822583.91 frames. ], batch size: 64, lr: 1.19e-02, grad_scale: 8.0 2023-04-01 09:56:39,862 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=43927.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 09:57:04,993 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.608e+02 5.923e+02 7.218e+02 9.278e+02 2.092e+03, threshold=1.444e+03, percent-clipped=1.0 2023-04-01 09:57:30,516 INFO [train.py:903] (1/4) Epoch 7, batch 3000, loss[loss=0.2467, simple_loss=0.3168, pruned_loss=0.08826, over 19615.00 frames. ], tot_loss[loss=0.2618, simple_loss=0.3279, pruned_loss=0.09781, over 3817743.08 frames. ], batch size: 50, lr: 1.19e-02, grad_scale: 8.0 2023-04-01 09:57:30,516 INFO [train.py:928] (1/4) Computing validation loss 2023-04-01 09:57:43,097 INFO [train.py:937] (1/4) Epoch 7, validation: loss=0.1917, simple_loss=0.2919, pruned_loss=0.04574, over 944034.00 frames. 2023-04-01 09:57:43,098 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18649MB 2023-04-01 09:57:49,876 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_rvb from training. Duration: 29.735 2023-04-01 09:57:54,965 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=43977.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 09:58:07,980 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9135, 3.4811, 1.7515, 2.2295, 2.9001, 1.6179, 1.1644, 1.8307], device='cuda:1'), covar=tensor([0.1108, 0.0422, 0.0999, 0.0584, 0.0406, 0.0892, 0.0886, 0.0559], device='cuda:1'), in_proj_covar=tensor([0.0277, 0.0277, 0.0315, 0.0237, 0.0224, 0.0305, 0.0276, 0.0248], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 09:58:13,872 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=43991.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 09:58:47,326 INFO [train.py:903] (1/4) Epoch 7, batch 3050, loss[loss=0.2419, simple_loss=0.317, pruned_loss=0.08333, over 19667.00 frames. ], tot_loss[loss=0.2631, simple_loss=0.329, pruned_loss=0.09857, over 3800873.97 frames. ], batch size: 53, lr: 1.19e-02, grad_scale: 8.0 2023-04-01 09:59:24,083 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.748e+02 5.923e+02 7.306e+02 8.819e+02 1.422e+03, threshold=1.461e+03, percent-clipped=0.0 2023-04-01 09:59:50,240 INFO [train.py:903] (1/4) Epoch 7, batch 3100, loss[loss=0.2873, simple_loss=0.3534, pruned_loss=0.1106, over 17396.00 frames. ], tot_loss[loss=0.2617, simple_loss=0.3282, pruned_loss=0.09758, over 3813946.94 frames. ], batch size: 101, lr: 1.19e-02, grad_scale: 8.0 2023-04-01 10:00:17,743 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2169, 1.2795, 1.8430, 1.4503, 2.7039, 2.2276, 2.8324, 1.1632], device='cuda:1'), covar=tensor([0.1863, 0.3287, 0.1720, 0.1468, 0.1170, 0.1525, 0.1277, 0.2970], device='cuda:1'), in_proj_covar=tensor([0.0457, 0.0524, 0.0518, 0.0409, 0.0562, 0.0457, 0.0631, 0.0460], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 10:00:50,925 INFO [train.py:903] (1/4) Epoch 7, batch 3150, loss[loss=0.2838, simple_loss=0.3443, pruned_loss=0.1116, over 17120.00 frames. ], tot_loss[loss=0.2617, simple_loss=0.3281, pruned_loss=0.09763, over 3822599.85 frames. ], batch size: 101, lr: 1.19e-02, grad_scale: 8.0 2023-04-01 10:01:18,559 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 2023-04-01 10:01:26,078 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.269e+02 6.470e+02 8.010e+02 1.018e+03 2.357e+03, threshold=1.602e+03, percent-clipped=4.0 2023-04-01 10:01:28,517 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=44149.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 10:01:51,332 INFO [train.py:903] (1/4) Epoch 7, batch 3200, loss[loss=0.2433, simple_loss=0.3103, pruned_loss=0.08814, over 19593.00 frames. ], tot_loss[loss=0.262, simple_loss=0.3279, pruned_loss=0.09808, over 3816132.39 frames. ], batch size: 52, lr: 1.19e-02, grad_scale: 8.0 2023-04-01 10:02:51,374 INFO [train.py:903] (1/4) Epoch 7, batch 3250, loss[loss=0.2762, simple_loss=0.3386, pruned_loss=0.1069, over 19675.00 frames. ], tot_loss[loss=0.2608, simple_loss=0.3268, pruned_loss=0.09737, over 3831075.28 frames. ], batch size: 60, lr: 1.19e-02, grad_scale: 8.0 2023-04-01 10:03:21,326 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4249, 1.1363, 1.5345, 1.0235, 2.4193, 3.2629, 3.0383, 3.4625], device='cuda:1'), covar=tensor([0.1393, 0.3183, 0.2957, 0.2168, 0.0491, 0.0182, 0.0225, 0.0154], device='cuda:1'), in_proj_covar=tensor([0.0255, 0.0287, 0.0316, 0.0251, 0.0205, 0.0135, 0.0206, 0.0164], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 10:03:27,995 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.707e+02 6.408e+02 8.261e+02 1.024e+03 1.757e+03, threshold=1.652e+03, percent-clipped=4.0 2023-04-01 10:03:28,408 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=44247.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 10:03:49,212 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=44264.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 10:03:52,723 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.8736, 1.9204, 1.9277, 2.9750, 1.8213, 2.6133, 2.4908, 1.8076], device='cuda:1'), covar=tensor([0.2850, 0.2207, 0.1108, 0.1249, 0.2759, 0.0961, 0.2386, 0.2111], device='cuda:1'), in_proj_covar=tensor([0.0697, 0.0697, 0.0597, 0.0844, 0.0720, 0.0611, 0.0735, 0.0641], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 10:03:53,351 INFO [train.py:903] (1/4) Epoch 7, batch 3300, loss[loss=0.2466, simple_loss=0.3178, pruned_loss=0.08768, over 19778.00 frames. ], tot_loss[loss=0.2612, simple_loss=0.327, pruned_loss=0.09773, over 3829470.61 frames. ], batch size: 56, lr: 1.19e-02, grad_scale: 8.0 2023-04-01 10:03:59,180 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=44271.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 10:04:00,253 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 2023-04-01 10:04:00,648 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=44272.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 10:04:57,169 INFO [train.py:903] (1/4) Epoch 7, batch 3350, loss[loss=0.2554, simple_loss=0.3169, pruned_loss=0.09698, over 19046.00 frames. ], tot_loss[loss=0.2612, simple_loss=0.3271, pruned_loss=0.09763, over 3842808.51 frames. ], batch size: 42, lr: 1.19e-02, grad_scale: 8.0 2023-04-01 10:05:00,808 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=44321.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 10:05:32,247 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.423e+02 6.124e+02 7.865e+02 1.014e+03 2.362e+03, threshold=1.573e+03, percent-clipped=3.0 2023-04-01 10:05:41,284 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.85 vs. limit=2.0 2023-04-01 10:05:58,468 INFO [train.py:903] (1/4) Epoch 7, batch 3400, loss[loss=0.2777, simple_loss=0.3341, pruned_loss=0.1107, over 19660.00 frames. ], tot_loss[loss=0.2598, simple_loss=0.3259, pruned_loss=0.09681, over 3849884.50 frames. ], batch size: 53, lr: 1.19e-02, grad_scale: 8.0 2023-04-01 10:06:20,860 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=44386.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 10:07:01,359 INFO [train.py:903] (1/4) Epoch 7, batch 3450, loss[loss=0.2882, simple_loss=0.3471, pruned_loss=0.1146, over 19738.00 frames. ], tot_loss[loss=0.2592, simple_loss=0.3252, pruned_loss=0.09658, over 3850635.83 frames. ], batch size: 63, lr: 1.19e-02, grad_scale: 8.0 2023-04-01 10:07:06,944 WARNING [train.py:1073] (1/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9_rvb from training. Duration: 25.2444375 2023-04-01 10:07:20,143 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.7953, 3.1799, 3.2417, 3.2626, 1.1047, 3.0913, 2.7631, 2.9717], device='cuda:1'), covar=tensor([0.1207, 0.0800, 0.0698, 0.0648, 0.4127, 0.0646, 0.0650, 0.1159], device='cuda:1'), in_proj_covar=tensor([0.0577, 0.0507, 0.0681, 0.0557, 0.0631, 0.0427, 0.0433, 0.0627], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1') 2023-04-01 10:07:25,063 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=44436.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 10:07:39,986 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.210e+02 5.624e+02 7.209e+02 9.228e+02 1.461e+03, threshold=1.442e+03, percent-clipped=0.0 2023-04-01 10:08:03,659 INFO [train.py:903] (1/4) Epoch 7, batch 3500, loss[loss=0.2332, simple_loss=0.3168, pruned_loss=0.07481, over 19778.00 frames. ], tot_loss[loss=0.2586, simple_loss=0.3252, pruned_loss=0.09604, over 3851615.62 frames. ], batch size: 56, lr: 1.19e-02, grad_scale: 4.0 2023-04-01 10:08:21,647 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.8803, 4.9065, 5.6509, 5.6131, 1.6160, 5.3133, 4.6008, 5.1945], device='cuda:1'), covar=tensor([0.1050, 0.0645, 0.0464, 0.0426, 0.4494, 0.0355, 0.0439, 0.0916], device='cuda:1'), in_proj_covar=tensor([0.0576, 0.0501, 0.0679, 0.0556, 0.0630, 0.0425, 0.0434, 0.0626], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1') 2023-04-01 10:09:07,867 INFO [train.py:903] (1/4) Epoch 7, batch 3550, loss[loss=0.2421, simple_loss=0.3079, pruned_loss=0.08817, over 19855.00 frames. ], tot_loss[loss=0.2588, simple_loss=0.3254, pruned_loss=0.09612, over 3850575.70 frames. ], batch size: 52, lr: 1.19e-02, grad_scale: 4.0 2023-04-01 10:09:10,436 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=44520.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 10:09:10,463 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3034, 1.2562, 1.8551, 1.4713, 2.4734, 2.0188, 2.6098, 1.2330], device='cuda:1'), covar=tensor([0.2067, 0.3485, 0.1849, 0.1726, 0.1280, 0.1799, 0.1365, 0.2968], device='cuda:1'), in_proj_covar=tensor([0.0457, 0.0532, 0.0522, 0.0415, 0.0571, 0.0463, 0.0639, 0.0462], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 10:09:40,628 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=44545.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 10:09:40,740 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=44545.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 10:09:43,709 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.445e+02 5.736e+02 7.487e+02 9.426e+02 2.431e+03, threshold=1.497e+03, percent-clipped=5.0 2023-04-01 10:09:57,365 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4739, 2.5502, 1.7283, 1.6042, 2.0713, 1.3220, 1.3144, 1.7646], device='cuda:1'), covar=tensor([0.0846, 0.0486, 0.0917, 0.0595, 0.0455, 0.0996, 0.0679, 0.0465], device='cuda:1'), in_proj_covar=tensor([0.0280, 0.0280, 0.0317, 0.0242, 0.0224, 0.0314, 0.0288, 0.0254], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 10:10:10,024 INFO [train.py:903] (1/4) Epoch 7, batch 3600, loss[loss=0.2113, simple_loss=0.2836, pruned_loss=0.06946, over 19756.00 frames. ], tot_loss[loss=0.2596, simple_loss=0.3261, pruned_loss=0.09649, over 3853510.56 frames. ], batch size: 47, lr: 1.19e-02, grad_scale: 8.0 2023-04-01 10:10:10,286 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=44568.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 10:11:12,091 INFO [train.py:903] (1/4) Epoch 7, batch 3650, loss[loss=0.299, simple_loss=0.3457, pruned_loss=0.1262, over 13246.00 frames. ], tot_loss[loss=0.259, simple_loss=0.3255, pruned_loss=0.09621, over 3846115.62 frames. ], batch size: 135, lr: 1.18e-02, grad_scale: 8.0 2023-04-01 10:11:32,686 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.7665, 2.1052, 2.0871, 1.7680, 4.1841, 1.1272, 2.4110, 4.5349], device='cuda:1'), covar=tensor([0.0361, 0.2314, 0.2153, 0.1672, 0.0626, 0.2520, 0.1230, 0.0244], device='cuda:1'), in_proj_covar=tensor([0.0307, 0.0314, 0.0322, 0.0293, 0.0320, 0.0315, 0.0295, 0.0312], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 10:11:42,322 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=44642.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 10:11:49,798 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.538e+02 6.222e+02 7.769e+02 9.647e+02 2.431e+03, threshold=1.554e+03, percent-clipped=4.0 2023-04-01 10:12:05,237 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.52 vs. limit=2.0 2023-04-01 10:12:12,966 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=44667.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 10:12:13,735 INFO [train.py:903] (1/4) Epoch 7, batch 3700, loss[loss=0.2407, simple_loss=0.3133, pruned_loss=0.08403, over 19779.00 frames. ], tot_loss[loss=0.2607, simple_loss=0.327, pruned_loss=0.09719, over 3821598.73 frames. ], batch size: 54, lr: 1.18e-02, grad_scale: 8.0 2023-04-01 10:12:45,589 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=44692.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 10:13:16,193 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=44717.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 10:13:16,968 INFO [train.py:903] (1/4) Epoch 7, batch 3750, loss[loss=0.2455, simple_loss=0.3073, pruned_loss=0.09182, over 19734.00 frames. ], tot_loss[loss=0.2614, simple_loss=0.3274, pruned_loss=0.09765, over 3818299.56 frames. ], batch size: 51, lr: 1.18e-02, grad_scale: 8.0 2023-04-01 10:13:53,509 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.417e+02 6.149e+02 7.382e+02 9.468e+02 1.650e+03, threshold=1.476e+03, percent-clipped=2.0 2023-04-01 10:14:18,148 INFO [train.py:903] (1/4) Epoch 7, batch 3800, loss[loss=0.2669, simple_loss=0.3121, pruned_loss=0.1108, over 19763.00 frames. ], tot_loss[loss=0.2633, simple_loss=0.3285, pruned_loss=0.09906, over 3797058.66 frames. ], batch size: 48, lr: 1.18e-02, grad_scale: 8.0 2023-04-01 10:14:23,282 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 2023-04-01 10:14:50,714 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9_rvb from training. Duration: 29.1166875 2023-04-01 10:15:04,142 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 2023-04-01 10:15:19,321 INFO [train.py:903] (1/4) Epoch 7, batch 3850, loss[loss=0.2918, simple_loss=0.358, pruned_loss=0.1128, over 19445.00 frames. ], tot_loss[loss=0.2627, simple_loss=0.3277, pruned_loss=0.09883, over 3800185.08 frames. ], batch size: 64, lr: 1.18e-02, grad_scale: 8.0 2023-04-01 10:15:57,070 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.826e+02 6.658e+02 8.285e+02 1.095e+03 3.075e+03, threshold=1.657e+03, percent-clipped=10.0 2023-04-01 10:16:20,989 INFO [train.py:903] (1/4) Epoch 7, batch 3900, loss[loss=0.3232, simple_loss=0.3707, pruned_loss=0.1379, over 19321.00 frames. ], tot_loss[loss=0.2624, simple_loss=0.3276, pruned_loss=0.09859, over 3808328.51 frames. ], batch size: 66, lr: 1.18e-02, grad_scale: 8.0 2023-04-01 10:16:48,352 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=44889.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 10:16:56,963 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 2023-04-01 10:17:15,631 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=44912.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 10:17:24,389 INFO [train.py:903] (1/4) Epoch 7, batch 3950, loss[loss=0.2334, simple_loss=0.3134, pruned_loss=0.07676, over 19657.00 frames. ], tot_loss[loss=0.261, simple_loss=0.3265, pruned_loss=0.09771, over 3815125.86 frames. ], batch size: 53, lr: 1.18e-02, grad_scale: 8.0 2023-04-01 10:17:29,145 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 2023-04-01 10:17:58,837 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5267, 2.4863, 1.7390, 1.6173, 2.1823, 1.2801, 1.2879, 1.7406], device='cuda:1'), covar=tensor([0.0775, 0.0568, 0.0956, 0.0617, 0.0493, 0.1053, 0.0741, 0.0489], device='cuda:1'), in_proj_covar=tensor([0.0283, 0.0280, 0.0314, 0.0240, 0.0225, 0.0309, 0.0285, 0.0253], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 10:18:00,740 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.332e+02 5.655e+02 7.377e+02 9.527e+02 2.304e+03, threshold=1.475e+03, percent-clipped=3.0 2023-04-01 10:18:26,678 INFO [train.py:903] (1/4) Epoch 7, batch 4000, loss[loss=0.2549, simple_loss=0.3249, pruned_loss=0.09247, over 19667.00 frames. ], tot_loss[loss=0.2598, simple_loss=0.3256, pruned_loss=0.09697, over 3809394.26 frames. ], batch size: 53, lr: 1.18e-02, grad_scale: 8.0 2023-04-01 10:19:12,015 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=45004.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 10:19:15,224 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1_rvb from training. Duration: 0.7545625 2023-04-01 10:19:27,779 INFO [train.py:903] (1/4) Epoch 7, batch 4050, loss[loss=0.2531, simple_loss=0.3263, pruned_loss=0.08995, over 19651.00 frames. ], tot_loss[loss=0.2599, simple_loss=0.3256, pruned_loss=0.0971, over 3788049.84 frames. ], batch size: 58, lr: 1.18e-02, grad_scale: 8.0 2023-04-01 10:19:39,284 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=45027.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 10:20:05,076 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.186e+02 6.019e+02 6.941e+02 8.524e+02 1.564e+03, threshold=1.388e+03, percent-clipped=1.0 2023-04-01 10:20:28,972 INFO [train.py:903] (1/4) Epoch 7, batch 4100, loss[loss=0.2286, simple_loss=0.3124, pruned_loss=0.07242, over 19679.00 frames. ], tot_loss[loss=0.2598, simple_loss=0.3259, pruned_loss=0.09685, over 3788096.87 frames. ], batch size: 59, lr: 1.18e-02, grad_scale: 8.0 2023-04-01 10:20:36,277 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=45074.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 10:20:47,565 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.7589, 4.2891, 2.4409, 3.9123, 0.9870, 3.9630, 4.0717, 4.1985], device='cuda:1'), covar=tensor([0.0649, 0.1107, 0.2209, 0.0677, 0.3941, 0.0789, 0.0733, 0.0932], device='cuda:1'), in_proj_covar=tensor([0.0384, 0.0334, 0.0397, 0.0297, 0.0361, 0.0324, 0.0309, 0.0343], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-01 10:21:05,605 WARNING [train.py:1073] (1/4) Exclude cut with ID 497-129325-0061-9566_sp1.1_rvb from training. Duration: 0.97725 2023-04-01 10:21:07,155 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=45099.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 10:21:30,604 INFO [train.py:903] (1/4) Epoch 7, batch 4150, loss[loss=0.2133, simple_loss=0.2892, pruned_loss=0.06868, over 19613.00 frames. ], tot_loss[loss=0.2604, simple_loss=0.3263, pruned_loss=0.0972, over 3799153.60 frames. ], batch size: 50, lr: 1.18e-02, grad_scale: 8.0 2023-04-01 10:22:07,171 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.699e+02 6.430e+02 8.235e+02 1.004e+03 2.607e+03, threshold=1.647e+03, percent-clipped=6.0 2023-04-01 10:22:32,972 INFO [train.py:903] (1/4) Epoch 7, batch 4200, loss[loss=0.2888, simple_loss=0.3533, pruned_loss=0.1122, over 19303.00 frames. ], tot_loss[loss=0.2598, simple_loss=0.326, pruned_loss=0.09675, over 3801446.97 frames. ], batch size: 66, lr: 1.18e-02, grad_scale: 8.0 2023-04-01 10:22:38,352 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 2023-04-01 10:23:13,784 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.12 vs. limit=5.0 2023-04-01 10:23:33,351 INFO [train.py:903] (1/4) Epoch 7, batch 4250, loss[loss=0.2483, simple_loss=0.3184, pruned_loss=0.08905, over 19776.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.3268, pruned_loss=0.09745, over 3781540.04 frames. ], batch size: 54, lr: 1.18e-02, grad_scale: 8.0 2023-04-01 10:23:49,642 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9_rvb from training. Duration: 29.816625 2023-04-01 10:24:01,342 WARNING [train.py:1073] (1/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 2023-04-01 10:24:12,485 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.934e+02 6.110e+02 8.094e+02 1.072e+03 2.309e+03, threshold=1.619e+03, percent-clipped=5.0 2023-04-01 10:24:27,382 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=45260.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 10:24:37,369 INFO [train.py:903] (1/4) Epoch 7, batch 4300, loss[loss=0.2584, simple_loss=0.33, pruned_loss=0.09342, over 19755.00 frames. ], tot_loss[loss=0.2602, simple_loss=0.326, pruned_loss=0.09713, over 3790167.17 frames. ], batch size: 63, lr: 1.18e-02, grad_scale: 8.0 2023-04-01 10:24:57,140 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=45283.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 10:24:59,372 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=45285.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 10:25:02,933 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.42 vs. limit=2.0 2023-04-01 10:25:27,055 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=45308.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 10:25:32,136 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 2023-04-01 10:25:39,640 INFO [train.py:903] (1/4) Epoch 7, batch 4350, loss[loss=0.2492, simple_loss=0.3075, pruned_loss=0.0954, over 19756.00 frames. ], tot_loss[loss=0.2624, simple_loss=0.3283, pruned_loss=0.09826, over 3784683.15 frames. ], batch size: 47, lr: 1.18e-02, grad_scale: 8.0 2023-04-01 10:26:16,053 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.214e+02 6.479e+02 7.493e+02 1.002e+03 2.532e+03, threshold=1.499e+03, percent-clipped=5.0 2023-04-01 10:26:39,970 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=45366.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 10:26:42,860 INFO [train.py:903] (1/4) Epoch 7, batch 4400, loss[loss=0.2399, simple_loss=0.3141, pruned_loss=0.08291, over 19532.00 frames. ], tot_loss[loss=0.2604, simple_loss=0.3269, pruned_loss=0.09699, over 3804140.95 frames. ], batch size: 56, lr: 1.17e-02, grad_scale: 8.0 2023-04-01 10:27:06,062 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_rvb from training. Duration: 25.285 2023-04-01 10:27:14,913 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 2023-04-01 10:27:36,669 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6729, 1.9168, 2.3372, 2.1181, 3.1498, 3.4622, 3.6153, 3.8936], device='cuda:1'), covar=tensor([0.1327, 0.2356, 0.2092, 0.1572, 0.0736, 0.0473, 0.0181, 0.0146], device='cuda:1'), in_proj_covar=tensor([0.0254, 0.0284, 0.0312, 0.0247, 0.0203, 0.0133, 0.0202, 0.0162], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 10:27:36,744 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.4213, 2.3599, 1.9116, 1.9268, 1.7657, 2.1059, 1.2021, 2.0009], device='cuda:1'), covar=tensor([0.0281, 0.0322, 0.0274, 0.0410, 0.0506, 0.0485, 0.0529, 0.0417], device='cuda:1'), in_proj_covar=tensor([0.0309, 0.0310, 0.0299, 0.0331, 0.0405, 0.0325, 0.0287, 0.0310], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 10:27:44,672 INFO [train.py:903] (1/4) Epoch 7, batch 4450, loss[loss=0.2479, simple_loss=0.3232, pruned_loss=0.08631, over 19356.00 frames. ], tot_loss[loss=0.2597, simple_loss=0.3263, pruned_loss=0.09657, over 3806375.09 frames. ], batch size: 70, lr: 1.17e-02, grad_scale: 4.0 2023-04-01 10:27:44,834 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=45418.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 10:28:16,050 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=45443.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 10:28:22,921 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.630e+02 6.151e+02 7.769e+02 9.586e+02 4.695e+03, threshold=1.554e+03, percent-clipped=4.0 2023-04-01 10:28:46,110 INFO [train.py:903] (1/4) Epoch 7, batch 4500, loss[loss=0.2422, simple_loss=0.2993, pruned_loss=0.09252, over 19735.00 frames. ], tot_loss[loss=0.2599, simple_loss=0.3261, pruned_loss=0.09688, over 3808284.27 frames. ], batch size: 51, lr: 1.17e-02, grad_scale: 4.0 2023-04-01 10:29:48,541 INFO [train.py:903] (1/4) Epoch 7, batch 4550, loss[loss=0.2874, simple_loss=0.3525, pruned_loss=0.1111, over 19702.00 frames. ], tot_loss[loss=0.2606, simple_loss=0.327, pruned_loss=0.0971, over 3827099.06 frames. ], batch size: 59, lr: 1.17e-02, grad_scale: 4.0 2023-04-01 10:30:00,887 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9_rvb from training. Duration: 28.72225 2023-04-01 10:30:09,105 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=45533.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 10:30:17,530 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 2023-04-01 10:30:23,717 WARNING [train.py:1073] (1/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 2023-04-01 10:30:27,034 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.783e+02 6.251e+02 7.662e+02 9.726e+02 2.453e+03, threshold=1.532e+03, percent-clipped=6.0 2023-04-01 10:30:38,163 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=45558.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 10:30:42,535 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.5658, 4.1315, 2.3774, 3.6244, 1.0937, 3.8168, 3.8393, 3.8365], device='cuda:1'), covar=tensor([0.0606, 0.0962, 0.2063, 0.0644, 0.3647, 0.0734, 0.0660, 0.0952], device='cuda:1'), in_proj_covar=tensor([0.0378, 0.0328, 0.0386, 0.0291, 0.0353, 0.0317, 0.0304, 0.0335], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-01 10:30:51,049 INFO [train.py:903] (1/4) Epoch 7, batch 4600, loss[loss=0.3537, simple_loss=0.3887, pruned_loss=0.1594, over 13319.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.3282, pruned_loss=0.09804, over 3815576.41 frames. ], batch size: 136, lr: 1.17e-02, grad_scale: 4.0 2023-04-01 10:31:21,662 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9340, 2.0484, 2.1455, 2.9418, 1.8432, 2.6429, 2.6615, 2.0594], device='cuda:1'), covar=tensor([0.2908, 0.2355, 0.1159, 0.1406, 0.2832, 0.1089, 0.2297, 0.1974], device='cuda:1'), in_proj_covar=tensor([0.0711, 0.0709, 0.0600, 0.0851, 0.0728, 0.0622, 0.0743, 0.0646], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 10:31:48,241 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.77 vs. limit=2.0 2023-04-01 10:31:54,302 INFO [train.py:903] (1/4) Epoch 7, batch 4650, loss[loss=0.2309, simple_loss=0.2968, pruned_loss=0.08247, over 19464.00 frames. ], tot_loss[loss=0.2614, simple_loss=0.3276, pruned_loss=0.09757, over 3820340.99 frames. ], batch size: 49, lr: 1.17e-02, grad_scale: 4.0 2023-04-01 10:31:57,360 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 2023-04-01 10:32:11,878 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_rvb from training. Duration: 0.92 2023-04-01 10:32:24,511 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_rvb from training. Duration: 0.83 2023-04-01 10:32:33,275 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.198e+02 5.859e+02 7.403e+02 8.847e+02 2.429e+03, threshold=1.481e+03, percent-clipped=2.0 2023-04-01 10:32:55,951 INFO [train.py:903] (1/4) Epoch 7, batch 4700, loss[loss=0.2476, simple_loss=0.3068, pruned_loss=0.09415, over 19757.00 frames. ], tot_loss[loss=0.2603, simple_loss=0.3265, pruned_loss=0.09708, over 3816049.99 frames. ], batch size: 48, lr: 1.17e-02, grad_scale: 4.0 2023-04-01 10:33:09,342 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 2023-04-01 10:33:18,882 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9_rvb from training. Duration: 0.92225 2023-04-01 10:33:47,801 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=45710.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 10:33:58,623 INFO [train.py:903] (1/4) Epoch 7, batch 4750, loss[loss=0.2977, simple_loss=0.3655, pruned_loss=0.1149, over 19116.00 frames. ], tot_loss[loss=0.2606, simple_loss=0.3268, pruned_loss=0.0972, over 3829137.43 frames. ], batch size: 69, lr: 1.17e-02, grad_scale: 4.0 2023-04-01 10:34:01,195 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=45720.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 10:34:11,843 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.1145, 2.1223, 1.6194, 1.6344, 1.3639, 1.5862, 0.4712, 0.9960], device='cuda:1'), covar=tensor([0.0528, 0.0454, 0.0375, 0.0538, 0.0944, 0.0702, 0.0752, 0.0770], device='cuda:1'), in_proj_covar=tensor([0.0311, 0.0313, 0.0310, 0.0332, 0.0404, 0.0327, 0.0288, 0.0316], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 10:34:36,016 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.623e+02 6.576e+02 8.008e+02 9.322e+02 2.457e+03, threshold=1.602e+03, percent-clipped=6.0 2023-04-01 10:35:01,462 INFO [train.py:903] (1/4) Epoch 7, batch 4800, loss[loss=0.3598, simple_loss=0.3953, pruned_loss=0.1621, over 13702.00 frames. ], tot_loss[loss=0.2607, simple_loss=0.327, pruned_loss=0.09715, over 3825639.46 frames. ], batch size: 135, lr: 1.17e-02, grad_scale: 8.0 2023-04-01 10:35:14,194 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 2023-04-01 10:35:26,476 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=45789.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 10:35:29,632 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=45792.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 10:35:30,830 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.0599, 4.3719, 4.6905, 4.7152, 1.5636, 4.3464, 3.9414, 4.3037], device='cuda:1'), covar=tensor([0.1051, 0.0678, 0.0513, 0.0418, 0.4578, 0.0420, 0.0469, 0.1013], device='cuda:1'), in_proj_covar=tensor([0.0574, 0.0504, 0.0681, 0.0557, 0.0633, 0.0428, 0.0433, 0.0632], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1') 2023-04-01 10:35:37,411 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5502, 1.3065, 1.3965, 1.8770, 1.2969, 1.6719, 1.7272, 1.5808], device='cuda:1'), covar=tensor([0.0713, 0.0931, 0.0948, 0.0670, 0.0864, 0.0715, 0.0787, 0.0669], device='cuda:1'), in_proj_covar=tensor([0.0220, 0.0237, 0.0233, 0.0264, 0.0253, 0.0220, 0.0217, 0.0215], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1') 2023-04-01 10:35:58,119 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=45814.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 10:35:58,149 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=45814.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 10:36:02,115 INFO [train.py:903] (1/4) Epoch 7, batch 4850, loss[loss=0.2568, simple_loss=0.3345, pruned_loss=0.08954, over 18662.00 frames. ], tot_loss[loss=0.2621, simple_loss=0.3278, pruned_loss=0.09823, over 3825328.58 frames. ], batch size: 74, lr: 1.17e-02, grad_scale: 8.0 2023-04-01 10:36:10,586 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=45825.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 10:36:23,725 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 2023-04-01 10:36:25,129 WARNING [train.py:1073] (1/4) Exclude cut with ID 774-127930-0014-48411_sp1.1_rvb from training. Duration: 0.95 2023-04-01 10:36:27,414 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=45839.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 10:36:40,692 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.989e+02 6.823e+02 8.718e+02 1.115e+03 2.265e+03, threshold=1.744e+03, percent-clipped=6.0 2023-04-01 10:36:48,811 WARNING [train.py:1073] (1/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1_rvb from training. Duration: 0.9409375 2023-04-01 10:36:54,261 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 2023-04-01 10:36:54,291 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 2023-04-01 10:37:03,556 INFO [train.py:903] (1/4) Epoch 7, batch 4900, loss[loss=0.2295, simple_loss=0.3052, pruned_loss=0.07687, over 18693.00 frames. ], tot_loss[loss=0.2617, simple_loss=0.3274, pruned_loss=0.09799, over 3829462.74 frames. ], batch size: 74, lr: 1.17e-02, grad_scale: 8.0 2023-04-01 10:37:04,793 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1_rvb from training. Duration: 25.3818125 2023-04-01 10:37:14,476 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=45877.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 10:37:26,127 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 2023-04-01 10:38:05,106 INFO [train.py:903] (1/4) Epoch 7, batch 4950, loss[loss=0.2233, simple_loss=0.2972, pruned_loss=0.07476, over 19384.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.3274, pruned_loss=0.09721, over 3832661.65 frames. ], batch size: 48, lr: 1.17e-02, grad_scale: 8.0 2023-04-01 10:38:24,756 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 2023-04-01 10:38:44,274 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.370e+02 6.277e+02 7.738e+02 9.356e+02 2.304e+03, threshold=1.548e+03, percent-clipped=1.0 2023-04-01 10:38:44,961 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=4.95 vs. limit=5.0 2023-04-01 10:38:47,739 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 2023-04-01 10:39:09,683 INFO [train.py:903] (1/4) Epoch 7, batch 5000, loss[loss=0.2361, simple_loss=0.2951, pruned_loss=0.08856, over 19760.00 frames. ], tot_loss[loss=0.2613, simple_loss=0.3274, pruned_loss=0.09758, over 3826389.18 frames. ], batch size: 46, lr: 1.17e-02, grad_scale: 8.0 2023-04-01 10:39:18,622 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1858, 0.9395, 1.1562, 1.8150, 1.2578, 1.2613, 1.5003, 1.3171], device='cuda:1'), covar=tensor([0.1098, 0.1693, 0.1328, 0.0826, 0.1038, 0.1336, 0.1098, 0.1038], device='cuda:1'), in_proj_covar=tensor([0.0219, 0.0235, 0.0234, 0.0264, 0.0252, 0.0217, 0.0217, 0.0213], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1') 2023-04-01 10:39:20,567 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_rvb from training. Duration: 27.14 2023-04-01 10:39:30,942 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 2023-04-01 10:40:12,336 INFO [train.py:903] (1/4) Epoch 7, batch 5050, loss[loss=0.2216, simple_loss=0.293, pruned_loss=0.07504, over 19593.00 frames. ], tot_loss[loss=0.2617, simple_loss=0.3281, pruned_loss=0.09763, over 3834239.73 frames. ], batch size: 50, lr: 1.17e-02, grad_scale: 8.0 2023-04-01 10:40:49,615 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9_rvb from training. Duration: 26.62775 2023-04-01 10:40:51,923 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.897e+02 6.208e+02 7.894e+02 9.516e+02 2.052e+03, threshold=1.579e+03, percent-clipped=3.0 2023-04-01 10:41:08,571 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=46064.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 10:41:13,141 INFO [train.py:903] (1/4) Epoch 7, batch 5100, loss[loss=0.2623, simple_loss=0.335, pruned_loss=0.09474, over 19772.00 frames. ], tot_loss[loss=0.2602, simple_loss=0.3269, pruned_loss=0.09672, over 3850091.90 frames. ], batch size: 63, lr: 1.17e-02, grad_scale: 4.0 2023-04-01 10:41:24,901 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 2023-04-01 10:41:28,160 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_rvb from training. Duration: 25.035 2023-04-01 10:41:28,598 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=46081.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 10:41:32,533 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_rvb from training. Duration: 27.92 2023-04-01 10:42:00,040 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=46106.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 10:42:08,080 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3599, 1.2608, 1.6886, 1.5706, 3.2238, 4.4425, 4.3994, 4.8562], device='cuda:1'), covar=tensor([0.1426, 0.3078, 0.2861, 0.1739, 0.0393, 0.0178, 0.0137, 0.0084], device='cuda:1'), in_proj_covar=tensor([0.0254, 0.0284, 0.0317, 0.0244, 0.0204, 0.0133, 0.0203, 0.0164], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 10:42:13,343 INFO [train.py:903] (1/4) Epoch 7, batch 5150, loss[loss=0.2217, simple_loss=0.2875, pruned_loss=0.07797, over 19701.00 frames. ], tot_loss[loss=0.2604, simple_loss=0.3271, pruned_loss=0.09688, over 3845123.35 frames. ], batch size: 45, lr: 1.17e-02, grad_scale: 4.0 2023-04-01 10:42:27,452 WARNING [train.py:1073] (1/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 2023-04-01 10:42:37,818 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=46136.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 10:42:53,642 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.315e+02 6.102e+02 7.228e+02 8.853e+02 1.806e+03, threshold=1.446e+03, percent-clipped=2.0 2023-04-01 10:43:00,551 WARNING [train.py:1073] (1/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1_rvb from training. Duration: 0.9681875 2023-04-01 10:43:15,665 INFO [train.py:903] (1/4) Epoch 7, batch 5200, loss[loss=0.2759, simple_loss=0.3341, pruned_loss=0.1088, over 19390.00 frames. ], tot_loss[loss=0.2592, simple_loss=0.3259, pruned_loss=0.09624, over 3849103.93 frames. ], batch size: 48, lr: 1.16e-02, grad_scale: 8.0 2023-04-01 10:43:30,884 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=46179.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 10:43:31,710 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 2023-04-01 10:44:18,556 WARNING [train.py:1073] (1/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 2023-04-01 10:44:19,699 INFO [train.py:903] (1/4) Epoch 7, batch 5250, loss[loss=0.2669, simple_loss=0.3397, pruned_loss=0.097, over 18194.00 frames. ], tot_loss[loss=0.2595, simple_loss=0.3264, pruned_loss=0.09629, over 3842319.61 frames. ], batch size: 83, lr: 1.16e-02, grad_scale: 8.0 2023-04-01 10:44:23,434 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=46221.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 10:44:52,979 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.1004, 1.6042, 1.6906, 2.0119, 2.0169, 1.8024, 1.8625, 1.9675], device='cuda:1'), covar=tensor([0.0790, 0.1548, 0.1176, 0.0909, 0.1040, 0.0468, 0.0878, 0.0565], device='cuda:1'), in_proj_covar=tensor([0.0244, 0.0356, 0.0285, 0.0240, 0.0299, 0.0242, 0.0272, 0.0228], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 10:44:58,427 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.945e+02 5.998e+02 7.245e+02 8.826e+02 1.462e+03, threshold=1.449e+03, percent-clipped=1.0 2023-04-01 10:45:00,901 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=46251.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 10:45:21,289 INFO [train.py:903] (1/4) Epoch 7, batch 5300, loss[loss=0.2586, simple_loss=0.3254, pruned_loss=0.09595, over 18996.00 frames. ], tot_loss[loss=0.2597, simple_loss=0.3263, pruned_loss=0.09653, over 3833978.91 frames. ], batch size: 42, lr: 1.16e-02, grad_scale: 8.0 2023-04-01 10:45:34,735 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.91 vs. limit=2.0 2023-04-01 10:45:39,519 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 2023-04-01 10:45:42,162 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.8272, 1.5458, 1.4979, 1.9744, 1.7230, 2.2264, 2.2541, 1.9144], device='cuda:1'), covar=tensor([0.0752, 0.0928, 0.1025, 0.0922, 0.0881, 0.0622, 0.0838, 0.0616], device='cuda:1'), in_proj_covar=tensor([0.0218, 0.0234, 0.0234, 0.0263, 0.0253, 0.0219, 0.0216, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1') 2023-04-01 10:45:53,397 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=46294.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 10:46:22,053 INFO [train.py:903] (1/4) Epoch 7, batch 5350, loss[loss=0.2844, simple_loss=0.3497, pruned_loss=0.1096, over 19667.00 frames. ], tot_loss[loss=0.2596, simple_loss=0.3259, pruned_loss=0.09663, over 3840263.13 frames. ], batch size: 58, lr: 1.16e-02, grad_scale: 8.0 2023-04-01 10:46:46,801 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=46336.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 10:47:00,334 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9_rvb from training. Duration: 26.438875 2023-04-01 10:47:03,770 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.963e+02 5.681e+02 7.157e+02 9.578e+02 3.754e+03, threshold=1.431e+03, percent-clipped=4.0 2023-04-01 10:47:12,091 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.7218, 1.7443, 1.4008, 1.3088, 1.2420, 1.4490, 0.1083, 0.5542], device='cuda:1'), covar=tensor([0.0357, 0.0342, 0.0222, 0.0305, 0.0749, 0.0326, 0.0616, 0.0617], device='cuda:1'), in_proj_covar=tensor([0.0308, 0.0307, 0.0304, 0.0328, 0.0397, 0.0317, 0.0287, 0.0310], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 10:47:26,532 INFO [train.py:903] (1/4) Epoch 7, batch 5400, loss[loss=0.2588, simple_loss=0.3249, pruned_loss=0.09639, over 19478.00 frames. ], tot_loss[loss=0.2589, simple_loss=0.3255, pruned_loss=0.09608, over 3837575.80 frames. ], batch size: 49, lr: 1.16e-02, grad_scale: 8.0 2023-04-01 10:47:37,299 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.91 vs. limit=5.0 2023-04-01 10:48:29,979 INFO [train.py:903] (1/4) Epoch 7, batch 5450, loss[loss=0.225, simple_loss=0.3021, pruned_loss=0.07392, over 19406.00 frames. ], tot_loss[loss=0.2614, simple_loss=0.3272, pruned_loss=0.09777, over 3833875.76 frames. ], batch size: 48, lr: 1.16e-02, grad_scale: 4.0 2023-04-01 10:48:35,955 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.0005, 4.4827, 4.7475, 4.6701, 1.5656, 4.3430, 3.8552, 4.3792], device='cuda:1'), covar=tensor([0.1174, 0.0667, 0.0486, 0.0453, 0.4708, 0.0407, 0.0479, 0.0975], device='cuda:1'), in_proj_covar=tensor([0.0578, 0.0511, 0.0691, 0.0573, 0.0642, 0.0438, 0.0437, 0.0644], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1') 2023-04-01 10:48:49,839 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=46435.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 10:49:10,723 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.600e+02 7.102e+02 8.361e+02 1.036e+03 2.875e+03, threshold=1.672e+03, percent-clipped=7.0 2023-04-01 10:49:23,131 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=46460.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 10:49:31,706 INFO [train.py:903] (1/4) Epoch 7, batch 5500, loss[loss=0.2188, simple_loss=0.2903, pruned_loss=0.07363, over 19752.00 frames. ], tot_loss[loss=0.2613, simple_loss=0.3274, pruned_loss=0.09761, over 3838540.24 frames. ], batch size: 47, lr: 1.16e-02, grad_scale: 4.0 2023-04-01 10:49:58,195 WARNING [train.py:1073] (1/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9_rvb from training. Duration: 27.47775 2023-04-01 10:50:20,852 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=46507.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 10:50:33,176 INFO [train.py:903] (1/4) Epoch 7, batch 5550, loss[loss=0.2456, simple_loss=0.3195, pruned_loss=0.08582, over 19784.00 frames. ], tot_loss[loss=0.2596, simple_loss=0.326, pruned_loss=0.09664, over 3838841.41 frames. ], batch size: 56, lr: 1.16e-02, grad_scale: 4.0 2023-04-01 10:50:43,592 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 2023-04-01 10:50:51,410 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=46532.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 10:51:02,849 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 2023-04-01 10:51:15,056 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.541e+02 5.815e+02 7.044e+02 9.146e+02 3.032e+03, threshold=1.409e+03, percent-clipped=3.0 2023-04-01 10:51:31,072 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 2023-04-01 10:51:35,753 INFO [train.py:903] (1/4) Epoch 7, batch 5600, loss[loss=0.2707, simple_loss=0.3417, pruned_loss=0.09981, over 19471.00 frames. ], tot_loss[loss=0.2573, simple_loss=0.3243, pruned_loss=0.09515, over 3848351.34 frames. ], batch size: 64, lr: 1.16e-02, grad_scale: 8.0 2023-04-01 10:52:06,682 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=46592.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 10:52:32,951 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=46613.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 10:52:38,683 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=46617.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 10:52:40,322 INFO [train.py:903] (1/4) Epoch 7, batch 5650, loss[loss=0.2585, simple_loss=0.3311, pruned_loss=0.09291, over 19667.00 frames. ], tot_loss[loss=0.2586, simple_loss=0.3253, pruned_loss=0.09593, over 3852176.34 frames. ], batch size: 60, lr: 1.16e-02, grad_scale: 8.0 2023-04-01 10:52:42,904 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=46620.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 10:53:03,526 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=46638.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 10:53:20,361 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.719e+02 5.976e+02 7.674e+02 9.559e+02 1.706e+03, threshold=1.535e+03, percent-clipped=4.0 2023-04-01 10:53:28,305 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9_rvb from training. Duration: 28.0944375 2023-04-01 10:53:42,080 INFO [train.py:903] (1/4) Epoch 7, batch 5700, loss[loss=0.2767, simple_loss=0.349, pruned_loss=0.1022, over 19671.00 frames. ], tot_loss[loss=0.2585, simple_loss=0.3254, pruned_loss=0.09586, over 3845239.43 frames. ], batch size: 55, lr: 1.16e-02, grad_scale: 8.0 2023-04-01 10:53:55,154 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4079, 2.2886, 1.6682, 1.4102, 2.0768, 1.1775, 1.2972, 1.6066], device='cuda:1'), covar=tensor([0.0824, 0.0472, 0.0884, 0.0625, 0.0443, 0.1042, 0.0655, 0.0498], device='cuda:1'), in_proj_covar=tensor([0.0284, 0.0288, 0.0320, 0.0243, 0.0227, 0.0317, 0.0287, 0.0255], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 10:54:43,275 INFO [train.py:903] (1/4) Epoch 7, batch 5750, loss[loss=0.2632, simple_loss=0.3392, pruned_loss=0.09358, over 19370.00 frames. ], tot_loss[loss=0.2584, simple_loss=0.3254, pruned_loss=0.09567, over 3852563.31 frames. ], batch size: 70, lr: 1.16e-02, grad_scale: 8.0 2023-04-01 10:54:45,352 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.83 vs. limit=5.0 2023-04-01 10:54:45,641 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9_rvb from training. Duration: 33.038875 2023-04-01 10:54:55,169 WARNING [train.py:1073] (1/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 2023-04-01 10:54:59,797 WARNING [train.py:1073] (1/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 2023-04-01 10:55:25,495 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.168e+02 6.148e+02 7.021e+02 8.191e+02 1.564e+03, threshold=1.404e+03, percent-clipped=1.0 2023-04-01 10:55:28,123 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=46753.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 10:55:43,512 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.8975, 4.4883, 2.6767, 3.9412, 1.3496, 4.1245, 4.1387, 4.3275], device='cuda:1'), covar=tensor([0.0501, 0.0948, 0.1946, 0.0673, 0.3737, 0.0627, 0.0718, 0.0702], device='cuda:1'), in_proj_covar=tensor([0.0389, 0.0341, 0.0402, 0.0298, 0.0367, 0.0323, 0.0318, 0.0350], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-01 10:55:45,401 INFO [train.py:903] (1/4) Epoch 7, batch 5800, loss[loss=0.2257, simple_loss=0.2907, pruned_loss=0.08038, over 19063.00 frames. ], tot_loss[loss=0.2573, simple_loss=0.3247, pruned_loss=0.09493, over 3849691.04 frames. ], batch size: 42, lr: 1.16e-02, grad_scale: 8.0 2023-04-01 10:56:49,191 INFO [train.py:903] (1/4) Epoch 7, batch 5850, loss[loss=0.1979, simple_loss=0.2734, pruned_loss=0.0612, over 19136.00 frames. ], tot_loss[loss=0.2572, simple_loss=0.3244, pruned_loss=0.09502, over 3844807.40 frames. ], batch size: 42, lr: 1.16e-02, grad_scale: 8.0 2023-04-01 10:57:02,083 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=46828.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 10:57:29,389 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.050e+02 5.967e+02 7.374e+02 9.303e+02 2.879e+03, threshold=1.475e+03, percent-clipped=6.0 2023-04-01 10:57:51,529 INFO [train.py:903] (1/4) Epoch 7, batch 5900, loss[loss=0.199, simple_loss=0.2725, pruned_loss=0.06272, over 19429.00 frames. ], tot_loss[loss=0.2576, simple_loss=0.3247, pruned_loss=0.09523, over 3840492.24 frames. ], batch size: 48, lr: 1.16e-02, grad_scale: 8.0 2023-04-01 10:57:57,281 WARNING [train.py:1073] (1/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 2023-04-01 10:58:16,962 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9_rvb from training. Duration: 27.511125 2023-04-01 10:58:52,945 INFO [train.py:903] (1/4) Epoch 7, batch 5950, loss[loss=0.207, simple_loss=0.2798, pruned_loss=0.06708, over 19383.00 frames. ], tot_loss[loss=0.258, simple_loss=0.3252, pruned_loss=0.09543, over 3829509.50 frames. ], batch size: 48, lr: 1.16e-02, grad_scale: 8.0 2023-04-01 10:59:34,060 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.877e+02 6.644e+02 7.805e+02 9.690e+02 1.794e+03, threshold=1.561e+03, percent-clipped=4.0 2023-04-01 10:59:40,934 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=46957.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 10:59:48,547 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=46964.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 10:59:52,803 INFO [train.py:903] (1/4) Epoch 7, batch 6000, loss[loss=0.2822, simple_loss=0.338, pruned_loss=0.1133, over 17476.00 frames. ], tot_loss[loss=0.2598, simple_loss=0.326, pruned_loss=0.09682, over 3829300.31 frames. ], batch size: 101, lr: 1.15e-02, grad_scale: 8.0 2023-04-01 10:59:52,804 INFO [train.py:928] (1/4) Computing validation loss 2023-04-01 11:00:05,284 INFO [train.py:937] (1/4) Epoch 7, validation: loss=0.1903, simple_loss=0.2902, pruned_loss=0.04516, over 944034.00 frames. 2023-04-01 11:00:05,285 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18649MB 2023-04-01 11:00:57,533 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=47009.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 11:01:09,015 INFO [train.py:903] (1/4) Epoch 7, batch 6050, loss[loss=0.2677, simple_loss=0.3362, pruned_loss=0.09954, over 19653.00 frames. ], tot_loss[loss=0.2599, simple_loss=0.3265, pruned_loss=0.09671, over 3820078.92 frames. ], batch size: 58, lr: 1.15e-02, grad_scale: 8.0 2023-04-01 11:01:30,886 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=47034.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 11:01:42,123 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.2263, 2.9200, 2.1033, 2.6787, 0.9989, 2.6800, 2.6358, 2.7785], device='cuda:1'), covar=tensor([0.1085, 0.1364, 0.1993, 0.0921, 0.3402, 0.1063, 0.1067, 0.1268], device='cuda:1'), in_proj_covar=tensor([0.0388, 0.0341, 0.0399, 0.0295, 0.0366, 0.0322, 0.0316, 0.0350], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-01 11:01:49,942 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.317e+02 6.018e+02 7.577e+02 1.005e+03 2.434e+03, threshold=1.515e+03, percent-clipped=7.0 2023-04-01 11:01:57,659 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.58 vs. limit=5.0 2023-04-01 11:02:04,379 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 2023-04-01 11:02:13,566 INFO [train.py:903] (1/4) Epoch 7, batch 6100, loss[loss=0.2637, simple_loss=0.3298, pruned_loss=0.09876, over 19735.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.3259, pruned_loss=0.0965, over 3817274.77 frames. ], batch size: 63, lr: 1.15e-02, grad_scale: 8.0 2023-04-01 11:02:18,752 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=47072.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 11:02:26,737 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=47079.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 11:02:34,929 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.1440, 2.0868, 1.6426, 1.4735, 1.2960, 1.6308, 0.3662, 0.9841], device='cuda:1'), covar=tensor([0.0294, 0.0307, 0.0252, 0.0400, 0.0768, 0.0420, 0.0643, 0.0598], device='cuda:1'), in_proj_covar=tensor([0.0309, 0.0312, 0.0309, 0.0329, 0.0405, 0.0323, 0.0290, 0.0315], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 11:03:09,039 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([5.2492, 2.7896, 2.9105, 2.6172, 4.7684, 1.7383, 3.3439, 4.9242], device='cuda:1'), covar=tensor([0.0269, 0.1891, 0.1852, 0.1497, 0.0539, 0.2356, 0.1014, 0.0227], device='cuda:1'), in_proj_covar=tensor([0.0316, 0.0319, 0.0327, 0.0301, 0.0331, 0.0322, 0.0300, 0.0319], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 11:03:15,748 INFO [train.py:903] (1/4) Epoch 7, batch 6150, loss[loss=0.2658, simple_loss=0.3298, pruned_loss=0.1009, over 17358.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.3242, pruned_loss=0.09532, over 3806857.55 frames. ], batch size: 101, lr: 1.15e-02, grad_scale: 8.0 2023-04-01 11:03:23,651 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.53 vs. limit=2.0 2023-04-01 11:03:41,639 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9_rvb from training. Duration: 31.02225 2023-04-01 11:03:56,564 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.178e+02 5.986e+02 7.602e+02 9.492e+02 2.168e+03, threshold=1.520e+03, percent-clipped=5.0 2023-04-01 11:04:15,681 INFO [train.py:903] (1/4) Epoch 7, batch 6200, loss[loss=0.2331, simple_loss=0.2937, pruned_loss=0.08627, over 19776.00 frames. ], tot_loss[loss=0.259, simple_loss=0.3253, pruned_loss=0.09633, over 3811555.26 frames. ], batch size: 47, lr: 1.15e-02, grad_scale: 8.0 2023-04-01 11:04:20,356 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=47172.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 11:05:17,520 INFO [train.py:903] (1/4) Epoch 7, batch 6250, loss[loss=0.2674, simple_loss=0.339, pruned_loss=0.09797, over 19673.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.3257, pruned_loss=0.09653, over 3803721.06 frames. ], batch size: 59, lr: 1.15e-02, grad_scale: 8.0 2023-04-01 11:05:26,814 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.31 vs. limit=2.0 2023-04-01 11:05:47,057 WARNING [train.py:1073] (1/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9_rvb from training. Duration: 25.988875 2023-04-01 11:05:57,378 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.252e+02 6.379e+02 7.672e+02 9.726e+02 2.182e+03, threshold=1.534e+03, percent-clipped=2.0 2023-04-01 11:06:08,316 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=47259.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 11:06:15,466 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0420, 3.6779, 2.0851, 1.8636, 3.2180, 1.6219, 1.1027, 2.0176], device='cuda:1'), covar=tensor([0.0699, 0.0263, 0.0633, 0.0559, 0.0291, 0.0794, 0.0756, 0.0385], device='cuda:1'), in_proj_covar=tensor([0.0280, 0.0285, 0.0317, 0.0241, 0.0226, 0.0314, 0.0288, 0.0253], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 11:06:19,225 INFO [train.py:903] (1/4) Epoch 7, batch 6300, loss[loss=0.2211, simple_loss=0.2897, pruned_loss=0.07627, over 19376.00 frames. ], tot_loss[loss=0.2596, simple_loss=0.3256, pruned_loss=0.09673, over 3814070.09 frames. ], batch size: 47, lr: 1.15e-02, grad_scale: 8.0 2023-04-01 11:06:43,112 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=47287.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 11:07:21,231 INFO [train.py:903] (1/4) Epoch 7, batch 6350, loss[loss=0.2347, simple_loss=0.3149, pruned_loss=0.0772, over 19580.00 frames. ], tot_loss[loss=0.2587, simple_loss=0.3249, pruned_loss=0.09624, over 3813228.22 frames. ], batch size: 61, lr: 1.15e-02, grad_scale: 8.0 2023-04-01 11:07:33,187 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=47328.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 11:07:41,331 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=47335.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 11:07:49,423 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.97 vs. limit=5.0 2023-04-01 11:08:02,955 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.061e+02 5.885e+02 7.911e+02 1.072e+03 3.322e+03, threshold=1.582e+03, percent-clipped=7.0 2023-04-01 11:08:05,650 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=47353.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 11:08:13,920 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=47360.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 11:08:22,746 INFO [train.py:903] (1/4) Epoch 7, batch 6400, loss[loss=0.2413, simple_loss=0.2952, pruned_loss=0.09364, over 19369.00 frames. ], tot_loss[loss=0.2575, simple_loss=0.3241, pruned_loss=0.09548, over 3819117.65 frames. ], batch size: 47, lr: 1.15e-02, grad_scale: 8.0 2023-04-01 11:08:48,795 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.48 vs. limit=5.0 2023-04-01 11:09:00,237 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=47397.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 11:09:25,579 INFO [train.py:903] (1/4) Epoch 7, batch 6450, loss[loss=0.2317, simple_loss=0.3084, pruned_loss=0.07747, over 19611.00 frames. ], tot_loss[loss=0.2576, simple_loss=0.3245, pruned_loss=0.09533, over 3811277.27 frames. ], batch size: 50, lr: 1.15e-02, grad_scale: 8.0 2023-04-01 11:10:05,689 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.425e+02 6.283e+02 7.588e+02 1.008e+03 1.535e+03, threshold=1.518e+03, percent-clipped=0.0 2023-04-01 11:10:08,034 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9_rvb from training. Duration: 28.638875 2023-04-01 11:10:26,639 INFO [train.py:903] (1/4) Epoch 7, batch 6500, loss[loss=0.2471, simple_loss=0.3036, pruned_loss=0.0953, over 19741.00 frames. ], tot_loss[loss=0.2573, simple_loss=0.3246, pruned_loss=0.09496, over 3825013.80 frames. ], batch size: 46, lr: 1.15e-02, grad_scale: 8.0 2023-04-01 11:10:29,891 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1_rvb from training. Duration: 0.836375 2023-04-01 11:11:27,822 INFO [train.py:903] (1/4) Epoch 7, batch 6550, loss[loss=0.2712, simple_loss=0.3417, pruned_loss=0.1003, over 19590.00 frames. ], tot_loss[loss=0.2573, simple_loss=0.3244, pruned_loss=0.09513, over 3829258.27 frames. ], batch size: 57, lr: 1.15e-02, grad_scale: 8.0 2023-04-01 11:11:58,243 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=47543.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 11:12:10,167 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.342e+02 6.347e+02 8.071e+02 1.082e+03 2.174e+03, threshold=1.614e+03, percent-clipped=4.0 2023-04-01 11:12:20,950 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3270, 1.2516, 1.7339, 1.2061, 2.8084, 3.8005, 3.5643, 3.9982], device='cuda:1'), covar=tensor([0.1447, 0.3137, 0.2799, 0.2092, 0.0438, 0.0142, 0.0183, 0.0144], device='cuda:1'), in_proj_covar=tensor([0.0255, 0.0286, 0.0316, 0.0247, 0.0208, 0.0136, 0.0205, 0.0170], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 11:12:29,849 INFO [train.py:903] (1/4) Epoch 7, batch 6600, loss[loss=0.2089, simple_loss=0.2759, pruned_loss=0.07098, over 19762.00 frames. ], tot_loss[loss=0.2572, simple_loss=0.3239, pruned_loss=0.09528, over 3820925.45 frames. ], batch size: 46, lr: 1.15e-02, grad_scale: 8.0 2023-04-01 11:12:30,268 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=47568.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 11:13:13,381 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=47603.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 11:13:31,696 INFO [train.py:903] (1/4) Epoch 7, batch 6650, loss[loss=0.2288, simple_loss=0.2949, pruned_loss=0.08129, over 19383.00 frames. ], tot_loss[loss=0.2576, simple_loss=0.3241, pruned_loss=0.09556, over 3815478.14 frames. ], batch size: 47, lr: 1.15e-02, grad_scale: 8.0 2023-04-01 11:14:11,832 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.526e+02 6.649e+02 7.986e+02 1.008e+03 1.623e+03, threshold=1.597e+03, percent-clipped=0.0 2023-04-01 11:14:32,253 INFO [train.py:903] (1/4) Epoch 7, batch 6700, loss[loss=0.2402, simple_loss=0.3035, pruned_loss=0.08842, over 19753.00 frames. ], tot_loss[loss=0.2568, simple_loss=0.3234, pruned_loss=0.09511, over 3830876.71 frames. ], batch size: 46, lr: 1.15e-02, grad_scale: 8.0 2023-04-01 11:15:30,803 INFO [train.py:903] (1/4) Epoch 7, batch 6750, loss[loss=0.3223, simple_loss=0.3796, pruned_loss=0.1325, over 19699.00 frames. ], tot_loss[loss=0.2576, simple_loss=0.3242, pruned_loss=0.09548, over 3836503.99 frames. ], batch size: 59, lr: 1.15e-02, grad_scale: 8.0 2023-04-01 11:15:31,133 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=47718.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 11:15:58,014 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=47741.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 11:16:09,115 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.932e+02 5.831e+02 7.200e+02 8.445e+02 1.747e+03, threshold=1.440e+03, percent-clipped=3.0 2023-04-01 11:16:15,178 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=47756.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 11:16:28,686 INFO [train.py:903] (1/4) Epoch 7, batch 6800, loss[loss=0.2292, simple_loss=0.3, pruned_loss=0.07922, over 19740.00 frames. ], tot_loss[loss=0.2587, simple_loss=0.3251, pruned_loss=0.09618, over 3809356.08 frames. ], batch size: 51, lr: 1.15e-02, grad_scale: 8.0 2023-04-01 11:16:34,644 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.7358, 1.5514, 1.5642, 2.0663, 1.6084, 2.1420, 2.2330, 1.9565], device='cuda:1'), covar=tensor([0.0826, 0.0983, 0.1081, 0.0927, 0.1015, 0.0646, 0.0825, 0.0634], device='cuda:1'), in_proj_covar=tensor([0.0219, 0.0234, 0.0233, 0.0263, 0.0248, 0.0218, 0.0211, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1') 2023-04-01 11:16:47,733 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=5.26 vs. limit=5.0 2023-04-01 11:17:14,699 WARNING [train.py:1073] (1/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 2023-04-01 11:17:15,198 WARNING [train.py:1073] (1/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 2023-04-01 11:17:18,389 INFO [train.py:903] (1/4) Epoch 8, batch 0, loss[loss=0.2448, simple_loss=0.3202, pruned_loss=0.08467, over 19375.00 frames. ], tot_loss[loss=0.2448, simple_loss=0.3202, pruned_loss=0.08467, over 19375.00 frames. ], batch size: 70, lr: 1.08e-02, grad_scale: 8.0 2023-04-01 11:17:18,389 INFO [train.py:928] (1/4) Computing validation loss 2023-04-01 11:17:30,985 INFO [train.py:937] (1/4) Epoch 8, validation: loss=0.1916, simple_loss=0.2915, pruned_loss=0.0458, over 944034.00 frames. 2023-04-01 11:17:30,987 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18649MB 2023-04-01 11:17:41,966 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 2023-04-01 11:17:43,326 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=47806.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 11:18:07,542 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=47827.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 11:18:08,104 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.42 vs. limit=2.0 2023-04-01 11:18:32,755 INFO [train.py:903] (1/4) Epoch 8, batch 50, loss[loss=0.2579, simple_loss=0.3279, pruned_loss=0.09401, over 19567.00 frames. ], tot_loss[loss=0.2549, simple_loss=0.3227, pruned_loss=0.09352, over 873366.38 frames. ], batch size: 52, lr: 1.08e-02, grad_scale: 8.0 2023-04-01 11:18:38,614 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.560e+02 5.925e+02 7.453e+02 9.478e+02 2.348e+03, threshold=1.491e+03, percent-clipped=8.0 2023-04-01 11:18:44,789 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=47856.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 11:19:06,088 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1_rvb from training. Duration: 27.0318125 2023-04-01 11:19:32,492 INFO [train.py:903] (1/4) Epoch 8, batch 100, loss[loss=0.3817, simple_loss=0.4061, pruned_loss=0.1787, over 13444.00 frames. ], tot_loss[loss=0.2551, simple_loss=0.3222, pruned_loss=0.09393, over 1519701.34 frames. ], batch size: 135, lr: 1.08e-02, grad_scale: 8.0 2023-04-01 11:19:42,621 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 2023-04-01 11:20:32,671 INFO [train.py:903] (1/4) Epoch 8, batch 150, loss[loss=0.2907, simple_loss=0.3415, pruned_loss=0.1199, over 13185.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.3232, pruned_loss=0.09385, over 2034019.98 frames. ], batch size: 136, lr: 1.08e-02, grad_scale: 8.0 2023-04-01 11:20:38,421 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.629e+02 6.208e+02 7.626e+02 9.440e+02 2.273e+03, threshold=1.525e+03, percent-clipped=3.0 2023-04-01 11:21:07,748 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=47974.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 11:21:29,212 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 2023-04-01 11:21:32,696 INFO [train.py:903] (1/4) Epoch 8, batch 200, loss[loss=0.2623, simple_loss=0.333, pruned_loss=0.09573, over 19053.00 frames. ], tot_loss[loss=0.2543, simple_loss=0.3226, pruned_loss=0.09301, over 2442455.40 frames. ], batch size: 69, lr: 1.08e-02, grad_scale: 8.0 2023-04-01 11:21:36,549 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=47999.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 11:22:15,575 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([0.9692, 1.1916, 1.4888, 0.5786, 2.1375, 2.4818, 2.0596, 2.5867], device='cuda:1'), covar=tensor([0.1364, 0.3046, 0.2842, 0.2136, 0.0437, 0.0216, 0.0379, 0.0229], device='cuda:1'), in_proj_covar=tensor([0.0254, 0.0284, 0.0317, 0.0247, 0.0208, 0.0136, 0.0204, 0.0172], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 11:22:35,821 INFO [train.py:903] (1/4) Epoch 8, batch 250, loss[loss=0.2508, simple_loss=0.3251, pruned_loss=0.08827, over 19702.00 frames. ], tot_loss[loss=0.2536, simple_loss=0.3218, pruned_loss=0.09268, over 2747953.26 frames. ], batch size: 59, lr: 1.08e-02, grad_scale: 8.0 2023-04-01 11:22:42,341 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.557e+02 6.073e+02 7.344e+02 8.973e+02 2.163e+03, threshold=1.469e+03, percent-clipped=4.0 2023-04-01 11:23:30,336 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=48090.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 11:23:31,705 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1913, 1.2925, 1.6977, 1.4223, 2.4765, 2.0567, 2.7117, 0.9378], device='cuda:1'), covar=tensor([0.1941, 0.3294, 0.1860, 0.1546, 0.1268, 0.1707, 0.1227, 0.3226], device='cuda:1'), in_proj_covar=tensor([0.0455, 0.0533, 0.0530, 0.0416, 0.0572, 0.0467, 0.0624, 0.0459], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 11:23:36,833 INFO [train.py:903] (1/4) Epoch 8, batch 300, loss[loss=0.1932, simple_loss=0.2668, pruned_loss=0.05979, over 19096.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.3229, pruned_loss=0.09338, over 2984969.22 frames. ], batch size: 42, lr: 1.08e-02, grad_scale: 8.0 2023-04-01 11:23:41,627 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=48100.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 11:23:55,915 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=48112.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 11:23:59,571 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2362, 1.4640, 2.0188, 1.5544, 2.8958, 2.5579, 3.1548, 1.2889], device='cuda:1'), covar=tensor([0.2156, 0.3662, 0.2080, 0.1645, 0.1624, 0.1769, 0.1790, 0.3441], device='cuda:1'), in_proj_covar=tensor([0.0458, 0.0536, 0.0532, 0.0418, 0.0575, 0.0468, 0.0629, 0.0461], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 11:24:14,333 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.8252, 1.8459, 1.9092, 2.5022, 1.8969, 2.3880, 2.2494, 1.8238], device='cuda:1'), covar=tensor([0.2153, 0.1812, 0.0857, 0.0912, 0.1778, 0.0703, 0.1513, 0.1457], device='cuda:1'), in_proj_covar=tensor([0.0721, 0.0723, 0.0607, 0.0852, 0.0734, 0.0632, 0.0748, 0.0657], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 11:24:27,784 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=48137.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 11:24:37,682 INFO [train.py:903] (1/4) Epoch 8, batch 350, loss[loss=0.2201, simple_loss=0.3042, pruned_loss=0.06797, over 19705.00 frames. ], tot_loss[loss=0.2536, simple_loss=0.3221, pruned_loss=0.09251, over 3176291.76 frames. ], batch size: 59, lr: 1.07e-02, grad_scale: 8.0 2023-04-01 11:24:39,954 WARNING [train.py:1073] (1/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1_rvb from training. Duration: 0.9681875 2023-04-01 11:24:42,106 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=48150.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 11:24:43,059 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.994e+02 6.192e+02 7.149e+02 9.949e+02 1.629e+03, threshold=1.430e+03, percent-clipped=6.0 2023-04-01 11:25:08,559 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=48171.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 11:25:22,164 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.0697, 0.8389, 1.0609, 1.4982, 1.0570, 0.9455, 1.2787, 1.0106], device='cuda:1'), covar=tensor([0.1145, 0.2059, 0.1448, 0.0762, 0.0960, 0.1599, 0.1061, 0.1250], device='cuda:1'), in_proj_covar=tensor([0.0217, 0.0235, 0.0232, 0.0262, 0.0246, 0.0217, 0.0211, 0.0209], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1') 2023-04-01 11:25:33,647 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0402, 1.3292, 2.0171, 1.9757, 3.0622, 4.5865, 4.6062, 4.9844], device='cuda:1'), covar=tensor([0.1120, 0.3286, 0.2737, 0.1591, 0.0461, 0.0148, 0.0131, 0.0086], device='cuda:1'), in_proj_covar=tensor([0.0252, 0.0284, 0.0313, 0.0244, 0.0207, 0.0136, 0.0202, 0.0171], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 11:25:37,719 INFO [train.py:903] (1/4) Epoch 8, batch 400, loss[loss=0.304, simple_loss=0.3674, pruned_loss=0.1203, over 19662.00 frames. ], tot_loss[loss=0.2565, simple_loss=0.3243, pruned_loss=0.09432, over 3327774.16 frames. ], batch size: 55, lr: 1.07e-02, grad_scale: 8.0 2023-04-01 11:26:03,419 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=48215.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 11:26:05,323 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.47 vs. limit=2.0 2023-04-01 11:26:21,197 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=48230.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 11:26:33,624 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.7244, 1.3930, 1.3517, 1.9711, 1.5715, 1.9248, 2.0209, 1.6985], device='cuda:1'), covar=tensor([0.0715, 0.0974, 0.1099, 0.0824, 0.0850, 0.0689, 0.0763, 0.0665], device='cuda:1'), in_proj_covar=tensor([0.0219, 0.0235, 0.0234, 0.0263, 0.0248, 0.0219, 0.0212, 0.0210], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1') 2023-04-01 11:26:38,936 INFO [train.py:903] (1/4) Epoch 8, batch 450, loss[loss=0.3024, simple_loss=0.3602, pruned_loss=0.1223, over 19487.00 frames. ], tot_loss[loss=0.255, simple_loss=0.3233, pruned_loss=0.09333, over 3449642.32 frames. ], batch size: 64, lr: 1.07e-02, grad_scale: 8.0 2023-04-01 11:26:45,631 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.339e+02 5.626e+02 7.034e+02 8.568e+02 1.629e+03, threshold=1.407e+03, percent-clipped=1.0 2023-04-01 11:27:03,636 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=48265.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 11:27:11,137 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 2023-04-01 11:27:12,251 WARNING [train.py:1073] (1/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 2023-04-01 11:27:27,200 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=48286.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 11:27:41,080 INFO [train.py:903] (1/4) Epoch 8, batch 500, loss[loss=0.2374, simple_loss=0.2986, pruned_loss=0.08814, over 19751.00 frames. ], tot_loss[loss=0.2558, simple_loss=0.3233, pruned_loss=0.09417, over 3525147.60 frames. ], batch size: 45, lr: 1.07e-02, grad_scale: 8.0 2023-04-01 11:28:37,119 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=48342.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 11:28:41,343 INFO [train.py:903] (1/4) Epoch 8, batch 550, loss[loss=0.2907, simple_loss=0.3531, pruned_loss=0.1141, over 19690.00 frames. ], tot_loss[loss=0.2551, simple_loss=0.3227, pruned_loss=0.09371, over 3602891.43 frames. ], batch size: 59, lr: 1.07e-02, grad_scale: 8.0 2023-04-01 11:28:47,050 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.262e+02 6.228e+02 7.563e+02 9.337e+02 1.593e+03, threshold=1.513e+03, percent-clipped=3.0 2023-04-01 11:28:47,302 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.7843, 4.3127, 2.7057, 3.6796, 1.1277, 3.9770, 4.0403, 4.1458], device='cuda:1'), covar=tensor([0.0544, 0.1009, 0.1833, 0.0786, 0.3752, 0.0738, 0.0706, 0.0825], device='cuda:1'), in_proj_covar=tensor([0.0393, 0.0344, 0.0398, 0.0302, 0.0366, 0.0325, 0.0318, 0.0352], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-01 11:29:23,945 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2317, 1.3357, 1.1878, 0.9699, 1.0140, 1.1309, 0.1126, 0.4133], device='cuda:1'), covar=tensor([0.0346, 0.0363, 0.0218, 0.0313, 0.0811, 0.0288, 0.0652, 0.0590], device='cuda:1'), in_proj_covar=tensor([0.0308, 0.0313, 0.0309, 0.0328, 0.0401, 0.0321, 0.0289, 0.0312], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 11:29:42,860 INFO [train.py:903] (1/4) Epoch 8, batch 600, loss[loss=0.3092, simple_loss=0.3622, pruned_loss=0.1281, over 19342.00 frames. ], tot_loss[loss=0.2536, simple_loss=0.3215, pruned_loss=0.09285, over 3652975.11 frames. ], batch size: 66, lr: 1.07e-02, grad_scale: 8.0 2023-04-01 11:30:05,023 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=48415.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 11:30:26,077 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 2023-04-01 11:30:31,029 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=48434.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 11:30:39,410 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=48441.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 11:30:44,669 INFO [train.py:903] (1/4) Epoch 8, batch 650, loss[loss=0.2534, simple_loss=0.3243, pruned_loss=0.09124, over 19747.00 frames. ], tot_loss[loss=0.2544, simple_loss=0.3223, pruned_loss=0.09328, over 3692346.09 frames. ], batch size: 51, lr: 1.07e-02, grad_scale: 16.0 2023-04-01 11:30:50,367 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.061e+02 6.183e+02 7.491e+02 9.829e+02 2.830e+03, threshold=1.498e+03, percent-clipped=3.0 2023-04-01 11:31:17,328 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=48471.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 11:31:45,460 INFO [train.py:903] (1/4) Epoch 8, batch 700, loss[loss=0.2684, simple_loss=0.3453, pruned_loss=0.09569, over 19667.00 frames. ], tot_loss[loss=0.2539, simple_loss=0.322, pruned_loss=0.09291, over 3724812.84 frames. ], batch size: 55, lr: 1.07e-02, grad_scale: 16.0 2023-04-01 11:31:45,787 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=48496.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 11:32:17,639 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=48521.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 11:32:44,017 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=48542.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 11:32:48,880 INFO [train.py:903] (1/4) Epoch 8, batch 750, loss[loss=0.2587, simple_loss=0.3291, pruned_loss=0.09419, over 19537.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.3234, pruned_loss=0.09386, over 3752117.31 frames. ], batch size: 56, lr: 1.07e-02, grad_scale: 8.0 2023-04-01 11:32:49,279 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=48546.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 11:32:52,550 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=48549.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 11:32:55,664 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.581e+02 6.054e+02 7.636e+02 9.380e+02 1.990e+03, threshold=1.527e+03, percent-clipped=3.0 2023-04-01 11:33:13,731 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=48567.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 11:33:21,448 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=48574.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 11:33:36,163 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4086, 1.5429, 1.9727, 1.8623, 3.0231, 3.9940, 3.9569, 4.4219], device='cuda:1'), covar=tensor([0.1528, 0.3057, 0.2732, 0.1752, 0.0577, 0.0257, 0.0165, 0.0112], device='cuda:1'), in_proj_covar=tensor([0.0253, 0.0285, 0.0316, 0.0246, 0.0209, 0.0135, 0.0202, 0.0172], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 11:33:38,457 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.6686, 1.9106, 2.1273, 2.6568, 2.3400, 2.3717, 2.0010, 2.6339], device='cuda:1'), covar=tensor([0.0777, 0.1843, 0.1237, 0.0914, 0.1323, 0.0438, 0.1114, 0.0539], device='cuda:1'), in_proj_covar=tensor([0.0247, 0.0358, 0.0285, 0.0238, 0.0303, 0.0243, 0.0273, 0.0231], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 11:33:49,543 INFO [train.py:903] (1/4) Epoch 8, batch 800, loss[loss=0.287, simple_loss=0.3531, pruned_loss=0.1104, over 19522.00 frames. ], tot_loss[loss=0.2551, simple_loss=0.3231, pruned_loss=0.09354, over 3767333.55 frames. ], batch size: 54, lr: 1.07e-02, grad_scale: 8.0 2023-04-01 11:34:02,658 WARNING [train.py:1073] (1/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 2023-04-01 11:34:34,465 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([6.2178, 5.4872, 2.9458, 4.7037, 1.2399, 5.3475, 5.3518, 5.5737], device='cuda:1'), covar=tensor([0.0414, 0.0964, 0.1779, 0.0592, 0.3660, 0.0547, 0.0584, 0.0600], device='cuda:1'), in_proj_covar=tensor([0.0394, 0.0344, 0.0401, 0.0302, 0.0366, 0.0328, 0.0318, 0.0355], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-01 11:34:51,086 INFO [train.py:903] (1/4) Epoch 8, batch 850, loss[loss=0.2726, simple_loss=0.3464, pruned_loss=0.09938, over 19509.00 frames. ], tot_loss[loss=0.2552, simple_loss=0.3236, pruned_loss=0.09336, over 3784065.46 frames. ], batch size: 64, lr: 1.07e-02, grad_scale: 8.0 2023-04-01 11:34:57,942 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.799e+02 5.922e+02 7.936e+02 9.993e+02 1.897e+03, threshold=1.587e+03, percent-clipped=5.0 2023-04-01 11:35:39,775 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9_rvb from training. Duration: 25.061125 2023-04-01 11:35:39,888 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=48686.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 11:35:43,310 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=48689.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 11:35:50,881 INFO [train.py:903] (1/4) Epoch 8, batch 900, loss[loss=0.2587, simple_loss=0.3307, pruned_loss=0.09336, over 19770.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3239, pruned_loss=0.09358, over 3793163.58 frames. ], batch size: 56, lr: 1.07e-02, grad_scale: 8.0 2023-04-01 11:36:35,744 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0261, 2.0194, 1.6856, 1.4788, 1.5045, 1.5662, 0.3051, 0.8989], device='cuda:1'), covar=tensor([0.0322, 0.0338, 0.0222, 0.0357, 0.0694, 0.0412, 0.0634, 0.0585], device='cuda:1'), in_proj_covar=tensor([0.0312, 0.0313, 0.0311, 0.0329, 0.0402, 0.0324, 0.0291, 0.0311], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 11:36:54,666 INFO [train.py:903] (1/4) Epoch 8, batch 950, loss[loss=0.2174, simple_loss=0.2933, pruned_loss=0.07075, over 19482.00 frames. ], tot_loss[loss=0.2544, simple_loss=0.3229, pruned_loss=0.0929, over 3812150.80 frames. ], batch size: 49, lr: 1.07e-02, grad_scale: 8.0 2023-04-01 11:36:56,554 WARNING [train.py:1073] (1/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9_rvb from training. Duration: 26.32775 2023-04-01 11:37:03,059 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.996e+02 5.931e+02 7.048e+02 8.289e+02 1.665e+03, threshold=1.410e+03, percent-clipped=1.0 2023-04-01 11:37:11,347 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=48759.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 11:37:42,377 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=48785.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 11:37:57,226 INFO [train.py:903] (1/4) Epoch 8, batch 1000, loss[loss=0.2498, simple_loss=0.3229, pruned_loss=0.08838, over 19471.00 frames. ], tot_loss[loss=0.253, simple_loss=0.3217, pruned_loss=0.09218, over 3804747.09 frames. ], batch size: 64, lr: 1.07e-02, grad_scale: 8.0 2023-04-01 11:38:03,535 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=48801.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 11:38:08,263 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=48805.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 11:38:40,036 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=48830.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 11:38:50,187 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 2023-04-01 11:38:51,464 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=48839.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 11:38:59,276 INFO [train.py:903] (1/4) Epoch 8, batch 1050, loss[loss=0.2113, simple_loss=0.2795, pruned_loss=0.07151, over 19714.00 frames. ], tot_loss[loss=0.2533, simple_loss=0.3217, pruned_loss=0.09243, over 3812257.43 frames. ], batch size: 45, lr: 1.07e-02, grad_scale: 8.0 2023-04-01 11:39:06,227 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.557e+02 5.500e+02 6.561e+02 8.180e+02 1.521e+03, threshold=1.312e+03, percent-clipped=1.0 2023-04-01 11:39:11,103 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=48856.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 11:39:30,878 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 2023-04-01 11:39:34,224 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=48874.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 11:39:45,604 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([5.0059, 4.9372, 5.7909, 5.7056, 1.7395, 5.4359, 4.5964, 5.3763], device='cuda:1'), covar=tensor([0.1125, 0.0613, 0.0437, 0.0448, 0.4852, 0.0368, 0.0485, 0.0920], device='cuda:1'), in_proj_covar=tensor([0.0584, 0.0514, 0.0691, 0.0581, 0.0643, 0.0439, 0.0442, 0.0642], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1') 2023-04-01 11:39:45,753 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.2688, 2.9896, 1.9030, 2.1190, 1.8483, 2.4534, 0.8166, 1.9636], device='cuda:1'), covar=tensor([0.0321, 0.0334, 0.0420, 0.0649, 0.0722, 0.0523, 0.0748, 0.0655], device='cuda:1'), in_proj_covar=tensor([0.0311, 0.0312, 0.0309, 0.0329, 0.0401, 0.0320, 0.0289, 0.0313], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 11:39:58,991 INFO [train.py:903] (1/4) Epoch 8, batch 1100, loss[loss=0.2351, simple_loss=0.3137, pruned_loss=0.07827, over 19532.00 frames. ], tot_loss[loss=0.2538, simple_loss=0.322, pruned_loss=0.09277, over 3823442.32 frames. ], batch size: 54, lr: 1.07e-02, grad_scale: 8.0 2023-04-01 11:40:04,070 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=48900.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 11:41:00,139 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=48945.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 11:41:00,832 INFO [train.py:903] (1/4) Epoch 8, batch 1150, loss[loss=0.2699, simple_loss=0.3447, pruned_loss=0.09757, over 19687.00 frames. ], tot_loss[loss=0.2532, simple_loss=0.3221, pruned_loss=0.09216, over 3839316.61 frames. ], batch size: 59, lr: 1.07e-02, grad_scale: 8.0 2023-04-01 11:41:09,120 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.525e+02 5.943e+02 6.952e+02 8.882e+02 1.618e+03, threshold=1.390e+03, percent-clipped=5.0 2023-04-01 11:41:30,935 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=48970.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 11:42:04,104 INFO [train.py:903] (1/4) Epoch 8, batch 1200, loss[loss=0.2236, simple_loss=0.2905, pruned_loss=0.07836, over 19785.00 frames. ], tot_loss[loss=0.2542, simple_loss=0.3228, pruned_loss=0.09273, over 3841145.54 frames. ], batch size: 47, lr: 1.07e-02, grad_scale: 8.0 2023-04-01 11:42:32,667 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 2023-04-01 11:43:05,008 INFO [train.py:903] (1/4) Epoch 8, batch 1250, loss[loss=0.2885, simple_loss=0.3552, pruned_loss=0.1109, over 19661.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3237, pruned_loss=0.0936, over 3827223.68 frames. ], batch size: 60, lr: 1.06e-02, grad_scale: 8.0 2023-04-01 11:43:11,744 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.529e+02 6.811e+02 8.521e+02 1.008e+03 2.064e+03, threshold=1.704e+03, percent-clipped=4.0 2023-04-01 11:43:17,968 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=49057.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 11:43:39,590 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.10 vs. limit=2.0 2023-04-01 11:43:50,271 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=49082.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 11:44:06,041 INFO [train.py:903] (1/4) Epoch 8, batch 1300, loss[loss=0.2294, simple_loss=0.3044, pruned_loss=0.07717, over 19663.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.325, pruned_loss=0.09493, over 3820733.80 frames. ], batch size: 55, lr: 1.06e-02, grad_scale: 8.0 2023-04-01 11:44:48,997 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=49130.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 11:45:07,322 INFO [train.py:903] (1/4) Epoch 8, batch 1350, loss[loss=0.3032, simple_loss=0.357, pruned_loss=0.1247, over 13612.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.3233, pruned_loss=0.09408, over 3826212.77 frames. ], batch size: 136, lr: 1.06e-02, grad_scale: 8.0 2023-04-01 11:45:08,145 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 2023-04-01 11:45:16,535 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.771e+02 5.835e+02 7.092e+02 8.908e+02 2.388e+03, threshold=1.418e+03, percent-clipped=3.0 2023-04-01 11:45:20,435 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=49155.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 11:45:22,504 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=49156.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 11:45:41,540 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.8681, 3.5850, 1.7772, 2.1868, 2.9819, 1.7747, 1.2053, 1.8552], device='cuda:1'), covar=tensor([0.1094, 0.0316, 0.0923, 0.0568, 0.0403, 0.0872, 0.0873, 0.0660], device='cuda:1'), in_proj_covar=tensor([0.0282, 0.0290, 0.0315, 0.0243, 0.0232, 0.0309, 0.0284, 0.0251], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 11:45:42,677 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([5.0825, 1.9732, 2.2732, 1.6322, 4.5800, 0.9265, 2.5288, 4.7938], device='cuda:1'), covar=tensor([0.0295, 0.2201, 0.2051, 0.1685, 0.0571, 0.2549, 0.1177, 0.0199], device='cuda:1'), in_proj_covar=tensor([0.0313, 0.0317, 0.0325, 0.0298, 0.0325, 0.0319, 0.0297, 0.0321], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 11:45:52,053 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=49181.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 11:45:54,119 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=49183.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 11:46:10,832 INFO [train.py:903] (1/4) Epoch 8, batch 1400, loss[loss=0.2462, simple_loss=0.3181, pruned_loss=0.08715, over 19580.00 frames. ], tot_loss[loss=0.2561, simple_loss=0.3234, pruned_loss=0.09436, over 3825227.88 frames. ], batch size: 61, lr: 1.06e-02, grad_scale: 8.0 2023-04-01 11:46:17,514 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=49200.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 11:47:12,312 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_rvb from training. Duration: 25.85 2023-04-01 11:47:13,438 INFO [train.py:903] (1/4) Epoch 8, batch 1450, loss[loss=0.2203, simple_loss=0.2818, pruned_loss=0.0794, over 19765.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.323, pruned_loss=0.09396, over 3841104.18 frames. ], batch size: 47, lr: 1.06e-02, grad_scale: 8.0 2023-04-01 11:47:19,909 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.194e+02 6.215e+02 8.146e+02 9.729e+02 2.293e+03, threshold=1.629e+03, percent-clipped=2.0 2023-04-01 11:47:22,582 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=49254.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 11:48:14,510 INFO [train.py:903] (1/4) Epoch 8, batch 1500, loss[loss=0.2613, simple_loss=0.3346, pruned_loss=0.094, over 19556.00 frames. ], tot_loss[loss=0.2541, simple_loss=0.3219, pruned_loss=0.09317, over 3840826.31 frames. ], batch size: 61, lr: 1.06e-02, grad_scale: 8.0 2023-04-01 11:48:16,759 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=49298.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 11:48:38,609 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=49315.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 11:49:05,432 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.5509, 1.8697, 2.1618, 2.5535, 2.2943, 1.8399, 1.7247, 2.3949], device='cuda:1'), covar=tensor([0.0765, 0.1718, 0.1185, 0.0902, 0.1198, 0.0677, 0.1241, 0.0597], device='cuda:1'), in_proj_covar=tensor([0.0241, 0.0346, 0.0278, 0.0232, 0.0290, 0.0238, 0.0266, 0.0225], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 11:49:14,529 INFO [train.py:903] (1/4) Epoch 8, batch 1550, loss[loss=0.2571, simple_loss=0.3276, pruned_loss=0.09333, over 19671.00 frames. ], tot_loss[loss=0.256, simple_loss=0.3233, pruned_loss=0.09436, over 3835751.98 frames. ], batch size: 53, lr: 1.06e-02, grad_scale: 8.0 2023-04-01 11:49:23,152 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.230e+02 6.392e+02 7.844e+02 9.464e+02 1.840e+03, threshold=1.569e+03, percent-clipped=1.0 2023-04-01 11:49:42,518 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.7867, 4.3590, 2.6044, 3.8461, 1.0324, 4.0205, 4.1147, 4.3041], device='cuda:1'), covar=tensor([0.0586, 0.0964, 0.2059, 0.0701, 0.4088, 0.0770, 0.0719, 0.0833], device='cuda:1'), in_proj_covar=tensor([0.0393, 0.0341, 0.0399, 0.0298, 0.0368, 0.0326, 0.0318, 0.0355], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-01 11:50:17,462 INFO [train.py:903] (1/4) Epoch 8, batch 1600, loss[loss=0.2713, simple_loss=0.3396, pruned_loss=0.1015, over 19077.00 frames. ], tot_loss[loss=0.2547, simple_loss=0.3224, pruned_loss=0.09357, over 3841077.49 frames. ], batch size: 75, lr: 1.06e-02, grad_scale: 8.0 2023-04-01 11:50:25,486 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.6915, 4.0904, 4.4435, 4.3756, 1.7008, 4.1323, 3.5742, 4.0495], device='cuda:1'), covar=tensor([0.1150, 0.0990, 0.0530, 0.0499, 0.4872, 0.0502, 0.0605, 0.0996], device='cuda:1'), in_proj_covar=tensor([0.0600, 0.0533, 0.0717, 0.0598, 0.0667, 0.0458, 0.0454, 0.0663], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1') 2023-04-01 11:50:38,471 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9_rvb from training. Duration: 30.1555625 2023-04-01 11:51:12,153 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=49440.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 11:51:20,005 INFO [train.py:903] (1/4) Epoch 8, batch 1650, loss[loss=0.2182, simple_loss=0.2924, pruned_loss=0.07199, over 19876.00 frames. ], tot_loss[loss=0.2549, simple_loss=0.3228, pruned_loss=0.09354, over 3844744.18 frames. ], batch size: 52, lr: 1.06e-02, grad_scale: 8.0 2023-04-01 11:51:24,764 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=49450.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 11:51:26,747 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.384e+02 6.230e+02 7.478e+02 9.229e+02 3.510e+03, threshold=1.496e+03, percent-clipped=3.0 2023-04-01 11:51:52,824 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0 2023-04-01 11:52:21,800 INFO [train.py:903] (1/4) Epoch 8, batch 1700, loss[loss=0.2916, simple_loss=0.3528, pruned_loss=0.1152, over 19744.00 frames. ], tot_loss[loss=0.252, simple_loss=0.3209, pruned_loss=0.09159, over 3856779.05 frames. ], batch size: 63, lr: 1.06e-02, grad_scale: 8.0 2023-04-01 11:53:02,520 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 2023-04-01 11:53:04,656 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 2023-04-01 11:53:23,394 INFO [train.py:903] (1/4) Epoch 8, batch 1750, loss[loss=0.2419, simple_loss=0.3139, pruned_loss=0.08494, over 19601.00 frames. ], tot_loss[loss=0.2523, simple_loss=0.3206, pruned_loss=0.09203, over 3830651.10 frames. ], batch size: 57, lr: 1.06e-02, grad_scale: 8.0 2023-04-01 11:53:31,458 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.581e+02 6.147e+02 7.390e+02 1.012e+03 1.809e+03, threshold=1.478e+03, percent-clipped=6.0 2023-04-01 11:53:35,052 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=49554.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 11:53:56,597 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=49571.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 11:54:06,069 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=49579.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 11:54:23,092 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 2023-04-01 11:54:27,936 INFO [train.py:903] (1/4) Epoch 8, batch 1800, loss[loss=0.2646, simple_loss=0.3372, pruned_loss=0.09602, over 19488.00 frames. ], tot_loss[loss=0.2523, simple_loss=0.3205, pruned_loss=0.09202, over 3816801.83 frames. ], batch size: 64, lr: 1.06e-02, grad_scale: 8.0 2023-04-01 11:54:28,341 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=49596.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 11:54:31,424 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=49598.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 11:54:32,681 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.5748, 1.6729, 1.7117, 1.7223, 4.1512, 1.0667, 2.3260, 4.2358], device='cuda:1'), covar=tensor([0.0345, 0.2297, 0.2407, 0.1571, 0.0600, 0.2448, 0.1350, 0.0263], device='cuda:1'), in_proj_covar=tensor([0.0316, 0.0322, 0.0331, 0.0299, 0.0331, 0.0321, 0.0303, 0.0324], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 11:55:18,163 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([0.6377, 0.8746, 0.6856, 0.6421, 0.7852, 0.5865, 0.6429, 0.8251], device='cuda:1'), covar=tensor([0.0299, 0.0369, 0.0518, 0.0308, 0.0291, 0.0614, 0.0308, 0.0248], device='cuda:1'), in_proj_covar=tensor([0.0281, 0.0288, 0.0313, 0.0241, 0.0230, 0.0311, 0.0285, 0.0253], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 11:55:25,414 WARNING [train.py:1073] (1/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9_rvb from training. Duration: 25.0944375 2023-04-01 11:55:30,998 INFO [train.py:903] (1/4) Epoch 8, batch 1850, loss[loss=0.2759, simple_loss=0.3409, pruned_loss=0.1055, over 17547.00 frames. ], tot_loss[loss=0.2538, simple_loss=0.3218, pruned_loss=0.09289, over 3815401.15 frames. ], batch size: 101, lr: 1.06e-02, grad_scale: 8.0 2023-04-01 11:55:38,000 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.438e+02 5.845e+02 7.519e+02 8.649e+02 2.522e+03, threshold=1.504e+03, percent-clipped=4.0 2023-04-01 11:55:45,094 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3271, 2.2653, 1.5610, 1.4139, 2.1486, 1.2184, 1.1314, 1.6270], device='cuda:1'), covar=tensor([0.0791, 0.0521, 0.0830, 0.0612, 0.0390, 0.0976, 0.0670, 0.0429], device='cuda:1'), in_proj_covar=tensor([0.0283, 0.0290, 0.0315, 0.0242, 0.0231, 0.0312, 0.0286, 0.0253], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 11:56:02,471 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9_rvb from training. Duration: 27.8166875 2023-04-01 11:56:31,218 INFO [train.py:903] (1/4) Epoch 8, batch 1900, loss[loss=0.2497, simple_loss=0.3254, pruned_loss=0.08696, over 19663.00 frames. ], tot_loss[loss=0.2546, simple_loss=0.3224, pruned_loss=0.09337, over 3824640.90 frames. ], batch size: 53, lr: 1.06e-02, grad_scale: 8.0 2023-04-01 11:56:48,641 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 2023-04-01 11:56:52,370 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=49713.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 11:56:54,371 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9_rvb from training. Duration: 27.02225 2023-04-01 11:57:18,977 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 2023-04-01 11:57:33,007 INFO [train.py:903] (1/4) Epoch 8, batch 1950, loss[loss=0.2491, simple_loss=0.3044, pruned_loss=0.09687, over 19327.00 frames. ], tot_loss[loss=0.2552, simple_loss=0.3227, pruned_loss=0.09381, over 3824370.52 frames. ], batch size: 44, lr: 1.06e-02, grad_scale: 8.0 2023-04-01 11:57:40,106 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.492e+02 5.402e+02 6.624e+02 8.916e+02 2.925e+03, threshold=1.325e+03, percent-clipped=4.0 2023-04-01 11:58:20,353 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=49784.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 11:58:20,512 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=49784.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 11:58:32,991 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=49794.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 11:58:35,919 INFO [train.py:903] (1/4) Epoch 8, batch 2000, loss[loss=0.2691, simple_loss=0.3426, pruned_loss=0.09776, over 19681.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.323, pruned_loss=0.0941, over 3805183.79 frames. ], batch size: 60, lr: 1.06e-02, grad_scale: 8.0 2023-04-01 11:59:35,610 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_rvb from training. Duration: 26.8349375 2023-04-01 11:59:38,066 INFO [train.py:903] (1/4) Epoch 8, batch 2050, loss[loss=0.2523, simple_loss=0.3094, pruned_loss=0.09758, over 19399.00 frames. ], tot_loss[loss=0.2552, simple_loss=0.3225, pruned_loss=0.09393, over 3802769.22 frames. ], batch size: 48, lr: 1.06e-02, grad_scale: 8.0 2023-04-01 11:59:45,935 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.480e+02 5.411e+02 7.156e+02 9.098e+02 3.444e+03, threshold=1.431e+03, percent-clipped=9.0 2023-04-01 11:59:51,700 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.7694, 4.3173, 2.6179, 3.8457, 1.1373, 3.9821, 4.0041, 4.1419], device='cuda:1'), covar=tensor([0.0538, 0.0950, 0.1967, 0.0699, 0.3746, 0.0745, 0.0763, 0.0804], device='cuda:1'), in_proj_covar=tensor([0.0393, 0.0340, 0.0402, 0.0298, 0.0363, 0.0328, 0.0314, 0.0356], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-01 11:59:53,833 WARNING [train.py:1073] (1/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 2023-04-01 11:59:55,105 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 2023-04-01 12:00:17,459 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 2023-04-01 12:00:29,260 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0985, 2.1687, 2.2027, 3.4999, 2.2689, 3.3206, 2.7984, 1.8842], device='cuda:1'), covar=tensor([0.3225, 0.2571, 0.1177, 0.1317, 0.3146, 0.0981, 0.2677, 0.2352], device='cuda:1'), in_proj_covar=tensor([0.0723, 0.0732, 0.0611, 0.0855, 0.0733, 0.0635, 0.0758, 0.0658], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 12:00:39,973 INFO [train.py:903] (1/4) Epoch 8, batch 2100, loss[loss=0.2705, simple_loss=0.3415, pruned_loss=0.09975, over 19707.00 frames. ], tot_loss[loss=0.254, simple_loss=0.3217, pruned_loss=0.09311, over 3808850.51 frames. ], batch size: 60, lr: 1.06e-02, grad_scale: 8.0 2023-04-01 12:00:43,849 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=49899.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 12:00:55,216 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=49909.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 12:01:04,920 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.6116, 4.1762, 2.8286, 3.8110, 0.9860, 3.9406, 3.8982, 4.0301], device='cuda:1'), covar=tensor([0.0562, 0.1070, 0.1728, 0.0677, 0.4194, 0.0744, 0.0692, 0.0811], device='cuda:1'), in_proj_covar=tensor([0.0393, 0.0338, 0.0398, 0.0295, 0.0363, 0.0325, 0.0315, 0.0355], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-01 12:01:09,457 WARNING [train.py:1073] (1/4) Exclude cut with ID 453-131332-0000-131866_sp0.9_rvb from training. Duration: 25.3333125 2023-04-01 12:01:32,579 WARNING [train.py:1073] (1/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9_rvb from training. Duration: 26.6166875 2023-04-01 12:01:40,040 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.7419, 1.7862, 1.8891, 2.4553, 1.5424, 2.1263, 2.2826, 1.8055], device='cuda:1'), covar=tensor([0.2612, 0.2179, 0.1181, 0.1304, 0.2531, 0.1117, 0.2436, 0.2079], device='cuda:1'), in_proj_covar=tensor([0.0724, 0.0734, 0.0612, 0.0859, 0.0731, 0.0636, 0.0760, 0.0658], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 12:01:41,831 INFO [train.py:903] (1/4) Epoch 8, batch 2150, loss[loss=0.2533, simple_loss=0.3247, pruned_loss=0.09095, over 18818.00 frames. ], tot_loss[loss=0.2541, simple_loss=0.3218, pruned_loss=0.09317, over 3800843.70 frames. ], batch size: 74, lr: 1.06e-02, grad_scale: 8.0 2023-04-01 12:01:48,291 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.004e+02 5.997e+02 7.086e+02 8.659e+02 2.224e+03, threshold=1.417e+03, percent-clipped=8.0 2023-04-01 12:01:48,672 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.3530, 2.1842, 1.6810, 1.6105, 1.4865, 1.7393, 0.5315, 1.2447], device='cuda:1'), covar=tensor([0.0284, 0.0316, 0.0299, 0.0490, 0.0664, 0.0454, 0.0640, 0.0534], device='cuda:1'), in_proj_covar=tensor([0.0307, 0.0307, 0.0308, 0.0327, 0.0397, 0.0319, 0.0288, 0.0306], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 12:02:06,546 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2360, 1.3766, 1.7535, 1.4541, 2.7380, 2.3263, 2.8763, 1.1419], device='cuda:1'), covar=tensor([0.1867, 0.3175, 0.1829, 0.1480, 0.1103, 0.1481, 0.1271, 0.2883], device='cuda:1'), in_proj_covar=tensor([0.0454, 0.0530, 0.0526, 0.0411, 0.0567, 0.0461, 0.0628, 0.0459], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 12:02:12,071 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=49969.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 12:02:40,685 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=49994.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 12:02:43,585 INFO [train.py:903] (1/4) Epoch 8, batch 2200, loss[loss=0.2261, simple_loss=0.2974, pruned_loss=0.07739, over 19791.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.3224, pruned_loss=0.09357, over 3809445.31 frames. ], batch size: 48, lr: 1.05e-02, grad_scale: 8.0 2023-04-01 12:03:48,609 INFO [train.py:903] (1/4) Epoch 8, batch 2250, loss[loss=0.2204, simple_loss=0.2977, pruned_loss=0.0715, over 19544.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.3228, pruned_loss=0.09418, over 3807196.17 frames. ], batch size: 56, lr: 1.05e-02, grad_scale: 8.0 2023-04-01 12:03:51,232 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.3574, 3.9515, 2.4641, 3.6011, 1.0516, 3.6431, 3.7021, 3.7467], device='cuda:1'), covar=tensor([0.0600, 0.0989, 0.1995, 0.0725, 0.3649, 0.0769, 0.0777, 0.1018], device='cuda:1'), in_proj_covar=tensor([0.0390, 0.0338, 0.0398, 0.0294, 0.0360, 0.0323, 0.0312, 0.0353], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-01 12:03:55,469 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.410e+02 6.238e+02 7.806e+02 1.014e+03 2.092e+03, threshold=1.561e+03, percent-clipped=8.0 2023-04-01 12:03:58,799 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.3190, 3.9139, 2.4776, 3.6033, 1.0039, 3.5901, 3.6720, 3.7373], device='cuda:1'), covar=tensor([0.0695, 0.1197, 0.2050, 0.0697, 0.3861, 0.0799, 0.0706, 0.0982], device='cuda:1'), in_proj_covar=tensor([0.0390, 0.0339, 0.0398, 0.0294, 0.0360, 0.0323, 0.0313, 0.0353], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-01 12:04:51,003 INFO [train.py:903] (1/4) Epoch 8, batch 2300, loss[loss=0.2429, simple_loss=0.3006, pruned_loss=0.09264, over 18999.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.3227, pruned_loss=0.09392, over 3818258.83 frames. ], batch size: 42, lr: 1.05e-02, grad_scale: 8.0 2023-04-01 12:05:03,035 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.4672, 1.7586, 2.0413, 2.9490, 2.1907, 2.2156, 2.6321, 2.3735], device='cuda:1'), covar=tensor([0.0722, 0.1032, 0.1013, 0.0886, 0.0954, 0.0943, 0.0925, 0.0737], device='cuda:1'), in_proj_covar=tensor([0.0217, 0.0232, 0.0229, 0.0258, 0.0245, 0.0217, 0.0208, 0.0208], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1') 2023-04-01 12:05:04,996 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_rvb from training. Duration: 26.205 2023-04-01 12:05:11,317 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.2925, 2.2008, 1.7732, 1.7374, 1.5497, 1.7664, 0.5575, 1.1234], device='cuda:1'), covar=tensor([0.0299, 0.0307, 0.0278, 0.0386, 0.0727, 0.0425, 0.0654, 0.0587], device='cuda:1'), in_proj_covar=tensor([0.0312, 0.0312, 0.0312, 0.0328, 0.0403, 0.0324, 0.0293, 0.0310], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 12:05:31,273 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=50128.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 12:05:37,134 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2305, 1.2263, 1.6244, 1.3422, 2.7336, 3.6399, 3.4578, 3.8941], device='cuda:1'), covar=tensor([0.1509, 0.3187, 0.2874, 0.1965, 0.0443, 0.0182, 0.0194, 0.0139], device='cuda:1'), in_proj_covar=tensor([0.0251, 0.0284, 0.0312, 0.0247, 0.0208, 0.0135, 0.0204, 0.0171], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 12:05:52,927 INFO [train.py:903] (1/4) Epoch 8, batch 2350, loss[loss=0.2536, simple_loss=0.3285, pruned_loss=0.08937, over 19597.00 frames. ], tot_loss[loss=0.2537, simple_loss=0.3214, pruned_loss=0.093, over 3826183.59 frames. ], batch size: 57, lr: 1.05e-02, grad_scale: 8.0 2023-04-01 12:05:53,229 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=50146.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 12:05:59,867 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.785e+02 5.938e+02 7.771e+02 9.106e+02 1.869e+03, threshold=1.554e+03, percent-clipped=2.0 2023-04-01 12:06:03,825 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=50155.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 12:06:18,070 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=50165.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 12:06:36,693 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=50180.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 12:06:38,680 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_rvb from training. Duration: 25.775 2023-04-01 12:06:47,984 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=50190.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 12:06:54,684 INFO [train.py:903] (1/4) Epoch 8, batch 2400, loss[loss=0.3306, simple_loss=0.3863, pruned_loss=0.1375, over 17223.00 frames. ], tot_loss[loss=0.2534, simple_loss=0.3214, pruned_loss=0.09268, over 3829310.35 frames. ], batch size: 101, lr: 1.05e-02, grad_scale: 8.0 2023-04-01 12:06:54,695 WARNING [train.py:1073] (1/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9_rvb from training. Duration: 25.45 2023-04-01 12:07:20,927 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=50215.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 12:07:54,636 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=50243.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 12:07:58,090 INFO [train.py:903] (1/4) Epoch 8, batch 2450, loss[loss=0.2349, simple_loss=0.3176, pruned_loss=0.07615, over 18725.00 frames. ], tot_loss[loss=0.2528, simple_loss=0.3211, pruned_loss=0.09223, over 3829618.40 frames. ], batch size: 74, lr: 1.05e-02, grad_scale: 8.0 2023-04-01 12:08:05,246 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.846e+02 6.354e+02 7.339e+02 9.160e+02 2.255e+03, threshold=1.468e+03, percent-clipped=3.0 2023-04-01 12:09:00,864 INFO [train.py:903] (1/4) Epoch 8, batch 2500, loss[loss=0.2442, simple_loss=0.3114, pruned_loss=0.08849, over 19590.00 frames. ], tot_loss[loss=0.2527, simple_loss=0.3211, pruned_loss=0.09219, over 3827424.43 frames. ], batch size: 52, lr: 1.05e-02, grad_scale: 8.0 2023-04-01 12:10:03,028 INFO [train.py:903] (1/4) Epoch 8, batch 2550, loss[loss=0.2381, simple_loss=0.3081, pruned_loss=0.08406, over 19847.00 frames. ], tot_loss[loss=0.2514, simple_loss=0.3205, pruned_loss=0.09116, over 3823501.35 frames. ], batch size: 52, lr: 1.05e-02, grad_scale: 8.0 2023-04-01 12:10:09,527 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.329e+02 5.479e+02 6.803e+02 8.076e+02 1.672e+03, threshold=1.361e+03, percent-clipped=2.0 2023-04-01 12:10:09,870 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([0.8891, 0.8945, 1.1578, 0.5518, 1.6300, 1.7062, 1.5337, 1.8075], device='cuda:1'), covar=tensor([0.1023, 0.2420, 0.2132, 0.1977, 0.0649, 0.0441, 0.0299, 0.0246], device='cuda:1'), in_proj_covar=tensor([0.0251, 0.0285, 0.0312, 0.0247, 0.0208, 0.0137, 0.0204, 0.0170], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 12:10:59,294 WARNING [train.py:1073] (1/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 2023-04-01 12:11:05,085 INFO [train.py:903] (1/4) Epoch 8, batch 2600, loss[loss=0.2569, simple_loss=0.3276, pruned_loss=0.09316, over 17446.00 frames. ], tot_loss[loss=0.2526, simple_loss=0.3214, pruned_loss=0.09191, over 3819985.40 frames. ], batch size: 101, lr: 1.05e-02, grad_scale: 8.0 2023-04-01 12:12:09,317 INFO [train.py:903] (1/4) Epoch 8, batch 2650, loss[loss=0.3097, simple_loss=0.3639, pruned_loss=0.1277, over 18223.00 frames. ], tot_loss[loss=0.2524, simple_loss=0.321, pruned_loss=0.09189, over 3826485.89 frames. ], batch size: 83, lr: 1.05e-02, grad_scale: 8.0 2023-04-01 12:12:15,975 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.098e+02 6.836e+02 8.198e+02 1.046e+03 1.620e+03, threshold=1.640e+03, percent-clipped=8.0 2023-04-01 12:12:27,572 WARNING [train.py:1073] (1/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9_rvb from training. Duration: 27.25 2023-04-01 12:12:52,968 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.53 vs. limit=2.0 2023-04-01 12:13:04,667 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=50490.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 12:13:11,329 INFO [train.py:903] (1/4) Epoch 8, batch 2700, loss[loss=0.3248, simple_loss=0.3784, pruned_loss=0.1356, over 19591.00 frames. ], tot_loss[loss=0.2534, simple_loss=0.3217, pruned_loss=0.0926, over 3817609.93 frames. ], batch size: 61, lr: 1.05e-02, grad_scale: 8.0 2023-04-01 12:13:16,526 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=50499.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 12:13:16,892 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0 2023-04-01 12:13:17,796 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9953, 2.0611, 2.1017, 2.8358, 1.8622, 2.6928, 2.6169, 1.9897], device='cuda:1'), covar=tensor([0.2862, 0.2244, 0.1159, 0.1449, 0.2832, 0.1052, 0.2445, 0.2074], device='cuda:1'), in_proj_covar=tensor([0.0723, 0.0729, 0.0611, 0.0851, 0.0731, 0.0635, 0.0759, 0.0655], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 12:13:46,639 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=50524.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 12:14:15,291 INFO [train.py:903] (1/4) Epoch 8, batch 2750, loss[loss=0.2337, simple_loss=0.2923, pruned_loss=0.08757, over 19349.00 frames. ], tot_loss[loss=0.2539, simple_loss=0.3222, pruned_loss=0.09278, over 3813676.47 frames. ], batch size: 47, lr: 1.05e-02, grad_scale: 8.0 2023-04-01 12:14:23,779 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.883e+02 5.912e+02 7.181e+02 9.047e+02 1.864e+03, threshold=1.436e+03, percent-clipped=1.0 2023-04-01 12:14:30,887 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=50559.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 12:14:37,754 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=50564.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 12:14:44,503 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.7074, 1.7305, 1.8339, 2.5183, 1.5672, 2.2531, 2.1552, 1.7519], device='cuda:1'), covar=tensor([0.2909, 0.2339, 0.1308, 0.1196, 0.2618, 0.1087, 0.2815, 0.2252], device='cuda:1'), in_proj_covar=tensor([0.0731, 0.0739, 0.0618, 0.0861, 0.0742, 0.0644, 0.0769, 0.0663], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 12:15:10,438 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.53 vs. limit=2.0 2023-04-01 12:15:18,011 INFO [train.py:903] (1/4) Epoch 8, batch 2800, loss[loss=0.2773, simple_loss=0.3256, pruned_loss=0.1145, over 19422.00 frames. ], tot_loss[loss=0.2543, simple_loss=0.3226, pruned_loss=0.093, over 3810555.23 frames. ], batch size: 48, lr: 1.05e-02, grad_scale: 8.0 2023-04-01 12:15:29,723 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=50605.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 12:16:21,989 INFO [train.py:903] (1/4) Epoch 8, batch 2850, loss[loss=0.3336, simple_loss=0.3751, pruned_loss=0.146, over 13315.00 frames. ], tot_loss[loss=0.2546, simple_loss=0.323, pruned_loss=0.09308, over 3805420.33 frames. ], batch size: 135, lr: 1.05e-02, grad_scale: 8.0 2023-04-01 12:16:31,181 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.810e+02 5.617e+02 7.073e+02 8.787e+02 1.544e+03, threshold=1.415e+03, percent-clipped=2.0 2023-04-01 12:16:57,548 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=50674.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 12:17:03,736 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9392, 1.9935, 2.0535, 2.9102, 1.9392, 2.6509, 2.5477, 1.9299], device='cuda:1'), covar=tensor([0.3064, 0.2428, 0.1209, 0.1327, 0.2710, 0.1089, 0.2489, 0.2158], device='cuda:1'), in_proj_covar=tensor([0.0724, 0.0731, 0.0611, 0.0846, 0.0729, 0.0633, 0.0752, 0.0652], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 12:17:25,975 INFO [train.py:903] (1/4) Epoch 8, batch 2900, loss[loss=0.3068, simple_loss=0.3688, pruned_loss=0.1224, over 19652.00 frames. ], tot_loss[loss=0.2552, simple_loss=0.3233, pruned_loss=0.09354, over 3805695.43 frames. ], batch size: 60, lr: 1.05e-02, grad_scale: 8.0 2023-04-01 12:17:26,012 WARNING [train.py:1073] (1/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 2023-04-01 12:17:36,254 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 2023-04-01 12:18:26,806 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=2.00 vs. limit=2.0 2023-04-01 12:18:29,362 INFO [train.py:903] (1/4) Epoch 8, batch 2950, loss[loss=0.2459, simple_loss=0.3208, pruned_loss=0.08548, over 19530.00 frames. ], tot_loss[loss=0.2568, simple_loss=0.3245, pruned_loss=0.09455, over 3784009.72 frames. ], batch size: 54, lr: 1.05e-02, grad_scale: 8.0 2023-04-01 12:18:37,517 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.796e+02 6.074e+02 7.951e+02 1.027e+03 2.467e+03, threshold=1.590e+03, percent-clipped=7.0 2023-04-01 12:19:20,726 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1109, 1.1909, 1.4838, 0.9919, 1.9282, 2.2265, 2.1325, 2.3640], device='cuda:1'), covar=tensor([0.1286, 0.2624, 0.2343, 0.2111, 0.0885, 0.0570, 0.0287, 0.0268], device='cuda:1'), in_proj_covar=tensor([0.0255, 0.0286, 0.0315, 0.0249, 0.0208, 0.0137, 0.0207, 0.0174], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 12:19:31,422 INFO [train.py:903] (1/4) Epoch 8, batch 3000, loss[loss=0.2635, simple_loss=0.3272, pruned_loss=0.09993, over 19538.00 frames. ], tot_loss[loss=0.2575, simple_loss=0.325, pruned_loss=0.09498, over 3773493.17 frames. ], batch size: 54, lr: 1.05e-02, grad_scale: 8.0 2023-04-01 12:19:31,423 INFO [train.py:928] (1/4) Computing validation loss 2023-04-01 12:19:38,338 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.8280, 3.4778, 2.3834, 3.4020, 0.7980, 3.2190, 3.1246, 3.5193], device='cuda:1'), covar=tensor([0.0765, 0.0875, 0.2478, 0.0598, 0.4861, 0.0975, 0.0828, 0.0859], device='cuda:1'), in_proj_covar=tensor([0.0394, 0.0343, 0.0405, 0.0296, 0.0365, 0.0329, 0.0319, 0.0358], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-01 12:19:44,057 INFO [train.py:937] (1/4) Epoch 8, validation: loss=0.1875, simple_loss=0.2879, pruned_loss=0.04358, over 944034.00 frames. 2023-04-01 12:19:44,058 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18649MB 2023-04-01 12:19:46,423 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_rvb from training. Duration: 29.735 2023-04-01 12:19:49,174 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=50800.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 12:20:21,742 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=50826.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 12:20:45,722 INFO [train.py:903] (1/4) Epoch 8, batch 3050, loss[loss=0.2019, simple_loss=0.2757, pruned_loss=0.06408, over 19709.00 frames. ], tot_loss[loss=0.2566, simple_loss=0.3245, pruned_loss=0.09431, over 3782912.16 frames. ], batch size: 45, lr: 1.05e-02, grad_scale: 8.0 2023-04-01 12:20:55,178 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.189e+02 5.734e+02 7.199e+02 9.163e+02 1.650e+03, threshold=1.440e+03, percent-clipped=2.0 2023-04-01 12:21:06,918 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=50861.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 12:21:36,767 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=50886.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 12:21:49,951 INFO [train.py:903] (1/4) Epoch 8, batch 3100, loss[loss=0.2254, simple_loss=0.293, pruned_loss=0.07897, over 19694.00 frames. ], tot_loss[loss=0.255, simple_loss=0.3227, pruned_loss=0.09361, over 3780494.30 frames. ], batch size: 53, lr: 1.05e-02, grad_scale: 8.0 2023-04-01 12:22:04,759 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=50908.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 12:22:32,041 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=50930.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 12:22:52,354 INFO [train.py:903] (1/4) Epoch 8, batch 3150, loss[loss=0.2346, simple_loss=0.3108, pruned_loss=0.07927, over 19794.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.3227, pruned_loss=0.09341, over 3792641.32 frames. ], batch size: 56, lr: 1.05e-02, grad_scale: 8.0 2023-04-01 12:23:00,490 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.154e+02 5.915e+02 7.023e+02 8.955e+02 1.571e+03, threshold=1.405e+03, percent-clipped=4.0 2023-04-01 12:23:03,134 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=50955.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 12:23:14,734 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=50964.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 12:23:20,380 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 2023-04-01 12:23:45,417 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.7570, 2.6543, 2.0398, 2.0067, 1.7737, 2.2076, 1.1741, 1.9947], device='cuda:1'), covar=tensor([0.0324, 0.0349, 0.0361, 0.0536, 0.0628, 0.0577, 0.0626, 0.0511], device='cuda:1'), in_proj_covar=tensor([0.0318, 0.0316, 0.0316, 0.0333, 0.0408, 0.0330, 0.0295, 0.0313], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 12:23:54,308 INFO [train.py:903] (1/4) Epoch 8, batch 3200, loss[loss=0.221, simple_loss=0.2908, pruned_loss=0.07562, over 19744.00 frames. ], tot_loss[loss=0.2538, simple_loss=0.3217, pruned_loss=0.09295, over 3807261.88 frames. ], batch size: 45, lr: 1.04e-02, grad_scale: 8.0 2023-04-01 12:24:10,783 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.69 vs. limit=2.0 2023-04-01 12:24:30,138 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=51023.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 12:24:57,557 INFO [train.py:903] (1/4) Epoch 8, batch 3250, loss[loss=0.2295, simple_loss=0.3099, pruned_loss=0.07454, over 19534.00 frames. ], tot_loss[loss=0.2539, simple_loss=0.3219, pruned_loss=0.09299, over 3805095.70 frames. ], batch size: 54, lr: 1.04e-02, grad_scale: 8.0 2023-04-01 12:25:00,409 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.8893, 1.9379, 1.9779, 2.8422, 2.0131, 2.6600, 2.3941, 1.9221], device='cuda:1'), covar=tensor([0.2814, 0.2375, 0.1173, 0.1209, 0.2536, 0.1044, 0.2428, 0.2088], device='cuda:1'), in_proj_covar=tensor([0.0727, 0.0738, 0.0615, 0.0856, 0.0731, 0.0641, 0.0758, 0.0660], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 12:25:05,782 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.083e+02 6.087e+02 7.883e+02 9.942e+02 3.174e+03, threshold=1.577e+03, percent-clipped=7.0 2023-04-01 12:25:10,903 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.8730, 4.4318, 2.3970, 3.9210, 1.2538, 4.0936, 4.0575, 4.3018], device='cuda:1'), covar=tensor([0.0565, 0.0970, 0.2164, 0.0643, 0.3724, 0.0708, 0.0778, 0.0879], device='cuda:1'), in_proj_covar=tensor([0.0401, 0.0348, 0.0407, 0.0299, 0.0372, 0.0332, 0.0326, 0.0362], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-01 12:25:20,241 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.49 vs. limit=5.0 2023-04-01 12:25:40,469 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=51080.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 12:26:00,812 INFO [train.py:903] (1/4) Epoch 8, batch 3300, loss[loss=0.2464, simple_loss=0.3062, pruned_loss=0.09328, over 19613.00 frames. ], tot_loss[loss=0.2539, simple_loss=0.3218, pruned_loss=0.09298, over 3816036.71 frames. ], batch size: 50, lr: 1.04e-02, grad_scale: 8.0 2023-04-01 12:26:01,340 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5168, 1.5957, 1.6844, 2.0894, 1.3441, 1.7413, 2.0017, 1.6110], device='cuda:1'), covar=tensor([0.2778, 0.2265, 0.1240, 0.1277, 0.2524, 0.1220, 0.2634, 0.2169], device='cuda:1'), in_proj_covar=tensor([0.0726, 0.0738, 0.0614, 0.0857, 0.0732, 0.0644, 0.0758, 0.0659], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 12:26:08,875 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 2023-04-01 12:26:26,570 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3662, 1.4041, 1.8534, 1.5199, 2.7291, 2.2412, 2.7676, 1.2617], device='cuda:1'), covar=tensor([0.1954, 0.3179, 0.1890, 0.1620, 0.1292, 0.1688, 0.1416, 0.3150], device='cuda:1'), in_proj_covar=tensor([0.0462, 0.0537, 0.0535, 0.0416, 0.0573, 0.0469, 0.0631, 0.0465], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 12:26:44,960 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0378, 3.5815, 1.9612, 2.2175, 3.1771, 1.9007, 1.1808, 1.9872], device='cuda:1'), covar=tensor([0.0890, 0.0416, 0.0793, 0.0600, 0.0343, 0.0855, 0.0849, 0.0593], device='cuda:1'), in_proj_covar=tensor([0.0280, 0.0292, 0.0312, 0.0239, 0.0230, 0.0311, 0.0280, 0.0255], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 12:27:02,452 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=51144.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 12:27:04,792 INFO [train.py:903] (1/4) Epoch 8, batch 3350, loss[loss=0.2581, simple_loss=0.3318, pruned_loss=0.09225, over 19666.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3229, pruned_loss=0.0941, over 3809145.24 frames. ], batch size: 60, lr: 1.04e-02, grad_scale: 8.0 2023-04-01 12:27:12,718 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.478e+02 5.942e+02 7.377e+02 9.279e+02 2.136e+03, threshold=1.475e+03, percent-clipped=2.0 2023-04-01 12:27:14,292 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1410, 1.2293, 1.5923, 1.3452, 2.1901, 1.9411, 2.2612, 0.6796], device='cuda:1'), covar=tensor([0.2007, 0.3442, 0.1957, 0.1599, 0.1255, 0.1728, 0.1256, 0.3421], device='cuda:1'), in_proj_covar=tensor([0.0464, 0.0541, 0.0537, 0.0418, 0.0574, 0.0470, 0.0633, 0.0467], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 12:27:34,025 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=51170.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 12:28:00,622 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=51190.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 12:28:07,180 INFO [train.py:903] (1/4) Epoch 8, batch 3400, loss[loss=0.2479, simple_loss=0.32, pruned_loss=0.0879, over 19666.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.3239, pruned_loss=0.09438, over 3819898.06 frames. ], batch size: 55, lr: 1.04e-02, grad_scale: 8.0 2023-04-01 12:28:26,744 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.46 vs. limit=2.0 2023-04-01 12:29:10,668 INFO [train.py:903] (1/4) Epoch 8, batch 3450, loss[loss=0.2781, simple_loss=0.3465, pruned_loss=0.1049, over 19693.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.3235, pruned_loss=0.09383, over 3822127.68 frames. ], batch size: 60, lr: 1.04e-02, grad_scale: 8.0 2023-04-01 12:29:16,194 WARNING [train.py:1073] (1/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9_rvb from training. Duration: 25.2444375 2023-04-01 12:29:18,574 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.243e+02 6.280e+02 7.477e+02 9.686e+02 1.820e+03, threshold=1.495e+03, percent-clipped=3.0 2023-04-01 12:29:27,053 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=51259.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 12:29:52,038 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=51279.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 12:30:00,084 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=51285.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 12:30:01,385 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.2294, 1.6895, 1.7882, 2.1744, 1.9518, 1.9146, 1.7983, 2.1578], device='cuda:1'), covar=tensor([0.0742, 0.1482, 0.1197, 0.0747, 0.1102, 0.0439, 0.0966, 0.0518], device='cuda:1'), in_proj_covar=tensor([0.0249, 0.0352, 0.0285, 0.0236, 0.0295, 0.0240, 0.0269, 0.0233], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 12:30:12,716 INFO [train.py:903] (1/4) Epoch 8, batch 3500, loss[loss=0.2674, simple_loss=0.3395, pruned_loss=0.09762, over 19615.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.3233, pruned_loss=0.09363, over 3827975.48 frames. ], batch size: 57, lr: 1.04e-02, grad_scale: 8.0 2023-04-01 12:30:25,552 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=51304.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 12:30:29,262 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([0.8586, 1.3391, 1.0545, 1.0066, 1.1911, 0.9547, 0.9311, 1.2411], device='cuda:1'), covar=tensor([0.0487, 0.0681, 0.0936, 0.0515, 0.0417, 0.1091, 0.0514, 0.0384], device='cuda:1'), in_proj_covar=tensor([0.0281, 0.0292, 0.0315, 0.0242, 0.0230, 0.0309, 0.0283, 0.0254], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 12:30:30,171 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=51308.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 12:31:19,055 INFO [train.py:903] (1/4) Epoch 8, batch 3550, loss[loss=0.2751, simple_loss=0.3421, pruned_loss=0.1041, over 19483.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.3232, pruned_loss=0.09385, over 3809680.36 frames. ], batch size: 64, lr: 1.04e-02, grad_scale: 8.0 2023-04-01 12:31:27,383 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.001e+02 5.209e+02 6.396e+02 8.561e+02 1.899e+03, threshold=1.279e+03, percent-clipped=3.0 2023-04-01 12:31:30,122 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1831, 1.1590, 1.3619, 1.3030, 1.6824, 1.7592, 1.7134, 0.5054], device='cuda:1'), covar=tensor([0.1996, 0.3376, 0.1988, 0.1593, 0.1254, 0.1794, 0.1209, 0.3325], device='cuda:1'), in_proj_covar=tensor([0.0466, 0.0539, 0.0540, 0.0419, 0.0575, 0.0469, 0.0634, 0.0468], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 12:32:21,070 INFO [train.py:903] (1/4) Epoch 8, batch 3600, loss[loss=0.2557, simple_loss=0.3229, pruned_loss=0.09429, over 17356.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.3228, pruned_loss=0.09391, over 3813878.78 frames. ], batch size: 101, lr: 1.04e-02, grad_scale: 8.0 2023-04-01 12:32:26,129 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=51400.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 12:32:38,753 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=51410.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 12:32:55,836 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=51423.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 12:32:56,815 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=51424.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 12:33:22,342 INFO [train.py:903] (1/4) Epoch 8, batch 3650, loss[loss=0.2619, simple_loss=0.3324, pruned_loss=0.09567, over 19090.00 frames. ], tot_loss[loss=0.2551, simple_loss=0.3229, pruned_loss=0.0936, over 3815197.01 frames. ], batch size: 69, lr: 1.04e-02, grad_scale: 8.0 2023-04-01 12:33:31,510 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.306e+02 6.415e+02 7.779e+02 9.918e+02 2.619e+03, threshold=1.556e+03, percent-clipped=14.0 2023-04-01 12:34:24,454 INFO [train.py:903] (1/4) Epoch 8, batch 3700, loss[loss=0.2708, simple_loss=0.3428, pruned_loss=0.09936, over 19765.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.3239, pruned_loss=0.0944, over 3822272.12 frames. ], batch size: 54, lr: 1.04e-02, grad_scale: 8.0 2023-04-01 12:34:31,565 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2131, 2.0519, 1.5110, 1.4660, 1.9139, 1.1367, 1.1924, 1.7689], device='cuda:1'), covar=tensor([0.0721, 0.0574, 0.0888, 0.0533, 0.0382, 0.1007, 0.0557, 0.0353], device='cuda:1'), in_proj_covar=tensor([0.0279, 0.0290, 0.0315, 0.0242, 0.0227, 0.0311, 0.0279, 0.0251], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 12:34:46,883 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=51513.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 12:34:49,337 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=51515.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 12:35:12,250 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=51534.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 12:35:18,876 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=51539.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 12:35:20,110 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=51540.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 12:35:22,153 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=51541.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 12:35:28,598 INFO [train.py:903] (1/4) Epoch 8, batch 3750, loss[loss=0.2436, simple_loss=0.3143, pruned_loss=0.08646, over 19602.00 frames. ], tot_loss[loss=0.2562, simple_loss=0.324, pruned_loss=0.09423, over 3809588.46 frames. ], batch size: 61, lr: 1.04e-02, grad_scale: 8.0 2023-04-01 12:35:36,643 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.985e+02 5.907e+02 7.282e+02 9.270e+02 2.268e+03, threshold=1.456e+03, percent-clipped=4.0 2023-04-01 12:35:51,385 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.39 vs. limit=5.0 2023-04-01 12:35:52,290 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=51566.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 12:36:30,191 INFO [train.py:903] (1/4) Epoch 8, batch 3800, loss[loss=0.241, simple_loss=0.3103, pruned_loss=0.08579, over 19695.00 frames. ], tot_loss[loss=0.2564, simple_loss=0.3241, pruned_loss=0.0944, over 3795166.92 frames. ], batch size: 59, lr: 1.04e-02, grad_scale: 8.0 2023-04-01 12:37:02,476 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9_rvb from training. Duration: 29.1166875 2023-04-01 12:37:31,362 INFO [train.py:903] (1/4) Epoch 8, batch 3850, loss[loss=0.291, simple_loss=0.3641, pruned_loss=0.109, over 19508.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.3234, pruned_loss=0.09394, over 3794299.63 frames. ], batch size: 64, lr: 1.04e-02, grad_scale: 8.0 2023-04-01 12:37:35,087 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=51649.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 12:37:40,053 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.568e+02 6.155e+02 7.716e+02 1.023e+03 2.199e+03, threshold=1.543e+03, percent-clipped=8.0 2023-04-01 12:38:13,424 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=51679.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 12:38:33,776 INFO [train.py:903] (1/4) Epoch 8, batch 3900, loss[loss=0.2184, simple_loss=0.2974, pruned_loss=0.06971, over 19685.00 frames. ], tot_loss[loss=0.2535, simple_loss=0.3218, pruned_loss=0.09253, over 3813776.10 frames. ], batch size: 53, lr: 1.04e-02, grad_scale: 8.0 2023-04-01 12:38:44,951 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=51704.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 12:39:28,917 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 2023-04-01 12:39:33,665 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=51744.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 12:39:37,151 INFO [train.py:903] (1/4) Epoch 8, batch 3950, loss[loss=0.2343, simple_loss=0.3101, pruned_loss=0.07927, over 19301.00 frames. ], tot_loss[loss=0.2531, simple_loss=0.3213, pruned_loss=0.09246, over 3808211.47 frames. ], batch size: 66, lr: 1.04e-02, grad_scale: 8.0 2023-04-01 12:39:41,707 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 2023-04-01 12:39:45,221 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.212e+02 5.815e+02 7.280e+02 9.203e+02 2.422e+03, threshold=1.456e+03, percent-clipped=4.0 2023-04-01 12:39:46,650 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=51754.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 12:39:59,509 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=51765.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 12:40:37,949 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=51795.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 12:40:38,725 INFO [train.py:903] (1/4) Epoch 8, batch 4000, loss[loss=0.2582, simple_loss=0.3329, pruned_loss=0.09177, over 19498.00 frames. ], tot_loss[loss=0.2528, simple_loss=0.3213, pruned_loss=0.09213, over 3803152.72 frames. ], batch size: 64, lr: 1.04e-02, grad_scale: 8.0 2023-04-01 12:41:09,893 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=51820.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 12:41:27,808 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1_rvb from training. Duration: 0.7545625 2023-04-01 12:41:41,697 INFO [train.py:903] (1/4) Epoch 8, batch 4050, loss[loss=0.2968, simple_loss=0.3587, pruned_loss=0.1175, over 17477.00 frames. ], tot_loss[loss=0.2535, simple_loss=0.3219, pruned_loss=0.09255, over 3781484.36 frames. ], batch size: 101, lr: 1.04e-02, grad_scale: 8.0 2023-04-01 12:41:50,759 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.576e+02 5.742e+02 7.614e+02 9.901e+02 2.045e+03, threshold=1.523e+03, percent-clipped=5.0 2023-04-01 12:41:55,464 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=51857.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 12:41:59,733 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=51859.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 12:42:11,936 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=51869.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 12:42:33,321 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.68 vs. limit=2.0 2023-04-01 12:42:43,847 INFO [train.py:903] (1/4) Epoch 8, batch 4100, loss[loss=0.2622, simple_loss=0.329, pruned_loss=0.09769, over 19786.00 frames. ], tot_loss[loss=0.2534, simple_loss=0.322, pruned_loss=0.09236, over 3800355.30 frames. ], batch size: 56, lr: 1.04e-02, grad_scale: 8.0 2023-04-01 12:42:54,390 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 2023-04-01 12:42:56,231 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=51905.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 12:43:21,444 WARNING [train.py:1073] (1/4) Exclude cut with ID 497-129325-0061-9566_sp1.1_rvb from training. Duration: 0.97725 2023-04-01 12:43:26,433 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=51930.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 12:43:47,778 INFO [train.py:903] (1/4) Epoch 8, batch 4150, loss[loss=0.2269, simple_loss=0.2898, pruned_loss=0.082, over 19781.00 frames. ], tot_loss[loss=0.2529, simple_loss=0.3214, pruned_loss=0.09218, over 3795912.63 frames. ], batch size: 47, lr: 1.03e-02, grad_scale: 8.0 2023-04-01 12:43:56,802 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.129e+02 6.343e+02 7.798e+02 9.790e+02 2.215e+03, threshold=1.560e+03, percent-clipped=4.0 2023-04-01 12:44:19,720 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=51972.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 12:44:50,678 INFO [train.py:903] (1/4) Epoch 8, batch 4200, loss[loss=0.2597, simple_loss=0.327, pruned_loss=0.09618, over 19412.00 frames. ], tot_loss[loss=0.2518, simple_loss=0.3205, pruned_loss=0.09159, over 3801714.29 frames. ], batch size: 70, lr: 1.03e-02, grad_scale: 8.0 2023-04-01 12:44:57,640 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 2023-04-01 12:45:04,991 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.3846, 2.3551, 2.4318, 3.2465, 2.6518, 3.2320, 2.9098, 2.3716], device='cuda:1'), covar=tensor([0.2470, 0.1956, 0.0974, 0.1104, 0.2096, 0.0790, 0.1754, 0.1572], device='cuda:1'), in_proj_covar=tensor([0.0727, 0.0737, 0.0616, 0.0860, 0.0734, 0.0643, 0.0759, 0.0662], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 12:45:53,322 INFO [train.py:903] (1/4) Epoch 8, batch 4250, loss[loss=0.2042, simple_loss=0.2723, pruned_loss=0.06802, over 18712.00 frames. ], tot_loss[loss=0.2514, simple_loss=0.3198, pruned_loss=0.09152, over 3802900.13 frames. ], batch size: 41, lr: 1.03e-02, grad_scale: 8.0 2023-04-01 12:46:01,321 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.676e+02 5.455e+02 6.472e+02 8.916e+02 2.597e+03, threshold=1.294e+03, percent-clipped=4.0 2023-04-01 12:46:08,438 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=52058.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 12:46:11,482 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9_rvb from training. Duration: 29.816625 2023-04-01 12:46:21,888 WARNING [train.py:1073] (1/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 2023-04-01 12:46:55,486 INFO [train.py:903] (1/4) Epoch 8, batch 4300, loss[loss=0.257, simple_loss=0.3226, pruned_loss=0.09571, over 19832.00 frames. ], tot_loss[loss=0.2521, simple_loss=0.3203, pruned_loss=0.09194, over 3782085.96 frames. ], batch size: 52, lr: 1.03e-02, grad_scale: 8.0 2023-04-01 12:47:14,058 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=52109.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 12:47:22,339 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=52115.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 12:47:29,334 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6090, 1.6450, 1.7821, 2.2035, 1.4416, 1.9207, 2.0737, 1.7345], device='cuda:1'), covar=tensor([0.2801, 0.2173, 0.1221, 0.1189, 0.2355, 0.1091, 0.2764, 0.2139], device='cuda:1'), in_proj_covar=tensor([0.0728, 0.0739, 0.0616, 0.0863, 0.0737, 0.0648, 0.0761, 0.0664], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 12:47:34,043 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=52125.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 12:47:50,782 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 2023-04-01 12:47:53,227 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=52140.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 12:48:00,440 INFO [train.py:903] (1/4) Epoch 8, batch 4350, loss[loss=0.2619, simple_loss=0.3135, pruned_loss=0.1051, over 19792.00 frames. ], tot_loss[loss=0.2514, simple_loss=0.3196, pruned_loss=0.09153, over 3780568.72 frames. ], batch size: 47, lr: 1.03e-02, grad_scale: 8.0 2023-04-01 12:48:06,434 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=52150.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 12:48:09,408 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.508e+02 5.672e+02 7.291e+02 9.101e+02 1.997e+03, threshold=1.458e+03, percent-clipped=8.0 2023-04-01 12:49:03,039 INFO [train.py:903] (1/4) Epoch 8, batch 4400, loss[loss=0.2926, simple_loss=0.349, pruned_loss=0.1181, over 17353.00 frames. ], tot_loss[loss=0.2524, simple_loss=0.3209, pruned_loss=0.09193, over 3778662.40 frames. ], batch size: 101, lr: 1.03e-02, grad_scale: 8.0 2023-04-01 12:49:21,135 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=52211.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 12:49:26,733 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_rvb from training. Duration: 25.285 2023-04-01 12:49:38,179 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 2023-04-01 12:49:38,557 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=52224.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 12:49:43,301 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=52228.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 12:50:04,685 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 2023-04-01 12:50:05,239 INFO [train.py:903] (1/4) Epoch 8, batch 4450, loss[loss=0.2299, simple_loss=0.3061, pruned_loss=0.07688, over 18815.00 frames. ], tot_loss[loss=0.2524, simple_loss=0.3212, pruned_loss=0.0918, over 3796657.10 frames. ], batch size: 74, lr: 1.03e-02, grad_scale: 8.0 2023-04-01 12:50:13,310 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.227e+02 5.880e+02 7.086e+02 8.839e+02 1.936e+03, threshold=1.417e+03, percent-clipped=3.0 2023-04-01 12:50:13,728 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=52253.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 12:51:06,386 INFO [train.py:903] (1/4) Epoch 8, batch 4500, loss[loss=0.2041, simple_loss=0.2713, pruned_loss=0.06847, over 19801.00 frames. ], tot_loss[loss=0.2526, simple_loss=0.3211, pruned_loss=0.09208, over 3797321.31 frames. ], batch size: 45, lr: 1.03e-02, grad_scale: 8.0 2023-04-01 12:51:08,366 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 2023-04-01 12:51:50,106 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=52330.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 12:51:57,139 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.3720, 1.7863, 1.8594, 2.5453, 1.9542, 2.4755, 2.4543, 2.2320], device='cuda:1'), covar=tensor([0.0707, 0.0962, 0.1003, 0.1016, 0.1000, 0.0681, 0.0953, 0.0642], device='cuda:1'), in_proj_covar=tensor([0.0215, 0.0234, 0.0229, 0.0259, 0.0247, 0.0219, 0.0210, 0.0208], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1') 2023-04-01 12:52:10,344 INFO [train.py:903] (1/4) Epoch 8, batch 4550, loss[loss=0.2499, simple_loss=0.3177, pruned_loss=0.09109, over 19528.00 frames. ], tot_loss[loss=0.2514, simple_loss=0.3199, pruned_loss=0.09145, over 3801277.41 frames. ], batch size: 56, lr: 1.03e-02, grad_scale: 8.0 2023-04-01 12:52:18,695 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.675e+02 5.922e+02 7.010e+02 8.869e+02 1.679e+03, threshold=1.402e+03, percent-clipped=2.0 2023-04-01 12:52:18,728 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9_rvb from training. Duration: 28.72225 2023-04-01 12:52:28,212 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.3657, 2.1519, 1.8192, 1.7743, 1.6475, 1.7070, 0.4501, 1.2222], device='cuda:1'), covar=tensor([0.0325, 0.0353, 0.0314, 0.0447, 0.0666, 0.0496, 0.0790, 0.0614], device='cuda:1'), in_proj_covar=tensor([0.0321, 0.0319, 0.0316, 0.0335, 0.0413, 0.0332, 0.0302, 0.0318], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 12:52:28,251 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2183, 1.2862, 1.6259, 1.4220, 2.6708, 2.3022, 2.8333, 1.1441], device='cuda:1'), covar=tensor([0.2102, 0.3499, 0.2087, 0.1630, 0.1358, 0.1654, 0.1352, 0.3320], device='cuda:1'), in_proj_covar=tensor([0.0470, 0.0546, 0.0545, 0.0422, 0.0584, 0.0475, 0.0646, 0.0472], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 12:52:41,942 WARNING [train.py:1073] (1/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 2023-04-01 12:52:51,954 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.87 vs. limit=5.0 2023-04-01 12:53:11,430 INFO [train.py:903] (1/4) Epoch 8, batch 4600, loss[loss=0.2255, simple_loss=0.306, pruned_loss=0.0725, over 19544.00 frames. ], tot_loss[loss=0.251, simple_loss=0.3198, pruned_loss=0.09108, over 3811263.75 frames. ], batch size: 56, lr: 1.03e-02, grad_scale: 8.0 2023-04-01 12:53:18,528 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=52402.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 12:54:12,886 INFO [train.py:903] (1/4) Epoch 8, batch 4650, loss[loss=0.2199, simple_loss=0.2848, pruned_loss=0.07755, over 19783.00 frames. ], tot_loss[loss=0.2513, simple_loss=0.3203, pruned_loss=0.09109, over 3815616.62 frames. ], batch size: 48, lr: 1.03e-02, grad_scale: 8.0 2023-04-01 12:54:21,260 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.655e+02 5.664e+02 6.903e+02 8.285e+02 1.576e+03, threshold=1.381e+03, percent-clipped=2.0 2023-04-01 12:54:30,512 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_rvb from training. Duration: 0.92 2023-04-01 12:54:42,674 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_rvb from training. Duration: 0.83 2023-04-01 12:54:56,903 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=52480.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 12:55:15,231 INFO [train.py:903] (1/4) Epoch 8, batch 4700, loss[loss=0.2023, simple_loss=0.2859, pruned_loss=0.05941, over 19874.00 frames. ], tot_loss[loss=0.2486, simple_loss=0.3178, pruned_loss=0.08974, over 3823401.03 frames. ], batch size: 52, lr: 1.03e-02, grad_scale: 8.0 2023-04-01 12:55:27,906 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=52505.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 12:55:39,731 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9_rvb from training. Duration: 0.92225 2023-04-01 12:55:43,520 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=52517.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 12:56:18,062 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 2023-04-01 12:56:18,360 INFO [train.py:903] (1/4) Epoch 8, batch 4750, loss[loss=0.2904, simple_loss=0.3523, pruned_loss=0.1142, over 19756.00 frames. ], tot_loss[loss=0.2503, simple_loss=0.319, pruned_loss=0.09083, over 3822288.81 frames. ], batch size: 54, lr: 1.03e-02, grad_scale: 8.0 2023-04-01 12:56:29,686 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.555e+02 6.316e+02 7.348e+02 9.529e+02 1.491e+03, threshold=1.470e+03, percent-clipped=3.0 2023-04-01 12:56:31,067 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=52555.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 12:56:52,298 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=52573.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 12:57:12,459 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.0085, 1.3296, 0.9866, 0.9972, 1.1514, 0.9345, 0.8732, 1.2366], device='cuda:1'), covar=tensor([0.0424, 0.0601, 0.0932, 0.0503, 0.0457, 0.0979, 0.0536, 0.0362], device='cuda:1'), in_proj_covar=tensor([0.0281, 0.0292, 0.0315, 0.0241, 0.0229, 0.0312, 0.0288, 0.0256], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 12:57:22,235 INFO [train.py:903] (1/4) Epoch 8, batch 4800, loss[loss=0.274, simple_loss=0.344, pruned_loss=0.102, over 17360.00 frames. ], tot_loss[loss=0.2512, simple_loss=0.3201, pruned_loss=0.09113, over 3823701.92 frames. ], batch size: 101, lr: 1.03e-02, grad_scale: 8.0 2023-04-01 12:58:16,379 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 2023-04-01 12:58:22,761 INFO [train.py:903] (1/4) Epoch 8, batch 4850, loss[loss=0.2526, simple_loss=0.3314, pruned_loss=0.08684, over 19097.00 frames. ], tot_loss[loss=0.2513, simple_loss=0.3202, pruned_loss=0.09119, over 3820011.20 frames. ], batch size: 69, lr: 1.03e-02, grad_scale: 8.0 2023-04-01 12:58:32,081 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.583e+02 6.021e+02 7.604e+02 9.872e+02 2.114e+03, threshold=1.521e+03, percent-clipped=8.0 2023-04-01 12:58:46,867 WARNING [train.py:1073] (1/4) Exclude cut with ID 774-127930-0014-48411_sp1.1_rvb from training. Duration: 0.95 2023-04-01 12:58:52,939 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=52670.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 12:58:58,261 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=52674.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 12:58:59,836 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.8124, 1.8830, 1.9490, 2.6694, 1.7256, 2.3668, 2.3743, 1.8919], device='cuda:1'), covar=tensor([0.2964, 0.2453, 0.1273, 0.1430, 0.2727, 0.1133, 0.2645, 0.2155], device='cuda:1'), in_proj_covar=tensor([0.0729, 0.0739, 0.0610, 0.0867, 0.0733, 0.0638, 0.0761, 0.0659], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 12:59:00,297 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.82 vs. limit=2.0 2023-04-01 12:59:08,337 WARNING [train.py:1073] (1/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1_rvb from training. Duration: 0.9409375 2023-04-01 12:59:14,086 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 2023-04-01 12:59:14,111 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 2023-04-01 12:59:23,224 INFO [train.py:903] (1/4) Epoch 8, batch 4900, loss[loss=0.2682, simple_loss=0.3196, pruned_loss=0.1084, over 19380.00 frames. ], tot_loss[loss=0.2516, simple_loss=0.3204, pruned_loss=0.09144, over 3813533.40 frames. ], batch size: 47, lr: 1.03e-02, grad_scale: 8.0 2023-04-01 12:59:24,410 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1_rvb from training. Duration: 25.3818125 2023-04-01 12:59:44,290 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 2023-04-01 13:00:22,662 INFO [train.py:903] (1/4) Epoch 8, batch 4950, loss[loss=0.2577, simple_loss=0.3326, pruned_loss=0.09136, over 19775.00 frames. ], tot_loss[loss=0.252, simple_loss=0.3207, pruned_loss=0.09165, over 3819649.95 frames. ], batch size: 56, lr: 1.03e-02, grad_scale: 8.0 2023-04-01 13:00:35,712 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.348e+02 6.374e+02 8.178e+02 1.048e+03 2.702e+03, threshold=1.636e+03, percent-clipped=11.0 2023-04-01 13:00:40,357 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 2023-04-01 13:00:53,296 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=52770.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 13:00:56,991 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=52773.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 13:01:04,909 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 2023-04-01 13:01:17,600 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=52789.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 13:01:26,606 INFO [train.py:903] (1/4) Epoch 8, batch 5000, loss[loss=0.2239, simple_loss=0.3006, pruned_loss=0.07362, over 19691.00 frames. ], tot_loss[loss=0.2521, simple_loss=0.3206, pruned_loss=0.09183, over 3813333.52 frames. ], batch size: 53, lr: 1.03e-02, grad_scale: 4.0 2023-04-01 13:01:29,394 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=52798.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 13:01:35,984 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_rvb from training. Duration: 27.14 2023-04-01 13:01:47,277 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 2023-04-01 13:02:28,523 INFO [train.py:903] (1/4) Epoch 8, batch 5050, loss[loss=0.2003, simple_loss=0.2767, pruned_loss=0.06191, over 17760.00 frames. ], tot_loss[loss=0.2508, simple_loss=0.3196, pruned_loss=0.09106, over 3807153.88 frames. ], batch size: 39, lr: 1.03e-02, grad_scale: 4.0 2023-04-01 13:02:39,042 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.635e+02 5.648e+02 7.062e+02 8.811e+02 1.795e+03, threshold=1.412e+03, percent-clipped=2.0 2023-04-01 13:03:04,976 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9_rvb from training. Duration: 26.62775 2023-04-01 13:03:22,581 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.0098, 0.9554, 1.3679, 0.6426, 2.2940, 2.4589, 2.2283, 2.7213], device='cuda:1'), covar=tensor([0.1564, 0.4449, 0.3945, 0.2292, 0.0465, 0.0260, 0.0422, 0.0267], device='cuda:1'), in_proj_covar=tensor([0.0252, 0.0287, 0.0316, 0.0249, 0.0207, 0.0140, 0.0205, 0.0175], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 13:03:30,507 INFO [train.py:903] (1/4) Epoch 8, batch 5100, loss[loss=0.2378, simple_loss=0.3188, pruned_loss=0.07841, over 19740.00 frames. ], tot_loss[loss=0.2522, simple_loss=0.3209, pruned_loss=0.09182, over 3800671.46 frames. ], batch size: 63, lr: 1.03e-02, grad_scale: 4.0 2023-04-01 13:03:41,015 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 2023-04-01 13:03:45,417 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_rvb from training. Duration: 25.035 2023-04-01 13:03:50,849 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_rvb from training. Duration: 27.92 2023-04-01 13:03:59,150 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=52917.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 13:04:06,195 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2552, 1.2865, 1.6521, 1.4321, 2.2708, 2.0933, 2.4500, 0.8148], device='cuda:1'), covar=tensor([0.2022, 0.3451, 0.2078, 0.1647, 0.1437, 0.1698, 0.1388, 0.3462], device='cuda:1'), in_proj_covar=tensor([0.0462, 0.0532, 0.0538, 0.0412, 0.0574, 0.0468, 0.0631, 0.0466], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 13:04:09,597 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=52926.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 13:04:23,604 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5642, 1.6123, 1.6895, 2.1276, 1.3048, 1.9297, 2.0062, 1.6251], device='cuda:1'), covar=tensor([0.2802, 0.2346, 0.1291, 0.1330, 0.2615, 0.1119, 0.2828, 0.2266], device='cuda:1'), in_proj_covar=tensor([0.0723, 0.0739, 0.0609, 0.0864, 0.0733, 0.0639, 0.0761, 0.0655], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 13:04:32,211 INFO [train.py:903] (1/4) Epoch 8, batch 5150, loss[loss=0.2207, simple_loss=0.2879, pruned_loss=0.07679, over 19745.00 frames. ], tot_loss[loss=0.2519, simple_loss=0.3209, pruned_loss=0.09149, over 3811711.95 frames. ], batch size: 46, lr: 1.03e-02, grad_scale: 4.0 2023-04-01 13:04:41,017 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=52951.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 13:04:45,911 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.391e+02 6.060e+02 7.983e+02 1.041e+03 2.368e+03, threshold=1.597e+03, percent-clipped=6.0 2023-04-01 13:04:47,183 WARNING [train.py:1073] (1/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 2023-04-01 13:05:19,618 WARNING [train.py:1073] (1/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1_rvb from training. Duration: 0.9681875 2023-04-01 13:05:37,038 INFO [train.py:903] (1/4) Epoch 8, batch 5200, loss[loss=0.2632, simple_loss=0.3445, pruned_loss=0.09098, over 17961.00 frames. ], tot_loss[loss=0.2518, simple_loss=0.3209, pruned_loss=0.09133, over 3813422.30 frames. ], batch size: 83, lr: 1.02e-02, grad_scale: 8.0 2023-04-01 13:05:51,348 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 2023-04-01 13:06:21,624 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=53032.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 13:06:36,725 WARNING [train.py:1073] (1/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 2023-04-01 13:06:38,262 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=53045.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 13:06:39,062 INFO [train.py:903] (1/4) Epoch 8, batch 5250, loss[loss=0.3123, simple_loss=0.3635, pruned_loss=0.1306, over 13652.00 frames. ], tot_loss[loss=0.251, simple_loss=0.32, pruned_loss=0.09098, over 3815430.69 frames. ], batch size: 136, lr: 1.02e-02, grad_scale: 8.0 2023-04-01 13:06:49,012 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.283e+02 5.973e+02 7.081e+02 8.822e+02 3.028e+03, threshold=1.416e+03, percent-clipped=2.0 2023-04-01 13:07:08,287 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=53070.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 13:07:13,776 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=53074.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 13:07:39,272 INFO [train.py:903] (1/4) Epoch 8, batch 5300, loss[loss=0.2538, simple_loss=0.317, pruned_loss=0.09529, over 19546.00 frames. ], tot_loss[loss=0.2496, simple_loss=0.3183, pruned_loss=0.09043, over 3815899.21 frames. ], batch size: 54, lr: 1.02e-02, grad_scale: 8.0 2023-04-01 13:07:57,431 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 2023-04-01 13:08:03,083 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=53114.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 13:08:41,250 INFO [train.py:903] (1/4) Epoch 8, batch 5350, loss[loss=0.2643, simple_loss=0.3357, pruned_loss=0.09648, over 19624.00 frames. ], tot_loss[loss=0.2494, simple_loss=0.3185, pruned_loss=0.09017, over 3827214.37 frames. ], batch size: 57, lr: 1.02e-02, grad_scale: 8.0 2023-04-01 13:08:52,773 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.210e+02 6.176e+02 7.478e+02 9.376e+02 1.338e+03, threshold=1.496e+03, percent-clipped=0.0 2023-04-01 13:09:14,972 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.55 vs. limit=5.0 2023-04-01 13:09:18,561 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9_rvb from training. Duration: 26.438875 2023-04-01 13:09:44,008 INFO [train.py:903] (1/4) Epoch 8, batch 5400, loss[loss=0.296, simple_loss=0.359, pruned_loss=0.1165, over 17098.00 frames. ], tot_loss[loss=0.2492, simple_loss=0.3187, pruned_loss=0.08988, over 3819738.59 frames. ], batch size: 101, lr: 1.02e-02, grad_scale: 8.0 2023-04-01 13:10:15,109 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9544, 1.4877, 1.6756, 2.4533, 1.9433, 2.0877, 2.3769, 1.9786], device='cuda:1'), covar=tensor([0.0905, 0.1237, 0.1091, 0.0988, 0.0987, 0.0812, 0.0948, 0.0769], device='cuda:1'), in_proj_covar=tensor([0.0215, 0.0232, 0.0228, 0.0258, 0.0245, 0.0217, 0.0208, 0.0207], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1') 2023-04-01 13:10:24,924 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=53229.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 13:10:47,201 INFO [train.py:903] (1/4) Epoch 8, batch 5450, loss[loss=0.2216, simple_loss=0.2871, pruned_loss=0.07806, over 19387.00 frames. ], tot_loss[loss=0.2505, simple_loss=0.3193, pruned_loss=0.09091, over 3802396.13 frames. ], batch size: 47, lr: 1.02e-02, grad_scale: 8.0 2023-04-01 13:10:53,232 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3549, 1.4392, 2.2817, 1.7286, 3.2277, 2.4557, 3.1855, 1.6909], device='cuda:1'), covar=tensor([0.2085, 0.3671, 0.1909, 0.1613, 0.1397, 0.1824, 0.1832, 0.3070], device='cuda:1'), in_proj_covar=tensor([0.0468, 0.0539, 0.0545, 0.0417, 0.0577, 0.0473, 0.0637, 0.0470], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 13:10:57,343 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.737e+02 5.768e+02 7.150e+02 9.218e+02 2.127e+03, threshold=1.430e+03, percent-clipped=3.0 2023-04-01 13:11:02,296 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=53259.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 13:11:40,430 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=53288.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 13:11:48,929 INFO [train.py:903] (1/4) Epoch 8, batch 5500, loss[loss=0.2547, simple_loss=0.3107, pruned_loss=0.09933, over 19748.00 frames. ], tot_loss[loss=0.252, simple_loss=0.3204, pruned_loss=0.09178, over 3805870.97 frames. ], batch size: 46, lr: 1.02e-02, grad_scale: 8.0 2023-04-01 13:12:11,194 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=53313.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 13:12:16,565 WARNING [train.py:1073] (1/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9_rvb from training. Duration: 27.47775 2023-04-01 13:12:23,269 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=53322.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 13:12:34,907 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.77 vs. limit=2.0 2023-04-01 13:12:35,593 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=53333.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 13:12:50,182 INFO [train.py:903] (1/4) Epoch 8, batch 5550, loss[loss=0.2196, simple_loss=0.2965, pruned_loss=0.07131, over 19599.00 frames. ], tot_loss[loss=0.249, simple_loss=0.3179, pruned_loss=0.09008, over 3818578.46 frames. ], batch size: 52, lr: 1.02e-02, grad_scale: 8.0 2023-04-01 13:12:56,429 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.4558, 2.2510, 1.8737, 1.7530, 1.6619, 1.8154, 0.4628, 1.1987], device='cuda:1'), covar=tensor([0.0321, 0.0377, 0.0315, 0.0486, 0.0739, 0.0530, 0.0767, 0.0625], device='cuda:1'), in_proj_covar=tensor([0.0320, 0.0319, 0.0316, 0.0331, 0.0411, 0.0331, 0.0296, 0.0312], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 13:13:00,594 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 2023-04-01 13:13:03,076 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.436e+02 6.394e+02 7.821e+02 9.803e+02 2.197e+03, threshold=1.564e+03, percent-clipped=2.0 2023-04-01 13:13:40,338 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=53385.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 13:13:50,350 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 2023-04-01 13:13:53,923 INFO [train.py:903] (1/4) Epoch 8, batch 5600, loss[loss=0.2087, simple_loss=0.2866, pruned_loss=0.06545, over 19741.00 frames. ], tot_loss[loss=0.248, simple_loss=0.3174, pruned_loss=0.08928, over 3821018.13 frames. ], batch size: 51, lr: 1.02e-02, grad_scale: 8.0 2023-04-01 13:14:21,365 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=53418.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 13:14:57,497 INFO [train.py:903] (1/4) Epoch 8, batch 5650, loss[loss=0.2595, simple_loss=0.3349, pruned_loss=0.09201, over 18166.00 frames. ], tot_loss[loss=0.2488, simple_loss=0.3182, pruned_loss=0.08976, over 3823845.20 frames. ], batch size: 83, lr: 1.02e-02, grad_scale: 8.0 2023-04-01 13:15:07,835 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.837e+02 5.961e+02 7.306e+02 9.270e+02 2.985e+03, threshold=1.461e+03, percent-clipped=1.0 2023-04-01 13:15:45,791 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9_rvb from training. Duration: 28.0944375 2023-04-01 13:15:46,224 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=53485.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 13:15:58,760 INFO [train.py:903] (1/4) Epoch 8, batch 5700, loss[loss=0.2443, simple_loss=0.3235, pruned_loss=0.08257, over 19662.00 frames. ], tot_loss[loss=0.2484, simple_loss=0.3181, pruned_loss=0.08934, over 3828686.88 frames. ], batch size: 55, lr: 1.02e-02, grad_scale: 8.0 2023-04-01 13:16:15,382 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=53510.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 13:16:44,894 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=53533.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 13:16:59,703 INFO [train.py:903] (1/4) Epoch 8, batch 5750, loss[loss=0.2025, simple_loss=0.2849, pruned_loss=0.06002, over 19692.00 frames. ], tot_loss[loss=0.2477, simple_loss=0.3171, pruned_loss=0.08913, over 3809638.20 frames. ], batch size: 53, lr: 1.02e-02, grad_scale: 8.0 2023-04-01 13:17:00,947 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9_rvb from training. Duration: 33.038875 2023-04-01 13:17:10,146 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 2023-04-01 13:17:10,455 WARNING [train.py:1073] (1/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 2023-04-01 13:17:11,664 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.475e+02 5.699e+02 6.647e+02 8.184e+02 1.829e+03, threshold=1.329e+03, percent-clipped=1.0 2023-04-01 13:17:15,818 WARNING [train.py:1073] (1/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 2023-04-01 13:18:02,105 INFO [train.py:903] (1/4) Epoch 8, batch 5800, loss[loss=0.2434, simple_loss=0.3249, pruned_loss=0.0809, over 19688.00 frames. ], tot_loss[loss=0.249, simple_loss=0.3181, pruned_loss=0.08997, over 3818457.31 frames. ], batch size: 60, lr: 1.02e-02, grad_scale: 8.0 2023-04-01 13:18:12,770 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=53603.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 13:18:56,128 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.66 vs. limit=2.0 2023-04-01 13:19:04,521 INFO [train.py:903] (1/4) Epoch 8, batch 5850, loss[loss=0.2493, simple_loss=0.3233, pruned_loss=0.0876, over 19790.00 frames. ], tot_loss[loss=0.2481, simple_loss=0.3174, pruned_loss=0.08937, over 3818661.80 frames. ], batch size: 56, lr: 1.02e-02, grad_scale: 8.0 2023-04-01 13:19:15,050 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.702e+02 6.371e+02 7.746e+02 9.220e+02 2.993e+03, threshold=1.549e+03, percent-clipped=10.0 2023-04-01 13:19:27,608 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=53666.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 13:19:42,004 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=53677.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 13:20:05,078 INFO [train.py:903] (1/4) Epoch 8, batch 5900, loss[loss=0.253, simple_loss=0.3185, pruned_loss=0.09378, over 19742.00 frames. ], tot_loss[loss=0.2479, simple_loss=0.317, pruned_loss=0.08945, over 3827493.01 frames. ], batch size: 51, lr: 1.02e-02, grad_scale: 8.0 2023-04-01 13:20:09,570 WARNING [train.py:1073] (1/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 2023-04-01 13:20:30,254 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9_rvb from training. Duration: 27.511125 2023-04-01 13:20:30,494 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.8924, 4.2066, 4.5564, 4.4957, 2.0051, 4.2002, 3.7218, 4.1902], device='cuda:1'), covar=tensor([0.1220, 0.1103, 0.0534, 0.0524, 0.4381, 0.0550, 0.0543, 0.1010], device='cuda:1'), in_proj_covar=tensor([0.0611, 0.0534, 0.0727, 0.0606, 0.0669, 0.0469, 0.0456, 0.0663], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1') 2023-04-01 13:20:32,883 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=53718.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 13:20:47,011 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=53729.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 13:21:06,832 INFO [train.py:903] (1/4) Epoch 8, batch 5950, loss[loss=0.2292, simple_loss=0.3072, pruned_loss=0.07562, over 19745.00 frames. ], tot_loss[loss=0.2493, simple_loss=0.3181, pruned_loss=0.09026, over 3812315.36 frames. ], batch size: 63, lr: 1.02e-02, grad_scale: 8.0 2023-04-01 13:21:19,050 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.652e+02 5.886e+02 7.195e+02 1.025e+03 2.007e+03, threshold=1.439e+03, percent-clipped=3.0 2023-04-01 13:21:50,794 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=53781.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 13:21:52,059 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3365, 1.3409, 1.8884, 1.5010, 3.1275, 2.5346, 3.2571, 1.3153], device='cuda:1'), covar=tensor([0.2203, 0.3596, 0.2244, 0.1766, 0.1432, 0.1742, 0.1574, 0.3432], device='cuda:1'), in_proj_covar=tensor([0.0469, 0.0543, 0.0549, 0.0418, 0.0579, 0.0478, 0.0641, 0.0476], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 13:22:00,062 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=53789.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 13:22:04,689 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=53792.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 13:22:09,974 INFO [train.py:903] (1/4) Epoch 8, batch 6000, loss[loss=0.2068, simple_loss=0.2781, pruned_loss=0.06769, over 19363.00 frames. ], tot_loss[loss=0.2501, simple_loss=0.3188, pruned_loss=0.09071, over 3819485.86 frames. ], batch size: 44, lr: 1.02e-02, grad_scale: 8.0 2023-04-01 13:22:09,975 INFO [train.py:928] (1/4) Computing validation loss 2023-04-01 13:22:18,966 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.8288, 3.3955, 2.6555, 3.3843, 0.8755, 3.2565, 3.3366, 3.4363], device='cuda:1'), covar=tensor([0.0788, 0.0927, 0.1873, 0.0729, 0.4143, 0.1107, 0.0747, 0.0889], device='cuda:1'), in_proj_covar=tensor([0.0406, 0.0347, 0.0409, 0.0303, 0.0369, 0.0336, 0.0329, 0.0364], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-01 13:22:22,632 INFO [train.py:937] (1/4) Epoch 8, validation: loss=0.1864, simple_loss=0.2865, pruned_loss=0.04314, over 944034.00 frames. 2023-04-01 13:22:22,633 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18649MB 2023-04-01 13:22:48,334 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=53814.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 13:22:54,263 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.44 vs. limit=5.0 2023-04-01 13:23:24,848 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=53844.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 13:23:26,617 INFO [train.py:903] (1/4) Epoch 8, batch 6050, loss[loss=0.2489, simple_loss=0.324, pruned_loss=0.08685, over 19652.00 frames. ], tot_loss[loss=0.2505, simple_loss=0.3194, pruned_loss=0.09084, over 3820298.95 frames. ], batch size: 55, lr: 1.02e-02, grad_scale: 8.0 2023-04-01 13:23:39,145 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.317e+02 5.583e+02 7.013e+02 9.917e+02 2.418e+03, threshold=1.403e+03, percent-clipped=8.0 2023-04-01 13:23:47,675 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2248, 1.5229, 2.0719, 1.5977, 3.2107, 4.9009, 4.6871, 5.0792], device='cuda:1'), covar=tensor([0.1503, 0.2955, 0.2685, 0.1750, 0.0381, 0.0111, 0.0130, 0.0079], device='cuda:1'), in_proj_covar=tensor([0.0253, 0.0286, 0.0316, 0.0246, 0.0208, 0.0141, 0.0206, 0.0175], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 13:24:30,145 INFO [train.py:903] (1/4) Epoch 8, batch 6100, loss[loss=0.283, simple_loss=0.3465, pruned_loss=0.1097, over 19434.00 frames. ], tot_loss[loss=0.2483, simple_loss=0.3176, pruned_loss=0.08951, over 3833180.86 frames. ], batch size: 70, lr: 1.02e-02, grad_scale: 8.0 2023-04-01 13:24:37,005 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.8850, 4.4362, 2.6879, 3.8090, 1.0400, 4.0348, 4.0785, 4.2784], device='cuda:1'), covar=tensor([0.0567, 0.1007, 0.1930, 0.0790, 0.3999, 0.0750, 0.0826, 0.0983], device='cuda:1'), in_proj_covar=tensor([0.0408, 0.0347, 0.0410, 0.0307, 0.0372, 0.0335, 0.0329, 0.0365], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-01 13:24:52,630 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2208, 1.2528, 1.5717, 1.4000, 2.2426, 1.9585, 2.2877, 0.7496], device='cuda:1'), covar=tensor([0.2033, 0.3548, 0.2040, 0.1645, 0.1316, 0.1752, 0.1300, 0.3402], device='cuda:1'), in_proj_covar=tensor([0.0465, 0.0540, 0.0548, 0.0416, 0.0576, 0.0472, 0.0636, 0.0473], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 13:25:31,563 INFO [train.py:903] (1/4) Epoch 8, batch 6150, loss[loss=0.2715, simple_loss=0.3404, pruned_loss=0.1013, over 19526.00 frames. ], tot_loss[loss=0.2483, simple_loss=0.318, pruned_loss=0.0893, over 3827064.58 frames. ], batch size: 54, lr: 1.02e-02, grad_scale: 8.0 2023-04-01 13:25:42,184 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.923e+02 5.703e+02 7.303e+02 8.849e+02 1.874e+03, threshold=1.461e+03, percent-clipped=4.0 2023-04-01 13:25:56,461 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9_rvb from training. Duration: 31.02225 2023-04-01 13:26:07,892 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=53974.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 13:26:33,372 INFO [train.py:903] (1/4) Epoch 8, batch 6200, loss[loss=0.2205, simple_loss=0.2928, pruned_loss=0.07414, over 19738.00 frames. ], tot_loss[loss=0.2486, simple_loss=0.3181, pruned_loss=0.08953, over 3818463.52 frames. ], batch size: 46, lr: 1.02e-02, grad_scale: 8.0 2023-04-01 13:26:37,440 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=53999.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 13:27:25,494 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=54037.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 13:27:37,612 INFO [train.py:903] (1/4) Epoch 8, batch 6250, loss[loss=0.2783, simple_loss=0.3485, pruned_loss=0.104, over 18072.00 frames. ], tot_loss[loss=0.2484, simple_loss=0.318, pruned_loss=0.08943, over 3819187.42 frames. ], batch size: 83, lr: 1.01e-02, grad_scale: 8.0 2023-04-01 13:27:40,337 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=54048.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 13:27:49,335 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.888e+02 5.723e+02 7.068e+02 9.572e+02 2.133e+03, threshold=1.414e+03, percent-clipped=6.0 2023-04-01 13:27:58,068 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=54062.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 13:28:04,583 WARNING [train.py:1073] (1/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9_rvb from training. Duration: 25.988875 2023-04-01 13:28:10,762 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=54073.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 13:28:40,882 INFO [train.py:903] (1/4) Epoch 8, batch 6300, loss[loss=0.234, simple_loss=0.3144, pruned_loss=0.07677, over 19582.00 frames. ], tot_loss[loss=0.2496, simple_loss=0.3193, pruned_loss=0.08994, over 3818574.29 frames. ], batch size: 61, lr: 1.01e-02, grad_scale: 8.0 2023-04-01 13:28:45,535 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=54100.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 13:29:16,204 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=54125.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 13:29:41,755 INFO [train.py:903] (1/4) Epoch 8, batch 6350, loss[loss=0.2138, simple_loss=0.2837, pruned_loss=0.07201, over 19385.00 frames. ], tot_loss[loss=0.2507, simple_loss=0.3204, pruned_loss=0.09052, over 3820221.99 frames. ], batch size: 48, lr: 1.01e-02, grad_scale: 8.0 2023-04-01 13:29:52,049 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.072e+02 6.055e+02 7.520e+02 8.667e+02 2.456e+03, threshold=1.504e+03, percent-clipped=3.0 2023-04-01 13:30:43,419 INFO [train.py:903] (1/4) Epoch 8, batch 6400, loss[loss=0.2797, simple_loss=0.3509, pruned_loss=0.1043, over 19320.00 frames. ], tot_loss[loss=0.2496, simple_loss=0.319, pruned_loss=0.09013, over 3808949.84 frames. ], batch size: 66, lr: 1.01e-02, grad_scale: 8.0 2023-04-01 13:31:45,991 INFO [train.py:903] (1/4) Epoch 8, batch 6450, loss[loss=0.2351, simple_loss=0.3136, pruned_loss=0.07833, over 19696.00 frames. ], tot_loss[loss=0.2489, simple_loss=0.3187, pruned_loss=0.08954, over 3823171.93 frames. ], batch size: 59, lr: 1.01e-02, grad_scale: 8.0 2023-04-01 13:31:58,338 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.099e+02 5.571e+02 6.761e+02 8.500e+02 1.702e+03, threshold=1.352e+03, percent-clipped=3.0 2023-04-01 13:32:21,619 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 2023-04-01 13:32:30,246 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9_rvb from training. Duration: 28.638875 2023-04-01 13:32:48,752 INFO [train.py:903] (1/4) Epoch 8, batch 6500, loss[loss=0.2204, simple_loss=0.2877, pruned_loss=0.07652, over 19784.00 frames. ], tot_loss[loss=0.2505, simple_loss=0.3203, pruned_loss=0.0904, over 3822392.27 frames. ], batch size: 48, lr: 1.01e-02, grad_scale: 8.0 2023-04-01 13:32:54,426 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1_rvb from training. Duration: 0.836375 2023-04-01 13:33:34,583 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.8909, 4.4099, 2.5275, 3.8089, 1.0043, 4.1569, 4.1283, 4.3306], device='cuda:1'), covar=tensor([0.0540, 0.0955, 0.2068, 0.0675, 0.4092, 0.0752, 0.0730, 0.0869], device='cuda:1'), in_proj_covar=tensor([0.0407, 0.0345, 0.0411, 0.0303, 0.0371, 0.0334, 0.0328, 0.0362], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-01 13:33:50,182 INFO [train.py:903] (1/4) Epoch 8, batch 6550, loss[loss=0.2867, simple_loss=0.3505, pruned_loss=0.1115, over 18824.00 frames. ], tot_loss[loss=0.2505, simple_loss=0.32, pruned_loss=0.09051, over 3814447.03 frames. ], batch size: 74, lr: 1.01e-02, grad_scale: 8.0 2023-04-01 13:34:00,555 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.237e+02 6.329e+02 8.151e+02 1.084e+03 2.341e+03, threshold=1.630e+03, percent-clipped=12.0 2023-04-01 13:34:51,146 INFO [train.py:903] (1/4) Epoch 8, batch 6600, loss[loss=0.2729, simple_loss=0.3382, pruned_loss=0.1038, over 19671.00 frames. ], tot_loss[loss=0.2516, simple_loss=0.3211, pruned_loss=0.09107, over 3807119.16 frames. ], batch size: 60, lr: 1.01e-02, grad_scale: 8.0 2023-04-01 13:35:53,829 INFO [train.py:903] (1/4) Epoch 8, batch 6650, loss[loss=0.2649, simple_loss=0.3381, pruned_loss=0.09585, over 19706.00 frames. ], tot_loss[loss=0.2522, simple_loss=0.3213, pruned_loss=0.0915, over 3800237.22 frames. ], batch size: 63, lr: 1.01e-02, grad_scale: 8.0 2023-04-01 13:35:56,425 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.3454, 3.7398, 3.8845, 3.8952, 1.5394, 3.5409, 3.2124, 3.5948], device='cuda:1'), covar=tensor([0.1300, 0.0756, 0.0650, 0.0655, 0.4441, 0.0680, 0.0687, 0.1122], device='cuda:1'), in_proj_covar=tensor([0.0611, 0.0532, 0.0726, 0.0605, 0.0659, 0.0470, 0.0455, 0.0663], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1') 2023-04-01 13:36:04,916 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.746e+02 5.865e+02 7.808e+02 1.010e+03 1.907e+03, threshold=1.562e+03, percent-clipped=2.0 2023-04-01 13:36:19,259 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.3293, 3.0170, 2.3385, 2.2960, 1.8552, 2.4617, 0.7147, 2.0537], device='cuda:1'), covar=tensor([0.0395, 0.0335, 0.0399, 0.0679, 0.0832, 0.0686, 0.0876, 0.0694], device='cuda:1'), in_proj_covar=tensor([0.0325, 0.0321, 0.0317, 0.0332, 0.0413, 0.0338, 0.0296, 0.0315], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 13:36:55,567 INFO [train.py:903] (1/4) Epoch 8, batch 6700, loss[loss=0.2155, simple_loss=0.2886, pruned_loss=0.07119, over 19592.00 frames. ], tot_loss[loss=0.2514, simple_loss=0.3202, pruned_loss=0.09129, over 3805467.38 frames. ], batch size: 52, lr: 1.01e-02, grad_scale: 8.0 2023-04-01 13:37:52,319 INFO [train.py:903] (1/4) Epoch 8, batch 6750, loss[loss=0.2513, simple_loss=0.333, pruned_loss=0.08473, over 19513.00 frames. ], tot_loss[loss=0.2516, simple_loss=0.3202, pruned_loss=0.09148, over 3794144.70 frames. ], batch size: 64, lr: 1.01e-02, grad_scale: 8.0 2023-04-01 13:38:03,630 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.992e+02 6.380e+02 7.224e+02 9.353e+02 2.017e+03, threshold=1.445e+03, percent-clipped=3.0 2023-04-01 13:38:50,507 INFO [train.py:903] (1/4) Epoch 8, batch 6800, loss[loss=0.2582, simple_loss=0.328, pruned_loss=0.09417, over 17187.00 frames. ], tot_loss[loss=0.2511, simple_loss=0.3194, pruned_loss=0.09137, over 3791421.84 frames. ], batch size: 101, lr: 1.01e-02, grad_scale: 8.0 2023-04-01 13:39:34,403 WARNING [train.py:1073] (1/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 2023-04-01 13:39:34,850 WARNING [train.py:1073] (1/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 2023-04-01 13:39:38,462 INFO [train.py:903] (1/4) Epoch 9, batch 0, loss[loss=0.2631, simple_loss=0.3346, pruned_loss=0.09575, over 19682.00 frames. ], tot_loss[loss=0.2631, simple_loss=0.3346, pruned_loss=0.09575, over 19682.00 frames. ], batch size: 58, lr: 9.56e-03, grad_scale: 8.0 2023-04-01 13:39:38,462 INFO [train.py:928] (1/4) Computing validation loss 2023-04-01 13:39:49,508 INFO [train.py:937] (1/4) Epoch 9, validation: loss=0.1866, simple_loss=0.2872, pruned_loss=0.04294, over 944034.00 frames. 2023-04-01 13:39:49,509 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18649MB 2023-04-01 13:40:03,797 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 2023-04-01 13:40:28,247 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.879e+02 5.500e+02 7.208e+02 8.930e+02 1.459e+03, threshold=1.442e+03, percent-clipped=1.0 2023-04-01 13:40:51,455 INFO [train.py:903] (1/4) Epoch 9, batch 50, loss[loss=0.2379, simple_loss=0.3037, pruned_loss=0.08599, over 19832.00 frames. ], tot_loss[loss=0.2467, simple_loss=0.3159, pruned_loss=0.08874, over 870420.51 frames. ], batch size: 52, lr: 9.55e-03, grad_scale: 8.0 2023-04-01 13:41:02,951 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=54682.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 13:41:26,449 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1_rvb from training. Duration: 27.0318125 2023-04-01 13:41:53,037 INFO [train.py:903] (1/4) Epoch 9, batch 100, loss[loss=0.2322, simple_loss=0.2894, pruned_loss=0.08752, over 18655.00 frames. ], tot_loss[loss=0.2447, simple_loss=0.314, pruned_loss=0.08766, over 1527815.80 frames. ], batch size: 41, lr: 9.55e-03, grad_scale: 8.0 2023-04-01 13:42:05,359 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 2023-04-01 13:42:31,168 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.400e+02 5.564e+02 7.190e+02 9.001e+02 2.877e+03, threshold=1.438e+03, percent-clipped=2.0 2023-04-01 13:42:53,294 INFO [train.py:903] (1/4) Epoch 9, batch 150, loss[loss=0.3229, simple_loss=0.3751, pruned_loss=0.1354, over 19282.00 frames. ], tot_loss[loss=0.2465, simple_loss=0.316, pruned_loss=0.08852, over 2050723.76 frames. ], batch size: 66, lr: 9.54e-03, grad_scale: 16.0 2023-04-01 13:43:53,897 INFO [train.py:903] (1/4) Epoch 9, batch 200, loss[loss=0.2931, simple_loss=0.3573, pruned_loss=0.1144, over 19593.00 frames. ], tot_loss[loss=0.2472, simple_loss=0.3169, pruned_loss=0.08874, over 2450984.75 frames. ], batch size: 61, lr: 9.54e-03, grad_scale: 8.0 2023-04-01 13:43:56,330 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 2023-04-01 13:44:03,609 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.8205, 1.7684, 1.6872, 2.0185, 1.9734, 1.6943, 1.7406, 1.8687], device='cuda:1'), covar=tensor([0.0711, 0.1135, 0.0955, 0.0593, 0.0772, 0.0409, 0.0812, 0.0484], device='cuda:1'), in_proj_covar=tensor([0.0247, 0.0355, 0.0290, 0.0239, 0.0297, 0.0244, 0.0269, 0.0234], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 13:44:36,452 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.193e+02 5.883e+02 7.738e+02 9.204e+02 1.688e+03, threshold=1.548e+03, percent-clipped=2.0 2023-04-01 13:44:47,200 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.9010, 1.6711, 1.8374, 1.7003, 4.3558, 1.0474, 2.4778, 4.5587], device='cuda:1'), covar=tensor([0.0313, 0.2440, 0.2430, 0.1795, 0.0624, 0.2513, 0.1237, 0.0240], device='cuda:1'), in_proj_covar=tensor([0.0332, 0.0330, 0.0339, 0.0309, 0.0339, 0.0323, 0.0313, 0.0334], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 13:44:57,168 INFO [train.py:903] (1/4) Epoch 9, batch 250, loss[loss=0.2622, simple_loss=0.3356, pruned_loss=0.09437, over 19659.00 frames. ], tot_loss[loss=0.2481, simple_loss=0.318, pruned_loss=0.08911, over 2761494.45 frames. ], batch size: 55, lr: 9.54e-03, grad_scale: 8.0 2023-04-01 13:45:06,558 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=54880.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 13:45:57,849 INFO [train.py:903] (1/4) Epoch 9, batch 300, loss[loss=0.2394, simple_loss=0.3207, pruned_loss=0.079, over 19693.00 frames. ], tot_loss[loss=0.2492, simple_loss=0.3185, pruned_loss=0.08997, over 2999747.61 frames. ], batch size: 59, lr: 9.53e-03, grad_scale: 8.0 2023-04-01 13:46:39,713 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.328e+02 5.660e+02 7.032e+02 9.457e+02 2.087e+03, threshold=1.406e+03, percent-clipped=3.0 2023-04-01 13:47:01,338 INFO [train.py:903] (1/4) Epoch 9, batch 350, loss[loss=0.2227, simple_loss=0.2923, pruned_loss=0.07652, over 19774.00 frames. ], tot_loss[loss=0.2491, simple_loss=0.3183, pruned_loss=0.08989, over 3180076.98 frames. ], batch size: 47, lr: 9.53e-03, grad_scale: 8.0 2023-04-01 13:47:07,244 WARNING [train.py:1073] (1/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1_rvb from training. Duration: 0.9681875 2023-04-01 13:48:03,063 INFO [train.py:903] (1/4) Epoch 9, batch 400, loss[loss=0.2169, simple_loss=0.3026, pruned_loss=0.06566, over 19661.00 frames. ], tot_loss[loss=0.2497, simple_loss=0.3189, pruned_loss=0.09031, over 3311386.04 frames. ], batch size: 58, lr: 9.52e-03, grad_scale: 8.0 2023-04-01 13:48:06,560 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=55026.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 13:48:38,431 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.99 vs. limit=2.0 2023-04-01 13:48:44,523 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.195e+02 5.719e+02 7.898e+02 1.018e+03 2.327e+03, threshold=1.580e+03, percent-clipped=4.0 2023-04-01 13:49:04,314 INFO [train.py:903] (1/4) Epoch 9, batch 450, loss[loss=0.261, simple_loss=0.3379, pruned_loss=0.09204, over 19616.00 frames. ], tot_loss[loss=0.2494, simple_loss=0.3187, pruned_loss=0.09005, over 3433701.51 frames. ], batch size: 57, lr: 9.52e-03, grad_scale: 8.0 2023-04-01 13:49:42,318 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 2023-04-01 13:49:43,519 WARNING [train.py:1073] (1/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 2023-04-01 13:50:07,765 INFO [train.py:903] (1/4) Epoch 9, batch 500, loss[loss=0.3412, simple_loss=0.3818, pruned_loss=0.1503, over 13660.00 frames. ], tot_loss[loss=0.2486, simple_loss=0.3179, pruned_loss=0.08961, over 3526656.37 frames. ], batch size: 136, lr: 9.51e-03, grad_scale: 8.0 2023-04-01 13:50:30,920 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=55141.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 13:50:47,944 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.663e+02 6.183e+02 7.261e+02 8.870e+02 1.589e+03, threshold=1.452e+03, percent-clipped=1.0 2023-04-01 13:51:01,602 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=55166.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 13:51:10,561 INFO [train.py:903] (1/4) Epoch 9, batch 550, loss[loss=0.2193, simple_loss=0.2929, pruned_loss=0.07284, over 19627.00 frames. ], tot_loss[loss=0.2502, simple_loss=0.3192, pruned_loss=0.09059, over 3591449.68 frames. ], batch size: 50, lr: 9.51e-03, grad_scale: 8.0 2023-04-01 13:52:14,553 INFO [train.py:903] (1/4) Epoch 9, batch 600, loss[loss=0.2758, simple_loss=0.3325, pruned_loss=0.1095, over 19468.00 frames. ], tot_loss[loss=0.2499, simple_loss=0.3193, pruned_loss=0.09026, over 3656351.99 frames. ], batch size: 49, lr: 9.51e-03, grad_scale: 8.0 2023-04-01 13:52:15,818 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=55224.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 13:52:21,746 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=55229.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 13:52:46,694 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.2817, 2.9294, 2.0967, 2.7000, 0.7712, 2.8444, 2.7634, 2.8557], device='cuda:1'), covar=tensor([0.1114, 0.1462, 0.2190, 0.0996, 0.4059, 0.1122, 0.1049, 0.1319], device='cuda:1'), in_proj_covar=tensor([0.0408, 0.0347, 0.0415, 0.0306, 0.0373, 0.0342, 0.0331, 0.0366], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-01 13:52:55,191 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.376e+02 5.838e+02 6.783e+02 8.586e+02 3.812e+03, threshold=1.357e+03, percent-clipped=5.0 2023-04-01 13:52:58,747 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 2023-04-01 13:53:16,399 INFO [train.py:903] (1/4) Epoch 9, batch 650, loss[loss=0.2491, simple_loss=0.3243, pruned_loss=0.08698, over 19662.00 frames. ], tot_loss[loss=0.2501, simple_loss=0.3195, pruned_loss=0.09029, over 3686702.67 frames. ], batch size: 60, lr: 9.50e-03, grad_scale: 4.0 2023-04-01 13:54:19,295 INFO [train.py:903] (1/4) Epoch 9, batch 700, loss[loss=0.2573, simple_loss=0.3182, pruned_loss=0.09823, over 19860.00 frames. ], tot_loss[loss=0.2501, simple_loss=0.3196, pruned_loss=0.0903, over 3722887.95 frames. ], batch size: 52, lr: 9.50e-03, grad_scale: 4.0 2023-04-01 13:54:41,865 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=55339.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 13:55:02,469 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.468e+02 5.553e+02 6.808e+02 9.255e+02 2.546e+03, threshold=1.362e+03, percent-clipped=4.0 2023-04-01 13:55:22,970 INFO [train.py:903] (1/4) Epoch 9, batch 750, loss[loss=0.2703, simple_loss=0.3407, pruned_loss=0.1, over 19580.00 frames. ], tot_loss[loss=0.2495, simple_loss=0.3194, pruned_loss=0.08985, over 3747936.43 frames. ], batch size: 61, lr: 9.49e-03, grad_scale: 4.0 2023-04-01 13:55:53,593 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=55397.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 13:56:08,129 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([6.0634, 5.4054, 2.8572, 4.7746, 1.3237, 5.2744, 5.2957, 5.5067], device='cuda:1'), covar=tensor([0.0413, 0.1049, 0.1872, 0.0623, 0.3831, 0.0559, 0.0584, 0.0888], device='cuda:1'), in_proj_covar=tensor([0.0407, 0.0347, 0.0414, 0.0306, 0.0372, 0.0339, 0.0331, 0.0366], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-01 13:56:25,354 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0 2023-04-01 13:56:25,919 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=55422.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 13:56:26,679 INFO [train.py:903] (1/4) Epoch 9, batch 800, loss[loss=0.2491, simple_loss=0.3314, pruned_loss=0.08343, over 19102.00 frames. ], tot_loss[loss=0.2498, simple_loss=0.3199, pruned_loss=0.08987, over 3765683.07 frames. ], batch size: 69, lr: 9.49e-03, grad_scale: 8.0 2023-04-01 13:56:41,967 WARNING [train.py:1073] (1/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 2023-04-01 13:57:07,881 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.102e+02 5.631e+02 7.179e+02 9.586e+02 1.610e+03, threshold=1.436e+03, percent-clipped=4.0 2023-04-01 13:57:28,767 INFO [train.py:903] (1/4) Epoch 9, batch 850, loss[loss=0.301, simple_loss=0.3525, pruned_loss=0.1248, over 13447.00 frames. ], tot_loss[loss=0.2487, simple_loss=0.3187, pruned_loss=0.08931, over 3776847.63 frames. ], batch size: 136, lr: 9.48e-03, grad_scale: 8.0 2023-04-01 13:57:46,893 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.81 vs. limit=2.0 2023-04-01 13:58:09,665 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 2023-04-01 13:58:14,385 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=55510.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 13:58:22,720 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9_rvb from training. Duration: 25.061125 2023-04-01 13:58:29,590 INFO [train.py:903] (1/4) Epoch 9, batch 900, loss[loss=0.2535, simple_loss=0.3146, pruned_loss=0.09619, over 19763.00 frames. ], tot_loss[loss=0.2495, simple_loss=0.3194, pruned_loss=0.08982, over 3791600.08 frames. ], batch size: 48, lr: 9.48e-03, grad_scale: 8.0 2023-04-01 13:59:12,469 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.962e+02 5.974e+02 7.143e+02 8.825e+02 2.413e+03, threshold=1.429e+03, percent-clipped=6.0 2023-04-01 13:59:32,016 INFO [train.py:903] (1/4) Epoch 9, batch 950, loss[loss=0.2171, simple_loss=0.2915, pruned_loss=0.07137, over 19762.00 frames. ], tot_loss[loss=0.2482, simple_loss=0.3183, pruned_loss=0.089, over 3810935.42 frames. ], batch size: 47, lr: 9.48e-03, grad_scale: 8.0 2023-04-01 13:59:32,175 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=55573.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 13:59:37,628 WARNING [train.py:1073] (1/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9_rvb from training. Duration: 26.32775 2023-04-01 13:59:49,679 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2019, 1.0109, 1.4649, 1.2370, 2.4586, 3.4958, 3.2664, 3.7992], device='cuda:1'), covar=tensor([0.1840, 0.4719, 0.4104, 0.2312, 0.0659, 0.0254, 0.0314, 0.0204], device='cuda:1'), in_proj_covar=tensor([0.0255, 0.0286, 0.0317, 0.0250, 0.0209, 0.0142, 0.0206, 0.0177], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 14:00:01,045 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=55595.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 14:00:32,397 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=55620.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 14:00:35,414 INFO [train.py:903] (1/4) Epoch 9, batch 1000, loss[loss=0.2406, simple_loss=0.309, pruned_loss=0.0861, over 19683.00 frames. ], tot_loss[loss=0.2483, simple_loss=0.3183, pruned_loss=0.08917, over 3808744.53 frames. ], batch size: 58, lr: 9.47e-03, grad_scale: 8.0 2023-04-01 14:00:37,993 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=55625.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 14:00:44,435 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 2023-04-01 14:01:18,077 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.465e+02 5.596e+02 6.834e+02 8.838e+02 1.578e+03, threshold=1.367e+03, percent-clipped=2.0 2023-04-01 14:01:29,787 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 2023-04-01 14:01:39,287 INFO [train.py:903] (1/4) Epoch 9, batch 1050, loss[loss=0.2876, simple_loss=0.3545, pruned_loss=0.1104, over 19687.00 frames. ], tot_loss[loss=0.2481, simple_loss=0.318, pruned_loss=0.0891, over 3802456.40 frames. ], batch size: 58, lr: 9.47e-03, grad_scale: 8.0 2023-04-01 14:01:57,382 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=55688.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 14:02:02,634 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.82 vs. limit=2.0 2023-04-01 14:02:10,954 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 2023-04-01 14:02:35,955 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.2378, 3.7473, 3.8765, 3.8376, 1.4255, 3.5847, 3.1358, 3.5553], device='cuda:1'), covar=tensor([0.1270, 0.0725, 0.0540, 0.0581, 0.4768, 0.0628, 0.0685, 0.1028], device='cuda:1'), in_proj_covar=tensor([0.0617, 0.0544, 0.0733, 0.0616, 0.0674, 0.0481, 0.0465, 0.0674], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1') 2023-04-01 14:02:42,619 INFO [train.py:903] (1/4) Epoch 9, batch 1100, loss[loss=0.2652, simple_loss=0.3322, pruned_loss=0.09911, over 19378.00 frames. ], tot_loss[loss=0.248, simple_loss=0.3182, pruned_loss=0.08888, over 3809247.32 frames. ], batch size: 70, lr: 9.46e-03, grad_scale: 8.0 2023-04-01 14:02:48,791 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.8876, 1.9979, 2.0776, 2.8193, 1.8288, 2.5029, 2.4530, 2.0216], device='cuda:1'), covar=tensor([0.3363, 0.2704, 0.1286, 0.1591, 0.3204, 0.1288, 0.2877, 0.2254], device='cuda:1'), in_proj_covar=tensor([0.0747, 0.0755, 0.0626, 0.0875, 0.0750, 0.0658, 0.0772, 0.0677], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 14:03:25,907 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.690e+02 6.007e+02 7.412e+02 9.315e+02 2.515e+03, threshold=1.482e+03, percent-clipped=6.0 2023-04-01 14:03:36,850 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=55766.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 14:03:45,595 INFO [train.py:903] (1/4) Epoch 9, batch 1150, loss[loss=0.2165, simple_loss=0.2973, pruned_loss=0.06783, over 19678.00 frames. ], tot_loss[loss=0.248, simple_loss=0.3179, pruned_loss=0.08907, over 3804788.08 frames. ], batch size: 53, lr: 9.46e-03, grad_scale: 8.0 2023-04-01 14:04:50,175 INFO [train.py:903] (1/4) Epoch 9, batch 1200, loss[loss=0.2044, simple_loss=0.2779, pruned_loss=0.0654, over 19720.00 frames. ], tot_loss[loss=0.248, simple_loss=0.3175, pruned_loss=0.08926, over 3815981.30 frames. ], batch size: 46, lr: 9.45e-03, grad_scale: 8.0 2023-04-01 14:05:18,983 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 2023-04-01 14:05:31,642 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.284e+02 5.926e+02 7.645e+02 1.010e+03 3.329e+03, threshold=1.529e+03, percent-clipped=6.0 2023-04-01 14:05:44,738 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=55866.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 14:05:44,893 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.1598, 2.0614, 1.7382, 1.5942, 1.5347, 1.7196, 0.2704, 0.8879], device='cuda:1'), covar=tensor([0.0327, 0.0359, 0.0255, 0.0433, 0.0745, 0.0423, 0.0746, 0.0662], device='cuda:1'), in_proj_covar=tensor([0.0314, 0.0315, 0.0311, 0.0327, 0.0403, 0.0330, 0.0290, 0.0311], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 14:05:53,923 INFO [train.py:903] (1/4) Epoch 9, batch 1250, loss[loss=0.2451, simple_loss=0.3204, pruned_loss=0.08489, over 19300.00 frames. ], tot_loss[loss=0.2485, simple_loss=0.318, pruned_loss=0.08949, over 3808428.83 frames. ], batch size: 66, lr: 9.45e-03, grad_scale: 8.0 2023-04-01 14:06:03,173 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=55881.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 14:06:09,655 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.47 vs. limit=2.0 2023-04-01 14:06:35,374 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=55906.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 14:06:36,769 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.23 vs. limit=5.0 2023-04-01 14:06:55,976 INFO [train.py:903] (1/4) Epoch 9, batch 1300, loss[loss=0.2316, simple_loss=0.2974, pruned_loss=0.08288, over 19579.00 frames. ], tot_loss[loss=0.2473, simple_loss=0.317, pruned_loss=0.08875, over 3819500.36 frames. ], batch size: 52, lr: 9.45e-03, grad_scale: 8.0 2023-04-01 14:07:23,894 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=55944.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 14:07:39,624 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.793e+02 5.793e+02 6.909e+02 8.321e+02 2.022e+03, threshold=1.382e+03, percent-clipped=2.0 2023-04-01 14:07:54,328 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=55969.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 14:07:58,594 INFO [train.py:903] (1/4) Epoch 9, batch 1350, loss[loss=0.2224, simple_loss=0.3015, pruned_loss=0.07169, over 19587.00 frames. ], tot_loss[loss=0.2477, simple_loss=0.3174, pruned_loss=0.08899, over 3814743.53 frames. ], batch size: 52, lr: 9.44e-03, grad_scale: 8.0 2023-04-01 14:08:43,438 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 2023-04-01 14:09:02,608 INFO [train.py:903] (1/4) Epoch 9, batch 1400, loss[loss=0.234, simple_loss=0.3146, pruned_loss=0.07666, over 19662.00 frames. ], tot_loss[loss=0.2482, simple_loss=0.3179, pruned_loss=0.08926, over 3821924.76 frames. ], batch size: 55, lr: 9.44e-03, grad_scale: 8.0 2023-04-01 14:09:19,733 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6415, 1.6897, 1.4195, 1.2710, 1.0781, 1.2741, 0.2705, 0.5319], device='cuda:1'), covar=tensor([0.0543, 0.0526, 0.0335, 0.0474, 0.1118, 0.0554, 0.0774, 0.0906], device='cuda:1'), in_proj_covar=tensor([0.0316, 0.0316, 0.0314, 0.0329, 0.0405, 0.0333, 0.0291, 0.0313], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 14:09:37,353 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.0671, 1.2283, 1.3260, 1.3677, 2.6404, 0.9404, 1.7817, 2.7826], device='cuda:1'), covar=tensor([0.0449, 0.2458, 0.2629, 0.1574, 0.0731, 0.2391, 0.1274, 0.0417], device='cuda:1'), in_proj_covar=tensor([0.0327, 0.0326, 0.0337, 0.0307, 0.0333, 0.0325, 0.0314, 0.0334], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 14:09:47,082 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.547e+02 5.536e+02 7.314e+02 8.995e+02 2.483e+03, threshold=1.463e+03, percent-clipped=9.0 2023-04-01 14:10:07,183 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_rvb from training. Duration: 25.85 2023-04-01 14:10:08,163 INFO [train.py:903] (1/4) Epoch 9, batch 1450, loss[loss=0.3054, simple_loss=0.3682, pruned_loss=0.1213, over 19412.00 frames. ], tot_loss[loss=0.2495, simple_loss=0.319, pruned_loss=0.09, over 3811682.56 frames. ], batch size: 70, lr: 9.43e-03, grad_scale: 8.0 2023-04-01 14:10:27,134 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.8555, 4.2721, 4.4980, 4.4807, 1.4667, 4.0773, 3.6938, 4.1469], device='cuda:1'), covar=tensor([0.1211, 0.0573, 0.0497, 0.0518, 0.5133, 0.0470, 0.0613, 0.0989], device='cuda:1'), in_proj_covar=tensor([0.0616, 0.0543, 0.0740, 0.0621, 0.0682, 0.0486, 0.0467, 0.0677], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1') 2023-04-01 14:10:55,232 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=56110.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 14:11:11,145 INFO [train.py:903] (1/4) Epoch 9, batch 1500, loss[loss=0.2491, simple_loss=0.3192, pruned_loss=0.08948, over 19742.00 frames. ], tot_loss[loss=0.2504, simple_loss=0.3198, pruned_loss=0.0905, over 3824139.52 frames. ], batch size: 51, lr: 9.43e-03, grad_scale: 8.0 2023-04-01 14:11:12,165 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.91 vs. limit=2.0 2023-04-01 14:11:39,374 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3880, 1.4371, 2.2816, 1.6510, 2.8049, 2.3096, 2.9444, 1.5315], device='cuda:1'), covar=tensor([0.2233, 0.3693, 0.1965, 0.1672, 0.1531, 0.1870, 0.1827, 0.3273], device='cuda:1'), in_proj_covar=tensor([0.0474, 0.0545, 0.0555, 0.0422, 0.0573, 0.0473, 0.0637, 0.0472], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 14:11:40,562 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0130, 0.9374, 1.0543, 0.9878, 1.6858, 0.7507, 1.4974, 1.7957], device='cuda:1'), covar=tensor([0.0525, 0.2063, 0.2055, 0.1260, 0.0790, 0.1671, 0.0810, 0.0436], device='cuda:1'), in_proj_covar=tensor([0.0329, 0.0328, 0.0338, 0.0308, 0.0333, 0.0326, 0.0316, 0.0334], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 14:11:52,401 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.776e+02 6.008e+02 7.164e+02 9.066e+02 2.093e+03, threshold=1.433e+03, percent-clipped=3.0 2023-04-01 14:11:54,314 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 2023-04-01 14:12:01,103 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=56163.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 14:12:12,300 INFO [train.py:903] (1/4) Epoch 9, batch 1550, loss[loss=0.2132, simple_loss=0.2905, pruned_loss=0.068, over 19565.00 frames. ], tot_loss[loss=0.2516, simple_loss=0.3207, pruned_loss=0.0912, over 3821741.02 frames. ], batch size: 61, lr: 9.43e-03, grad_scale: 8.0 2023-04-01 14:12:19,309 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([6.0520, 5.4177, 2.7174, 4.7060, 1.1010, 5.2737, 5.3064, 5.5164], device='cuda:1'), covar=tensor([0.0432, 0.0952, 0.2081, 0.0690, 0.4195, 0.0598, 0.0657, 0.0841], device='cuda:1'), in_proj_covar=tensor([0.0408, 0.0344, 0.0413, 0.0306, 0.0374, 0.0341, 0.0334, 0.0367], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-01 14:12:59,841 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=56210.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 14:13:15,315 INFO [train.py:903] (1/4) Epoch 9, batch 1600, loss[loss=0.2257, simple_loss=0.2968, pruned_loss=0.07734, over 19858.00 frames. ], tot_loss[loss=0.2497, simple_loss=0.3194, pruned_loss=0.09001, over 3831524.92 frames. ], batch size: 52, lr: 9.42e-03, grad_scale: 8.0 2023-04-01 14:13:18,970 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=56225.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 14:13:41,610 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9_rvb from training. Duration: 30.1555625 2023-04-01 14:13:59,883 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.177e+02 5.781e+02 6.769e+02 8.617e+02 2.222e+03, threshold=1.354e+03, percent-clipped=2.0 2023-04-01 14:14:16,067 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2273, 1.2898, 1.1655, 1.0111, 1.0754, 1.1095, 0.1042, 0.4201], device='cuda:1'), covar=tensor([0.0413, 0.0424, 0.0242, 0.0322, 0.0812, 0.0350, 0.0678, 0.0647], device='cuda:1'), in_proj_covar=tensor([0.0316, 0.0315, 0.0315, 0.0331, 0.0406, 0.0334, 0.0293, 0.0316], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 14:14:19,018 INFO [train.py:903] (1/4) Epoch 9, batch 1650, loss[loss=0.2082, simple_loss=0.2681, pruned_loss=0.07411, over 19777.00 frames. ], tot_loss[loss=0.2499, simple_loss=0.3198, pruned_loss=0.09003, over 3834392.93 frames. ], batch size: 46, lr: 9.42e-03, grad_scale: 4.0 2023-04-01 14:14:21,984 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9041, 2.0335, 2.0732, 2.8437, 1.9456, 2.5659, 2.4783, 1.9021], device='cuda:1'), covar=tensor([0.3113, 0.2505, 0.1280, 0.1346, 0.2877, 0.1213, 0.2781, 0.2315], device='cuda:1'), in_proj_covar=tensor([0.0743, 0.0751, 0.0624, 0.0864, 0.0747, 0.0661, 0.0764, 0.0673], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 14:15:22,544 INFO [train.py:903] (1/4) Epoch 9, batch 1700, loss[loss=0.2272, simple_loss=0.31, pruned_loss=0.0722, over 19400.00 frames. ], tot_loss[loss=0.2511, simple_loss=0.3207, pruned_loss=0.09075, over 3825558.25 frames. ], batch size: 70, lr: 9.41e-03, grad_scale: 4.0 2023-04-01 14:15:25,162 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=56325.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 14:16:02,358 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 2023-04-01 14:16:05,919 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.642e+02 5.619e+02 6.768e+02 8.904e+02 2.101e+03, threshold=1.354e+03, percent-clipped=6.0 2023-04-01 14:16:24,636 INFO [train.py:903] (1/4) Epoch 9, batch 1750, loss[loss=0.2499, simple_loss=0.3232, pruned_loss=0.08826, over 19635.00 frames. ], tot_loss[loss=0.2509, simple_loss=0.3202, pruned_loss=0.09076, over 3808759.96 frames. ], batch size: 57, lr: 9.41e-03, grad_scale: 4.0 2023-04-01 14:17:26,692 INFO [train.py:903] (1/4) Epoch 9, batch 1800, loss[loss=0.2219, simple_loss=0.3007, pruned_loss=0.0715, over 19832.00 frames. ], tot_loss[loss=0.2514, simple_loss=0.3204, pruned_loss=0.09119, over 3793962.79 frames. ], batch size: 52, lr: 9.40e-03, grad_scale: 4.0 2023-04-01 14:18:09,953 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.374e+02 5.839e+02 7.002e+02 8.564e+02 1.629e+03, threshold=1.400e+03, percent-clipped=1.0 2023-04-01 14:18:25,587 WARNING [train.py:1073] (1/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9_rvb from training. Duration: 25.0944375 2023-04-01 14:18:29,983 INFO [train.py:903] (1/4) Epoch 9, batch 1850, loss[loss=0.3003, simple_loss=0.3556, pruned_loss=0.1225, over 19581.00 frames. ], tot_loss[loss=0.2514, simple_loss=0.3206, pruned_loss=0.09108, over 3799672.76 frames. ], batch size: 61, lr: 9.40e-03, grad_scale: 4.0 2023-04-01 14:18:36,522 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.64 vs. limit=2.0 2023-04-01 14:18:40,562 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=56481.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 14:19:02,323 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9_rvb from training. Duration: 27.8166875 2023-04-01 14:19:06,824 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.9197, 4.3248, 4.6113, 4.6313, 1.7889, 4.3422, 3.8436, 4.2495], device='cuda:1'), covar=tensor([0.1090, 0.0652, 0.0515, 0.0446, 0.4182, 0.0477, 0.0528, 0.0972], device='cuda:1'), in_proj_covar=tensor([0.0620, 0.0543, 0.0742, 0.0619, 0.0686, 0.0485, 0.0463, 0.0684], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1') 2023-04-01 14:19:11,098 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=56506.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 14:19:12,018 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=56507.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 14:19:33,000 INFO [train.py:903] (1/4) Epoch 9, batch 1900, loss[loss=0.2395, simple_loss=0.2954, pruned_loss=0.09185, over 19192.00 frames. ], tot_loss[loss=0.2496, simple_loss=0.3192, pruned_loss=0.09002, over 3807966.36 frames. ], batch size: 42, lr: 9.40e-03, grad_scale: 4.0 2023-04-01 14:19:48,395 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 2023-04-01 14:19:54,892 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9_rvb from training. Duration: 27.02225 2023-04-01 14:19:55,176 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=56541.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 14:20:16,637 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.412e+02 5.665e+02 6.760e+02 8.403e+02 1.758e+03, threshold=1.352e+03, percent-clipped=2.0 2023-04-01 14:20:18,933 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 2023-04-01 14:20:35,969 INFO [train.py:903] (1/4) Epoch 9, batch 1950, loss[loss=0.2453, simple_loss=0.3159, pruned_loss=0.08742, over 19611.00 frames. ], tot_loss[loss=0.2473, simple_loss=0.3169, pruned_loss=0.08889, over 3811043.01 frames. ], batch size: 50, lr: 9.39e-03, grad_scale: 4.0 2023-04-01 14:20:46,780 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=56581.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 14:21:19,019 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=56606.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 14:21:38,528 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=56622.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 14:21:39,356 INFO [train.py:903] (1/4) Epoch 9, batch 2000, loss[loss=0.2778, simple_loss=0.3444, pruned_loss=0.1055, over 19680.00 frames. ], tot_loss[loss=0.2472, simple_loss=0.317, pruned_loss=0.08868, over 3808680.36 frames. ], batch size: 58, lr: 9.39e-03, grad_scale: 8.0 2023-04-01 14:22:22,876 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.621e+02 5.506e+02 7.081e+02 9.090e+02 3.144e+03, threshold=1.416e+03, percent-clipped=7.0 2023-04-01 14:22:33,281 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9985, 2.0167, 2.1233, 2.8690, 2.0540, 2.7182, 2.5203, 2.0253], device='cuda:1'), covar=tensor([0.3029, 0.2559, 0.1237, 0.1494, 0.2844, 0.1173, 0.2533, 0.2177], device='cuda:1'), in_proj_covar=tensor([0.0745, 0.0749, 0.0620, 0.0859, 0.0742, 0.0659, 0.0763, 0.0668], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 14:22:36,101 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_rvb from training. Duration: 26.8349375 2023-04-01 14:22:42,513 INFO [train.py:903] (1/4) Epoch 9, batch 2050, loss[loss=0.2715, simple_loss=0.3409, pruned_loss=0.1011, over 19533.00 frames. ], tot_loss[loss=0.2488, simple_loss=0.3181, pruned_loss=0.08978, over 3805636.01 frames. ], batch size: 56, lr: 9.38e-03, grad_scale: 8.0 2023-04-01 14:22:56,328 WARNING [train.py:1073] (1/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 2023-04-01 14:22:57,511 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 2023-04-01 14:23:17,614 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.70 vs. limit=2.0 2023-04-01 14:23:19,221 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 2023-04-01 14:23:44,269 INFO [train.py:903] (1/4) Epoch 9, batch 2100, loss[loss=0.2504, simple_loss=0.3271, pruned_loss=0.08689, over 18128.00 frames. ], tot_loss[loss=0.2485, simple_loss=0.3181, pruned_loss=0.08944, over 3816872.74 frames. ], batch size: 83, lr: 9.38e-03, grad_scale: 8.0 2023-04-01 14:23:49,639 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 2023-04-01 14:23:56,216 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.4440, 2.5016, 2.5999, 3.5134, 2.4434, 3.5415, 3.1731, 2.5345], device='cuda:1'), covar=tensor([0.3245, 0.2708, 0.1169, 0.1567, 0.3185, 0.1150, 0.2561, 0.2060], device='cuda:1'), in_proj_covar=tensor([0.0754, 0.0758, 0.0628, 0.0871, 0.0749, 0.0668, 0.0775, 0.0678], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 14:24:12,117 WARNING [train.py:1073] (1/4) Exclude cut with ID 453-131332-0000-131866_sp0.9_rvb from training. Duration: 25.3333125 2023-04-01 14:24:29,329 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.187e+02 5.573e+02 6.906e+02 8.990e+02 1.566e+03, threshold=1.381e+03, percent-clipped=3.0 2023-04-01 14:24:35,327 WARNING [train.py:1073] (1/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9_rvb from training. Duration: 26.6166875 2023-04-01 14:24:48,312 INFO [train.py:903] (1/4) Epoch 9, batch 2150, loss[loss=0.2512, simple_loss=0.329, pruned_loss=0.0867, over 19783.00 frames. ], tot_loss[loss=0.2477, simple_loss=0.3179, pruned_loss=0.08875, over 3816791.22 frames. ], batch size: 63, lr: 9.38e-03, grad_scale: 8.0 2023-04-01 14:25:11,566 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.1323, 1.8276, 1.6985, 2.1516, 1.9280, 1.8263, 1.6737, 2.0697], device='cuda:1'), covar=tensor([0.0840, 0.1448, 0.1302, 0.0864, 0.1190, 0.0499, 0.1142, 0.0592], device='cuda:1'), in_proj_covar=tensor([0.0247, 0.0348, 0.0285, 0.0236, 0.0296, 0.0241, 0.0269, 0.0231], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 14:25:46,482 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6220, 1.8729, 2.2309, 2.0519, 3.1413, 3.6696, 3.6185, 3.8666], device='cuda:1'), covar=tensor([0.1371, 0.2640, 0.2419, 0.1620, 0.0708, 0.0144, 0.0207, 0.0200], device='cuda:1'), in_proj_covar=tensor([0.0251, 0.0289, 0.0319, 0.0248, 0.0208, 0.0142, 0.0204, 0.0179], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 14:25:48,792 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=56821.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 14:25:50,901 INFO [train.py:903] (1/4) Epoch 9, batch 2200, loss[loss=0.2073, simple_loss=0.2724, pruned_loss=0.07115, over 19717.00 frames. ], tot_loss[loss=0.2468, simple_loss=0.3166, pruned_loss=0.08844, over 3818912.42 frames. ], batch size: 46, lr: 9.37e-03, grad_scale: 8.0 2023-04-01 14:26:36,146 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.021e+02 5.854e+02 7.300e+02 9.690e+02 2.298e+03, threshold=1.460e+03, percent-clipped=5.0 2023-04-01 14:26:57,251 INFO [train.py:903] (1/4) Epoch 9, batch 2250, loss[loss=0.2376, simple_loss=0.3127, pruned_loss=0.08128, over 19609.00 frames. ], tot_loss[loss=0.2458, simple_loss=0.3163, pruned_loss=0.08767, over 3823365.63 frames. ], batch size: 57, lr: 9.37e-03, grad_scale: 8.0 2023-04-01 14:27:03,680 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=56878.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 14:27:08,852 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.86 vs. limit=5.0 2023-04-01 14:27:11,763 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=56885.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 14:27:34,021 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=56903.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 14:28:00,647 INFO [train.py:903] (1/4) Epoch 9, batch 2300, loss[loss=0.2329, simple_loss=0.2992, pruned_loss=0.0833, over 19728.00 frames. ], tot_loss[loss=0.2442, simple_loss=0.3148, pruned_loss=0.08685, over 3824396.43 frames. ], batch size: 51, lr: 9.36e-03, grad_scale: 8.0 2023-04-01 14:28:05,834 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=56926.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 14:28:13,792 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_rvb from training. Duration: 26.205 2023-04-01 14:28:24,092 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2561, 1.2967, 1.6295, 1.4237, 2.6393, 2.2319, 2.9151, 1.1670], device='cuda:1'), covar=tensor([0.2149, 0.3567, 0.2266, 0.1742, 0.1345, 0.1746, 0.1443, 0.3315], device='cuda:1'), in_proj_covar=tensor([0.0466, 0.0537, 0.0553, 0.0417, 0.0572, 0.0470, 0.0633, 0.0469], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 14:28:46,771 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.182e+02 5.809e+02 7.207e+02 9.233e+02 1.673e+03, threshold=1.441e+03, percent-clipped=4.0 2023-04-01 14:29:05,091 INFO [train.py:903] (1/4) Epoch 9, batch 2350, loss[loss=0.253, simple_loss=0.3127, pruned_loss=0.09666, over 19472.00 frames. ], tot_loss[loss=0.2445, simple_loss=0.315, pruned_loss=0.08695, over 3815507.86 frames. ], batch size: 49, lr: 9.36e-03, grad_scale: 8.0 2023-04-01 14:29:40,044 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57000.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 14:29:46,431 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_rvb from training. Duration: 25.775 2023-04-01 14:29:53,646 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57011.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 14:29:55,788 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57013.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 14:30:01,354 WARNING [train.py:1073] (1/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9_rvb from training. Duration: 25.45 2023-04-01 14:30:07,031 INFO [train.py:903] (1/4) Epoch 9, batch 2400, loss[loss=0.2122, simple_loss=0.2912, pruned_loss=0.06659, over 19677.00 frames. ], tot_loss[loss=0.2449, simple_loss=0.3152, pruned_loss=0.08731, over 3815510.99 frames. ], batch size: 53, lr: 9.36e-03, grad_scale: 8.0 2023-04-01 14:30:31,225 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.6737, 1.8265, 2.2419, 2.7555, 2.4100, 2.0825, 2.2021, 2.8567], device='cuda:1'), covar=tensor([0.0718, 0.1628, 0.1109, 0.0853, 0.1075, 0.0482, 0.0951, 0.0449], device='cuda:1'), in_proj_covar=tensor([0.0248, 0.0349, 0.0290, 0.0239, 0.0297, 0.0243, 0.0270, 0.0230], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 14:30:51,626 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.434e+02 5.803e+02 7.402e+02 8.988e+02 1.700e+03, threshold=1.480e+03, percent-clipped=2.0 2023-04-01 14:31:11,535 INFO [train.py:903] (1/4) Epoch 9, batch 2450, loss[loss=0.219, simple_loss=0.2807, pruned_loss=0.07868, over 19759.00 frames. ], tot_loss[loss=0.2448, simple_loss=0.3157, pruned_loss=0.08694, over 3824335.68 frames. ], batch size: 46, lr: 9.35e-03, grad_scale: 8.0 2023-04-01 14:31:34,284 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.2763, 2.9659, 2.2353, 2.7541, 0.7853, 2.7918, 2.7505, 2.8453], device='cuda:1'), covar=tensor([0.1180, 0.1534, 0.2010, 0.0959, 0.3932, 0.1135, 0.1009, 0.1252], device='cuda:1'), in_proj_covar=tensor([0.0407, 0.0348, 0.0413, 0.0304, 0.0371, 0.0339, 0.0332, 0.0366], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-01 14:32:15,742 INFO [train.py:903] (1/4) Epoch 9, batch 2500, loss[loss=0.2511, simple_loss=0.3277, pruned_loss=0.08729, over 19618.00 frames. ], tot_loss[loss=0.2445, simple_loss=0.3152, pruned_loss=0.0869, over 3833661.93 frames. ], batch size: 57, lr: 9.35e-03, grad_scale: 8.0 2023-04-01 14:33:00,911 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.384e+02 5.325e+02 6.954e+02 9.918e+02 1.981e+03, threshold=1.391e+03, percent-clipped=3.0 2023-04-01 14:33:09,586 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=57165.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 14:33:19,772 INFO [train.py:903] (1/4) Epoch 9, batch 2550, loss[loss=0.2116, simple_loss=0.2916, pruned_loss=0.06582, over 18257.00 frames. ], tot_loss[loss=0.2434, simple_loss=0.3141, pruned_loss=0.08629, over 3836634.93 frames. ], batch size: 83, lr: 9.34e-03, grad_scale: 8.0 2023-04-01 14:33:59,381 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57204.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 14:34:13,726 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57215.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 14:34:15,575 WARNING [train.py:1073] (1/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 2023-04-01 14:34:22,473 INFO [train.py:903] (1/4) Epoch 9, batch 2600, loss[loss=0.2708, simple_loss=0.3437, pruned_loss=0.09895, over 19508.00 frames. ], tot_loss[loss=0.2438, simple_loss=0.3147, pruned_loss=0.0864, over 3825350.67 frames. ], batch size: 56, lr: 9.34e-03, grad_scale: 8.0 2023-04-01 14:35:05,270 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57256.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 14:35:07,158 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.285e+02 5.816e+02 6.932e+02 7.774e+02 1.592e+03, threshold=1.386e+03, percent-clipped=3.0 2023-04-01 14:35:12,160 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57262.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 14:35:22,411 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=57270.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 14:35:25,689 INFO [train.py:903] (1/4) Epoch 9, batch 2650, loss[loss=0.3496, simple_loss=0.3941, pruned_loss=0.1526, over 13157.00 frames. ], tot_loss[loss=0.2424, simple_loss=0.3134, pruned_loss=0.08568, over 3827486.85 frames. ], batch size: 136, lr: 9.34e-03, grad_scale: 8.0 2023-04-01 14:35:36,477 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57280.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 14:35:37,721 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=57281.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 14:35:45,907 WARNING [train.py:1073] (1/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9_rvb from training. Duration: 27.25 2023-04-01 14:36:31,183 INFO [train.py:903] (1/4) Epoch 9, batch 2700, loss[loss=0.2766, simple_loss=0.3482, pruned_loss=0.1025, over 19614.00 frames. ], tot_loss[loss=0.2442, simple_loss=0.315, pruned_loss=0.08671, over 3825028.68 frames. ], batch size: 57, lr: 9.33e-03, grad_scale: 8.0 2023-04-01 14:37:12,076 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=57355.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 14:37:13,795 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.67 vs. limit=2.0 2023-04-01 14:37:14,482 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=57357.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 14:37:15,519 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.587e+02 5.592e+02 7.209e+02 8.892e+02 3.755e+03, threshold=1.442e+03, percent-clipped=7.0 2023-04-01 14:37:28,473 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.9685, 1.8058, 2.0606, 1.9772, 4.5222, 1.1484, 2.5546, 4.7731], device='cuda:1'), covar=tensor([0.0265, 0.2160, 0.2189, 0.1457, 0.0535, 0.2332, 0.1136, 0.0176], device='cuda:1'), in_proj_covar=tensor([0.0333, 0.0324, 0.0337, 0.0302, 0.0333, 0.0321, 0.0308, 0.0330], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 14:37:30,697 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.3137, 2.2470, 1.8934, 1.7262, 1.6567, 1.8067, 0.4209, 1.0694], device='cuda:1'), covar=tensor([0.0344, 0.0314, 0.0268, 0.0407, 0.0667, 0.0442, 0.0693, 0.0618], device='cuda:1'), in_proj_covar=tensor([0.0316, 0.0316, 0.0314, 0.0328, 0.0412, 0.0332, 0.0294, 0.0314], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 14:37:33,719 INFO [train.py:903] (1/4) Epoch 9, batch 2750, loss[loss=0.2223, simple_loss=0.3054, pruned_loss=0.06956, over 19793.00 frames. ], tot_loss[loss=0.2462, simple_loss=0.3165, pruned_loss=0.0879, over 3804658.59 frames. ], batch size: 56, lr: 9.33e-03, grad_scale: 8.0 2023-04-01 14:37:49,387 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57385.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 14:38:30,590 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57416.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 14:38:38,517 INFO [train.py:903] (1/4) Epoch 9, batch 2800, loss[loss=0.2407, simple_loss=0.315, pruned_loss=0.08323, over 19311.00 frames. ], tot_loss[loss=0.2465, simple_loss=0.3173, pruned_loss=0.08783, over 3800043.89 frames. ], batch size: 66, lr: 9.32e-03, grad_scale: 8.0 2023-04-01 14:39:23,168 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.000e+02 5.525e+02 6.599e+02 8.446e+02 1.316e+03, threshold=1.320e+03, percent-clipped=0.0 2023-04-01 14:39:38,423 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57470.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 14:39:41,722 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57472.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 14:39:42,554 INFO [train.py:903] (1/4) Epoch 9, batch 2850, loss[loss=0.2208, simple_loss=0.3034, pruned_loss=0.06911, over 17138.00 frames. ], tot_loss[loss=0.2472, simple_loss=0.3176, pruned_loss=0.0884, over 3792935.85 frames. ], batch size: 101, lr: 9.32e-03, grad_scale: 8.0 2023-04-01 14:39:51,241 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57480.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 14:40:00,530 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([0.8632, 1.3261, 0.9551, 0.9916, 1.1556, 0.9538, 0.9054, 1.2231], device='cuda:1'), covar=tensor([0.0498, 0.0673, 0.0985, 0.0553, 0.0472, 0.1105, 0.0519, 0.0407], device='cuda:1'), in_proj_covar=tensor([0.0287, 0.0289, 0.0318, 0.0239, 0.0231, 0.0315, 0.0287, 0.0258], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 14:40:33,729 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57513.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 14:40:41,907 WARNING [train.py:1073] (1/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 2023-04-01 14:40:46,553 INFO [train.py:903] (1/4) Epoch 9, batch 2900, loss[loss=0.2387, simple_loss=0.3215, pruned_loss=0.0779, over 19622.00 frames. ], tot_loss[loss=0.2474, simple_loss=0.3174, pruned_loss=0.08866, over 3783727.40 frames. ], batch size: 57, lr: 9.32e-03, grad_scale: 8.0 2023-04-01 14:41:03,523 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57536.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 14:41:16,949 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=57548.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 14:41:30,793 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.787e+02 5.829e+02 7.475e+02 8.920e+02 2.516e+03, threshold=1.495e+03, percent-clipped=6.0 2023-04-01 14:41:32,226 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=57559.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 14:41:34,839 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=57561.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 14:41:49,398 INFO [train.py:903] (1/4) Epoch 9, batch 2950, loss[loss=0.2205, simple_loss=0.3098, pruned_loss=0.06564, over 19587.00 frames. ], tot_loss[loss=0.2457, simple_loss=0.3164, pruned_loss=0.0875, over 3788021.44 frames. ], batch size: 57, lr: 9.31e-03, grad_scale: 8.0 2023-04-01 14:41:53,377 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 2023-04-01 14:42:32,119 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=57606.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 14:42:53,540 INFO [train.py:903] (1/4) Epoch 9, batch 3000, loss[loss=0.2548, simple_loss=0.3238, pruned_loss=0.09292, over 19791.00 frames. ], tot_loss[loss=0.2467, simple_loss=0.3169, pruned_loss=0.08825, over 3812962.18 frames. ], batch size: 56, lr: 9.31e-03, grad_scale: 8.0 2023-04-01 14:42:53,541 INFO [train.py:928] (1/4) Computing validation loss 2023-04-01 14:43:06,203 INFO [train.py:937] (1/4) Epoch 9, validation: loss=0.1831, simple_loss=0.2838, pruned_loss=0.04122, over 944034.00 frames. 2023-04-01 14:43:06,204 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18649MB 2023-04-01 14:43:08,508 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_rvb from training. Duration: 29.735 2023-04-01 14:43:28,402 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57641.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 14:43:50,442 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.712e+02 6.061e+02 7.914e+02 9.800e+02 2.087e+03, threshold=1.583e+03, percent-clipped=4.0 2023-04-01 14:43:56,568 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57663.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 14:43:57,062 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=4.99 vs. limit=5.0 2023-04-01 14:44:00,186 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=57666.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 14:44:00,214 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.7578, 1.7800, 1.5445, 1.4192, 1.3561, 1.3941, 0.2454, 0.5582], device='cuda:1'), covar=tensor([0.0339, 0.0342, 0.0228, 0.0338, 0.0744, 0.0387, 0.0662, 0.0665], device='cuda:1'), in_proj_covar=tensor([0.0318, 0.0322, 0.0318, 0.0335, 0.0412, 0.0338, 0.0296, 0.0319], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 14:44:08,768 INFO [train.py:903] (1/4) Epoch 9, batch 3050, loss[loss=0.2904, simple_loss=0.3515, pruned_loss=0.1147, over 19327.00 frames. ], tot_loss[loss=0.2476, simple_loss=0.3176, pruned_loss=0.08879, over 3810884.29 frames. ], batch size: 66, lr: 9.30e-03, grad_scale: 8.0 2023-04-01 14:44:10,220 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57674.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 14:44:33,016 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.4677, 3.1106, 2.1669, 2.2941, 2.3191, 2.5061, 0.9608, 2.1461], device='cuda:1'), covar=tensor([0.0449, 0.0397, 0.0495, 0.0655, 0.0708, 0.0778, 0.0852, 0.0781], device='cuda:1'), in_proj_covar=tensor([0.0317, 0.0322, 0.0318, 0.0335, 0.0411, 0.0338, 0.0296, 0.0320], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 14:45:09,137 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57721.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 14:45:11,163 INFO [train.py:903] (1/4) Epoch 9, batch 3100, loss[loss=0.267, simple_loss=0.3307, pruned_loss=0.1017, over 18736.00 frames. ], tot_loss[loss=0.2465, simple_loss=0.3168, pruned_loss=0.08808, over 3826283.87 frames. ], batch size: 74, lr: 9.30e-03, grad_scale: 8.0 2023-04-01 14:45:16,014 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57726.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 14:45:19,259 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57728.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 14:45:47,435 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57751.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 14:45:47,537 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=57751.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 14:45:49,820 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=57753.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 14:45:54,416 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.2872, 2.3738, 2.4758, 3.1676, 2.1872, 3.1394, 2.9522, 2.3546], device='cuda:1'), covar=tensor([0.3078, 0.2681, 0.1146, 0.1803, 0.3221, 0.1225, 0.2604, 0.2121], device='cuda:1'), in_proj_covar=tensor([0.0749, 0.0759, 0.0625, 0.0874, 0.0747, 0.0659, 0.0772, 0.0677], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 14:45:54,923 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.860e+02 5.906e+02 7.408e+02 1.029e+03 2.368e+03, threshold=1.482e+03, percent-clipped=3.0 2023-04-01 14:45:57,422 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=57760.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 14:46:06,785 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.2376, 2.8849, 2.2934, 2.1068, 1.9030, 2.3010, 0.7609, 2.0038], device='cuda:1'), covar=tensor([0.0387, 0.0378, 0.0399, 0.0666, 0.0812, 0.0657, 0.0889, 0.0700], device='cuda:1'), in_proj_covar=tensor([0.0315, 0.0320, 0.0316, 0.0334, 0.0409, 0.0336, 0.0294, 0.0316], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 14:46:14,163 INFO [train.py:903] (1/4) Epoch 9, batch 3150, loss[loss=0.2029, simple_loss=0.2784, pruned_loss=0.06363, over 19312.00 frames. ], tot_loss[loss=0.2464, simple_loss=0.3171, pruned_loss=0.08782, over 3829012.93 frames. ], batch size: 44, lr: 9.30e-03, grad_scale: 8.0 2023-04-01 14:46:42,107 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 2023-04-01 14:47:14,703 INFO [train.py:903] (1/4) Epoch 9, batch 3200, loss[loss=0.2356, simple_loss=0.3122, pruned_loss=0.07946, over 19752.00 frames. ], tot_loss[loss=0.2485, simple_loss=0.3189, pruned_loss=0.08907, over 3819173.30 frames. ], batch size: 54, lr: 9.29e-03, grad_scale: 8.0 2023-04-01 14:47:16,009 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=57824.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 14:47:55,926 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=57857.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 14:47:56,966 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.598e+02 5.791e+02 7.100e+02 9.261e+02 4.038e+03, threshold=1.420e+03, percent-clipped=7.0 2023-04-01 14:48:14,924 INFO [train.py:903] (1/4) Epoch 9, batch 3250, loss[loss=0.2193, simple_loss=0.2933, pruned_loss=0.07265, over 19707.00 frames. ], tot_loss[loss=0.2483, simple_loss=0.3185, pruned_loss=0.08904, over 3810477.15 frames. ], batch size: 51, lr: 9.29e-03, grad_scale: 8.0 2023-04-01 14:48:18,527 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57875.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 14:49:12,236 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57919.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 14:49:16,424 INFO [train.py:903] (1/4) Epoch 9, batch 3300, loss[loss=0.2361, simple_loss=0.3105, pruned_loss=0.08087, over 18006.00 frames. ], tot_loss[loss=0.2485, simple_loss=0.3188, pruned_loss=0.08906, over 3816177.45 frames. ], batch size: 83, lr: 9.28e-03, grad_scale: 8.0 2023-04-01 14:49:23,985 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 2023-04-01 14:49:26,386 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57930.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 14:49:36,426 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57939.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 14:49:43,145 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=57944.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 14:49:56,565 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=57955.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 14:49:59,469 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.480e+02 5.470e+02 6.783e+02 8.234e+02 1.579e+03, threshold=1.357e+03, percent-clipped=1.0 2023-04-01 14:50:17,158 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57972.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 14:50:17,941 INFO [train.py:903] (1/4) Epoch 9, batch 3350, loss[loss=0.2471, simple_loss=0.3159, pruned_loss=0.08918, over 19788.00 frames. ], tot_loss[loss=0.2478, simple_loss=0.3182, pruned_loss=0.08871, over 3820904.22 frames. ], batch size: 56, lr: 9.28e-03, grad_scale: 8.0 2023-04-01 14:50:22,883 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57977.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 14:50:33,023 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.1746, 2.1608, 2.2071, 3.1142, 2.1944, 3.1104, 2.6173, 2.0845], device='cuda:1'), covar=tensor([0.3307, 0.2952, 0.1343, 0.1825, 0.3364, 0.1226, 0.2970, 0.2300], device='cuda:1'), in_proj_covar=tensor([0.0748, 0.0762, 0.0628, 0.0871, 0.0749, 0.0660, 0.0771, 0.0679], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 14:50:45,335 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.6757, 4.8029, 5.4701, 5.4156, 2.0598, 5.0541, 4.3520, 5.0309], device='cuda:1'), covar=tensor([0.1151, 0.0837, 0.0473, 0.0453, 0.4721, 0.0460, 0.0596, 0.0947], device='cuda:1'), in_proj_covar=tensor([0.0624, 0.0550, 0.0747, 0.0619, 0.0686, 0.0490, 0.0468, 0.0683], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1') 2023-04-01 14:50:54,297 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=58002.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 14:51:20,070 INFO [train.py:903] (1/4) Epoch 9, batch 3400, loss[loss=0.2045, simple_loss=0.2748, pruned_loss=0.06707, over 18190.00 frames. ], tot_loss[loss=0.2471, simple_loss=0.3175, pruned_loss=0.08836, over 3831624.14 frames. ], batch size: 40, lr: 9.28e-03, grad_scale: 8.0 2023-04-01 14:51:56,100 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.7060, 1.6121, 1.3786, 1.7079, 1.6332, 1.0624, 1.0471, 1.5081], device='cuda:1'), covar=tensor([0.0946, 0.1492, 0.1537, 0.0937, 0.1354, 0.1402, 0.1607, 0.0963], device='cuda:1'), in_proj_covar=tensor([0.0251, 0.0353, 0.0291, 0.0240, 0.0300, 0.0241, 0.0272, 0.0236], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 14:52:02,957 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.171e+02 5.309e+02 6.592e+02 8.930e+02 1.711e+03, threshold=1.318e+03, percent-clipped=4.0 2023-04-01 14:52:20,816 INFO [train.py:903] (1/4) Epoch 9, batch 3450, loss[loss=0.2711, simple_loss=0.3303, pruned_loss=0.106, over 19528.00 frames. ], tot_loss[loss=0.2467, simple_loss=0.3169, pruned_loss=0.08824, over 3816063.16 frames. ], batch size: 54, lr: 9.27e-03, grad_scale: 8.0 2023-04-01 14:52:25,119 WARNING [train.py:1073] (1/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9_rvb from training. Duration: 25.2444375 2023-04-01 14:52:48,527 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=58095.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 14:53:10,922 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.92 vs. limit=5.0 2023-04-01 14:53:12,077 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 2023-04-01 14:53:22,376 INFO [train.py:903] (1/4) Epoch 9, batch 3500, loss[loss=0.2288, simple_loss=0.289, pruned_loss=0.08435, over 19751.00 frames. ], tot_loss[loss=0.246, simple_loss=0.316, pruned_loss=0.08799, over 3817314.59 frames. ], batch size: 46, lr: 9.27e-03, grad_scale: 8.0 2023-04-01 14:53:33,963 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=58131.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 14:54:03,544 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=58156.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 14:54:05,472 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.510e+02 6.015e+02 7.407e+02 9.414e+02 2.837e+03, threshold=1.481e+03, percent-clipped=3.0 2023-04-01 14:54:06,978 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=58159.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 14:54:10,374 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=58162.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 14:54:22,219 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([5.0324, 5.0606, 5.7777, 5.7703, 1.9796, 5.4417, 4.7345, 5.3860], device='cuda:1'), covar=tensor([0.1178, 0.0687, 0.0478, 0.0452, 0.4647, 0.0415, 0.0454, 0.0964], device='cuda:1'), in_proj_covar=tensor([0.0627, 0.0554, 0.0752, 0.0622, 0.0692, 0.0495, 0.0469, 0.0691], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1') 2023-04-01 14:54:24,280 INFO [train.py:903] (1/4) Epoch 9, batch 3550, loss[loss=0.2532, simple_loss=0.327, pruned_loss=0.08968, over 18156.00 frames. ], tot_loss[loss=0.2462, simple_loss=0.3163, pruned_loss=0.0881, over 3825334.27 frames. ], batch size: 83, lr: 9.26e-03, grad_scale: 8.0 2023-04-01 14:54:50,587 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=58195.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 14:55:09,179 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=58210.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 14:55:20,433 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=58220.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 14:55:24,267 INFO [train.py:903] (1/4) Epoch 9, batch 3600, loss[loss=0.2905, simple_loss=0.3545, pruned_loss=0.1133, over 19491.00 frames. ], tot_loss[loss=0.2467, simple_loss=0.3166, pruned_loss=0.08837, over 3835850.98 frames. ], batch size: 64, lr: 9.26e-03, grad_scale: 8.0 2023-04-01 14:55:30,320 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=58228.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 14:55:54,914 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0910, 2.0539, 1.7679, 1.5776, 1.5283, 1.6162, 0.2736, 0.8130], device='cuda:1'), covar=tensor([0.0305, 0.0339, 0.0233, 0.0350, 0.0712, 0.0425, 0.0671, 0.0638], device='cuda:1'), in_proj_covar=tensor([0.0317, 0.0321, 0.0317, 0.0334, 0.0407, 0.0333, 0.0297, 0.0315], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 14:56:01,039 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=58253.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 14:56:06,161 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.641e+02 5.487e+02 6.500e+02 7.997e+02 2.924e+03, threshold=1.300e+03, percent-clipped=2.0 2023-04-01 14:56:17,831 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9552, 2.0118, 2.1709, 2.8307, 1.8524, 2.8445, 2.5978, 1.9688], device='cuda:1'), covar=tensor([0.3149, 0.2656, 0.1232, 0.1512, 0.3130, 0.1104, 0.2749, 0.2318], device='cuda:1'), in_proj_covar=tensor([0.0744, 0.0757, 0.0627, 0.0866, 0.0750, 0.0663, 0.0767, 0.0679], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 14:56:23,762 INFO [train.py:903] (1/4) Epoch 9, batch 3650, loss[loss=0.2385, simple_loss=0.3128, pruned_loss=0.08211, over 18841.00 frames. ], tot_loss[loss=0.2472, simple_loss=0.3169, pruned_loss=0.08873, over 3827779.88 frames. ], batch size: 75, lr: 9.26e-03, grad_scale: 16.0 2023-04-01 14:57:24,454 INFO [train.py:903] (1/4) Epoch 9, batch 3700, loss[loss=0.1842, simple_loss=0.2556, pruned_loss=0.0564, over 19745.00 frames. ], tot_loss[loss=0.2456, simple_loss=0.3156, pruned_loss=0.08783, over 3825958.60 frames. ], batch size: 47, lr: 9.25e-03, grad_scale: 8.0 2023-04-01 14:58:07,749 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.190e+02 6.060e+02 7.736e+02 9.843e+02 2.060e+03, threshold=1.547e+03, percent-clipped=9.0 2023-04-01 14:58:23,939 INFO [train.py:903] (1/4) Epoch 9, batch 3750, loss[loss=0.2305, simple_loss=0.2975, pruned_loss=0.08172, over 19688.00 frames. ], tot_loss[loss=0.2461, simple_loss=0.316, pruned_loss=0.08804, over 3833303.67 frames. ], batch size: 53, lr: 9.25e-03, grad_scale: 8.0 2023-04-01 14:58:28,588 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.7485, 1.5340, 1.4041, 2.1645, 1.6613, 2.1016, 2.2339, 1.8950], device='cuda:1'), covar=tensor([0.0781, 0.0915, 0.1040, 0.0855, 0.0902, 0.0681, 0.0777, 0.0643], device='cuda:1'), in_proj_covar=tensor([0.0217, 0.0233, 0.0228, 0.0258, 0.0243, 0.0218, 0.0204, 0.0206], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-01 14:58:37,437 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.13 vs. limit=5.0 2023-04-01 14:59:09,814 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.2884, 3.8463, 2.6728, 3.4497, 1.2119, 3.7084, 3.6703, 3.7114], device='cuda:1'), covar=tensor([0.0655, 0.1021, 0.1747, 0.0797, 0.3347, 0.0772, 0.0762, 0.0997], device='cuda:1'), in_proj_covar=tensor([0.0410, 0.0349, 0.0418, 0.0314, 0.0368, 0.0344, 0.0337, 0.0371], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-01 14:59:24,637 INFO [train.py:903] (1/4) Epoch 9, batch 3800, loss[loss=0.2427, simple_loss=0.318, pruned_loss=0.08375, over 18693.00 frames. ], tot_loss[loss=0.2455, simple_loss=0.3156, pruned_loss=0.08771, over 3820440.50 frames. ], batch size: 74, lr: 9.24e-03, grad_scale: 8.0 2023-04-01 14:59:38,865 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.6200, 1.9775, 2.0878, 2.8883, 2.4807, 2.2778, 2.3576, 2.7698], device='cuda:1'), covar=tensor([0.0813, 0.1816, 0.1368, 0.0910, 0.1225, 0.0459, 0.0936, 0.0516], device='cuda:1'), in_proj_covar=tensor([0.0249, 0.0353, 0.0291, 0.0240, 0.0299, 0.0241, 0.0272, 0.0233], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 14:59:54,353 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9_rvb from training. Duration: 29.1166875 2023-04-01 15:00:08,584 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.327e+02 5.230e+02 6.601e+02 8.580e+02 1.875e+03, threshold=1.320e+03, percent-clipped=3.0 2023-04-01 15:00:17,603 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=58466.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 15:00:24,959 INFO [train.py:903] (1/4) Epoch 9, batch 3850, loss[loss=0.2736, simple_loss=0.3349, pruned_loss=0.1062, over 13888.00 frames. ], tot_loss[loss=0.2446, simple_loss=0.3153, pruned_loss=0.08695, over 3815658.37 frames. ], batch size: 136, lr: 9.24e-03, grad_scale: 8.0 2023-04-01 15:00:27,780 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3759, 1.4326, 1.7359, 1.5378, 2.4363, 2.0950, 2.5240, 1.2248], device='cuda:1'), covar=tensor([0.1915, 0.3302, 0.1954, 0.1581, 0.1243, 0.1731, 0.1224, 0.3076], device='cuda:1'), in_proj_covar=tensor([0.0471, 0.0550, 0.0564, 0.0423, 0.0581, 0.0476, 0.0638, 0.0473], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 15:00:46,674 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=58491.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 15:01:01,848 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=58503.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 15:01:06,098 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=58506.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 15:01:25,307 INFO [train.py:903] (1/4) Epoch 9, batch 3900, loss[loss=0.2925, simple_loss=0.3588, pruned_loss=0.1131, over 19751.00 frames. ], tot_loss[loss=0.2459, simple_loss=0.3165, pruned_loss=0.0877, over 3824750.37 frames. ], batch size: 63, lr: 9.24e-03, grad_scale: 8.0 2023-04-01 15:02:06,362 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5173, 1.3708, 1.9839, 1.6187, 3.0155, 4.6487, 4.6347, 5.0205], device='cuda:1'), covar=tensor([0.1381, 0.3271, 0.2727, 0.1865, 0.0445, 0.0143, 0.0132, 0.0098], device='cuda:1'), in_proj_covar=tensor([0.0247, 0.0285, 0.0316, 0.0245, 0.0208, 0.0144, 0.0203, 0.0182], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 15:02:08,186 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.315e+02 6.109e+02 7.308e+02 8.933e+02 2.200e+03, threshold=1.462e+03, percent-clipped=5.0 2023-04-01 15:02:26,012 INFO [train.py:903] (1/4) Epoch 9, batch 3950, loss[loss=0.1904, simple_loss=0.2641, pruned_loss=0.05836, over 19752.00 frames. ], tot_loss[loss=0.2451, simple_loss=0.3155, pruned_loss=0.08733, over 3826284.14 frames. ], batch size: 45, lr: 9.23e-03, grad_scale: 8.0 2023-04-01 15:02:28,973 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 2023-04-01 15:03:21,996 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=58618.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 15:03:25,501 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=58621.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 15:03:28,125 INFO [train.py:903] (1/4) Epoch 9, batch 4000, loss[loss=0.2522, simple_loss=0.3068, pruned_loss=0.09886, over 19589.00 frames. ], tot_loss[loss=0.2441, simple_loss=0.3149, pruned_loss=0.08663, over 3816520.91 frames. ], batch size: 52, lr: 9.23e-03, grad_scale: 8.0 2023-04-01 15:04:10,730 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.392e+02 5.642e+02 6.845e+02 8.578e+02 2.087e+03, threshold=1.369e+03, percent-clipped=3.0 2023-04-01 15:04:10,788 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1_rvb from training. Duration: 0.7545625 2023-04-01 15:04:27,006 INFO [train.py:903] (1/4) Epoch 9, batch 4050, loss[loss=0.2141, simple_loss=0.2975, pruned_loss=0.06533, over 19774.00 frames. ], tot_loss[loss=0.2435, simple_loss=0.3142, pruned_loss=0.0864, over 3833584.43 frames. ], batch size: 54, lr: 9.22e-03, grad_scale: 8.0 2023-04-01 15:04:46,279 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.9178, 1.6763, 2.0476, 1.8712, 4.4613, 1.1482, 2.3736, 4.6427], device='cuda:1'), covar=tensor([0.0307, 0.2327, 0.2298, 0.1573, 0.0619, 0.2437, 0.1277, 0.0196], device='cuda:1'), in_proj_covar=tensor([0.0334, 0.0325, 0.0336, 0.0306, 0.0338, 0.0322, 0.0309, 0.0333], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 15:04:59,268 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=58699.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 15:05:28,158 INFO [train.py:903] (1/4) Epoch 9, batch 4100, loss[loss=0.2904, simple_loss=0.3617, pruned_loss=0.1096, over 19785.00 frames. ], tot_loss[loss=0.244, simple_loss=0.3145, pruned_loss=0.0867, over 3840203.51 frames. ], batch size: 56, lr: 9.22e-03, grad_scale: 8.0 2023-04-01 15:06:03,196 WARNING [train.py:1073] (1/4) Exclude cut with ID 497-129325-0061-9566_sp1.1_rvb from training. Duration: 0.97725 2023-04-01 15:06:11,192 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.329e+02 5.572e+02 6.953e+02 8.904e+02 1.888e+03, threshold=1.391e+03, percent-clipped=3.0 2023-04-01 15:06:28,251 INFO [train.py:903] (1/4) Epoch 9, batch 4150, loss[loss=0.3085, simple_loss=0.3657, pruned_loss=0.1256, over 13432.00 frames. ], tot_loss[loss=0.2453, simple_loss=0.3156, pruned_loss=0.08755, over 3827098.79 frames. ], batch size: 136, lr: 9.22e-03, grad_scale: 8.0 2023-04-01 15:06:54,425 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=58793.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 15:07:30,066 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.7351, 1.5238, 1.4072, 2.1005, 1.7230, 1.9468, 2.0696, 1.8288], device='cuda:1'), covar=tensor([0.0773, 0.0978, 0.1112, 0.0797, 0.0863, 0.0749, 0.0822, 0.0643], device='cuda:1'), in_proj_covar=tensor([0.0215, 0.0231, 0.0227, 0.0255, 0.0242, 0.0214, 0.0201, 0.0205], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1') 2023-04-01 15:07:30,859 INFO [train.py:903] (1/4) Epoch 9, batch 4200, loss[loss=0.2012, simple_loss=0.2768, pruned_loss=0.06284, over 19411.00 frames. ], tot_loss[loss=0.2442, simple_loss=0.3149, pruned_loss=0.08674, over 3838499.23 frames. ], batch size: 47, lr: 9.21e-03, grad_scale: 8.0 2023-04-01 15:07:34,200 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 2023-04-01 15:07:44,062 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 2023-04-01 15:08:14,668 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.798e+02 5.991e+02 7.229e+02 9.145e+02 1.621e+03, threshold=1.446e+03, percent-clipped=3.0 2023-04-01 15:08:32,532 INFO [train.py:903] (1/4) Epoch 9, batch 4250, loss[loss=0.2672, simple_loss=0.3318, pruned_loss=0.1013, over 19654.00 frames. ], tot_loss[loss=0.2419, simple_loss=0.3132, pruned_loss=0.08528, over 3846120.47 frames. ], batch size: 60, lr: 9.21e-03, grad_scale: 8.0 2023-04-01 15:08:34,223 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=58874.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 15:08:37,507 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=58877.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 15:08:48,676 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9_rvb from training. Duration: 29.816625 2023-04-01 15:08:59,992 WARNING [train.py:1073] (1/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 2023-04-01 15:09:03,829 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=58899.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 15:09:08,232 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=58902.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 15:09:32,789 INFO [train.py:903] (1/4) Epoch 9, batch 4300, loss[loss=0.1982, simple_loss=0.2691, pruned_loss=0.06359, over 19778.00 frames. ], tot_loss[loss=0.2428, simple_loss=0.3142, pruned_loss=0.08567, over 3849486.86 frames. ], batch size: 48, lr: 9.20e-03, grad_scale: 8.0 2023-04-01 15:09:34,166 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=58924.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 15:10:05,294 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4465, 2.3526, 1.6348, 1.5950, 2.2500, 1.2691, 1.1271, 1.7861], device='cuda:1'), covar=tensor([0.0883, 0.0572, 0.0953, 0.0596, 0.0357, 0.1049, 0.0830, 0.0529], device='cuda:1'), in_proj_covar=tensor([0.0286, 0.0296, 0.0322, 0.0242, 0.0229, 0.0318, 0.0288, 0.0256], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 15:10:17,294 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.360e+02 5.584e+02 7.321e+02 9.114e+02 2.155e+03, threshold=1.464e+03, percent-clipped=5.0 2023-04-01 15:10:24,490 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1704, 1.1103, 1.1325, 1.4006, 1.1308, 1.2685, 1.3259, 1.2394], device='cuda:1'), covar=tensor([0.0791, 0.0904, 0.0982, 0.0603, 0.0723, 0.0714, 0.0694, 0.0654], device='cuda:1'), in_proj_covar=tensor([0.0213, 0.0229, 0.0224, 0.0254, 0.0240, 0.0212, 0.0201, 0.0203], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1') 2023-04-01 15:10:26,452 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 2023-04-01 15:10:33,821 INFO [train.py:903] (1/4) Epoch 9, batch 4350, loss[loss=0.2233, simple_loss=0.3006, pruned_loss=0.07295, over 19736.00 frames. ], tot_loss[loss=0.2426, simple_loss=0.314, pruned_loss=0.0856, over 3840508.19 frames. ], batch size: 51, lr: 9.20e-03, grad_scale: 8.0 2023-04-01 15:11:34,343 INFO [train.py:903] (1/4) Epoch 9, batch 4400, loss[loss=0.2459, simple_loss=0.3123, pruned_loss=0.08975, over 18239.00 frames. ], tot_loss[loss=0.2448, simple_loss=0.3154, pruned_loss=0.08711, over 3829120.59 frames. ], batch size: 83, lr: 9.20e-03, grad_scale: 8.0 2023-04-01 15:11:58,695 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_rvb from training. Duration: 25.285 2023-04-01 15:11:58,796 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=59043.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 15:12:08,641 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 2023-04-01 15:12:18,355 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.517e+02 5.626e+02 6.682e+02 1.014e+03 2.125e+03, threshold=1.336e+03, percent-clipped=6.0 2023-04-01 15:12:35,840 INFO [train.py:903] (1/4) Epoch 9, batch 4450, loss[loss=0.2337, simple_loss=0.2982, pruned_loss=0.08461, over 16129.00 frames. ], tot_loss[loss=0.2443, simple_loss=0.3149, pruned_loss=0.08684, over 3834578.54 frames. ], batch size: 35, lr: 9.19e-03, grad_scale: 8.0 2023-04-01 15:13:36,621 INFO [train.py:903] (1/4) Epoch 9, batch 4500, loss[loss=0.24, simple_loss=0.2986, pruned_loss=0.09068, over 19726.00 frames. ], tot_loss[loss=0.2441, simple_loss=0.3141, pruned_loss=0.08701, over 3825624.76 frames. ], batch size: 45, lr: 9.19e-03, grad_scale: 8.0 2023-04-01 15:13:54,291 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=59137.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 15:14:20,653 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=59158.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 15:14:21,306 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 2023-04-01 15:14:21,456 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.665e+02 5.638e+02 7.209e+02 9.049e+02 2.430e+03, threshold=1.442e+03, percent-clipped=5.0 2023-04-01 15:14:38,162 INFO [train.py:903] (1/4) Epoch 9, batch 4550, loss[loss=0.2419, simple_loss=0.3113, pruned_loss=0.08623, over 19621.00 frames. ], tot_loss[loss=0.2426, simple_loss=0.3134, pruned_loss=0.08593, over 3829626.24 frames. ], batch size: 50, lr: 9.19e-03, grad_scale: 8.0 2023-04-01 15:14:46,053 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9_rvb from training. Duration: 28.72225 2023-04-01 15:14:49,760 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=59182.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 15:15:11,272 WARNING [train.py:1073] (1/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 2023-04-01 15:15:39,255 INFO [train.py:903] (1/4) Epoch 9, batch 4600, loss[loss=0.2215, simple_loss=0.2961, pruned_loss=0.07346, over 19772.00 frames. ], tot_loss[loss=0.2439, simple_loss=0.3147, pruned_loss=0.08657, over 3834614.83 frames. ], batch size: 48, lr: 9.18e-03, grad_scale: 8.0 2023-04-01 15:16:15,344 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=59252.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 15:16:24,167 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.716e+02 5.555e+02 6.855e+02 8.371e+02 1.742e+03, threshold=1.371e+03, percent-clipped=2.0 2023-04-01 15:16:35,123 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=59268.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 15:16:40,781 INFO [train.py:903] (1/4) Epoch 9, batch 4650, loss[loss=0.2531, simple_loss=0.32, pruned_loss=0.09309, over 19607.00 frames. ], tot_loss[loss=0.2438, simple_loss=0.315, pruned_loss=0.08628, over 3827343.41 frames. ], batch size: 50, lr: 9.18e-03, grad_scale: 8.0 2023-04-01 15:16:51,378 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.12 vs. limit=2.0 2023-04-01 15:16:56,380 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_rvb from training. Duration: 0.92 2023-04-01 15:17:05,559 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.55 vs. limit=5.0 2023-04-01 15:17:09,147 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_rvb from training. Duration: 0.83 2023-04-01 15:17:41,933 INFO [train.py:903] (1/4) Epoch 9, batch 4700, loss[loss=0.2319, simple_loss=0.3043, pruned_loss=0.07974, over 19765.00 frames. ], tot_loss[loss=0.2425, simple_loss=0.3139, pruned_loss=0.08549, over 3840240.59 frames. ], batch size: 54, lr: 9.17e-03, grad_scale: 8.0 2023-04-01 15:18:03,974 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9_rvb from training. Duration: 0.92225 2023-04-01 15:18:25,859 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.928e+02 5.586e+02 6.716e+02 8.168e+02 1.499e+03, threshold=1.343e+03, percent-clipped=1.0 2023-04-01 15:18:41,883 INFO [train.py:903] (1/4) Epoch 9, batch 4750, loss[loss=0.2215, simple_loss=0.2933, pruned_loss=0.07489, over 19757.00 frames. ], tot_loss[loss=0.2428, simple_loss=0.3143, pruned_loss=0.08566, over 3838737.42 frames. ], batch size: 51, lr: 9.17e-03, grad_scale: 8.0 2023-04-01 15:18:51,377 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.3934, 2.2455, 1.8018, 1.6871, 1.6060, 1.7787, 0.4127, 1.3167], device='cuda:1'), covar=tensor([0.0373, 0.0368, 0.0343, 0.0552, 0.0814, 0.0601, 0.0840, 0.0690], device='cuda:1'), in_proj_covar=tensor([0.0325, 0.0322, 0.0323, 0.0339, 0.0413, 0.0340, 0.0300, 0.0320], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 15:18:54,742 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=59383.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 15:19:32,457 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=59414.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 15:19:43,261 INFO [train.py:903] (1/4) Epoch 9, batch 4800, loss[loss=0.2604, simple_loss=0.3214, pruned_loss=0.09964, over 19751.00 frames. ], tot_loss[loss=0.2432, simple_loss=0.3143, pruned_loss=0.08602, over 3840229.08 frames. ], batch size: 63, lr: 9.17e-03, grad_scale: 8.0 2023-04-01 15:19:51,183 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.9066, 4.3117, 4.6059, 4.5875, 1.7336, 4.2557, 3.7268, 4.2829], device='cuda:1'), covar=tensor([0.1211, 0.0660, 0.0525, 0.0508, 0.4718, 0.0483, 0.0564, 0.1019], device='cuda:1'), in_proj_covar=tensor([0.0624, 0.0557, 0.0746, 0.0626, 0.0690, 0.0504, 0.0460, 0.0692], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1') 2023-04-01 15:20:03,379 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=59439.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 15:20:13,002 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.50 vs. limit=5.0 2023-04-01 15:20:27,560 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.770e+02 5.762e+02 6.893e+02 8.818e+02 1.836e+03, threshold=1.379e+03, percent-clipped=7.0 2023-04-01 15:20:43,942 INFO [train.py:903] (1/4) Epoch 9, batch 4850, loss[loss=0.2063, simple_loss=0.2941, pruned_loss=0.05922, over 19631.00 frames. ], tot_loss[loss=0.2428, simple_loss=0.3138, pruned_loss=0.0859, over 3832380.65 frames. ], batch size: 57, lr: 9.16e-03, grad_scale: 8.0 2023-04-01 15:21:08,916 WARNING [train.py:1073] (1/4) Exclude cut with ID 774-127930-0014-48411_sp1.1_rvb from training. Duration: 0.95 2023-04-01 15:21:27,157 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=59508.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 15:21:29,003 WARNING [train.py:1073] (1/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1_rvb from training. Duration: 0.9409375 2023-04-01 15:21:35,286 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 2023-04-01 15:21:36,461 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 2023-04-01 15:21:42,201 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=59520.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 15:21:45,373 INFO [train.py:903] (1/4) Epoch 9, batch 4900, loss[loss=0.2529, simple_loss=0.3261, pruned_loss=0.08984, over 19622.00 frames. ], tot_loss[loss=0.2413, simple_loss=0.3126, pruned_loss=0.08498, over 3836741.47 frames. ], batch size: 50, lr: 9.16e-03, grad_scale: 8.0 2023-04-01 15:21:46,551 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1_rvb from training. Duration: 25.3818125 2023-04-01 15:21:49,055 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=59526.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 15:21:56,968 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=59533.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 15:22:05,289 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 2023-04-01 15:22:29,458 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.724e+02 5.274e+02 6.528e+02 8.023e+02 2.606e+03, threshold=1.306e+03, percent-clipped=3.0 2023-04-01 15:22:45,479 INFO [train.py:903] (1/4) Epoch 9, batch 4950, loss[loss=0.3137, simple_loss=0.3693, pruned_loss=0.129, over 18843.00 frames. ], tot_loss[loss=0.2432, simple_loss=0.3139, pruned_loss=0.08626, over 3813162.81 frames. ], batch size: 74, lr: 9.15e-03, grad_scale: 8.0 2023-04-01 15:22:48,367 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 2023-04-01 15:23:01,071 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 2023-04-01 15:23:24,587 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 2023-04-01 15:23:46,155 INFO [train.py:903] (1/4) Epoch 9, batch 5000, loss[loss=0.227, simple_loss=0.287, pruned_loss=0.08347, over 19764.00 frames. ], tot_loss[loss=0.2433, simple_loss=0.314, pruned_loss=0.0863, over 3804515.53 frames. ], batch size: 46, lr: 9.15e-03, grad_scale: 4.0 2023-04-01 15:23:53,577 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_rvb from training. Duration: 27.14 2023-04-01 15:24:04,778 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 2023-04-01 15:24:06,384 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=59639.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 15:24:08,534 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=59641.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 15:24:30,504 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.264e+02 5.818e+02 6.890e+02 9.185e+02 2.943e+03, threshold=1.378e+03, percent-clipped=3.0 2023-04-01 15:24:35,434 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=59664.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 15:24:46,099 INFO [train.py:903] (1/4) Epoch 9, batch 5050, loss[loss=0.2693, simple_loss=0.3428, pruned_loss=0.09792, over 19792.00 frames. ], tot_loss[loss=0.2441, simple_loss=0.3149, pruned_loss=0.08671, over 3790431.33 frames. ], batch size: 56, lr: 9.15e-03, grad_scale: 4.0 2023-04-01 15:24:47,599 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1160, 1.1964, 1.7631, 1.2221, 2.7793, 3.7347, 3.4524, 3.9050], device='cuda:1'), covar=tensor([0.1508, 0.3265, 0.2829, 0.1944, 0.0415, 0.0134, 0.0200, 0.0141], device='cuda:1'), in_proj_covar=tensor([0.0252, 0.0289, 0.0321, 0.0246, 0.0212, 0.0147, 0.0205, 0.0185], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 15:25:21,702 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9_rvb from training. Duration: 26.62775 2023-04-01 15:25:47,319 INFO [train.py:903] (1/4) Epoch 9, batch 5100, loss[loss=0.1974, simple_loss=0.2617, pruned_loss=0.06654, over 19728.00 frames. ], tot_loss[loss=0.2419, simple_loss=0.3128, pruned_loss=0.08552, over 3809317.27 frames. ], batch size: 46, lr: 9.14e-03, grad_scale: 4.0 2023-04-01 15:25:56,495 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 2023-04-01 15:25:59,769 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_rvb from training. Duration: 25.035 2023-04-01 15:26:05,154 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_rvb from training. Duration: 27.92 2023-04-01 15:26:33,076 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.640e+02 5.962e+02 7.316e+02 9.170e+02 1.645e+03, threshold=1.463e+03, percent-clipped=5.0 2023-04-01 15:26:47,515 INFO [train.py:903] (1/4) Epoch 9, batch 5150, loss[loss=0.3427, simple_loss=0.3712, pruned_loss=0.1571, over 13601.00 frames. ], tot_loss[loss=0.2422, simple_loss=0.3128, pruned_loss=0.08574, over 3794085.77 frames. ], batch size: 135, lr: 9.14e-03, grad_scale: 4.0 2023-04-01 15:26:58,278 WARNING [train.py:1073] (1/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 2023-04-01 15:27:32,211 WARNING [train.py:1073] (1/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1_rvb from training. Duration: 0.9681875 2023-04-01 15:27:49,842 INFO [train.py:903] (1/4) Epoch 9, batch 5200, loss[loss=0.2687, simple_loss=0.3468, pruned_loss=0.09525, over 19812.00 frames. ], tot_loss[loss=0.2444, simple_loss=0.3147, pruned_loss=0.08706, over 3792008.17 frames. ], batch size: 56, lr: 9.14e-03, grad_scale: 8.0 2023-04-01 15:27:59,709 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 2023-04-01 15:28:34,610 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.759e+02 5.545e+02 6.726e+02 8.723e+02 1.623e+03, threshold=1.345e+03, percent-clipped=2.0 2023-04-01 15:28:39,329 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=59864.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 15:28:41,610 WARNING [train.py:1073] (1/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 2023-04-01 15:28:41,869 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=59866.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 15:28:50,361 INFO [train.py:903] (1/4) Epoch 9, batch 5250, loss[loss=0.2319, simple_loss=0.3158, pruned_loss=0.07397, over 19623.00 frames. ], tot_loss[loss=0.2447, simple_loss=0.3149, pruned_loss=0.08722, over 3801782.13 frames. ], batch size: 57, lr: 9.13e-03, grad_scale: 8.0 2023-04-01 15:29:19,305 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=59897.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 15:29:31,328 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=59907.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 15:29:50,070 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=59922.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 15:29:50,795 INFO [train.py:903] (1/4) Epoch 9, batch 5300, loss[loss=0.265, simple_loss=0.3361, pruned_loss=0.09691, over 19648.00 frames. ], tot_loss[loss=0.2447, simple_loss=0.3151, pruned_loss=0.08717, over 3802961.70 frames. ], batch size: 60, lr: 9.13e-03, grad_scale: 8.0 2023-04-01 15:30:04,495 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 2023-04-01 15:30:36,748 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.396e+02 5.525e+02 7.204e+02 8.995e+02 2.228e+03, threshold=1.441e+03, percent-clipped=7.0 2023-04-01 15:30:50,971 INFO [train.py:903] (1/4) Epoch 9, batch 5350, loss[loss=0.2475, simple_loss=0.319, pruned_loss=0.08795, over 18144.00 frames. ], tot_loss[loss=0.2458, simple_loss=0.3161, pruned_loss=0.08778, over 3814431.50 frames. ], batch size: 83, lr: 9.12e-03, grad_scale: 8.0 2023-04-01 15:30:58,316 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=59979.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 15:31:24,016 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9_rvb from training. Duration: 26.438875 2023-04-01 15:31:52,157 INFO [train.py:903] (1/4) Epoch 9, batch 5400, loss[loss=0.2735, simple_loss=0.3419, pruned_loss=0.1025, over 19544.00 frames. ], tot_loss[loss=0.2461, simple_loss=0.3164, pruned_loss=0.0879, over 3829958.04 frames. ], batch size: 56, lr: 9.12e-03, grad_scale: 8.0 2023-04-01 15:32:34,980 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=60058.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 15:32:36,999 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.978e+02 5.587e+02 7.112e+02 9.365e+02 1.948e+03, threshold=1.422e+03, percent-clipped=3.0 2023-04-01 15:32:49,986 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=60070.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 15:32:53,203 INFO [train.py:903] (1/4) Epoch 9, batch 5450, loss[loss=0.2177, simple_loss=0.2812, pruned_loss=0.07713, over 15815.00 frames. ], tot_loss[loss=0.2445, simple_loss=0.3148, pruned_loss=0.08708, over 3828513.04 frames. ], batch size: 35, lr: 9.12e-03, grad_scale: 8.0 2023-04-01 15:33:06,621 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=60083.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 15:33:21,813 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=60096.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 15:33:55,444 INFO [train.py:903] (1/4) Epoch 9, batch 5500, loss[loss=0.2591, simple_loss=0.3331, pruned_loss=0.09252, over 19856.00 frames. ], tot_loss[loss=0.2448, simple_loss=0.315, pruned_loss=0.08735, over 3820830.33 frames. ], batch size: 52, lr: 9.11e-03, grad_scale: 8.0 2023-04-01 15:34:17,039 WARNING [train.py:1073] (1/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9_rvb from training. Duration: 27.47775 2023-04-01 15:34:40,129 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.116e+02 5.485e+02 6.858e+02 8.862e+02 1.983e+03, threshold=1.372e+03, percent-clipped=4.0 2023-04-01 15:34:56,050 INFO [train.py:903] (1/4) Epoch 9, batch 5550, loss[loss=0.2385, simple_loss=0.3068, pruned_loss=0.08513, over 19617.00 frames. ], tot_loss[loss=0.2436, simple_loss=0.314, pruned_loss=0.08658, over 3823811.41 frames. ], batch size: 50, lr: 9.11e-03, grad_scale: 8.0 2023-04-01 15:35:01,729 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 2023-04-01 15:35:42,127 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=60210.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 15:35:48,964 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=60216.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 15:35:49,877 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 2023-04-01 15:35:51,216 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9225, 1.6662, 1.7368, 2.1057, 1.8320, 1.8644, 1.7447, 1.9467], device='cuda:1'), covar=tensor([0.1059, 0.1719, 0.1367, 0.0866, 0.1274, 0.0480, 0.1137, 0.0666], device='cuda:1'), in_proj_covar=tensor([0.0251, 0.0351, 0.0290, 0.0240, 0.0298, 0.0241, 0.0275, 0.0236], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 15:35:57,509 INFO [train.py:903] (1/4) Epoch 9, batch 5600, loss[loss=0.2409, simple_loss=0.3155, pruned_loss=0.08314, over 19664.00 frames. ], tot_loss[loss=0.2438, simple_loss=0.3138, pruned_loss=0.08687, over 3818789.99 frames. ], batch size: 55, lr: 9.11e-03, grad_scale: 8.0 2023-04-01 15:36:12,978 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=60235.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 15:36:31,698 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=60251.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 15:36:36,797 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.64 vs. limit=5.0 2023-04-01 15:36:41,623 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.287e+02 5.903e+02 7.092e+02 9.595e+02 1.671e+03, threshold=1.418e+03, percent-clipped=4.0 2023-04-01 15:36:42,034 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=60260.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 15:36:58,114 INFO [train.py:903] (1/4) Epoch 9, batch 5650, loss[loss=0.2128, simple_loss=0.2962, pruned_loss=0.06467, over 19665.00 frames. ], tot_loss[loss=0.2441, simple_loss=0.3141, pruned_loss=0.08701, over 3821201.18 frames. ], batch size: 53, lr: 9.10e-03, grad_scale: 8.0 2023-04-01 15:37:25,224 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5267, 1.6192, 1.7919, 2.1128, 1.4335, 1.8633, 1.9876, 1.6718], device='cuda:1'), covar=tensor([0.3252, 0.2658, 0.1366, 0.1438, 0.2787, 0.1350, 0.3337, 0.2532], device='cuda:1'), in_proj_covar=tensor([0.0755, 0.0764, 0.0628, 0.0871, 0.0748, 0.0668, 0.0775, 0.0686], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 15:37:45,050 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9_rvb from training. Duration: 28.0944375 2023-04-01 15:38:00,339 INFO [train.py:903] (1/4) Epoch 9, batch 5700, loss[loss=0.2333, simple_loss=0.3047, pruned_loss=0.08088, over 19576.00 frames. ], tot_loss[loss=0.2429, simple_loss=0.3132, pruned_loss=0.08624, over 3829997.40 frames. ], batch size: 52, lr: 9.10e-03, grad_scale: 8.0 2023-04-01 15:38:02,936 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=60325.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 15:38:45,033 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.384e+02 5.631e+02 6.670e+02 7.834e+02 2.342e+03, threshold=1.334e+03, percent-clipped=2.0 2023-04-01 15:38:51,870 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.8624, 1.5570, 1.4786, 2.0716, 1.6322, 2.1813, 2.1164, 1.9990], device='cuda:1'), covar=tensor([0.0682, 0.0889, 0.0926, 0.0750, 0.0793, 0.0591, 0.0731, 0.0561], device='cuda:1'), in_proj_covar=tensor([0.0216, 0.0232, 0.0230, 0.0256, 0.0245, 0.0217, 0.0202, 0.0208], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-01 15:38:53,003 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=60366.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 15:38:59,668 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9_rvb from training. Duration: 33.038875 2023-04-01 15:39:00,760 INFO [train.py:903] (1/4) Epoch 9, batch 5750, loss[loss=0.2537, simple_loss=0.3168, pruned_loss=0.09528, over 19486.00 frames. ], tot_loss[loss=0.2443, simple_loss=0.3148, pruned_loss=0.08688, over 3818851.63 frames. ], batch size: 49, lr: 9.09e-03, grad_scale: 8.0 2023-04-01 15:39:07,410 WARNING [train.py:1073] (1/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 2023-04-01 15:39:12,813 WARNING [train.py:1073] (1/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 2023-04-01 15:39:36,134 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=60402.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 15:39:50,840 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=60414.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 15:40:00,911 INFO [train.py:903] (1/4) Epoch 9, batch 5800, loss[loss=0.2496, simple_loss=0.3125, pruned_loss=0.09332, over 19854.00 frames. ], tot_loss[loss=0.244, simple_loss=0.3146, pruned_loss=0.08666, over 3823804.61 frames. ], batch size: 52, lr: 9.09e-03, grad_scale: 8.0 2023-04-01 15:40:06,306 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=60427.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 15:40:22,426 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=60440.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 15:40:45,417 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.333e+02 6.303e+02 7.771e+02 9.979e+02 2.257e+03, threshold=1.554e+03, percent-clipped=12.0 2023-04-01 15:41:01,367 INFO [train.py:903] (1/4) Epoch 9, batch 5850, loss[loss=0.2748, simple_loss=0.3463, pruned_loss=0.1017, over 19326.00 frames. ], tot_loss[loss=0.247, simple_loss=0.317, pruned_loss=0.08851, over 3824911.97 frames. ], batch size: 66, lr: 9.09e-03, grad_scale: 8.0 2023-04-01 15:41:56,059 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=60517.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 15:42:02,434 WARNING [train.py:1073] (1/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 2023-04-01 15:42:03,583 INFO [train.py:903] (1/4) Epoch 9, batch 5900, loss[loss=0.1948, simple_loss=0.271, pruned_loss=0.0593, over 19403.00 frames. ], tot_loss[loss=0.246, simple_loss=0.3162, pruned_loss=0.08789, over 3834223.94 frames. ], batch size: 48, lr: 9.08e-03, grad_scale: 8.0 2023-04-01 15:42:10,681 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=60529.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 15:42:22,848 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9_rvb from training. Duration: 27.511125 2023-04-01 15:42:25,420 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=60542.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 15:42:42,871 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=60555.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 15:42:46,398 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.57 vs. limit=2.0 2023-04-01 15:42:47,839 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.985e+02 5.460e+02 7.799e+02 9.723e+02 2.713e+03, threshold=1.560e+03, percent-clipped=1.0 2023-04-01 15:42:48,020 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=60560.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 15:42:58,283 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2439, 1.9968, 1.5525, 1.2252, 1.8999, 1.1274, 1.1515, 1.6895], device='cuda:1'), covar=tensor([0.0792, 0.0668, 0.0910, 0.0647, 0.0409, 0.1090, 0.0673, 0.0405], device='cuda:1'), in_proj_covar=tensor([0.0291, 0.0299, 0.0327, 0.0246, 0.0233, 0.0324, 0.0294, 0.0259], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 15:43:02,773 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3547, 1.1302, 1.0867, 1.2949, 1.1133, 1.2708, 1.1122, 1.2694], device='cuda:1'), covar=tensor([0.1013, 0.1231, 0.1329, 0.0801, 0.1056, 0.0521, 0.1140, 0.0688], device='cuda:1'), in_proj_covar=tensor([0.0248, 0.0349, 0.0288, 0.0239, 0.0296, 0.0239, 0.0274, 0.0234], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 15:43:03,516 INFO [train.py:903] (1/4) Epoch 9, batch 5950, loss[loss=0.2667, simple_loss=0.3356, pruned_loss=0.09888, over 18098.00 frames. ], tot_loss[loss=0.2462, simple_loss=0.3164, pruned_loss=0.08803, over 3844057.77 frames. ], batch size: 83, lr: 9.08e-03, grad_scale: 8.0 2023-04-01 15:43:13,171 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=60581.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 15:43:28,346 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.2176, 3.8115, 2.6971, 3.4481, 0.9974, 3.4527, 3.5786, 3.7532], device='cuda:1'), covar=tensor([0.0746, 0.1038, 0.1900, 0.0786, 0.3902, 0.0974, 0.0803, 0.0964], device='cuda:1'), in_proj_covar=tensor([0.0406, 0.0346, 0.0415, 0.0305, 0.0368, 0.0341, 0.0332, 0.0369], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-01 15:43:45,276 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=60606.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 15:44:03,702 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=60622.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 15:44:04,423 INFO [train.py:903] (1/4) Epoch 9, batch 6000, loss[loss=0.2327, simple_loss=0.3162, pruned_loss=0.07464, over 19783.00 frames. ], tot_loss[loss=0.2443, simple_loss=0.3152, pruned_loss=0.08674, over 3847637.24 frames. ], batch size: 56, lr: 9.08e-03, grad_scale: 8.0 2023-04-01 15:44:04,423 INFO [train.py:928] (1/4) Computing validation loss 2023-04-01 15:44:16,868 INFO [train.py:937] (1/4) Epoch 9, validation: loss=0.1828, simple_loss=0.2835, pruned_loss=0.04105, over 944034.00 frames. 2023-04-01 15:44:16,869 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18649MB 2023-04-01 15:44:28,522 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.3711, 1.2284, 1.4427, 1.5228, 2.9176, 1.0307, 2.2835, 3.1787], device='cuda:1'), covar=tensor([0.0460, 0.2558, 0.2515, 0.1611, 0.0696, 0.2304, 0.1039, 0.0336], device='cuda:1'), in_proj_covar=tensor([0.0337, 0.0327, 0.0336, 0.0309, 0.0334, 0.0323, 0.0315, 0.0335], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 15:44:47,014 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=60647.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 15:45:00,686 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0206, 1.6967, 1.6065, 2.1280, 1.8857, 1.8975, 1.7184, 2.0085], device='cuda:1'), covar=tensor([0.0913, 0.1621, 0.1363, 0.0840, 0.1138, 0.0450, 0.1143, 0.0601], device='cuda:1'), in_proj_covar=tensor([0.0249, 0.0351, 0.0290, 0.0239, 0.0297, 0.0241, 0.0274, 0.0236], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 15:45:02,308 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.543e+02 5.785e+02 7.329e+02 9.590e+02 1.552e+03, threshold=1.466e+03, percent-clipped=0.0 2023-04-01 15:45:05,837 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.7874, 3.1883, 3.2793, 3.2945, 1.2092, 3.1268, 2.7861, 2.9983], device='cuda:1'), covar=tensor([0.1305, 0.0894, 0.0774, 0.0744, 0.4726, 0.0704, 0.0705, 0.1251], device='cuda:1'), in_proj_covar=tensor([0.0631, 0.0567, 0.0755, 0.0633, 0.0695, 0.0500, 0.0466, 0.0702], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1') 2023-04-01 15:45:17,223 INFO [train.py:903] (1/4) Epoch 9, batch 6050, loss[loss=0.2714, simple_loss=0.3412, pruned_loss=0.1008, over 19759.00 frames. ], tot_loss[loss=0.2438, simple_loss=0.3147, pruned_loss=0.08648, over 3850261.56 frames. ], batch size: 63, lr: 9.07e-03, grad_scale: 8.0 2023-04-01 15:45:19,798 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=60675.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 15:46:18,288 INFO [train.py:903] (1/4) Epoch 9, batch 6100, loss[loss=0.23, simple_loss=0.2963, pruned_loss=0.08183, over 19780.00 frames. ], tot_loss[loss=0.2438, simple_loss=0.3146, pruned_loss=0.0865, over 3837459.81 frames. ], batch size: 48, lr: 9.07e-03, grad_scale: 8.0 2023-04-01 15:46:34,575 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=60736.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 15:47:03,004 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.337e+02 5.339e+02 6.367e+02 8.531e+02 1.806e+03, threshold=1.273e+03, percent-clipped=4.0 2023-04-01 15:47:18,987 INFO [train.py:903] (1/4) Epoch 9, batch 6150, loss[loss=0.2493, simple_loss=0.3116, pruned_loss=0.09345, over 16559.00 frames. ], tot_loss[loss=0.245, simple_loss=0.3157, pruned_loss=0.08715, over 3841876.01 frames. ], batch size: 36, lr: 9.06e-03, grad_scale: 8.0 2023-04-01 15:47:19,408 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=60773.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 15:47:23,770 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0808, 2.0529, 1.6678, 1.5381, 1.3243, 1.5341, 0.4254, 0.8869], device='cuda:1'), covar=tensor([0.0351, 0.0364, 0.0297, 0.0400, 0.0768, 0.0501, 0.0684, 0.0615], device='cuda:1'), in_proj_covar=tensor([0.0322, 0.0317, 0.0318, 0.0333, 0.0413, 0.0337, 0.0295, 0.0316], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 15:47:33,341 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=60785.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 15:47:44,098 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9_rvb from training. Duration: 31.02225 2023-04-01 15:47:48,615 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=60798.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 15:47:48,645 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=60798.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 15:48:03,546 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=60810.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 15:48:04,568 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=60811.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 15:48:18,466 INFO [train.py:903] (1/4) Epoch 9, batch 6200, loss[loss=0.2941, simple_loss=0.3473, pruned_loss=0.1204, over 13181.00 frames. ], tot_loss[loss=0.2451, simple_loss=0.3158, pruned_loss=0.0872, over 3838253.82 frames. ], batch size: 136, lr: 9.06e-03, grad_scale: 8.0 2023-04-01 15:48:18,859 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=60823.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 15:48:34,245 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=60836.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 15:49:03,645 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.706e+02 5.965e+02 7.614e+02 9.330e+02 2.107e+03, threshold=1.523e+03, percent-clipped=6.0 2023-04-01 15:49:06,425 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0681, 1.6997, 1.6956, 1.9927, 1.8001, 1.8012, 1.6276, 2.0106], device='cuda:1'), covar=tensor([0.0798, 0.1522, 0.1315, 0.0839, 0.1120, 0.0474, 0.1125, 0.0592], device='cuda:1'), in_proj_covar=tensor([0.0249, 0.0351, 0.0288, 0.0237, 0.0295, 0.0240, 0.0271, 0.0234], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 15:49:19,605 INFO [train.py:903] (1/4) Epoch 9, batch 6250, loss[loss=0.2417, simple_loss=0.3027, pruned_loss=0.09041, over 16939.00 frames. ], tot_loss[loss=0.2447, simple_loss=0.3153, pruned_loss=0.08704, over 3836296.06 frames. ], batch size: 37, lr: 9.06e-03, grad_scale: 8.0 2023-04-01 15:49:49,667 WARNING [train.py:1073] (1/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9_rvb from training. Duration: 25.988875 2023-04-01 15:50:20,310 INFO [train.py:903] (1/4) Epoch 9, batch 6300, loss[loss=0.2128, simple_loss=0.2974, pruned_loss=0.06413, over 19655.00 frames. ], tot_loss[loss=0.2444, simple_loss=0.3155, pruned_loss=0.08665, over 3840077.08 frames. ], batch size: 58, lr: 9.05e-03, grad_scale: 8.0 2023-04-01 15:50:31,337 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=60931.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 15:51:01,572 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=60956.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 15:51:05,847 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.132e+02 5.456e+02 6.999e+02 8.896e+02 1.665e+03, threshold=1.400e+03, percent-clipped=1.0 2023-04-01 15:51:21,557 INFO [train.py:903] (1/4) Epoch 9, batch 6350, loss[loss=0.2176, simple_loss=0.2963, pruned_loss=0.06948, over 19605.00 frames. ], tot_loss[loss=0.2434, simple_loss=0.3147, pruned_loss=0.08604, over 3836840.07 frames. ], batch size: 57, lr: 9.05e-03, grad_scale: 8.0 2023-04-01 15:52:22,396 INFO [train.py:903] (1/4) Epoch 9, batch 6400, loss[loss=0.2312, simple_loss=0.3088, pruned_loss=0.07683, over 19659.00 frames. ], tot_loss[loss=0.243, simple_loss=0.314, pruned_loss=0.08596, over 3844504.76 frames. ], batch size: 53, lr: 9.05e-03, grad_scale: 8.0 2023-04-01 15:52:25,890 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.5893, 2.9614, 3.0370, 3.0413, 1.3530, 2.8187, 2.5397, 2.7978], device='cuda:1'), covar=tensor([0.1347, 0.1258, 0.0717, 0.0770, 0.4349, 0.0882, 0.0693, 0.1187], device='cuda:1'), in_proj_covar=tensor([0.0628, 0.0562, 0.0754, 0.0627, 0.0692, 0.0499, 0.0464, 0.0697], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1') 2023-04-01 15:52:39,073 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2891, 1.3948, 1.6182, 1.5048, 2.3261, 2.1268, 2.2806, 0.7992], device='cuda:1'), covar=tensor([0.2051, 0.3567, 0.2115, 0.1678, 0.1229, 0.1712, 0.1279, 0.3528], device='cuda:1'), in_proj_covar=tensor([0.0469, 0.0549, 0.0565, 0.0422, 0.0583, 0.0471, 0.0631, 0.0475], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 15:52:48,100 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.91 vs. limit=2.0 2023-04-01 15:53:05,775 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 2023-04-01 15:53:07,179 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.915e+02 6.196e+02 7.384e+02 8.672e+02 1.804e+03, threshold=1.477e+03, percent-clipped=4.0 2023-04-01 15:53:22,489 INFO [train.py:903] (1/4) Epoch 9, batch 6450, loss[loss=0.2865, simple_loss=0.3474, pruned_loss=0.1128, over 19660.00 frames. ], tot_loss[loss=0.2446, simple_loss=0.3155, pruned_loss=0.08684, over 3835662.35 frames. ], batch size: 60, lr: 9.04e-03, grad_scale: 8.0 2023-04-01 15:53:30,374 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=61079.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 15:53:31,426 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=61080.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 15:54:05,092 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9_rvb from training. Duration: 28.638875 2023-04-01 15:54:22,077 INFO [train.py:903] (1/4) Epoch 9, batch 6500, loss[loss=0.233, simple_loss=0.3102, pruned_loss=0.07795, over 19787.00 frames. ], tot_loss[loss=0.2456, simple_loss=0.3159, pruned_loss=0.08761, over 3848484.81 frames. ], batch size: 56, lr: 9.04e-03, grad_scale: 8.0 2023-04-01 15:54:27,554 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1_rvb from training. Duration: 0.836375 2023-04-01 15:55:06,066 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.712e+02 6.706e+02 8.302e+02 1.012e+03 2.679e+03, threshold=1.660e+03, percent-clipped=7.0 2023-04-01 15:55:21,994 INFO [train.py:903] (1/4) Epoch 9, batch 6550, loss[loss=0.2994, simple_loss=0.3713, pruned_loss=0.1138, over 19578.00 frames. ], tot_loss[loss=0.2451, simple_loss=0.3155, pruned_loss=0.08735, over 3844246.46 frames. ], batch size: 61, lr: 9.03e-03, grad_scale: 8.0 2023-04-01 15:55:50,217 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=61195.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 15:56:24,270 INFO [train.py:903] (1/4) Epoch 9, batch 6600, loss[loss=0.2069, simple_loss=0.2803, pruned_loss=0.06677, over 19395.00 frames. ], tot_loss[loss=0.2448, simple_loss=0.315, pruned_loss=0.08724, over 3836927.54 frames. ], batch size: 48, lr: 9.03e-03, grad_scale: 8.0 2023-04-01 15:57:09,244 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.208e+02 5.654e+02 6.784e+02 8.392e+02 1.741e+03, threshold=1.357e+03, percent-clipped=1.0 2023-04-01 15:57:24,976 INFO [train.py:903] (1/4) Epoch 9, batch 6650, loss[loss=0.2652, simple_loss=0.3395, pruned_loss=0.0955, over 19536.00 frames. ], tot_loss[loss=0.2451, simple_loss=0.3158, pruned_loss=0.08716, over 3838952.23 frames. ], batch size: 54, lr: 9.03e-03, grad_scale: 8.0 2023-04-01 15:58:25,574 INFO [train.py:903] (1/4) Epoch 9, batch 6700, loss[loss=0.2815, simple_loss=0.346, pruned_loss=0.1085, over 19508.00 frames. ], tot_loss[loss=0.2435, simple_loss=0.3144, pruned_loss=0.08633, over 3826919.42 frames. ], batch size: 64, lr: 9.02e-03, grad_scale: 8.0 2023-04-01 15:59:08,505 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.478e+02 5.683e+02 7.501e+02 9.618e+02 2.603e+03, threshold=1.500e+03, percent-clipped=7.0 2023-04-01 15:59:23,018 INFO [train.py:903] (1/4) Epoch 9, batch 6750, loss[loss=0.2567, simple_loss=0.3361, pruned_loss=0.08865, over 19604.00 frames. ], tot_loss[loss=0.2443, simple_loss=0.3149, pruned_loss=0.08685, over 3823268.53 frames. ], batch size: 57, lr: 9.02e-03, grad_scale: 8.0 2023-04-01 16:00:18,969 INFO [train.py:903] (1/4) Epoch 9, batch 6800, loss[loss=0.2525, simple_loss=0.3299, pruned_loss=0.08755, over 19607.00 frames. ], tot_loss[loss=0.2442, simple_loss=0.315, pruned_loss=0.08666, over 3820518.25 frames. ], batch size: 57, lr: 9.02e-03, grad_scale: 8.0 2023-04-01 16:00:19,095 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=61423.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:01:02,913 WARNING [train.py:1073] (1/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 2023-04-01 16:01:03,886 WARNING [train.py:1073] (1/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 2023-04-01 16:01:06,621 INFO [train.py:903] (1/4) Epoch 10, batch 0, loss[loss=0.2375, simple_loss=0.3102, pruned_loss=0.08246, over 19570.00 frames. ], tot_loss[loss=0.2375, simple_loss=0.3102, pruned_loss=0.08246, over 19570.00 frames. ], batch size: 52, lr: 8.57e-03, grad_scale: 8.0 2023-04-01 16:01:06,622 INFO [train.py:928] (1/4) Computing validation loss 2023-04-01 16:01:17,499 INFO [train.py:937] (1/4) Epoch 10, validation: loss=0.1825, simple_loss=0.2836, pruned_loss=0.04072, over 944034.00 frames. 2023-04-01 16:01:17,500 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18649MB 2023-04-01 16:01:17,929 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=61451.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:01:27,610 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.284e+02 5.689e+02 6.760e+02 8.116e+02 1.440e+03, threshold=1.352e+03, percent-clipped=0.0 2023-04-01 16:01:29,691 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 2023-04-01 16:01:48,111 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=61476.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:01:50,677 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.96 vs. limit=5.0 2023-04-01 16:02:17,430 INFO [train.py:903] (1/4) Epoch 10, batch 50, loss[loss=0.2499, simple_loss=0.3298, pruned_loss=0.08497, over 19538.00 frames. ], tot_loss[loss=0.2503, simple_loss=0.3208, pruned_loss=0.08991, over 861903.62 frames. ], batch size: 56, lr: 8.57e-03, grad_scale: 8.0 2023-04-01 16:02:50,354 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1_rvb from training. Duration: 27.0318125 2023-04-01 16:02:55,443 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([0.8290, 1.3378, 1.0559, 0.8746, 1.1628, 0.8789, 0.8948, 1.2167], device='cuda:1'), covar=tensor([0.0510, 0.0682, 0.0991, 0.0584, 0.0460, 0.1151, 0.0543, 0.0399], device='cuda:1'), in_proj_covar=tensor([0.0290, 0.0300, 0.0324, 0.0245, 0.0234, 0.0326, 0.0289, 0.0261], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 16:03:03,261 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=61538.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:03:18,484 INFO [train.py:903] (1/4) Epoch 10, batch 100, loss[loss=0.2346, simple_loss=0.3143, pruned_loss=0.07746, over 18936.00 frames. ], tot_loss[loss=0.2545, simple_loss=0.3238, pruned_loss=0.09262, over 1506611.73 frames. ], batch size: 74, lr: 8.57e-03, grad_scale: 8.0 2023-04-01 16:03:25,125 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 2023-04-01 16:03:29,318 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.995e+02 6.266e+02 7.755e+02 9.384e+02 2.029e+03, threshold=1.551e+03, percent-clipped=6.0 2023-04-01 16:04:19,613 INFO [train.py:903] (1/4) Epoch 10, batch 150, loss[loss=0.218, simple_loss=0.2945, pruned_loss=0.07072, over 19622.00 frames. ], tot_loss[loss=0.2498, simple_loss=0.32, pruned_loss=0.08977, over 2009327.33 frames. ], batch size: 50, lr: 8.56e-03, grad_scale: 16.0 2023-04-01 16:05:12,419 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 2023-04-01 16:05:20,114 INFO [train.py:903] (1/4) Epoch 10, batch 200, loss[loss=0.2497, simple_loss=0.3238, pruned_loss=0.08776, over 19546.00 frames. ], tot_loss[loss=0.2464, simple_loss=0.3172, pruned_loss=0.08784, over 2403810.55 frames. ], batch size: 56, lr: 8.56e-03, grad_scale: 8.0 2023-04-01 16:05:32,340 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.478e+02 5.325e+02 6.934e+02 9.117e+02 1.602e+03, threshold=1.387e+03, percent-clipped=3.0 2023-04-01 16:05:34,722 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9831, 1.7665, 1.5640, 2.1274, 1.9408, 1.8322, 1.6839, 1.8827], device='cuda:1'), covar=tensor([0.0885, 0.1435, 0.1357, 0.0904, 0.1083, 0.0457, 0.1051, 0.0616], device='cuda:1'), in_proj_covar=tensor([0.0250, 0.0354, 0.0289, 0.0238, 0.0297, 0.0243, 0.0274, 0.0239], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 16:05:35,096 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.51 vs. limit=2.0 2023-04-01 16:05:45,492 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([5.0146, 5.1340, 5.9012, 5.8572, 1.9880, 5.5262, 4.6968, 5.4568], device='cuda:1'), covar=tensor([0.1302, 0.0600, 0.0417, 0.0438, 0.4986, 0.0401, 0.0530, 0.1003], device='cuda:1'), in_proj_covar=tensor([0.0632, 0.0563, 0.0751, 0.0633, 0.0694, 0.0506, 0.0465, 0.0701], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1') 2023-04-01 16:06:20,989 INFO [train.py:903] (1/4) Epoch 10, batch 250, loss[loss=0.2502, simple_loss=0.3244, pruned_loss=0.088, over 19680.00 frames. ], tot_loss[loss=0.2425, simple_loss=0.3136, pruned_loss=0.08573, over 2727842.28 frames. ], batch size: 60, lr: 8.56e-03, grad_scale: 8.0 2023-04-01 16:07:19,627 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=61750.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:07:20,498 INFO [train.py:903] (1/4) Epoch 10, batch 300, loss[loss=0.2666, simple_loss=0.3377, pruned_loss=0.09776, over 19782.00 frames. ], tot_loss[loss=0.2441, simple_loss=0.3149, pruned_loss=0.08664, over 2971985.68 frames. ], batch size: 56, lr: 8.55e-03, grad_scale: 8.0 2023-04-01 16:07:32,773 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.696e+02 5.563e+02 6.785e+02 8.281e+02 1.821e+03, threshold=1.357e+03, percent-clipped=1.0 2023-04-01 16:07:33,012 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=61761.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:08:12,847 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=61794.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:08:20,594 WARNING [train.py:1073] (1/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1_rvb from training. Duration: 0.9681875 2023-04-01 16:08:21,878 INFO [train.py:903] (1/4) Epoch 10, batch 350, loss[loss=0.1957, simple_loss=0.2667, pruned_loss=0.06238, over 19306.00 frames. ], tot_loss[loss=0.2448, simple_loss=0.3153, pruned_loss=0.08716, over 3161159.80 frames. ], batch size: 44, lr: 8.55e-03, grad_scale: 8.0 2023-04-01 16:08:44,831 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=61819.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:09:23,130 INFO [train.py:903] (1/4) Epoch 10, batch 400, loss[loss=0.2827, simple_loss=0.3417, pruned_loss=0.1119, over 13589.00 frames. ], tot_loss[loss=0.2461, simple_loss=0.3166, pruned_loss=0.08778, over 3297592.11 frames. ], batch size: 137, lr: 8.54e-03, grad_scale: 8.0 2023-04-01 16:09:36,145 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.162e+02 5.438e+02 6.846e+02 8.745e+02 2.106e+03, threshold=1.369e+03, percent-clipped=7.0 2023-04-01 16:10:26,723 INFO [train.py:903] (1/4) Epoch 10, batch 450, loss[loss=0.2899, simple_loss=0.3516, pruned_loss=0.1141, over 19294.00 frames. ], tot_loss[loss=0.2458, simple_loss=0.3161, pruned_loss=0.08774, over 3421229.01 frames. ], batch size: 66, lr: 8.54e-03, grad_scale: 8.0 2023-04-01 16:10:29,713 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 2023-04-01 16:10:50,948 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 2023-04-01 16:10:50,978 WARNING [train.py:1073] (1/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 2023-04-01 16:11:05,169 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=61932.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:11:19,229 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.8821, 4.3541, 4.6086, 4.5647, 1.5845, 4.3038, 3.7466, 4.2541], device='cuda:1'), covar=tensor([0.1232, 0.0707, 0.0461, 0.0532, 0.5142, 0.0583, 0.0564, 0.0943], device='cuda:1'), in_proj_covar=tensor([0.0624, 0.0559, 0.0743, 0.0631, 0.0687, 0.0500, 0.0460, 0.0690], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1') 2023-04-01 16:11:27,992 INFO [train.py:903] (1/4) Epoch 10, batch 500, loss[loss=0.2556, simple_loss=0.3332, pruned_loss=0.08903, over 19461.00 frames. ], tot_loss[loss=0.2435, simple_loss=0.3146, pruned_loss=0.08619, over 3508363.67 frames. ], batch size: 49, lr: 8.54e-03, grad_scale: 8.0 2023-04-01 16:11:30,094 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.45 vs. limit=2.0 2023-04-01 16:11:39,882 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.771e+02 5.473e+02 6.472e+02 7.882e+02 1.512e+03, threshold=1.294e+03, percent-clipped=2.0 2023-04-01 16:12:30,839 INFO [train.py:903] (1/4) Epoch 10, batch 550, loss[loss=0.2276, simple_loss=0.2924, pruned_loss=0.08139, over 19343.00 frames. ], tot_loss[loss=0.2415, simple_loss=0.3128, pruned_loss=0.0851, over 3592109.64 frames. ], batch size: 44, lr: 8.53e-03, grad_scale: 8.0 2023-04-01 16:13:32,125 INFO [train.py:903] (1/4) Epoch 10, batch 600, loss[loss=0.3382, simple_loss=0.3735, pruned_loss=0.1515, over 12885.00 frames. ], tot_loss[loss=0.2423, simple_loss=0.3135, pruned_loss=0.08551, over 3640389.63 frames. ], batch size: 135, lr: 8.53e-03, grad_scale: 8.0 2023-04-01 16:13:42,201 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.31 vs. limit=2.0 2023-04-01 16:13:46,038 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.499e+02 5.366e+02 7.096e+02 8.541e+02 1.663e+03, threshold=1.419e+03, percent-clipped=2.0 2023-04-01 16:14:11,915 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 2023-04-01 16:14:26,432 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=62094.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:14:35,324 INFO [train.py:903] (1/4) Epoch 10, batch 650, loss[loss=0.2397, simple_loss=0.3253, pruned_loss=0.07707, over 19674.00 frames. ], tot_loss[loss=0.2439, simple_loss=0.3149, pruned_loss=0.08651, over 3679514.42 frames. ], batch size: 55, lr: 8.53e-03, grad_scale: 8.0 2023-04-01 16:14:40,022 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=62105.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:15:24,625 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=62140.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:15:39,373 INFO [train.py:903] (1/4) Epoch 10, batch 700, loss[loss=0.2329, simple_loss=0.3014, pruned_loss=0.08217, over 19477.00 frames. ], tot_loss[loss=0.2427, simple_loss=0.3135, pruned_loss=0.08591, over 3708722.16 frames. ], batch size: 49, lr: 8.52e-03, grad_scale: 8.0 2023-04-01 16:15:51,155 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.410e+02 5.986e+02 7.007e+02 9.230e+02 2.462e+03, threshold=1.401e+03, percent-clipped=6.0 2023-04-01 16:15:51,631 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1926, 1.3134, 1.7656, 1.5503, 2.5031, 1.9125, 2.4331, 1.0694], device='cuda:1'), covar=tensor([0.2276, 0.3916, 0.2078, 0.1743, 0.1384, 0.2121, 0.1647, 0.3613], device='cuda:1'), in_proj_covar=tensor([0.0476, 0.0557, 0.0575, 0.0426, 0.0585, 0.0480, 0.0639, 0.0475], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 16:16:41,935 INFO [train.py:903] (1/4) Epoch 10, batch 750, loss[loss=0.2609, simple_loss=0.3334, pruned_loss=0.09416, over 19343.00 frames. ], tot_loss[loss=0.2423, simple_loss=0.3133, pruned_loss=0.08563, over 3741649.78 frames. ], batch size: 70, lr: 8.52e-03, grad_scale: 8.0 2023-04-01 16:16:51,035 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=62209.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:17:04,548 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=62220.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:17:11,177 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.8017, 1.5088, 1.7675, 1.7521, 4.2521, 0.9019, 2.2132, 4.5048], device='cuda:1'), covar=tensor([0.0342, 0.2554, 0.2716, 0.1679, 0.0705, 0.2688, 0.1496, 0.0238], device='cuda:1'), in_proj_covar=tensor([0.0338, 0.0326, 0.0337, 0.0311, 0.0338, 0.0322, 0.0317, 0.0334], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 16:17:35,231 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.81 vs. limit=5.0 2023-04-01 16:17:42,578 INFO [train.py:903] (1/4) Epoch 10, batch 800, loss[loss=0.2624, simple_loss=0.3256, pruned_loss=0.09962, over 19478.00 frames. ], tot_loss[loss=0.2429, simple_loss=0.314, pruned_loss=0.08585, over 3766113.46 frames. ], batch size: 49, lr: 8.52e-03, grad_scale: 8.0 2023-04-01 16:17:54,854 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.040e+02 5.237e+02 6.930e+02 8.487e+02 1.526e+03, threshold=1.386e+03, percent-clipped=2.0 2023-04-01 16:17:58,877 WARNING [train.py:1073] (1/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 2023-04-01 16:18:14,965 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=62276.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:18:32,368 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=62291.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:18:43,719 INFO [train.py:903] (1/4) Epoch 10, batch 850, loss[loss=0.2361, simple_loss=0.3121, pruned_loss=0.08007, over 19666.00 frames. ], tot_loss[loss=0.2436, simple_loss=0.3147, pruned_loss=0.08624, over 3772253.61 frames. ], batch size: 53, lr: 8.51e-03, grad_scale: 8.0 2023-04-01 16:19:37,721 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9_rvb from training. Duration: 25.061125 2023-04-01 16:19:45,734 INFO [train.py:903] (1/4) Epoch 10, batch 900, loss[loss=0.2749, simple_loss=0.3371, pruned_loss=0.1063, over 19710.00 frames. ], tot_loss[loss=0.2437, simple_loss=0.3146, pruned_loss=0.08635, over 3777740.42 frames. ], batch size: 63, lr: 8.51e-03, grad_scale: 8.0 2023-04-01 16:19:59,162 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.268e+02 6.023e+02 7.076e+02 9.770e+02 2.916e+03, threshold=1.415e+03, percent-clipped=7.0 2023-04-01 16:20:16,796 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.1288, 2.1849, 2.2460, 3.1874, 2.1676, 3.1757, 2.6466, 1.9285], device='cuda:1'), covar=tensor([0.3475, 0.2899, 0.1400, 0.1683, 0.3327, 0.1241, 0.3123, 0.2666], device='cuda:1'), in_proj_covar=tensor([0.0762, 0.0769, 0.0629, 0.0879, 0.0759, 0.0674, 0.0774, 0.0687], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 16:20:35,854 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=62391.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:20:48,986 INFO [train.py:903] (1/4) Epoch 10, batch 950, loss[loss=0.2491, simple_loss=0.3228, pruned_loss=0.08765, over 17218.00 frames. ], tot_loss[loss=0.2441, simple_loss=0.3147, pruned_loss=0.08672, over 3789319.70 frames. ], batch size: 101, lr: 8.51e-03, grad_scale: 8.0 2023-04-01 16:20:50,168 WARNING [train.py:1073] (1/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9_rvb from training. Duration: 26.32775 2023-04-01 16:20:58,654 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=62409.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:21:50,240 INFO [train.py:903] (1/4) Epoch 10, batch 1000, loss[loss=0.2575, simple_loss=0.3256, pruned_loss=0.09464, over 19347.00 frames. ], tot_loss[loss=0.2444, simple_loss=0.3151, pruned_loss=0.08681, over 3797791.90 frames. ], batch size: 66, lr: 8.50e-03, grad_scale: 8.0 2023-04-01 16:22:01,606 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.402e+02 5.469e+02 6.659e+02 8.311e+02 1.987e+03, threshold=1.332e+03, percent-clipped=4.0 2023-04-01 16:22:07,646 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=62465.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:22:21,435 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=62476.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:22:31,379 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=62484.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:22:38,715 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=62490.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:22:43,042 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 2023-04-01 16:22:51,162 INFO [train.py:903] (1/4) Epoch 10, batch 1050, loss[loss=0.2382, simple_loss=0.3146, pruned_loss=0.08092, over 19349.00 frames. ], tot_loss[loss=0.2426, simple_loss=0.3139, pruned_loss=0.0857, over 3806501.11 frames. ], batch size: 66, lr: 8.50e-03, grad_scale: 8.0 2023-04-01 16:22:51,512 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=62501.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:23:23,752 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 2023-04-01 16:23:54,502 INFO [train.py:903] (1/4) Epoch 10, batch 1100, loss[loss=0.2844, simple_loss=0.3567, pruned_loss=0.1061, over 19767.00 frames. ], tot_loss[loss=0.2431, simple_loss=0.3142, pruned_loss=0.08597, over 3809161.86 frames. ], batch size: 56, lr: 8.50e-03, grad_scale: 8.0 2023-04-01 16:24:07,773 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.238e+02 5.569e+02 6.927e+02 9.101e+02 1.941e+03, threshold=1.385e+03, percent-clipped=3.0 2023-04-01 16:24:10,208 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=62563.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:24:54,297 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=62599.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:24:56,396 INFO [train.py:903] (1/4) Epoch 10, batch 1150, loss[loss=0.2308, simple_loss=0.2854, pruned_loss=0.08813, over 19741.00 frames. ], tot_loss[loss=0.2433, simple_loss=0.3147, pruned_loss=0.08593, over 3813981.41 frames. ], batch size: 46, lr: 8.49e-03, grad_scale: 8.0 2023-04-01 16:24:59,688 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=62603.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:25:15,637 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2789, 2.0869, 1.6526, 1.3787, 1.9414, 1.2032, 1.1536, 1.7383], device='cuda:1'), covar=tensor([0.0694, 0.0531, 0.0858, 0.0617, 0.0351, 0.1102, 0.0558, 0.0373], device='cuda:1'), in_proj_covar=tensor([0.0287, 0.0296, 0.0321, 0.0239, 0.0231, 0.0325, 0.0285, 0.0255], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 16:25:37,686 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=62635.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:25:47,589 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.91 vs. limit=2.0 2023-04-01 16:25:53,819 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=62647.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:25:58,136 INFO [train.py:903] (1/4) Epoch 10, batch 1200, loss[loss=0.297, simple_loss=0.3584, pruned_loss=0.1178, over 19361.00 frames. ], tot_loss[loss=0.2425, simple_loss=0.3139, pruned_loss=0.08552, over 3819723.90 frames. ], batch size: 66, lr: 8.49e-03, grad_scale: 8.0 2023-04-01 16:26:09,506 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.384e+02 5.581e+02 6.902e+02 9.588e+02 2.703e+03, threshold=1.380e+03, percent-clipped=8.0 2023-04-01 16:26:24,936 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=62672.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:26:32,549 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 2023-04-01 16:26:39,718 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=62684.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:26:50,277 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.66 vs. limit=5.0 2023-04-01 16:26:52,246 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.1348, 1.8185, 1.5925, 2.1124, 1.9434, 1.7466, 1.7351, 1.9605], device='cuda:1'), covar=tensor([0.0869, 0.1512, 0.1395, 0.0937, 0.1210, 0.0538, 0.1120, 0.0721], device='cuda:1'), in_proj_covar=tensor([0.0252, 0.0355, 0.0288, 0.0241, 0.0297, 0.0244, 0.0273, 0.0239], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 16:26:53,300 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.8934, 1.7083, 1.4764, 2.0629, 1.8191, 1.6200, 1.6068, 1.8204], device='cuda:1'), covar=tensor([0.0904, 0.1526, 0.1359, 0.0875, 0.1132, 0.0529, 0.1078, 0.0700], device='cuda:1'), in_proj_covar=tensor([0.0252, 0.0355, 0.0288, 0.0241, 0.0297, 0.0244, 0.0273, 0.0239], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 16:26:59,719 INFO [train.py:903] (1/4) Epoch 10, batch 1250, loss[loss=0.2512, simple_loss=0.3254, pruned_loss=0.08852, over 19443.00 frames. ], tot_loss[loss=0.2438, simple_loss=0.315, pruned_loss=0.08626, over 3813286.68 frames. ], batch size: 64, lr: 8.49e-03, grad_scale: 4.0 2023-04-01 16:27:08,182 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.7242, 1.4740, 1.3806, 2.2127, 1.7941, 1.9891, 2.1712, 1.8331], device='cuda:1'), covar=tensor([0.0800, 0.0995, 0.1117, 0.0770, 0.0762, 0.0762, 0.0806, 0.0707], device='cuda:1'), in_proj_covar=tensor([0.0213, 0.0229, 0.0227, 0.0255, 0.0239, 0.0215, 0.0199, 0.0208], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-01 16:28:00,887 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=62750.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:28:01,682 INFO [train.py:903] (1/4) Epoch 10, batch 1300, loss[loss=0.2055, simple_loss=0.2759, pruned_loss=0.06754, over 19769.00 frames. ], tot_loss[loss=0.2426, simple_loss=0.3138, pruned_loss=0.08566, over 3831606.16 frames. ], batch size: 46, lr: 8.48e-03, grad_scale: 4.0 2023-04-01 16:28:05,045 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=62753.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:28:16,587 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.133e+02 5.064e+02 6.596e+02 8.862e+02 1.920e+03, threshold=1.319e+03, percent-clipped=1.0 2023-04-01 16:28:51,852 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=62791.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:29:05,059 INFO [train.py:903] (1/4) Epoch 10, batch 1350, loss[loss=0.2733, simple_loss=0.3394, pruned_loss=0.1037, over 19675.00 frames. ], tot_loss[loss=0.2428, simple_loss=0.3141, pruned_loss=0.08578, over 3835469.39 frames. ], batch size: 58, lr: 8.48e-03, grad_scale: 4.0 2023-04-01 16:29:58,563 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.62 vs. limit=5.0 2023-04-01 16:30:07,948 INFO [train.py:903] (1/4) Epoch 10, batch 1400, loss[loss=0.2902, simple_loss=0.3502, pruned_loss=0.1151, over 17238.00 frames. ], tot_loss[loss=0.2431, simple_loss=0.3144, pruned_loss=0.08591, over 3817580.66 frames. ], batch size: 101, lr: 8.48e-03, grad_scale: 4.0 2023-04-01 16:30:13,139 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=62855.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:30:20,948 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.159e+02 5.591e+02 6.749e+02 8.217e+02 1.554e+03, threshold=1.350e+03, percent-clipped=4.0 2023-04-01 16:30:28,002 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=62868.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:30:44,897 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=62880.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:31:07,150 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_rvb from training. Duration: 25.85 2023-04-01 16:31:09,395 INFO [train.py:903] (1/4) Epoch 10, batch 1450, loss[loss=0.2228, simple_loss=0.3054, pruned_loss=0.07006, over 19687.00 frames. ], tot_loss[loss=0.2418, simple_loss=0.3132, pruned_loss=0.08523, over 3812844.07 frames. ], batch size: 59, lr: 8.47e-03, grad_scale: 4.0 2023-04-01 16:31:11,268 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.99 vs. limit=5.0 2023-04-01 16:31:16,536 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=62907.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:32:07,379 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=62947.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:32:11,880 INFO [train.py:903] (1/4) Epoch 10, batch 1500, loss[loss=0.2252, simple_loss=0.3059, pruned_loss=0.0723, over 19347.00 frames. ], tot_loss[loss=0.2412, simple_loss=0.3125, pruned_loss=0.08499, over 3818437.44 frames. ], batch size: 70, lr: 8.47e-03, grad_scale: 4.0 2023-04-01 16:32:27,777 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.889e+02 5.675e+02 6.883e+02 8.252e+02 2.690e+03, threshold=1.377e+03, percent-clipped=4.0 2023-04-01 16:33:00,313 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6887, 1.9021, 2.2719, 1.9560, 2.9599, 3.3305, 3.2774, 3.5169], device='cuda:1'), covar=tensor([0.1272, 0.2449, 0.2176, 0.1674, 0.0768, 0.0288, 0.0171, 0.0191], device='cuda:1'), in_proj_covar=tensor([0.0258, 0.0293, 0.0323, 0.0249, 0.0213, 0.0149, 0.0206, 0.0189], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 16:33:17,111 INFO [train.py:903] (1/4) Epoch 10, batch 1550, loss[loss=0.2933, simple_loss=0.3596, pruned_loss=0.1135, over 19323.00 frames. ], tot_loss[loss=0.241, simple_loss=0.3124, pruned_loss=0.08483, over 3814660.88 frames. ], batch size: 66, lr: 8.47e-03, grad_scale: 4.0 2023-04-01 16:33:23,635 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63006.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:33:42,967 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=63022.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:33:49,635 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=63028.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:33:53,330 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=63031.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:34:20,243 INFO [train.py:903] (1/4) Epoch 10, batch 1600, loss[loss=0.2816, simple_loss=0.3409, pruned_loss=0.1111, over 19593.00 frames. ], tot_loss[loss=0.2402, simple_loss=0.3117, pruned_loss=0.08438, over 3803945.84 frames. ], batch size: 57, lr: 8.46e-03, grad_scale: 8.0 2023-04-01 16:34:31,099 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.8068, 1.8985, 1.9723, 2.4911, 1.6809, 2.2331, 2.2575, 1.9955], device='cuda:1'), covar=tensor([0.3082, 0.2492, 0.1300, 0.1465, 0.2822, 0.1328, 0.2898, 0.2208], device='cuda:1'), in_proj_covar=tensor([0.0763, 0.0772, 0.0630, 0.0875, 0.0753, 0.0674, 0.0771, 0.0685], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 16:34:33,007 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.609e+02 5.367e+02 6.713e+02 8.573e+02 1.582e+03, threshold=1.343e+03, percent-clipped=2.0 2023-04-01 16:34:33,387 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=63062.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:34:42,288 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9_rvb from training. Duration: 30.1555625 2023-04-01 16:35:21,330 INFO [train.py:903] (1/4) Epoch 10, batch 1650, loss[loss=0.2567, simple_loss=0.3337, pruned_loss=0.08984, over 17089.00 frames. ], tot_loss[loss=0.2417, simple_loss=0.313, pruned_loss=0.08519, over 3797709.02 frames. ], batch size: 101, lr: 8.46e-03, grad_scale: 8.0 2023-04-01 16:35:32,994 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0 2023-04-01 16:35:51,935 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63124.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:36:05,475 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=63135.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:36:14,686 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=63143.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:36:21,892 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=63149.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:36:23,862 INFO [train.py:903] (1/4) Epoch 10, batch 1700, loss[loss=0.2004, simple_loss=0.2711, pruned_loss=0.06485, over 19746.00 frames. ], tot_loss[loss=0.2414, simple_loss=0.3128, pruned_loss=0.08497, over 3808746.68 frames. ], batch size: 45, lr: 8.46e-03, grad_scale: 8.0 2023-04-01 16:36:38,454 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.334e+02 5.819e+02 7.178e+02 9.040e+02 2.117e+03, threshold=1.436e+03, percent-clipped=7.0 2023-04-01 16:37:04,077 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 2023-04-01 16:37:28,653 INFO [train.py:903] (1/4) Epoch 10, batch 1750, loss[loss=0.2207, simple_loss=0.28, pruned_loss=0.08069, over 19084.00 frames. ], tot_loss[loss=0.242, simple_loss=0.3132, pruned_loss=0.08536, over 3818553.18 frames. ], batch size: 42, lr: 8.45e-03, grad_scale: 4.0 2023-04-01 16:37:36,123 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2355, 1.3033, 1.1529, 0.9803, 1.0418, 1.0280, 0.0443, 0.3735], device='cuda:1'), covar=tensor([0.0414, 0.0409, 0.0244, 0.0356, 0.0783, 0.0359, 0.0737, 0.0677], device='cuda:1'), in_proj_covar=tensor([0.0326, 0.0322, 0.0326, 0.0338, 0.0413, 0.0338, 0.0303, 0.0318], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 16:38:14,933 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=63239.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:38:30,194 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=63250.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:38:30,879 INFO [train.py:903] (1/4) Epoch 10, batch 1800, loss[loss=0.2492, simple_loss=0.3294, pruned_loss=0.08453, over 19686.00 frames. ], tot_loss[loss=0.2409, simple_loss=0.3123, pruned_loss=0.08476, over 3824952.17 frames. ], batch size: 59, lr: 8.45e-03, grad_scale: 4.0 2023-04-01 16:38:44,559 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.600e+02 5.870e+02 7.659e+02 9.293e+02 2.596e+03, threshold=1.532e+03, percent-clipped=8.0 2023-04-01 16:39:03,322 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63278.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:39:16,053 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.7074, 1.4932, 1.6854, 1.8414, 4.1300, 1.1104, 2.1000, 4.3872], device='cuda:1'), covar=tensor([0.0417, 0.2600, 0.2777, 0.1732, 0.0751, 0.2579, 0.1643, 0.0262], device='cuda:1'), in_proj_covar=tensor([0.0343, 0.0327, 0.0341, 0.0315, 0.0340, 0.0325, 0.0322, 0.0343], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 16:39:30,783 WARNING [train.py:1073] (1/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9_rvb from training. Duration: 25.0944375 2023-04-01 16:39:32,958 INFO [train.py:903] (1/4) Epoch 10, batch 1850, loss[loss=0.1981, simple_loss=0.2699, pruned_loss=0.06312, over 19769.00 frames. ], tot_loss[loss=0.2399, simple_loss=0.3115, pruned_loss=0.0842, over 3820030.05 frames. ], batch size: 47, lr: 8.45e-03, grad_scale: 4.0 2023-04-01 16:39:35,786 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=63303.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:39:54,126 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63318.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:40:10,163 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9_rvb from training. Duration: 27.8166875 2023-04-01 16:40:26,813 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=63343.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:40:35,815 INFO [train.py:903] (1/4) Epoch 10, batch 1900, loss[loss=0.267, simple_loss=0.3426, pruned_loss=0.09577, over 19340.00 frames. ], tot_loss[loss=0.2376, simple_loss=0.3097, pruned_loss=0.08272, over 3828255.73 frames. ], batch size: 70, lr: 8.44e-03, grad_scale: 4.0 2023-04-01 16:40:45,541 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 2023-04-01 16:40:52,653 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.343e+02 5.562e+02 7.038e+02 8.618e+02 1.834e+03, threshold=1.408e+03, percent-clipped=3.0 2023-04-01 16:40:55,973 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 2023-04-01 16:41:01,866 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9_rvb from training. Duration: 27.02225 2023-04-01 16:41:24,920 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 2023-04-01 16:41:37,241 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63399.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:41:40,426 INFO [train.py:903] (1/4) Epoch 10, batch 1950, loss[loss=0.2171, simple_loss=0.2806, pruned_loss=0.07682, over 19311.00 frames. ], tot_loss[loss=0.2386, simple_loss=0.3104, pruned_loss=0.08341, over 3826823.43 frames. ], batch size: 44, lr: 8.44e-03, grad_scale: 4.0 2023-04-01 16:41:58,525 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.6970, 3.9127, 4.2623, 4.2212, 2.3680, 3.8766, 3.6663, 3.9925], device='cuda:1'), covar=tensor([0.1065, 0.1997, 0.0524, 0.0545, 0.3887, 0.0891, 0.0525, 0.0945], device='cuda:1'), in_proj_covar=tensor([0.0642, 0.0583, 0.0763, 0.0644, 0.0712, 0.0516, 0.0478, 0.0708], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1') 2023-04-01 16:42:10,539 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=63424.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:42:27,978 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=63438.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:42:31,429 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=63440.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:42:44,769 INFO [train.py:903] (1/4) Epoch 10, batch 2000, loss[loss=0.2993, simple_loss=0.3551, pruned_loss=0.1218, over 19424.00 frames. ], tot_loss[loss=0.2409, simple_loss=0.3125, pruned_loss=0.08468, over 3796409.93 frames. ], batch size: 70, lr: 8.44e-03, grad_scale: 8.0 2023-04-01 16:43:00,248 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.731e+02 5.221e+02 6.641e+02 8.776e+02 2.044e+03, threshold=1.328e+03, percent-clipped=3.0 2023-04-01 16:43:43,946 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_rvb from training. Duration: 26.8349375 2023-04-01 16:43:47,387 INFO [train.py:903] (1/4) Epoch 10, batch 2050, loss[loss=0.2692, simple_loss=0.3364, pruned_loss=0.101, over 13624.00 frames. ], tot_loss[loss=0.2408, simple_loss=0.3125, pruned_loss=0.08453, over 3798148.45 frames. ], batch size: 136, lr: 8.43e-03, grad_scale: 8.0 2023-04-01 16:43:53,649 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63506.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:44:03,084 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.6533, 1.4559, 1.6413, 1.6661, 3.1843, 1.1623, 2.3106, 3.5098], device='cuda:1'), covar=tensor([0.0371, 0.2477, 0.2505, 0.1698, 0.0672, 0.2515, 0.1314, 0.0291], device='cuda:1'), in_proj_covar=tensor([0.0343, 0.0328, 0.0344, 0.0313, 0.0342, 0.0328, 0.0323, 0.0346], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 16:44:04,853 WARNING [train.py:1073] (1/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 2023-04-01 16:44:06,009 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 2023-04-01 16:44:26,383 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=63531.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:44:28,443 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 2023-04-01 16:44:50,592 INFO [train.py:903] (1/4) Epoch 10, batch 2100, loss[loss=0.2421, simple_loss=0.3056, pruned_loss=0.08928, over 19742.00 frames. ], tot_loss[loss=0.2407, simple_loss=0.3122, pruned_loss=0.08456, over 3810142.72 frames. ], batch size: 46, lr: 8.43e-03, grad_scale: 8.0 2023-04-01 16:45:06,392 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.339e+02 5.645e+02 7.348e+02 9.688e+02 2.351e+03, threshold=1.470e+03, percent-clipped=4.0 2023-04-01 16:45:22,180 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1898, 1.2731, 1.8414, 1.5719, 3.0790, 4.5254, 4.4909, 4.9798], device='cuda:1'), covar=tensor([0.1650, 0.3283, 0.2988, 0.1904, 0.0461, 0.0150, 0.0150, 0.0100], device='cuda:1'), in_proj_covar=tensor([0.0255, 0.0292, 0.0321, 0.0249, 0.0212, 0.0151, 0.0205, 0.0188], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 16:45:26,359 WARNING [train.py:1073] (1/4) Exclude cut with ID 453-131332-0000-131866_sp0.9_rvb from training. Duration: 25.3333125 2023-04-01 16:45:32,499 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=63583.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:45:33,931 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.4137, 2.9758, 2.1206, 2.2453, 2.0902, 2.4789, 0.7448, 2.0942], device='cuda:1'), covar=tensor([0.0401, 0.0394, 0.0527, 0.0699, 0.0784, 0.0715, 0.0992, 0.0780], device='cuda:1'), in_proj_covar=tensor([0.0330, 0.0328, 0.0329, 0.0343, 0.0420, 0.0342, 0.0306, 0.0323], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 16:45:37,323 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=63587.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:45:46,595 WARNING [train.py:1073] (1/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9_rvb from training. Duration: 26.6166875 2023-04-01 16:45:54,843 INFO [train.py:903] (1/4) Epoch 10, batch 2150, loss[loss=0.2609, simple_loss=0.3336, pruned_loss=0.09405, over 19535.00 frames. ], tot_loss[loss=0.2414, simple_loss=0.3128, pruned_loss=0.085, over 3812480.02 frames. ], batch size: 54, lr: 8.43e-03, grad_scale: 8.0 2023-04-01 16:46:00,142 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=63604.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:46:58,770 INFO [train.py:903] (1/4) Epoch 10, batch 2200, loss[loss=0.2158, simple_loss=0.2954, pruned_loss=0.06808, over 19701.00 frames. ], tot_loss[loss=0.2421, simple_loss=0.3132, pruned_loss=0.08545, over 3808634.27 frames. ], batch size: 59, lr: 8.42e-03, grad_scale: 8.0 2023-04-01 16:47:13,922 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.200e+02 5.616e+02 6.875e+02 8.680e+02 1.983e+03, threshold=1.375e+03, percent-clipped=4.0 2023-04-01 16:48:00,194 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=63698.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:48:03,390 INFO [train.py:903] (1/4) Epoch 10, batch 2250, loss[loss=0.2898, simple_loss=0.355, pruned_loss=0.1123, over 19628.00 frames. ], tot_loss[loss=0.2432, simple_loss=0.314, pruned_loss=0.08624, over 3809460.41 frames. ], batch size: 57, lr: 8.42e-03, grad_scale: 8.0 2023-04-01 16:49:04,928 INFO [train.py:903] (1/4) Epoch 10, batch 2300, loss[loss=0.2441, simple_loss=0.324, pruned_loss=0.08208, over 17980.00 frames. ], tot_loss[loss=0.2427, simple_loss=0.3135, pruned_loss=0.08598, over 3814970.45 frames. ], batch size: 83, lr: 8.42e-03, grad_scale: 4.0 2023-04-01 16:49:18,697 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_rvb from training. Duration: 26.205 2023-04-01 16:49:23,044 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.591e+02 5.837e+02 6.965e+02 8.642e+02 2.205e+03, threshold=1.393e+03, percent-clipped=3.0 2023-04-01 16:49:45,520 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=63782.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:49:47,900 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=63784.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:50:09,268 INFO [train.py:903] (1/4) Epoch 10, batch 2350, loss[loss=0.2448, simple_loss=0.3197, pruned_loss=0.08495, over 19759.00 frames. ], tot_loss[loss=0.2431, simple_loss=0.3142, pruned_loss=0.08596, over 3808322.90 frames. ], batch size: 54, lr: 8.41e-03, grad_scale: 4.0 2023-04-01 16:50:44,299 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=63828.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:50:50,893 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_rvb from training. Duration: 25.775 2023-04-01 16:51:10,053 WARNING [train.py:1073] (1/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9_rvb from training. Duration: 25.45 2023-04-01 16:51:11,956 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.69 vs. limit=2.0 2023-04-01 16:51:13,625 INFO [train.py:903] (1/4) Epoch 10, batch 2400, loss[loss=0.2975, simple_loss=0.3594, pruned_loss=0.1178, over 19064.00 frames. ], tot_loss[loss=0.2451, simple_loss=0.3162, pruned_loss=0.08704, over 3786298.90 frames. ], batch size: 69, lr: 8.41e-03, grad_scale: 8.0 2023-04-01 16:51:29,381 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=63863.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:51:30,159 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.763e+02 5.915e+02 7.022e+02 9.244e+02 2.907e+03, threshold=1.404e+03, percent-clipped=10.0 2023-04-01 16:51:55,379 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6917, 1.7838, 1.8894, 2.4631, 1.6869, 2.2509, 2.1532, 1.8350], device='cuda:1'), covar=tensor([0.3254, 0.2632, 0.1358, 0.1495, 0.2783, 0.1286, 0.3206, 0.2468], device='cuda:1'), in_proj_covar=tensor([0.0772, 0.0780, 0.0638, 0.0882, 0.0763, 0.0683, 0.0781, 0.0695], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 16:52:03,536 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.65 vs. limit=2.0 2023-04-01 16:52:12,815 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=63897.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:52:15,946 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=63899.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:52:17,962 INFO [train.py:903] (1/4) Epoch 10, batch 2450, loss[loss=0.2485, simple_loss=0.3306, pruned_loss=0.08321, over 19677.00 frames. ], tot_loss[loss=0.2445, simple_loss=0.3159, pruned_loss=0.08662, over 3792335.72 frames. ], batch size: 60, lr: 8.41e-03, grad_scale: 8.0 2023-04-01 16:52:55,530 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=63931.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:52:59,294 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=63934.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:53:16,595 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=63948.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:53:19,961 INFO [train.py:903] (1/4) Epoch 10, batch 2500, loss[loss=0.2093, simple_loss=0.2878, pruned_loss=0.06537, over 19747.00 frames. ], tot_loss[loss=0.2426, simple_loss=0.3142, pruned_loss=0.08548, over 3801674.70 frames. ], batch size: 51, lr: 8.40e-03, grad_scale: 8.0 2023-04-01 16:53:24,570 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63954.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:53:35,920 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.566e+02 5.261e+02 6.826e+02 9.233e+02 1.687e+03, threshold=1.365e+03, percent-clipped=6.0 2023-04-01 16:53:56,277 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=63979.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:54:00,242 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0 2023-04-01 16:54:23,067 INFO [train.py:903] (1/4) Epoch 10, batch 2550, loss[loss=0.2366, simple_loss=0.2947, pruned_loss=0.08924, over 19750.00 frames. ], tot_loss[loss=0.2419, simple_loss=0.3133, pruned_loss=0.08526, over 3790700.32 frames. ], batch size: 46, lr: 8.40e-03, grad_scale: 8.0 2023-04-01 16:54:23,594 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.2187, 2.2180, 2.2840, 3.2956, 2.1331, 3.0886, 2.6876, 2.1431], device='cuda:1'), covar=tensor([0.3668, 0.3250, 0.1456, 0.1734, 0.3745, 0.1362, 0.3371, 0.2631], device='cuda:1'), in_proj_covar=tensor([0.0770, 0.0778, 0.0636, 0.0876, 0.0760, 0.0679, 0.0777, 0.0692], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 16:54:39,877 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.8056, 1.1786, 1.4426, 1.5587, 3.2719, 1.0326, 2.3129, 3.5616], device='cuda:1'), covar=tensor([0.0372, 0.2724, 0.2720, 0.1744, 0.0725, 0.2457, 0.1228, 0.0305], device='cuda:1'), in_proj_covar=tensor([0.0343, 0.0331, 0.0348, 0.0316, 0.0342, 0.0329, 0.0325, 0.0346], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 16:54:44,604 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4399, 1.4309, 1.6033, 1.5227, 2.3402, 2.0056, 2.3132, 1.2634], device='cuda:1'), covar=tensor([0.1652, 0.3018, 0.1810, 0.1430, 0.1009, 0.1476, 0.1072, 0.2917], device='cuda:1'), in_proj_covar=tensor([0.0483, 0.0563, 0.0582, 0.0428, 0.0584, 0.0481, 0.0647, 0.0483], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 16:54:56,132 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9496, 1.9790, 1.6972, 2.1462, 1.9743, 1.7731, 1.6590, 1.9620], device='cuda:1'), covar=tensor([0.0851, 0.1249, 0.1296, 0.0835, 0.1081, 0.0499, 0.1125, 0.0626], device='cuda:1'), in_proj_covar=tensor([0.0246, 0.0347, 0.0285, 0.0236, 0.0293, 0.0241, 0.0271, 0.0232], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 16:55:13,524 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64041.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:55:15,683 WARNING [train.py:1073] (1/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 2023-04-01 16:55:20,789 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64046.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:55:26,447 INFO [train.py:903] (1/4) Epoch 10, batch 2600, loss[loss=0.2573, simple_loss=0.3206, pruned_loss=0.09698, over 19763.00 frames. ], tot_loss[loss=0.2409, simple_loss=0.3125, pruned_loss=0.08459, over 3785097.93 frames. ], batch size: 63, lr: 8.40e-03, grad_scale: 8.0 2023-04-01 16:55:41,485 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64063.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:55:42,246 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.625e+02 5.188e+02 6.411e+02 7.864e+02 1.888e+03, threshold=1.282e+03, percent-clipped=1.0 2023-04-01 16:55:48,731 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.8143, 1.4644, 1.3862, 2.0480, 1.6445, 2.0356, 2.0908, 1.7737], device='cuda:1'), covar=tensor([0.0792, 0.1037, 0.1031, 0.0880, 0.0964, 0.0684, 0.0874, 0.0715], device='cuda:1'), in_proj_covar=tensor([0.0214, 0.0228, 0.0226, 0.0256, 0.0241, 0.0214, 0.0201, 0.0208], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1') 2023-04-01 16:56:27,599 INFO [train.py:903] (1/4) Epoch 10, batch 2650, loss[loss=0.2714, simple_loss=0.3351, pruned_loss=0.1039, over 18801.00 frames. ], tot_loss[loss=0.2428, simple_loss=0.3138, pruned_loss=0.08586, over 3779477.82 frames. ], batch size: 74, lr: 8.39e-03, grad_scale: 8.0 2023-04-01 16:56:44,902 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0 2023-04-01 16:56:45,322 WARNING [train.py:1073] (1/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9_rvb from training. Duration: 27.25 2023-04-01 16:57:23,755 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.8693, 4.4020, 2.6998, 3.9343, 0.9445, 4.0849, 4.1086, 4.3242], device='cuda:1'), covar=tensor([0.0537, 0.0883, 0.1879, 0.0689, 0.3872, 0.0767, 0.0661, 0.0823], device='cuda:1'), in_proj_covar=tensor([0.0411, 0.0351, 0.0417, 0.0311, 0.0370, 0.0345, 0.0341, 0.0372], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-01 16:57:29,262 INFO [train.py:903] (1/4) Epoch 10, batch 2700, loss[loss=0.1999, simple_loss=0.276, pruned_loss=0.06188, over 19611.00 frames. ], tot_loss[loss=0.2418, simple_loss=0.3133, pruned_loss=0.08521, over 3792925.86 frames. ], batch size: 50, lr: 8.39e-03, grad_scale: 8.0 2023-04-01 16:57:32,857 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64153.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:57:35,332 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64155.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:57:45,503 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.698e+02 5.891e+02 6.747e+02 8.779e+02 3.257e+03, threshold=1.349e+03, percent-clipped=11.0 2023-04-01 16:57:56,919 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=64172.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:58:04,115 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64178.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:58:07,364 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64180.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:58:33,079 INFO [train.py:903] (1/4) Epoch 10, batch 2750, loss[loss=0.2141, simple_loss=0.28, pruned_loss=0.0741, over 19785.00 frames. ], tot_loss[loss=0.2432, simple_loss=0.3145, pruned_loss=0.08593, over 3791348.30 frames. ], batch size: 47, lr: 8.39e-03, grad_scale: 8.0 2023-04-01 16:58:41,444 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=64207.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:59:36,110 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64250.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 16:59:36,899 INFO [train.py:903] (1/4) Epoch 10, batch 2800, loss[loss=0.2461, simple_loss=0.3208, pruned_loss=0.08574, over 19523.00 frames. ], tot_loss[loss=0.2427, simple_loss=0.3139, pruned_loss=0.08578, over 3794508.36 frames. ], batch size: 54, lr: 8.38e-03, grad_scale: 8.0 2023-04-01 16:59:42,949 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.8372, 2.0659, 2.1763, 2.1779, 3.5441, 1.7298, 2.8176, 3.5381], device='cuda:1'), covar=tensor([0.0395, 0.1915, 0.1962, 0.1370, 0.0599, 0.2011, 0.1642, 0.0336], device='cuda:1'), in_proj_covar=tensor([0.0339, 0.0327, 0.0342, 0.0313, 0.0339, 0.0326, 0.0319, 0.0341], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 16:59:45,208 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.5437, 4.0926, 2.6985, 3.6752, 1.0544, 3.8076, 3.8256, 3.8911], device='cuda:1'), covar=tensor([0.0580, 0.1005, 0.1808, 0.0672, 0.3670, 0.0751, 0.0765, 0.0927], device='cuda:1'), in_proj_covar=tensor([0.0409, 0.0353, 0.0415, 0.0309, 0.0368, 0.0343, 0.0341, 0.0372], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-01 16:59:52,862 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.388e+02 5.243e+02 6.695e+02 7.923e+02 2.040e+03, threshold=1.339e+03, percent-clipped=2.0 2023-04-01 17:00:10,662 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=64278.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:00:21,813 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([0.9928, 1.2406, 1.6513, 0.8388, 2.3410, 2.9910, 2.6961, 3.1563], device='cuda:1'), covar=tensor([0.1630, 0.3294, 0.2971, 0.2236, 0.0493, 0.0235, 0.0255, 0.0220], device='cuda:1'), in_proj_covar=tensor([0.0255, 0.0291, 0.0320, 0.0249, 0.0211, 0.0149, 0.0203, 0.0190], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 17:00:21,826 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64287.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:00:40,324 INFO [train.py:903] (1/4) Epoch 10, batch 2850, loss[loss=0.2001, simple_loss=0.2804, pruned_loss=0.05989, over 19861.00 frames. ], tot_loss[loss=0.2409, simple_loss=0.3122, pruned_loss=0.08484, over 3795561.77 frames. ], batch size: 52, lr: 8.38e-03, grad_scale: 8.0 2023-04-01 17:00:41,955 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64302.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:01:03,188 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64319.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:01:06,424 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64322.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:01:12,465 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64327.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:01:35,615 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64344.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:01:39,751 WARNING [train.py:1073] (1/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 2023-04-01 17:01:43,209 INFO [train.py:903] (1/4) Epoch 10, batch 2900, loss[loss=0.2482, simple_loss=0.3246, pruned_loss=0.08596, over 19663.00 frames. ], tot_loss[loss=0.2411, simple_loss=0.3125, pruned_loss=0.08484, over 3807167.46 frames. ], batch size: 58, lr: 8.38e-03, grad_scale: 8.0 2023-04-01 17:01:58,180 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64363.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:01:58,968 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.152e+02 6.015e+02 7.349e+02 1.031e+03 2.008e+03, threshold=1.470e+03, percent-clipped=12.0 2023-04-01 17:02:26,622 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=64385.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:02:28,058 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2852, 1.9743, 1.5421, 1.2253, 1.9046, 1.1082, 1.1918, 1.6748], device='cuda:1'), covar=tensor([0.0819, 0.0691, 0.0912, 0.0703, 0.0430, 0.1148, 0.0639, 0.0408], device='cuda:1'), in_proj_covar=tensor([0.0288, 0.0296, 0.0323, 0.0242, 0.0231, 0.0322, 0.0288, 0.0253], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 17:02:36,424 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64393.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:02:44,659 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5701, 1.3004, 1.1723, 1.4469, 1.2177, 1.3557, 1.1563, 1.3718], device='cuda:1'), covar=tensor([0.0880, 0.1058, 0.1408, 0.0840, 0.1031, 0.0573, 0.1191, 0.0730], device='cuda:1'), in_proj_covar=tensor([0.0249, 0.0349, 0.0288, 0.0237, 0.0296, 0.0243, 0.0273, 0.0233], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 17:02:46,686 INFO [train.py:903] (1/4) Epoch 10, batch 2950, loss[loss=0.2465, simple_loss=0.3298, pruned_loss=0.08159, over 19019.00 frames. ], tot_loss[loss=0.2426, simple_loss=0.3137, pruned_loss=0.08577, over 3805369.39 frames. ], batch size: 69, lr: 8.37e-03, grad_scale: 8.0 2023-04-01 17:03:48,250 INFO [train.py:903] (1/4) Epoch 10, batch 3000, loss[loss=0.2073, simple_loss=0.2965, pruned_loss=0.0591, over 19777.00 frames. ], tot_loss[loss=0.243, simple_loss=0.3139, pruned_loss=0.08607, over 3809453.60 frames. ], batch size: 56, lr: 8.37e-03, grad_scale: 8.0 2023-04-01 17:03:48,250 INFO [train.py:928] (1/4) Computing validation loss 2023-04-01 17:04:00,866 INFO [train.py:937] (1/4) Epoch 10, validation: loss=0.1811, simple_loss=0.2816, pruned_loss=0.04036, over 944034.00 frames. 2023-04-01 17:04:00,867 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18649MB 2023-04-01 17:04:01,547 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 2023-04-01 17:04:04,327 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_rvb from training. Duration: 29.735 2023-04-01 17:04:18,119 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.319e+02 5.533e+02 6.661e+02 8.174e+02 1.809e+03, threshold=1.332e+03, percent-clipped=2.0 2023-04-01 17:04:26,738 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.64 vs. limit=2.0 2023-04-01 17:04:41,353 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64483.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:05:03,230 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64500.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:05:04,025 INFO [train.py:903] (1/4) Epoch 10, batch 3050, loss[loss=0.2331, simple_loss=0.3139, pruned_loss=0.07616, over 19534.00 frames. ], tot_loss[loss=0.2424, simple_loss=0.3136, pruned_loss=0.08563, over 3820105.52 frames. ], batch size: 56, lr: 8.37e-03, grad_scale: 8.0 2023-04-01 17:05:48,594 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.7354, 1.4807, 1.4727, 1.9906, 1.7406, 2.0656, 2.0220, 1.8700], device='cuda:1'), covar=tensor([0.0840, 0.1004, 0.1037, 0.0919, 0.0864, 0.0698, 0.0899, 0.0695], device='cuda:1'), in_proj_covar=tensor([0.0213, 0.0226, 0.0225, 0.0253, 0.0239, 0.0216, 0.0202, 0.0208], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-01 17:05:57,668 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9086, 1.6599, 1.6073, 2.1406, 1.8943, 2.1366, 2.0981, 1.9391], device='cuda:1'), covar=tensor([0.0678, 0.0811, 0.0853, 0.0738, 0.0758, 0.0613, 0.0805, 0.0578], device='cuda:1'), in_proj_covar=tensor([0.0213, 0.0226, 0.0225, 0.0254, 0.0240, 0.0216, 0.0202, 0.0208], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-01 17:05:57,692 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64543.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:06:06,595 INFO [train.py:903] (1/4) Epoch 10, batch 3100, loss[loss=0.2024, simple_loss=0.2744, pruned_loss=0.06518, over 19817.00 frames. ], tot_loss[loss=0.2425, simple_loss=0.3136, pruned_loss=0.08564, over 3817676.95 frames. ], batch size: 47, lr: 8.37e-03, grad_scale: 8.0 2023-04-01 17:06:22,857 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.367e+02 5.961e+02 7.216e+02 8.690e+02 2.208e+03, threshold=1.443e+03, percent-clipped=3.0 2023-04-01 17:06:27,950 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64568.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:06:40,848 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64578.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:07:01,270 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=64594.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:07:10,478 INFO [train.py:903] (1/4) Epoch 10, batch 3150, loss[loss=0.2195, simple_loss=0.3034, pruned_loss=0.06783, over 18791.00 frames. ], tot_loss[loss=0.2396, simple_loss=0.3112, pruned_loss=0.08401, over 3826368.85 frames. ], batch size: 74, lr: 8.36e-03, grad_scale: 8.0 2023-04-01 17:07:12,974 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64603.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:07:17,461 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64607.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:07:21,730 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.9517, 4.4193, 4.6640, 4.6431, 1.5321, 4.2915, 3.7840, 4.3340], device='cuda:1'), covar=tensor([0.1196, 0.0576, 0.0446, 0.0469, 0.4929, 0.0533, 0.0539, 0.0908], device='cuda:1'), in_proj_covar=tensor([0.0645, 0.0581, 0.0768, 0.0645, 0.0713, 0.0518, 0.0476, 0.0712], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1') 2023-04-01 17:07:31,624 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.80 vs. limit=2.0 2023-04-01 17:07:39,764 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 2023-04-01 17:08:10,683 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64649.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:08:12,597 INFO [train.py:903] (1/4) Epoch 10, batch 3200, loss[loss=0.1973, simple_loss=0.2711, pruned_loss=0.06172, over 19405.00 frames. ], tot_loss[loss=0.2394, simple_loss=0.3112, pruned_loss=0.08378, over 3835453.70 frames. ], batch size: 48, lr: 8.36e-03, grad_scale: 8.0 2023-04-01 17:08:30,175 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.006e+02 5.483e+02 6.754e+02 8.204e+02 1.644e+03, threshold=1.351e+03, percent-clipped=2.0 2023-04-01 17:08:32,806 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64666.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:08:43,804 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64674.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:09:16,013 INFO [train.py:903] (1/4) Epoch 10, batch 3250, loss[loss=0.2169, simple_loss=0.291, pruned_loss=0.07138, over 19488.00 frames. ], tot_loss[loss=0.2393, simple_loss=0.3112, pruned_loss=0.08367, over 3822338.17 frames. ], batch size: 49, lr: 8.36e-03, grad_scale: 8.0 2023-04-01 17:09:24,035 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=64707.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:09:26,663 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64709.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:09:42,846 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64721.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:10:20,326 INFO [train.py:903] (1/4) Epoch 10, batch 3300, loss[loss=0.2567, simple_loss=0.3296, pruned_loss=0.09188, over 19525.00 frames. ], tot_loss[loss=0.2403, simple_loss=0.3124, pruned_loss=0.08415, over 3826064.52 frames. ], batch size: 54, lr: 8.35e-03, grad_scale: 8.0 2023-04-01 17:10:26,789 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64756.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:10:28,774 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 2023-04-01 17:10:35,550 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64763.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:10:37,425 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.278e+02 5.384e+02 6.605e+02 8.281e+02 2.311e+03, threshold=1.321e+03, percent-clipped=9.0 2023-04-01 17:10:43,935 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64770.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:10:57,704 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64781.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:11:23,411 INFO [train.py:903] (1/4) Epoch 10, batch 3350, loss[loss=0.2772, simple_loss=0.3488, pruned_loss=0.1028, over 19541.00 frames. ], tot_loss[loss=0.2383, simple_loss=0.3107, pruned_loss=0.08296, over 3834408.12 frames. ], batch size: 56, lr: 8.35e-03, grad_scale: 4.0 2023-04-01 17:11:49,526 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64822.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:11:55,093 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=64827.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:12:04,573 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4933, 1.3685, 1.3055, 1.9169, 1.5225, 1.8397, 1.8383, 1.7315], device='cuda:1'), covar=tensor([0.0857, 0.1000, 0.1069, 0.0813, 0.0881, 0.0753, 0.0856, 0.0673], device='cuda:1'), in_proj_covar=tensor([0.0213, 0.0227, 0.0225, 0.0253, 0.0239, 0.0215, 0.0201, 0.0207], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-01 17:12:24,009 INFO [train.py:903] (1/4) Epoch 10, batch 3400, loss[loss=0.2284, simple_loss=0.2946, pruned_loss=0.08109, over 15255.00 frames. ], tot_loss[loss=0.2407, simple_loss=0.3127, pruned_loss=0.08437, over 3821918.64 frames. ], batch size: 33, lr: 8.35e-03, grad_scale: 4.0 2023-04-01 17:12:42,252 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.893e+02 5.724e+02 7.342e+02 8.665e+02 1.913e+03, threshold=1.468e+03, percent-clipped=6.0 2023-04-01 17:13:27,825 INFO [train.py:903] (1/4) Epoch 10, batch 3450, loss[loss=0.252, simple_loss=0.324, pruned_loss=0.09001, over 19745.00 frames. ], tot_loss[loss=0.2407, simple_loss=0.3128, pruned_loss=0.08432, over 3826973.16 frames. ], batch size: 51, lr: 8.34e-03, grad_scale: 4.0 2023-04-01 17:13:35,091 WARNING [train.py:1073] (1/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9_rvb from training. Duration: 25.2444375 2023-04-01 17:13:36,885 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.13 vs. limit=2.0 2023-04-01 17:14:20,594 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64942.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:14:30,654 INFO [train.py:903] (1/4) Epoch 10, batch 3500, loss[loss=0.2288, simple_loss=0.3032, pruned_loss=0.07725, over 19828.00 frames. ], tot_loss[loss=0.2413, simple_loss=0.3131, pruned_loss=0.08473, over 3831502.21 frames. ], batch size: 52, lr: 8.34e-03, grad_scale: 4.0 2023-04-01 17:14:31,773 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=64951.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:14:48,656 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.495e+02 5.616e+02 6.822e+02 8.996e+02 1.764e+03, threshold=1.364e+03, percent-clipped=1.0 2023-04-01 17:14:49,112 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64965.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:14:50,321 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2872, 1.4218, 1.7762, 1.4895, 2.4972, 2.1977, 2.7798, 1.1010], device='cuda:1'), covar=tensor([0.2108, 0.3635, 0.2244, 0.1604, 0.1433, 0.1765, 0.1350, 0.3470], device='cuda:1'), in_proj_covar=tensor([0.0485, 0.0569, 0.0588, 0.0430, 0.0589, 0.0486, 0.0648, 0.0486], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 17:15:05,373 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.45 vs. limit=5.0 2023-04-01 17:15:19,929 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64990.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:15:25,301 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64994.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:15:34,047 INFO [train.py:903] (1/4) Epoch 10, batch 3550, loss[loss=0.239, simple_loss=0.2958, pruned_loss=0.09116, over 19751.00 frames. ], tot_loss[loss=0.2394, simple_loss=0.3115, pruned_loss=0.08369, over 3836550.44 frames. ], batch size: 47, lr: 8.34e-03, grad_scale: 4.0 2023-04-01 17:15:44,360 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=65010.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:16:13,757 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.8661, 1.5294, 1.7145, 1.8341, 4.2111, 1.0171, 2.2730, 4.4070], device='cuda:1'), covar=tensor([0.0357, 0.2690, 0.2730, 0.1806, 0.0769, 0.2744, 0.1517, 0.0267], device='cuda:1'), in_proj_covar=tensor([0.0342, 0.0330, 0.0347, 0.0318, 0.0341, 0.0329, 0.0321, 0.0345], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 17:16:34,666 INFO [train.py:903] (1/4) Epoch 10, batch 3600, loss[loss=0.2327, simple_loss=0.3018, pruned_loss=0.08184, over 19472.00 frames. ], tot_loss[loss=0.2382, simple_loss=0.3105, pruned_loss=0.08298, over 3844232.70 frames. ], batch size: 49, lr: 8.33e-03, grad_scale: 8.0 2023-04-01 17:16:50,081 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9965, 1.4837, 1.6246, 2.5354, 1.7203, 2.3305, 2.5698, 2.3174], device='cuda:1'), covar=tensor([0.0858, 0.1071, 0.1051, 0.0927, 0.1059, 0.0768, 0.0891, 0.0674], device='cuda:1'), in_proj_covar=tensor([0.0212, 0.0225, 0.0224, 0.0251, 0.0240, 0.0215, 0.0200, 0.0205], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-01 17:16:52,000 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.218e+02 5.791e+02 6.813e+02 8.568e+02 1.743e+03, threshold=1.363e+03, percent-clipped=4.0 2023-04-01 17:16:52,169 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=65065.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:16:53,525 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.5680, 1.3395, 1.4778, 1.5080, 3.1187, 1.0393, 2.2471, 3.4708], device='cuda:1'), covar=tensor([0.0416, 0.2356, 0.2489, 0.1664, 0.0747, 0.2340, 0.1168, 0.0266], device='cuda:1'), in_proj_covar=tensor([0.0342, 0.0330, 0.0347, 0.0316, 0.0341, 0.0327, 0.0321, 0.0344], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 17:16:53,570 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65066.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:17:02,292 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=65073.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:17:08,111 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=65078.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:17:35,954 INFO [train.py:903] (1/4) Epoch 10, batch 3650, loss[loss=0.2818, simple_loss=0.3556, pruned_loss=0.1041, over 17484.00 frames. ], tot_loss[loss=0.2381, simple_loss=0.3104, pruned_loss=0.08287, over 3849568.94 frames. ], batch size: 101, lr: 8.33e-03, grad_scale: 8.0 2023-04-01 17:17:38,638 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=65103.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:17:42,760 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=65107.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:17:44,238 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5196, 1.6888, 2.0881, 1.7927, 3.0427, 2.7286, 3.5438, 1.4487], device='cuda:1'), covar=tensor([0.2136, 0.3708, 0.2273, 0.1706, 0.1586, 0.1751, 0.1686, 0.3607], device='cuda:1'), in_proj_covar=tensor([0.0486, 0.0570, 0.0590, 0.0432, 0.0592, 0.0486, 0.0650, 0.0488], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 17:17:51,899 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=65114.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:18:05,473 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65125.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:18:21,258 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.7597, 1.5332, 1.4047, 1.7717, 1.5009, 1.5561, 1.3385, 1.6885], device='cuda:1'), covar=tensor([0.0954, 0.1288, 0.1313, 0.0901, 0.1183, 0.0515, 0.1175, 0.0637], device='cuda:1'), in_proj_covar=tensor([0.0250, 0.0348, 0.0287, 0.0237, 0.0295, 0.0241, 0.0274, 0.0234], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 17:18:26,952 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 2023-04-01 17:18:36,262 INFO [train.py:903] (1/4) Epoch 10, batch 3700, loss[loss=0.2391, simple_loss=0.3208, pruned_loss=0.07869, over 19676.00 frames. ], tot_loss[loss=0.239, simple_loss=0.3109, pruned_loss=0.08358, over 3837951.00 frames. ], batch size: 60, lr: 8.33e-03, grad_scale: 8.0 2023-04-01 17:18:47,859 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=4.03 vs. limit=5.0 2023-04-01 17:18:53,963 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.395e+02 5.629e+02 7.088e+02 8.754e+02 1.818e+03, threshold=1.418e+03, percent-clipped=5.0 2023-04-01 17:19:05,500 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0 2023-04-01 17:19:12,161 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65180.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:19:21,096 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.7986, 4.3926, 2.5351, 3.9082, 1.0785, 4.1278, 4.1696, 4.2560], device='cuda:1'), covar=tensor([0.0542, 0.0937, 0.1903, 0.0708, 0.3681, 0.0676, 0.0724, 0.0854], device='cuda:1'), in_proj_covar=tensor([0.0419, 0.0359, 0.0420, 0.0312, 0.0373, 0.0352, 0.0348, 0.0376], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-01 17:19:33,111 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.8679, 4.2566, 4.5718, 4.5147, 1.5568, 4.2586, 3.6999, 4.2124], device='cuda:1'), covar=tensor([0.1229, 0.0713, 0.0533, 0.0519, 0.5108, 0.0543, 0.0603, 0.1070], device='cuda:1'), in_proj_covar=tensor([0.0636, 0.0575, 0.0765, 0.0647, 0.0704, 0.0520, 0.0473, 0.0705], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1') 2023-04-01 17:19:33,232 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=65198.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:19:37,048 INFO [train.py:903] (1/4) Epoch 10, batch 3750, loss[loss=0.2003, simple_loss=0.2675, pruned_loss=0.06653, over 19131.00 frames. ], tot_loss[loss=0.2398, simple_loss=0.3115, pruned_loss=0.08407, over 3826331.94 frames. ], batch size: 42, lr: 8.32e-03, grad_scale: 8.0 2023-04-01 17:20:02,486 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65222.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:20:03,643 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=65223.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:20:10,521 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65229.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:20:36,852 INFO [train.py:903] (1/4) Epoch 10, batch 3800, loss[loss=0.2101, simple_loss=0.2787, pruned_loss=0.07075, over 19757.00 frames. ], tot_loss[loss=0.2395, simple_loss=0.3111, pruned_loss=0.08392, over 3820712.34 frames. ], batch size: 46, lr: 8.32e-03, grad_scale: 8.0 2023-04-01 17:20:53,996 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.166e+02 5.620e+02 6.620e+02 8.005e+02 1.692e+03, threshold=1.324e+03, percent-clipped=4.0 2023-04-01 17:21:06,342 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9_rvb from training. Duration: 29.1166875 2023-04-01 17:21:22,121 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.4049, 3.1216, 2.2004, 2.2004, 2.0216, 2.4396, 0.8599, 2.0846], device='cuda:1'), covar=tensor([0.0427, 0.0389, 0.0479, 0.0788, 0.0799, 0.0872, 0.1002, 0.0783], device='cuda:1'), in_proj_covar=tensor([0.0325, 0.0325, 0.0322, 0.0343, 0.0416, 0.0341, 0.0303, 0.0319], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 17:21:37,758 INFO [train.py:903] (1/4) Epoch 10, batch 3850, loss[loss=0.2168, simple_loss=0.2875, pruned_loss=0.07302, over 19609.00 frames. ], tot_loss[loss=0.2403, simple_loss=0.312, pruned_loss=0.08429, over 3810205.85 frames. ], batch size: 52, lr: 8.32e-03, grad_scale: 8.0 2023-04-01 17:21:40,425 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=65303.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:22:03,660 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=65322.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:22:21,616 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=65338.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:22:33,979 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=65347.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:22:37,970 INFO [train.py:903] (1/4) Epoch 10, batch 3900, loss[loss=0.2486, simple_loss=0.3193, pruned_loss=0.08895, over 18700.00 frames. ], tot_loss[loss=0.2403, simple_loss=0.3121, pruned_loss=0.08427, over 3802095.46 frames. ], batch size: 74, lr: 8.31e-03, grad_scale: 8.0 2023-04-01 17:22:55,821 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.027e+02 5.653e+02 7.180e+02 9.093e+02 1.633e+03, threshold=1.436e+03, percent-clipped=2.0 2023-04-01 17:22:57,717 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.02 vs. limit=5.0 2023-04-01 17:23:14,882 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=65381.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:23:40,377 INFO [train.py:903] (1/4) Epoch 10, batch 3950, loss[loss=0.2497, simple_loss=0.3192, pruned_loss=0.09015, over 19382.00 frames. ], tot_loss[loss=0.2396, simple_loss=0.3118, pruned_loss=0.08372, over 3809338.37 frames. ], batch size: 70, lr: 8.31e-03, grad_scale: 8.0 2023-04-01 17:23:40,410 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 2023-04-01 17:23:46,512 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=65406.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:23:58,951 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=65417.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:24:23,308 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=65436.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:24:41,317 INFO [train.py:903] (1/4) Epoch 10, batch 4000, loss[loss=0.2322, simple_loss=0.3032, pruned_loss=0.08057, over 19480.00 frames. ], tot_loss[loss=0.2378, simple_loss=0.3105, pruned_loss=0.08256, over 3825728.71 frames. ], batch size: 49, lr: 8.31e-03, grad_scale: 8.0 2023-04-01 17:24:44,090 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65453.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:24:54,616 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=65461.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:24:58,711 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.016e+02 5.448e+02 7.265e+02 9.508e+02 1.942e+03, threshold=1.453e+03, percent-clipped=2.0 2023-04-01 17:25:14,565 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=65478.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:25:23,065 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1_rvb from training. Duration: 0.7545625 2023-04-01 17:25:23,367 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=65485.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:25:41,796 INFO [train.py:903] (1/4) Epoch 10, batch 4050, loss[loss=0.301, simple_loss=0.3504, pruned_loss=0.1258, over 13120.00 frames. ], tot_loss[loss=0.2388, simple_loss=0.3112, pruned_loss=0.08317, over 3809956.88 frames. ], batch size: 136, lr: 8.30e-03, grad_scale: 8.0 2023-04-01 17:25:45,307 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=65503.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:25:52,229 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4095, 1.8602, 1.4352, 1.4185, 1.8022, 1.2102, 1.3180, 1.4672], device='cuda:1'), covar=tensor([0.0668, 0.0549, 0.0711, 0.0519, 0.0373, 0.0877, 0.0486, 0.0393], device='cuda:1'), in_proj_covar=tensor([0.0290, 0.0298, 0.0326, 0.0243, 0.0233, 0.0320, 0.0291, 0.0255], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 17:25:53,331 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=65510.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:26:19,253 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65532.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:26:40,654 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=65550.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 17:26:41,421 INFO [train.py:903] (1/4) Epoch 10, batch 4100, loss[loss=0.3357, simple_loss=0.3795, pruned_loss=0.146, over 13736.00 frames. ], tot_loss[loss=0.2399, simple_loss=0.3119, pruned_loss=0.08402, over 3810123.63 frames. ], batch size: 136, lr: 8.30e-03, grad_scale: 8.0 2023-04-01 17:26:59,038 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.532e+02 5.925e+02 7.032e+02 8.463e+02 2.911e+03, threshold=1.406e+03, percent-clipped=6.0 2023-04-01 17:27:11,836 WARNING [train.py:1073] (1/4) Exclude cut with ID 497-129325-0061-9566_sp1.1_rvb from training. Duration: 0.97725 2023-04-01 17:27:42,541 INFO [train.py:903] (1/4) Epoch 10, batch 4150, loss[loss=0.2556, simple_loss=0.3295, pruned_loss=0.09084, over 17484.00 frames. ], tot_loss[loss=0.2396, simple_loss=0.3115, pruned_loss=0.08382, over 3805644.66 frames. ], batch size: 101, lr: 8.30e-03, grad_scale: 8.0 2023-04-01 17:28:03,940 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.8316, 1.3965, 1.4090, 1.7497, 1.5799, 1.5573, 1.4118, 1.6599], device='cuda:1'), covar=tensor([0.0880, 0.1385, 0.1422, 0.0811, 0.1075, 0.0530, 0.1197, 0.0686], device='cuda:1'), in_proj_covar=tensor([0.0252, 0.0352, 0.0292, 0.0240, 0.0298, 0.0245, 0.0277, 0.0238], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 17:28:10,369 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.6998, 2.6821, 1.9827, 1.9799, 1.8090, 2.2463, 1.0362, 1.9409], device='cuda:1'), covar=tensor([0.0491, 0.0430, 0.0471, 0.0669, 0.0776, 0.0779, 0.0872, 0.0690], device='cuda:1'), in_proj_covar=tensor([0.0332, 0.0333, 0.0332, 0.0352, 0.0427, 0.0350, 0.0309, 0.0325], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 17:28:39,191 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=65647.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:28:43,358 INFO [train.py:903] (1/4) Epoch 10, batch 4200, loss[loss=0.2468, simple_loss=0.3124, pruned_loss=0.09061, over 19687.00 frames. ], tot_loss[loss=0.2387, simple_loss=0.311, pruned_loss=0.08323, over 3817288.21 frames. ], batch size: 60, lr: 8.30e-03, grad_scale: 8.0 2023-04-01 17:28:43,401 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 2023-04-01 17:28:59,413 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.075e+02 5.851e+02 6.725e+02 9.221e+02 2.199e+03, threshold=1.345e+03, percent-clipped=6.0 2023-04-01 17:29:02,678 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9632, 1.6022, 1.6305, 2.0493, 1.9189, 1.8253, 1.6487, 1.9674], device='cuda:1'), covar=tensor([0.0890, 0.1587, 0.1307, 0.0886, 0.1122, 0.0468, 0.1122, 0.0597], device='cuda:1'), in_proj_covar=tensor([0.0249, 0.0349, 0.0289, 0.0237, 0.0294, 0.0243, 0.0274, 0.0234], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 17:29:42,453 INFO [train.py:903] (1/4) Epoch 10, batch 4250, loss[loss=0.2513, simple_loss=0.3269, pruned_loss=0.08787, over 18708.00 frames. ], tot_loss[loss=0.2395, simple_loss=0.3115, pruned_loss=0.08374, over 3817953.41 frames. ], batch size: 74, lr: 8.29e-03, grad_scale: 8.0 2023-04-01 17:29:54,079 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=65709.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:29:55,747 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9_rvb from training. Duration: 29.816625 2023-04-01 17:29:56,212 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9468, 1.9664, 2.0718, 2.7234, 1.8642, 2.5793, 2.4611, 2.0286], device='cuda:1'), covar=tensor([0.3269, 0.2840, 0.1366, 0.1531, 0.3023, 0.1294, 0.2997, 0.2354], device='cuda:1'), in_proj_covar=tensor([0.0769, 0.0787, 0.0638, 0.0880, 0.0767, 0.0690, 0.0780, 0.0697], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 17:30:06,597 WARNING [train.py:1073] (1/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 2023-04-01 17:30:22,811 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=65734.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:30:43,320 INFO [train.py:903] (1/4) Epoch 10, batch 4300, loss[loss=0.3157, simple_loss=0.3636, pruned_loss=0.1339, over 12916.00 frames. ], tot_loss[loss=0.2408, simple_loss=0.3125, pruned_loss=0.08453, over 3826020.62 frames. ], batch size: 136, lr: 8.29e-03, grad_scale: 8.0 2023-04-01 17:30:56,175 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65762.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:31:00,063 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.946e+02 5.358e+02 7.223e+02 8.854e+02 2.636e+03, threshold=1.445e+03, percent-clipped=3.0 2023-04-01 17:31:27,975 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=65788.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:31:35,310 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 2023-04-01 17:31:43,144 INFO [train.py:903] (1/4) Epoch 10, batch 4350, loss[loss=0.2388, simple_loss=0.317, pruned_loss=0.08033, over 19299.00 frames. ], tot_loss[loss=0.2427, simple_loss=0.3144, pruned_loss=0.08548, over 3816462.21 frames. ], batch size: 70, lr: 8.29e-03, grad_scale: 8.0 2023-04-01 17:31:58,612 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=65813.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:32:03,757 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 2023-04-01 17:32:44,686 INFO [train.py:903] (1/4) Epoch 10, batch 4400, loss[loss=0.2218, simple_loss=0.2943, pruned_loss=0.07461, over 19727.00 frames. ], tot_loss[loss=0.2418, simple_loss=0.3134, pruned_loss=0.08508, over 3822120.16 frames. ], batch size: 51, lr: 8.28e-03, grad_scale: 8.0 2023-04-01 17:32:50,628 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=65856.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 17:33:00,106 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.366e+02 5.748e+02 7.554e+02 9.760e+02 1.805e+03, threshold=1.511e+03, percent-clipped=4.0 2023-04-01 17:33:10,941 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_rvb from training. Duration: 25.285 2023-04-01 17:33:19,850 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 2023-04-01 17:33:21,078 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.4440, 0.9988, 1.2924, 1.1741, 1.9236, 0.8360, 2.0795, 2.1258], device='cuda:1'), covar=tensor([0.0893, 0.3338, 0.3012, 0.1856, 0.1307, 0.2395, 0.0965, 0.0759], device='cuda:1'), in_proj_covar=tensor([0.0347, 0.0332, 0.0350, 0.0318, 0.0342, 0.0328, 0.0323, 0.0345], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 17:33:28,210 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.98 vs. limit=2.0 2023-04-01 17:33:36,075 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=65894.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 17:33:44,056 INFO [train.py:903] (1/4) Epoch 10, batch 4450, loss[loss=0.2517, simple_loss=0.3285, pruned_loss=0.08743, over 19159.00 frames. ], tot_loss[loss=0.2415, simple_loss=0.3129, pruned_loss=0.08511, over 3826712.41 frames. ], batch size: 69, lr: 8.28e-03, grad_scale: 8.0 2023-04-01 17:34:12,379 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.84 vs. limit=2.0 2023-04-01 17:34:33,958 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.4494, 1.2233, 1.2680, 1.3946, 2.2163, 1.0456, 1.8138, 2.3772], device='cuda:1'), covar=tensor([0.0476, 0.2072, 0.2159, 0.1351, 0.0612, 0.1958, 0.1413, 0.0428], device='cuda:1'), in_proj_covar=tensor([0.0350, 0.0334, 0.0353, 0.0320, 0.0344, 0.0332, 0.0325, 0.0348], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 17:34:45,019 INFO [train.py:903] (1/4) Epoch 10, batch 4500, loss[loss=0.2493, simple_loss=0.3291, pruned_loss=0.08476, over 19317.00 frames. ], tot_loss[loss=0.2419, simple_loss=0.3136, pruned_loss=0.08512, over 3822479.47 frames. ], batch size: 66, lr: 8.28e-03, grad_scale: 8.0 2023-04-01 17:35:01,306 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.423e+02 5.376e+02 6.626e+02 8.325e+02 1.832e+03, threshold=1.325e+03, percent-clipped=3.0 2023-04-01 17:35:47,221 INFO [train.py:903] (1/4) Epoch 10, batch 4550, loss[loss=0.2708, simple_loss=0.341, pruned_loss=0.1002, over 19744.00 frames. ], tot_loss[loss=0.2422, simple_loss=0.3141, pruned_loss=0.08512, over 3834109.50 frames. ], batch size: 63, lr: 8.27e-03, grad_scale: 8.0 2023-04-01 17:35:56,020 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9_rvb from training. Duration: 28.72225 2023-04-01 17:35:56,361 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=66009.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 17:36:02,809 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3573, 1.4076, 1.9057, 1.5454, 3.4522, 2.8749, 3.8628, 1.4510], device='cuda:1'), covar=tensor([0.1961, 0.3508, 0.2179, 0.1479, 0.1148, 0.1485, 0.1100, 0.3174], device='cuda:1'), in_proj_covar=tensor([0.0479, 0.0558, 0.0582, 0.0428, 0.0584, 0.0483, 0.0641, 0.0481], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 17:36:06,997 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=66018.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:36:18,445 WARNING [train.py:1073] (1/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 2023-04-01 17:36:36,237 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=66043.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:36:45,869 INFO [train.py:903] (1/4) Epoch 10, batch 4600, loss[loss=0.2771, simple_loss=0.3503, pruned_loss=0.102, over 19783.00 frames. ], tot_loss[loss=0.2426, simple_loss=0.3141, pruned_loss=0.0855, over 3831656.65 frames. ], batch size: 56, lr: 8.27e-03, grad_scale: 8.0 2023-04-01 17:37:00,874 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.957e+02 5.689e+02 6.858e+02 9.462e+02 1.667e+03, threshold=1.372e+03, percent-clipped=8.0 2023-04-01 17:37:43,789 INFO [train.py:903] (1/4) Epoch 10, batch 4650, loss[loss=0.2536, simple_loss=0.3203, pruned_loss=0.09349, over 12986.00 frames. ], tot_loss[loss=0.2435, simple_loss=0.315, pruned_loss=0.08597, over 3840459.82 frames. ], batch size: 136, lr: 8.27e-03, grad_scale: 8.0 2023-04-01 17:38:00,793 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_rvb from training. Duration: 0.92 2023-04-01 17:38:10,776 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_rvb from training. Duration: 0.83 2023-04-01 17:38:44,404 INFO [train.py:903] (1/4) Epoch 10, batch 4700, loss[loss=0.2392, simple_loss=0.3125, pruned_loss=0.08292, over 19734.00 frames. ], tot_loss[loss=0.243, simple_loss=0.3144, pruned_loss=0.08578, over 3835037.73 frames. ], batch size: 63, lr: 8.26e-03, grad_scale: 8.0 2023-04-01 17:38:57,788 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=66162.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:39:00,970 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.007e+02 6.063e+02 7.892e+02 1.039e+03 2.104e+03, threshold=1.578e+03, percent-clipped=6.0 2023-04-01 17:39:05,606 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9_rvb from training. Duration: 0.92225 2023-04-01 17:39:17,520 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=66178.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 17:39:17,930 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.13 vs. limit=5.0 2023-04-01 17:39:24,684 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 2023-04-01 17:39:43,296 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=66200.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 17:39:44,101 INFO [train.py:903] (1/4) Epoch 10, batch 4750, loss[loss=0.2106, simple_loss=0.279, pruned_loss=0.07107, over 19754.00 frames. ], tot_loss[loss=0.2426, simple_loss=0.314, pruned_loss=0.08562, over 3843911.57 frames. ], batch size: 46, lr: 8.26e-03, grad_scale: 8.0 2023-04-01 17:40:33,538 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([0.8796, 1.1566, 1.4096, 0.6210, 2.0863, 2.4847, 2.1935, 2.5551], device='cuda:1'), covar=tensor([0.1496, 0.3290, 0.2966, 0.2202, 0.0470, 0.0230, 0.0314, 0.0279], device='cuda:1'), in_proj_covar=tensor([0.0252, 0.0293, 0.0316, 0.0248, 0.0211, 0.0151, 0.0203, 0.0194], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 17:40:43,461 INFO [train.py:903] (1/4) Epoch 10, batch 4800, loss[loss=0.2341, simple_loss=0.2991, pruned_loss=0.08461, over 19408.00 frames. ], tot_loss[loss=0.2419, simple_loss=0.3132, pruned_loss=0.08535, over 3851117.01 frames. ], batch size: 48, lr: 8.26e-03, grad_scale: 8.0 2023-04-01 17:41:01,216 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.573e+02 5.914e+02 7.080e+02 8.206e+02 1.527e+03, threshold=1.416e+03, percent-clipped=0.0 2023-04-01 17:41:01,686 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=66265.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 17:41:15,546 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=66276.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:41:31,327 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=66290.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 17:41:44,760 INFO [train.py:903] (1/4) Epoch 10, batch 4850, loss[loss=0.2354, simple_loss=0.3146, pruned_loss=0.07814, over 19716.00 frames. ], tot_loss[loss=0.2411, simple_loss=0.3125, pruned_loss=0.08486, over 3832079.31 frames. ], batch size: 63, lr: 8.25e-03, grad_scale: 8.0 2023-04-01 17:41:58,682 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.82 vs. limit=5.0 2023-04-01 17:42:01,807 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=66315.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 17:42:10,351 WARNING [train.py:1073] (1/4) Exclude cut with ID 774-127930-0014-48411_sp1.1_rvb from training. Duration: 0.95 2023-04-01 17:42:30,018 WARNING [train.py:1073] (1/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1_rvb from training. Duration: 0.9409375 2023-04-01 17:42:35,763 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 2023-04-01 17:42:35,785 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 2023-04-01 17:42:45,284 INFO [train.py:903] (1/4) Epoch 10, batch 4900, loss[loss=0.2668, simple_loss=0.3369, pruned_loss=0.09834, over 18473.00 frames. ], tot_loss[loss=0.2411, simple_loss=0.3127, pruned_loss=0.08478, over 3815613.04 frames. ], batch size: 84, lr: 8.25e-03, grad_scale: 8.0 2023-04-01 17:42:45,298 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1_rvb from training. Duration: 25.3818125 2023-04-01 17:43:01,822 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.050e+02 5.967e+02 7.269e+02 9.008e+02 2.184e+03, threshold=1.454e+03, percent-clipped=11.0 2023-04-01 17:43:03,045 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 2023-04-01 17:43:14,216 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.2158, 2.0264, 1.7423, 1.6465, 1.4519, 1.6973, 0.4030, 1.1029], device='cuda:1'), covar=tensor([0.0363, 0.0404, 0.0317, 0.0467, 0.0844, 0.0525, 0.0820, 0.0649], device='cuda:1'), in_proj_covar=tensor([0.0328, 0.0326, 0.0325, 0.0345, 0.0415, 0.0345, 0.0301, 0.0317], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 17:43:33,665 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.81 vs. limit=2.0 2023-04-01 17:43:45,939 INFO [train.py:903] (1/4) Epoch 10, batch 4950, loss[loss=0.1848, simple_loss=0.2682, pruned_loss=0.05066, over 19736.00 frames. ], tot_loss[loss=0.2411, simple_loss=0.3131, pruned_loss=0.08458, over 3824772.18 frames. ], batch size: 51, lr: 8.25e-03, grad_scale: 8.0 2023-04-01 17:44:02,376 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 2023-04-01 17:44:16,778 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([6.0583, 5.4163, 2.9862, 4.6939, 1.1639, 5.2312, 5.2888, 5.5065], device='cuda:1'), covar=tensor([0.0391, 0.0876, 0.1766, 0.0654, 0.3776, 0.0562, 0.0626, 0.0810], device='cuda:1'), in_proj_covar=tensor([0.0426, 0.0360, 0.0425, 0.0316, 0.0375, 0.0354, 0.0349, 0.0383], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-01 17:44:19,200 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.4593, 2.2367, 1.7934, 1.8062, 1.7788, 1.8746, 0.7130, 1.3084], device='cuda:1'), covar=tensor([0.0342, 0.0424, 0.0351, 0.0485, 0.0736, 0.0582, 0.0753, 0.0667], device='cuda:1'), in_proj_covar=tensor([0.0328, 0.0323, 0.0324, 0.0344, 0.0412, 0.0343, 0.0300, 0.0318], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 17:44:26,190 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 2023-04-01 17:44:44,953 INFO [train.py:903] (1/4) Epoch 10, batch 5000, loss[loss=0.2837, simple_loss=0.3604, pruned_loss=0.1035, over 19730.00 frames. ], tot_loss[loss=0.2418, simple_loss=0.3136, pruned_loss=0.08495, over 3833935.57 frames. ], batch size: 63, lr: 8.25e-03, grad_scale: 8.0 2023-04-01 17:44:56,395 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_rvb from training. Duration: 27.14 2023-04-01 17:45:01,937 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.596e+02 5.419e+02 6.464e+02 8.305e+02 1.628e+03, threshold=1.293e+03, percent-clipped=3.0 2023-04-01 17:45:02,443 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3661, 1.6388, 2.0681, 1.5866, 3.3200, 2.6520, 3.7017, 1.4915], device='cuda:1'), covar=tensor([0.2097, 0.3451, 0.2076, 0.1606, 0.1369, 0.1711, 0.1334, 0.3369], device='cuda:1'), in_proj_covar=tensor([0.0479, 0.0566, 0.0585, 0.0430, 0.0586, 0.0487, 0.0643, 0.0484], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 17:45:06,363 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 2023-04-01 17:45:45,181 INFO [train.py:903] (1/4) Epoch 10, batch 5050, loss[loss=0.2419, simple_loss=0.3142, pruned_loss=0.08482, over 19728.00 frames. ], tot_loss[loss=0.2413, simple_loss=0.3132, pruned_loss=0.08472, over 3843286.87 frames. ], batch size: 63, lr: 8.24e-03, grad_scale: 8.0 2023-04-01 17:45:51,099 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=66506.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:46:10,885 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=66522.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 17:46:16,357 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([0.8319, 1.3231, 1.0280, 0.9674, 1.1659, 0.9746, 0.7846, 1.2184], device='cuda:1'), covar=tensor([0.0561, 0.0693, 0.0976, 0.0496, 0.0445, 0.1034, 0.0570, 0.0366], device='cuda:1'), in_proj_covar=tensor([0.0287, 0.0298, 0.0325, 0.0243, 0.0234, 0.0319, 0.0289, 0.0254], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 17:46:19,332 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9_rvb from training. Duration: 26.62775 2023-04-01 17:46:31,364 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=66539.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 17:46:44,532 INFO [train.py:903] (1/4) Epoch 10, batch 5100, loss[loss=0.2479, simple_loss=0.3243, pruned_loss=0.08574, over 19530.00 frames. ], tot_loss[loss=0.2418, simple_loss=0.3133, pruned_loss=0.08519, over 3837342.92 frames. ], batch size: 56, lr: 8.24e-03, grad_scale: 8.0 2023-04-01 17:46:56,052 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 2023-04-01 17:46:57,213 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 2023-04-01 17:46:58,791 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.2781, 2.2814, 2.4090, 3.5047, 2.2387, 3.2968, 2.9885, 2.3699], device='cuda:1'), covar=tensor([0.3467, 0.3093, 0.1365, 0.1537, 0.3466, 0.1275, 0.2779, 0.2447], device='cuda:1'), in_proj_covar=tensor([0.0774, 0.0785, 0.0640, 0.0885, 0.0764, 0.0690, 0.0774, 0.0701], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 17:46:59,502 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_rvb from training. Duration: 25.035 2023-04-01 17:47:01,862 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.567e+02 5.864e+02 7.142e+02 9.030e+02 2.803e+03, threshold=1.428e+03, percent-clipped=6.0 2023-04-01 17:47:04,003 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_rvb from training. Duration: 27.92 2023-04-01 17:47:08,998 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=66571.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 17:47:39,575 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=66596.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 17:47:45,734 INFO [train.py:903] (1/4) Epoch 10, batch 5150, loss[loss=0.2448, simple_loss=0.3, pruned_loss=0.09479, over 19753.00 frames. ], tot_loss[loss=0.2413, simple_loss=0.3122, pruned_loss=0.08518, over 3832868.28 frames. ], batch size: 46, lr: 8.24e-03, grad_scale: 8.0 2023-04-01 17:47:56,914 WARNING [train.py:1073] (1/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 2023-04-01 17:48:09,055 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=66620.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:48:10,475 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=66621.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:48:29,316 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=66637.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 17:48:33,116 WARNING [train.py:1073] (1/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1_rvb from training. Duration: 0.9681875 2023-04-01 17:48:45,641 INFO [train.py:903] (1/4) Epoch 10, batch 5200, loss[loss=0.2895, simple_loss=0.341, pruned_loss=0.119, over 19302.00 frames. ], tot_loss[loss=0.2409, simple_loss=0.3123, pruned_loss=0.08478, over 3838986.37 frames. ], batch size: 66, lr: 8.23e-03, grad_scale: 8.0 2023-04-01 17:49:00,675 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 2023-04-01 17:49:02,887 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.107e+02 5.688e+02 7.032e+02 8.430e+02 1.656e+03, threshold=1.406e+03, percent-clipped=2.0 2023-04-01 17:49:44,614 WARNING [train.py:1073] (1/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 2023-04-01 17:49:46,893 INFO [train.py:903] (1/4) Epoch 10, batch 5250, loss[loss=0.2092, simple_loss=0.2904, pruned_loss=0.06401, over 19609.00 frames. ], tot_loss[loss=0.2405, simple_loss=0.3118, pruned_loss=0.08462, over 3819254.63 frames. ], batch size: 61, lr: 8.23e-03, grad_scale: 8.0 2023-04-01 17:49:59,399 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.44 vs. limit=2.0 2023-04-01 17:50:28,121 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=66735.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:50:44,987 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=66749.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 17:50:47,099 INFO [train.py:903] (1/4) Epoch 10, batch 5300, loss[loss=0.1993, simple_loss=0.2703, pruned_loss=0.06412, over 16905.00 frames. ], tot_loss[loss=0.2394, simple_loss=0.3108, pruned_loss=0.08398, over 3804633.61 frames. ], batch size: 37, lr: 8.23e-03, grad_scale: 8.0 2023-04-01 17:51:03,221 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 2023-04-01 17:51:04,359 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.476e+02 5.985e+02 7.858e+02 1.069e+03 1.957e+03, threshold=1.572e+03, percent-clipped=7.0 2023-04-01 17:51:47,694 INFO [train.py:903] (1/4) Epoch 10, batch 5350, loss[loss=0.2364, simple_loss=0.3136, pruned_loss=0.07955, over 19703.00 frames. ], tot_loss[loss=0.2405, simple_loss=0.312, pruned_loss=0.08451, over 3812056.44 frames. ], batch size: 59, lr: 8.22e-03, grad_scale: 8.0 2023-04-01 17:52:18,694 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.1138, 1.2459, 1.4475, 1.3232, 2.6668, 0.7859, 1.9563, 2.8985], device='cuda:1'), covar=tensor([0.0447, 0.2602, 0.2556, 0.1620, 0.0715, 0.2481, 0.1105, 0.0363], device='cuda:1'), in_proj_covar=tensor([0.0345, 0.0333, 0.0349, 0.0313, 0.0341, 0.0330, 0.0319, 0.0342], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 17:52:19,367 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9_rvb from training. Duration: 26.438875 2023-04-01 17:52:46,931 INFO [train.py:903] (1/4) Epoch 10, batch 5400, loss[loss=0.2307, simple_loss=0.3112, pruned_loss=0.07508, over 19680.00 frames. ], tot_loss[loss=0.2401, simple_loss=0.3119, pruned_loss=0.08416, over 3818755.96 frames. ], batch size: 58, lr: 8.22e-03, grad_scale: 8.0 2023-04-01 17:53:05,689 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.674e+02 5.633e+02 6.871e+02 8.512e+02 1.525e+03, threshold=1.374e+03, percent-clipped=0.0 2023-04-01 17:53:18,085 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=66877.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:53:26,395 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=66883.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 17:53:37,766 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=66893.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 17:53:47,836 INFO [train.py:903] (1/4) Epoch 10, batch 5450, loss[loss=0.2717, simple_loss=0.3335, pruned_loss=0.105, over 13648.00 frames. ], tot_loss[loss=0.2401, simple_loss=0.3115, pruned_loss=0.08437, over 3798665.45 frames. ], batch size: 136, lr: 8.22e-03, grad_scale: 8.0 2023-04-01 17:53:49,463 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=66902.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:54:08,222 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=66918.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 17:54:21,771 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 2023-04-01 17:54:47,620 INFO [train.py:903] (1/4) Epoch 10, batch 5500, loss[loss=0.2414, simple_loss=0.3105, pruned_loss=0.08613, over 19842.00 frames. ], tot_loss[loss=0.2399, simple_loss=0.3113, pruned_loss=0.0843, over 3820046.07 frames. ], batch size: 52, lr: 8.21e-03, grad_scale: 8.0 2023-04-01 17:55:06,145 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.718e+02 5.851e+02 7.428e+02 9.674e+02 2.066e+03, threshold=1.486e+03, percent-clipped=6.0 2023-04-01 17:55:10,698 WARNING [train.py:1073] (1/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9_rvb from training. Duration: 27.47775 2023-04-01 17:55:16,777 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=66975.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:55:36,714 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=66991.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:55:44,732 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=66998.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 17:55:48,005 INFO [train.py:903] (1/4) Epoch 10, batch 5550, loss[loss=0.1996, simple_loss=0.2704, pruned_loss=0.06439, over 19606.00 frames. ], tot_loss[loss=0.2397, simple_loss=0.3113, pruned_loss=0.08407, over 3832421.73 frames. ], batch size: 50, lr: 8.21e-03, grad_scale: 8.0 2023-04-01 17:55:54,300 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 2023-04-01 17:56:05,776 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=67016.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:56:11,216 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 2023-04-01 17:56:42,308 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 2023-04-01 17:56:47,795 INFO [train.py:903] (1/4) Epoch 10, batch 5600, loss[loss=0.2739, simple_loss=0.3378, pruned_loss=0.105, over 19531.00 frames. ], tot_loss[loss=0.2412, simple_loss=0.3126, pruned_loss=0.0849, over 3821109.11 frames. ], batch size: 56, lr: 8.21e-03, grad_scale: 8.0 2023-04-01 17:57:06,481 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.953e+02 5.918e+02 7.804e+02 1.015e+03 2.269e+03, threshold=1.561e+03, percent-clipped=7.0 2023-04-01 17:57:38,374 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=67093.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 17:57:47,807 INFO [train.py:903] (1/4) Epoch 10, batch 5650, loss[loss=0.2363, simple_loss=0.3034, pruned_loss=0.0846, over 19401.00 frames. ], tot_loss[loss=0.2428, simple_loss=0.3138, pruned_loss=0.08584, over 3829117.03 frames. ], batch size: 48, lr: 8.21e-03, grad_scale: 8.0 2023-04-01 17:57:48,106 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=67101.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 17:58:33,177 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9_rvb from training. Duration: 28.0944375 2023-04-01 17:58:47,422 INFO [train.py:903] (1/4) Epoch 10, batch 5700, loss[loss=0.2525, simple_loss=0.3245, pruned_loss=0.09021, over 17637.00 frames. ], tot_loss[loss=0.2429, simple_loss=0.3141, pruned_loss=0.08584, over 3829339.12 frames. ], batch size: 101, lr: 8.20e-03, grad_scale: 8.0 2023-04-01 17:59:05,253 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.236e+02 6.145e+02 7.592e+02 1.064e+03 2.520e+03, threshold=1.518e+03, percent-clipped=7.0 2023-04-01 17:59:32,594 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.5181, 2.2316, 2.2208, 2.4840, 2.3838, 1.8986, 2.0020, 2.4530], device='cuda:1'), covar=tensor([0.0940, 0.1778, 0.1444, 0.1050, 0.1442, 0.0726, 0.1349, 0.0789], device='cuda:1'), in_proj_covar=tensor([0.0252, 0.0350, 0.0288, 0.0237, 0.0292, 0.0241, 0.0275, 0.0238], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 17:59:47,163 INFO [train.py:903] (1/4) Epoch 10, batch 5750, loss[loss=0.251, simple_loss=0.3315, pruned_loss=0.08531, over 19629.00 frames. ], tot_loss[loss=0.2418, simple_loss=0.3131, pruned_loss=0.08522, over 3838241.66 frames. ], batch size: 57, lr: 8.20e-03, grad_scale: 8.0 2023-04-01 17:59:48,371 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9_rvb from training. Duration: 33.038875 2023-04-01 17:59:57,174 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=67208.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 17:59:57,967 WARNING [train.py:1073] (1/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 2023-04-01 18:00:02,610 WARNING [train.py:1073] (1/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 2023-04-01 18:00:47,882 INFO [train.py:903] (1/4) Epoch 10, batch 5800, loss[loss=0.2293, simple_loss=0.315, pruned_loss=0.07174, over 19618.00 frames. ], tot_loss[loss=0.2421, simple_loss=0.3139, pruned_loss=0.08514, over 3843680.42 frames. ], batch size: 57, lr: 8.20e-03, grad_scale: 8.0 2023-04-01 18:00:51,514 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=67254.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 18:01:06,014 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.662e+02 5.628e+02 6.923e+02 8.923e+02 2.275e+03, threshold=1.385e+03, percent-clipped=6.0 2023-04-01 18:01:20,796 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=67279.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 18:01:47,312 INFO [train.py:903] (1/4) Epoch 10, batch 5850, loss[loss=0.2528, simple_loss=0.3287, pruned_loss=0.08849, over 19480.00 frames. ], tot_loss[loss=0.2414, simple_loss=0.3133, pruned_loss=0.0848, over 3847134.89 frames. ], batch size: 64, lr: 8.19e-03, grad_scale: 8.0 2023-04-01 18:02:07,169 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.9341, 4.4925, 2.7656, 4.0216, 1.0480, 4.1912, 4.2979, 4.3471], device='cuda:1'), covar=tensor([0.0476, 0.0837, 0.1691, 0.0599, 0.3628, 0.0652, 0.0640, 0.0754], device='cuda:1'), in_proj_covar=tensor([0.0419, 0.0356, 0.0420, 0.0308, 0.0372, 0.0352, 0.0345, 0.0376], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-01 18:02:09,356 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=67319.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 18:02:48,429 INFO [train.py:903] (1/4) Epoch 10, batch 5900, loss[loss=0.1875, simple_loss=0.2678, pruned_loss=0.05357, over 19793.00 frames. ], tot_loss[loss=0.2421, simple_loss=0.3141, pruned_loss=0.0851, over 3836699.04 frames. ], batch size: 49, lr: 8.19e-03, grad_scale: 8.0 2023-04-01 18:02:52,959 WARNING [train.py:1073] (1/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 2023-04-01 18:03:05,150 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.548e+02 5.688e+02 6.588e+02 8.588e+02 1.646e+03, threshold=1.318e+03, percent-clipped=2.0 2023-04-01 18:03:11,561 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9_rvb from training. Duration: 27.511125 2023-04-01 18:03:47,173 INFO [train.py:903] (1/4) Epoch 10, batch 5950, loss[loss=0.2372, simple_loss=0.3143, pruned_loss=0.0801, over 17346.00 frames. ], tot_loss[loss=0.2403, simple_loss=0.3129, pruned_loss=0.08387, over 3824400.67 frames. ], batch size: 101, lr: 8.19e-03, grad_scale: 8.0 2023-04-01 18:04:26,265 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=67434.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 18:04:38,305 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=67445.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 18:04:45,816 INFO [train.py:903] (1/4) Epoch 10, batch 6000, loss[loss=0.2467, simple_loss=0.3277, pruned_loss=0.08283, over 19725.00 frames. ], tot_loss[loss=0.242, simple_loss=0.3142, pruned_loss=0.08492, over 3828689.38 frames. ], batch size: 63, lr: 8.18e-03, grad_scale: 8.0 2023-04-01 18:04:45,816 INFO [train.py:928] (1/4) Computing validation loss 2023-04-01 18:04:58,242 INFO [train.py:937] (1/4) Epoch 10, validation: loss=0.1798, simple_loss=0.2805, pruned_loss=0.03952, over 944034.00 frames. 2023-04-01 18:04:58,243 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18649MB 2023-04-01 18:05:07,593 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4749, 2.4151, 1.7765, 1.4575, 2.3011, 1.2959, 1.2044, 1.7720], device='cuda:1'), covar=tensor([0.0885, 0.0609, 0.0913, 0.0741, 0.0400, 0.1096, 0.0778, 0.0446], device='cuda:1'), in_proj_covar=tensor([0.0291, 0.0300, 0.0328, 0.0248, 0.0236, 0.0319, 0.0291, 0.0259], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 18:05:15,092 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=67464.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 18:05:17,958 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.621e+02 5.414e+02 6.867e+02 8.657e+02 1.897e+03, threshold=1.373e+03, percent-clipped=4.0 2023-04-01 18:05:44,666 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=67489.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 18:05:59,570 INFO [train.py:903] (1/4) Epoch 10, batch 6050, loss[loss=0.2279, simple_loss=0.3103, pruned_loss=0.07275, over 17118.00 frames. ], tot_loss[loss=0.244, simple_loss=0.3157, pruned_loss=0.08608, over 3815862.28 frames. ], batch size: 101, lr: 8.18e-03, grad_scale: 8.0 2023-04-01 18:06:35,124 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=67531.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 18:06:59,147 INFO [train.py:903] (1/4) Epoch 10, batch 6100, loss[loss=0.2397, simple_loss=0.3116, pruned_loss=0.08392, over 19667.00 frames. ], tot_loss[loss=0.2439, simple_loss=0.3157, pruned_loss=0.08608, over 3818966.25 frames. ], batch size: 53, lr: 8.18e-03, grad_scale: 4.0 2023-04-01 18:07:10,580 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=67560.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 18:07:19,955 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.598e+02 5.925e+02 6.739e+02 8.410e+02 2.337e+03, threshold=1.348e+03, percent-clipped=5.0 2023-04-01 18:07:23,153 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.88 vs. limit=5.0 2023-04-01 18:07:59,354 INFO [train.py:903] (1/4) Epoch 10, batch 6150, loss[loss=0.283, simple_loss=0.339, pruned_loss=0.1135, over 19386.00 frames. ], tot_loss[loss=0.2426, simple_loss=0.3145, pruned_loss=0.0854, over 3819421.12 frames. ], batch size: 70, lr: 8.18e-03, grad_scale: 4.0 2023-04-01 18:08:10,099 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.31 vs. limit=5.0 2023-04-01 18:08:28,205 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9_rvb from training. Duration: 31.02225 2023-04-01 18:08:59,751 INFO [train.py:903] (1/4) Epoch 10, batch 6200, loss[loss=0.2825, simple_loss=0.3489, pruned_loss=0.1081, over 18188.00 frames. ], tot_loss[loss=0.2414, simple_loss=0.3135, pruned_loss=0.08462, over 3813740.74 frames. ], batch size: 83, lr: 8.17e-03, grad_scale: 4.0 2023-04-01 18:09:20,095 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.127e+02 5.059e+02 6.776e+02 9.553e+02 2.024e+03, threshold=1.355e+03, percent-clipped=7.0 2023-04-01 18:09:46,813 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=67690.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 18:09:56,514 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=67698.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 18:09:59,567 INFO [train.py:903] (1/4) Epoch 10, batch 6250, loss[loss=0.2715, simple_loss=0.3404, pruned_loss=0.1013, over 19099.00 frames. ], tot_loss[loss=0.242, simple_loss=0.314, pruned_loss=0.08501, over 3811453.47 frames. ], batch size: 69, lr: 8.17e-03, grad_scale: 4.0 2023-04-01 18:10:09,915 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=67710.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 18:10:15,805 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=67715.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 18:10:30,840 WARNING [train.py:1073] (1/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9_rvb from training. Duration: 25.988875 2023-04-01 18:10:59,324 INFO [train.py:903] (1/4) Epoch 10, batch 6300, loss[loss=0.2441, simple_loss=0.3132, pruned_loss=0.08754, over 19693.00 frames. ], tot_loss[loss=0.2424, simple_loss=0.3139, pruned_loss=0.08547, over 3817054.88 frames. ], batch size: 60, lr: 8.17e-03, grad_scale: 4.0 2023-04-01 18:11:19,321 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.616e+02 5.371e+02 6.798e+02 8.698e+02 1.915e+03, threshold=1.360e+03, percent-clipped=7.0 2023-04-01 18:11:49,380 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.13 vs. limit=2.0 2023-04-01 18:11:58,677 INFO [train.py:903] (1/4) Epoch 10, batch 6350, loss[loss=0.2471, simple_loss=0.3185, pruned_loss=0.08792, over 19372.00 frames. ], tot_loss[loss=0.2438, simple_loss=0.3152, pruned_loss=0.08623, over 3812886.90 frames. ], batch size: 47, lr: 8.16e-03, grad_scale: 4.0 2023-04-01 18:12:16,848 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=67816.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 18:12:47,280 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=67841.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 18:12:58,574 INFO [train.py:903] (1/4) Epoch 10, batch 6400, loss[loss=0.2728, simple_loss=0.3405, pruned_loss=0.1025, over 19505.00 frames. ], tot_loss[loss=0.2422, simple_loss=0.3138, pruned_loss=0.0853, over 3810040.39 frames. ], batch size: 64, lr: 8.16e-03, grad_scale: 8.0 2023-04-01 18:13:18,757 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.354e+02 5.283e+02 6.601e+02 9.298e+02 1.582e+03, threshold=1.320e+03, percent-clipped=4.0 2023-04-01 18:13:26,990 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=67875.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 18:13:59,327 INFO [train.py:903] (1/4) Epoch 10, batch 6450, loss[loss=0.2707, simple_loss=0.3305, pruned_loss=0.1054, over 19486.00 frames. ], tot_loss[loss=0.2416, simple_loss=0.3133, pruned_loss=0.08497, over 3808181.35 frames. ], batch size: 64, lr: 8.16e-03, grad_scale: 8.0 2023-04-01 18:14:09,732 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4013, 2.3271, 1.7424, 1.4692, 2.1705, 1.3154, 1.2898, 1.9291], device='cuda:1'), covar=tensor([0.0951, 0.0663, 0.0950, 0.0737, 0.0411, 0.1055, 0.0666, 0.0400], device='cuda:1'), in_proj_covar=tensor([0.0286, 0.0298, 0.0328, 0.0246, 0.0234, 0.0317, 0.0286, 0.0257], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 18:14:42,669 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9_rvb from training. Duration: 28.638875 2023-04-01 18:14:59,316 INFO [train.py:903] (1/4) Epoch 10, batch 6500, loss[loss=0.2203, simple_loss=0.2862, pruned_loss=0.07718, over 19754.00 frames. ], tot_loss[loss=0.241, simple_loss=0.313, pruned_loss=0.08451, over 3820165.91 frames. ], batch size: 46, lr: 8.15e-03, grad_scale: 8.0 2023-04-01 18:15:05,075 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1_rvb from training. Duration: 0.836375 2023-04-01 18:15:19,446 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.752e+02 5.758e+02 7.165e+02 9.309e+02 2.290e+03, threshold=1.433e+03, percent-clipped=7.0 2023-04-01 18:15:27,696 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4251, 1.4729, 2.0396, 1.7171, 3.4175, 2.8747, 3.6107, 1.5789], device='cuda:1'), covar=tensor([0.2117, 0.3826, 0.2235, 0.1597, 0.1239, 0.1614, 0.1381, 0.3296], device='cuda:1'), in_proj_covar=tensor([0.0482, 0.0568, 0.0587, 0.0432, 0.0589, 0.0488, 0.0646, 0.0487], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 18:15:27,740 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.2236, 2.3197, 2.5068, 3.4118, 2.2392, 3.4185, 2.9682, 2.2776], device='cuda:1'), covar=tensor([0.3569, 0.2986, 0.1290, 0.1664, 0.3504, 0.1242, 0.3062, 0.2512], device='cuda:1'), in_proj_covar=tensor([0.0770, 0.0786, 0.0638, 0.0881, 0.0765, 0.0689, 0.0776, 0.0692], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 18:15:47,549 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=67990.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 18:15:57,573 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=67999.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 18:16:01,761 INFO [train.py:903] (1/4) Epoch 10, batch 6550, loss[loss=0.2481, simple_loss=0.322, pruned_loss=0.0871, over 19361.00 frames. ], tot_loss[loss=0.24, simple_loss=0.3122, pruned_loss=0.08391, over 3816765.79 frames. ], batch size: 66, lr: 8.15e-03, grad_scale: 8.0 2023-04-01 18:16:51,306 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=68042.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 18:17:02,072 INFO [train.py:903] (1/4) Epoch 10, batch 6600, loss[loss=0.2059, simple_loss=0.2795, pruned_loss=0.06622, over 19736.00 frames. ], tot_loss[loss=0.2397, simple_loss=0.3121, pruned_loss=0.08358, over 3820245.89 frames. ], batch size: 51, lr: 8.15e-03, grad_scale: 8.0 2023-04-01 18:17:05,663 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=68054.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 18:17:23,331 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.670e+02 5.913e+02 7.324e+02 8.709e+02 1.479e+03, threshold=1.465e+03, percent-clipped=1.0 2023-04-01 18:17:49,969 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.2032, 2.1218, 1.8179, 1.6714, 1.7515, 1.7691, 0.3964, 1.0104], device='cuda:1'), covar=tensor([0.0399, 0.0384, 0.0283, 0.0463, 0.0747, 0.0503, 0.0824, 0.0734], device='cuda:1'), in_proj_covar=tensor([0.0326, 0.0323, 0.0324, 0.0343, 0.0416, 0.0339, 0.0302, 0.0316], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 18:18:02,155 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2564, 1.3090, 1.2044, 1.0972, 1.0777, 1.0672, 0.0165, 0.3068], device='cuda:1'), covar=tensor([0.0389, 0.0396, 0.0256, 0.0301, 0.0792, 0.0377, 0.0754, 0.0692], device='cuda:1'), in_proj_covar=tensor([0.0327, 0.0324, 0.0325, 0.0344, 0.0417, 0.0341, 0.0303, 0.0317], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 18:18:02,821 INFO [train.py:903] (1/4) Epoch 10, batch 6650, loss[loss=0.2722, simple_loss=0.3264, pruned_loss=0.109, over 19759.00 frames. ], tot_loss[loss=0.2403, simple_loss=0.3125, pruned_loss=0.08408, over 3820260.38 frames. ], batch size: 54, lr: 8.15e-03, grad_scale: 8.0 2023-04-01 18:18:07,714 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=68105.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 18:18:24,118 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.83 vs. limit=2.0 2023-04-01 18:19:02,659 INFO [train.py:903] (1/4) Epoch 10, batch 6700, loss[loss=0.1768, simple_loss=0.2563, pruned_loss=0.04862, over 19352.00 frames. ], tot_loss[loss=0.241, simple_loss=0.3131, pruned_loss=0.08444, over 3821242.21 frames. ], batch size: 47, lr: 8.14e-03, grad_scale: 8.0 2023-04-01 18:19:09,865 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=68157.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 18:19:21,305 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.58 vs. limit=5.0 2023-04-01 18:19:22,649 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.762e+02 5.919e+02 7.224e+02 9.287e+02 2.274e+03, threshold=1.445e+03, percent-clipped=4.0 2023-04-01 18:19:24,182 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=68169.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 18:19:28,459 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0055, 1.9069, 1.6770, 1.5020, 1.4500, 1.5295, 0.1784, 0.8201], device='cuda:1'), covar=tensor([0.0400, 0.0439, 0.0290, 0.0443, 0.0835, 0.0543, 0.0882, 0.0761], device='cuda:1'), in_proj_covar=tensor([0.0326, 0.0323, 0.0325, 0.0346, 0.0418, 0.0341, 0.0304, 0.0317], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 18:19:35,687 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0 2023-04-01 18:20:00,033 INFO [train.py:903] (1/4) Epoch 10, batch 6750, loss[loss=0.2394, simple_loss=0.3185, pruned_loss=0.08016, over 19549.00 frames. ], tot_loss[loss=0.241, simple_loss=0.3129, pruned_loss=0.08451, over 3820738.98 frames. ], batch size: 56, lr: 8.14e-03, grad_scale: 8.0 2023-04-01 18:20:16,563 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.8917, 1.9982, 2.1162, 2.8294, 1.9408, 2.6969, 2.4380, 1.9974], device='cuda:1'), covar=tensor([0.3689, 0.3029, 0.1373, 0.1837, 0.3349, 0.1404, 0.3231, 0.2573], device='cuda:1'), in_proj_covar=tensor([0.0775, 0.0788, 0.0641, 0.0881, 0.0767, 0.0692, 0.0775, 0.0697], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 18:20:51,822 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=68246.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 18:20:56,747 INFO [train.py:903] (1/4) Epoch 10, batch 6800, loss[loss=0.2077, simple_loss=0.2946, pruned_loss=0.06042, over 19611.00 frames. ], tot_loss[loss=0.2405, simple_loss=0.3125, pruned_loss=0.08424, over 3808752.57 frames. ], batch size: 57, lr: 8.14e-03, grad_scale: 8.0 2023-04-01 18:21:15,043 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.210e+02 6.150e+02 7.610e+02 9.082e+02 1.904e+03, threshold=1.522e+03, percent-clipped=4.0 2023-04-01 18:21:18,279 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=68271.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 18:21:40,721 WARNING [train.py:1073] (1/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 2023-04-01 18:21:41,163 WARNING [train.py:1073] (1/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 2023-04-01 18:21:44,654 INFO [train.py:903] (1/4) Epoch 11, batch 0, loss[loss=0.2256, simple_loss=0.3086, pruned_loss=0.07134, over 17218.00 frames. ], tot_loss[loss=0.2256, simple_loss=0.3086, pruned_loss=0.07134, over 17218.00 frames. ], batch size: 101, lr: 7.77e-03, grad_scale: 8.0 2023-04-01 18:21:44,654 INFO [train.py:928] (1/4) Computing validation loss 2023-04-01 18:21:56,763 INFO [train.py:937] (1/4) Epoch 11, validation: loss=0.181, simple_loss=0.2818, pruned_loss=0.04012, over 944034.00 frames. 2023-04-01 18:21:56,763 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18649MB 2023-04-01 18:22:09,361 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 2023-04-01 18:22:57,935 INFO [train.py:903] (1/4) Epoch 11, batch 50, loss[loss=0.228, simple_loss=0.294, pruned_loss=0.08102, over 18636.00 frames. ], tot_loss[loss=0.2399, simple_loss=0.3125, pruned_loss=0.08369, over 863862.14 frames. ], batch size: 41, lr: 7.76e-03, grad_scale: 8.0 2023-04-01 18:22:58,158 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([6.0976, 5.4209, 2.9553, 4.7892, 1.0884, 5.4289, 5.4209, 5.6178], device='cuda:1'), covar=tensor([0.0394, 0.0895, 0.1795, 0.0615, 0.3918, 0.0512, 0.0533, 0.0816], device='cuda:1'), in_proj_covar=tensor([0.0421, 0.0358, 0.0425, 0.0313, 0.0375, 0.0355, 0.0348, 0.0379], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-01 18:23:15,260 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=68343.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 18:23:23,486 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.9841, 1.8590, 2.0246, 1.6369, 4.4358, 0.9588, 2.3646, 4.7270], device='cuda:1'), covar=tensor([0.0361, 0.2467, 0.2404, 0.1840, 0.0701, 0.2710, 0.1378, 0.0217], device='cuda:1'), in_proj_covar=tensor([0.0348, 0.0333, 0.0349, 0.0313, 0.0344, 0.0328, 0.0324, 0.0345], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 18:23:35,570 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1_rvb from training. Duration: 27.0318125 2023-04-01 18:23:46,894 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.915e+02 5.747e+02 7.027e+02 9.557e+02 1.564e+03, threshold=1.405e+03, percent-clipped=1.0 2023-04-01 18:24:00,203 INFO [train.py:903] (1/4) Epoch 11, batch 100, loss[loss=0.2532, simple_loss=0.325, pruned_loss=0.09073, over 17367.00 frames. ], tot_loss[loss=0.2348, simple_loss=0.3076, pruned_loss=0.08097, over 1530020.68 frames. ], batch size: 101, lr: 7.76e-03, grad_scale: 8.0 2023-04-01 18:24:13,686 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 2023-04-01 18:24:42,554 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=68413.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 18:24:47,941 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=68417.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 18:24:57,058 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=68425.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 18:25:01,040 INFO [train.py:903] (1/4) Epoch 11, batch 150, loss[loss=0.2426, simple_loss=0.2965, pruned_loss=0.09432, over 19743.00 frames. ], tot_loss[loss=0.2414, simple_loss=0.3125, pruned_loss=0.08515, over 2032099.87 frames. ], batch size: 45, lr: 7.76e-03, grad_scale: 8.0 2023-04-01 18:25:11,925 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=68438.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 18:25:25,314 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=68449.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 18:25:26,562 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=68450.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 18:25:36,615 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=68458.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 18:25:47,784 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.848e+02 5.281e+02 6.719e+02 8.986e+02 1.619e+03, threshold=1.344e+03, percent-clipped=5.0 2023-04-01 18:26:00,680 INFO [train.py:903] (1/4) Epoch 11, batch 200, loss[loss=0.203, simple_loss=0.2777, pruned_loss=0.06413, over 19388.00 frames. ], tot_loss[loss=0.24, simple_loss=0.3112, pruned_loss=0.08435, over 2435673.36 frames. ], batch size: 48, lr: 7.76e-03, grad_scale: 8.0 2023-04-01 18:26:02,038 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 2023-04-01 18:27:03,586 INFO [train.py:903] (1/4) Epoch 11, batch 250, loss[loss=0.3064, simple_loss=0.3591, pruned_loss=0.1269, over 13167.00 frames. ], tot_loss[loss=0.2379, simple_loss=0.3096, pruned_loss=0.08309, over 2749832.44 frames. ], batch size: 136, lr: 7.75e-03, grad_scale: 8.0 2023-04-01 18:27:46,862 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=68564.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 18:27:50,296 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2796, 1.1535, 1.6751, 1.2581, 2.7896, 3.7048, 3.4606, 3.9365], device='cuda:1'), covar=tensor([0.1523, 0.3566, 0.2970, 0.2086, 0.0442, 0.0156, 0.0189, 0.0155], device='cuda:1'), in_proj_covar=tensor([0.0256, 0.0295, 0.0321, 0.0251, 0.0212, 0.0154, 0.0204, 0.0193], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 18:27:51,040 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.001e+02 5.512e+02 6.613e+02 8.406e+02 1.798e+03, threshold=1.323e+03, percent-clipped=1.0 2023-04-01 18:28:07,068 INFO [train.py:903] (1/4) Epoch 11, batch 300, loss[loss=0.2922, simple_loss=0.351, pruned_loss=0.1167, over 13471.00 frames. ], tot_loss[loss=0.2377, simple_loss=0.309, pruned_loss=0.08318, over 2990931.51 frames. ], batch size: 137, lr: 7.75e-03, grad_scale: 8.0 2023-04-01 18:28:27,740 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=68596.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 18:28:30,603 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 2023-04-01 18:29:09,810 INFO [train.py:903] (1/4) Epoch 11, batch 350, loss[loss=0.243, simple_loss=0.3217, pruned_loss=0.08214, over 19729.00 frames. ], tot_loss[loss=0.2374, simple_loss=0.3091, pruned_loss=0.08282, over 3175133.30 frames. ], batch size: 63, lr: 7.75e-03, grad_scale: 8.0 2023-04-01 18:29:16,666 WARNING [train.py:1073] (1/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1_rvb from training. Duration: 0.9681875 2023-04-01 18:29:40,809 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.4192, 2.1910, 1.8632, 1.7469, 1.4280, 1.5826, 0.5340, 1.2190], device='cuda:1'), covar=tensor([0.0356, 0.0409, 0.0359, 0.0565, 0.0862, 0.0685, 0.0841, 0.0744], device='cuda:1'), in_proj_covar=tensor([0.0329, 0.0326, 0.0327, 0.0349, 0.0420, 0.0347, 0.0305, 0.0321], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 18:29:57,402 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.055e+02 5.409e+02 6.235e+02 7.842e+02 1.948e+03, threshold=1.247e+03, percent-clipped=7.0 2023-04-01 18:30:09,871 INFO [train.py:903] (1/4) Epoch 11, batch 400, loss[loss=0.2051, simple_loss=0.2749, pruned_loss=0.06766, over 19333.00 frames. ], tot_loss[loss=0.2388, simple_loss=0.3104, pruned_loss=0.08355, over 3322446.99 frames. ], batch size: 44, lr: 7.74e-03, grad_scale: 8.0 2023-04-01 18:30:10,265 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2216, 1.9510, 1.5318, 1.3458, 1.7626, 1.1075, 1.3077, 1.6011], device='cuda:1'), covar=tensor([0.0804, 0.0687, 0.0973, 0.0632, 0.0524, 0.1105, 0.0554, 0.0400], device='cuda:1'), in_proj_covar=tensor([0.0282, 0.0294, 0.0326, 0.0244, 0.0233, 0.0317, 0.0286, 0.0255], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 18:30:40,453 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=68703.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 18:30:54,263 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=68714.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 18:31:11,162 INFO [train.py:903] (1/4) Epoch 11, batch 450, loss[loss=0.2227, simple_loss=0.2903, pruned_loss=0.07759, over 19756.00 frames. ], tot_loss[loss=0.2378, simple_loss=0.31, pruned_loss=0.08283, over 3449944.95 frames. ], batch size: 46, lr: 7.74e-03, grad_scale: 8.0 2023-04-01 18:31:25,476 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=68739.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 18:31:49,551 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 2023-04-01 18:31:50,648 WARNING [train.py:1073] (1/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 2023-04-01 18:31:51,656 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=68761.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 18:31:59,522 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.166e+02 5.180e+02 6.491e+02 8.751e+02 1.660e+03, threshold=1.298e+03, percent-clipped=7.0 2023-04-01 18:32:13,197 INFO [train.py:903] (1/4) Epoch 11, batch 500, loss[loss=0.1728, simple_loss=0.2499, pruned_loss=0.04783, over 19749.00 frames. ], tot_loss[loss=0.2376, simple_loss=0.3099, pruned_loss=0.08268, over 3537187.26 frames. ], batch size: 45, lr: 7.74e-03, grad_scale: 8.0 2023-04-01 18:32:31,375 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3709, 1.5088, 1.8329, 1.6367, 3.0918, 2.4556, 3.2909, 1.5118], device='cuda:1'), covar=tensor([0.2176, 0.3667, 0.2335, 0.1705, 0.1399, 0.1800, 0.1617, 0.3509], device='cuda:1'), in_proj_covar=tensor([0.0479, 0.0565, 0.0588, 0.0432, 0.0585, 0.0485, 0.0644, 0.0485], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 18:33:02,552 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.75 vs. limit=5.0 2023-04-01 18:33:05,583 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=68820.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 18:33:11,256 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.5612, 1.1787, 1.4045, 1.1455, 2.2080, 0.9033, 2.0207, 2.3476], device='cuda:1'), covar=tensor([0.0677, 0.2567, 0.2428, 0.1615, 0.0910, 0.2127, 0.0890, 0.0484], device='cuda:1'), in_proj_covar=tensor([0.0354, 0.0336, 0.0351, 0.0319, 0.0347, 0.0335, 0.0331, 0.0348], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 18:33:17,981 INFO [train.py:903] (1/4) Epoch 11, batch 550, loss[loss=0.1803, simple_loss=0.2551, pruned_loss=0.05277, over 19721.00 frames. ], tot_loss[loss=0.2358, simple_loss=0.3084, pruned_loss=0.08162, over 3616123.41 frames. ], batch size: 45, lr: 7.74e-03, grad_scale: 8.0 2023-04-01 18:33:36,721 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=68845.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 18:34:06,690 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.095e+02 5.206e+02 6.523e+02 8.559e+02 1.532e+03, threshold=1.305e+03, percent-clipped=5.0 2023-04-01 18:34:18,030 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=68876.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 18:34:21,114 INFO [train.py:903] (1/4) Epoch 11, batch 600, loss[loss=0.2192, simple_loss=0.2949, pruned_loss=0.0718, over 19846.00 frames. ], tot_loss[loss=0.2351, simple_loss=0.3077, pruned_loss=0.08121, over 3666328.74 frames. ], batch size: 52, lr: 7.73e-03, grad_scale: 8.0 2023-04-01 18:35:07,407 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 2023-04-01 18:35:23,576 INFO [train.py:903] (1/4) Epoch 11, batch 650, loss[loss=0.2496, simple_loss=0.3203, pruned_loss=0.08945, over 17213.00 frames. ], tot_loss[loss=0.2362, simple_loss=0.3089, pruned_loss=0.08179, over 3698779.89 frames. ], batch size: 101, lr: 7.73e-03, grad_scale: 4.0 2023-04-01 18:35:36,883 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=68940.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 18:36:14,702 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.863e+02 5.561e+02 7.157e+02 8.945e+02 2.089e+03, threshold=1.431e+03, percent-clipped=8.0 2023-04-01 18:36:26,501 INFO [train.py:903] (1/4) Epoch 11, batch 700, loss[loss=0.2355, simple_loss=0.3095, pruned_loss=0.08076, over 19581.00 frames. ], tot_loss[loss=0.2365, simple_loss=0.3093, pruned_loss=0.08184, over 3720689.12 frames. ], batch size: 52, lr: 7.73e-03, grad_scale: 4.0 2023-04-01 18:36:46,877 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4005, 1.6344, 2.2443, 1.8492, 2.9490, 2.4473, 3.1366, 1.4799], device='cuda:1'), covar=tensor([0.2364, 0.3884, 0.2318, 0.1740, 0.1683, 0.2023, 0.2008, 0.3617], device='cuda:1'), in_proj_covar=tensor([0.0481, 0.0566, 0.0585, 0.0432, 0.0580, 0.0486, 0.0641, 0.0483], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 18:37:00,609 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.7348, 1.8801, 1.6190, 2.8428, 1.8572, 2.5836, 1.9380, 1.4097], device='cuda:1'), covar=tensor([0.4180, 0.3430, 0.2231, 0.2213, 0.3934, 0.1731, 0.4668, 0.4062], device='cuda:1'), in_proj_covar=tensor([0.0781, 0.0795, 0.0642, 0.0889, 0.0776, 0.0698, 0.0783, 0.0703], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 18:37:30,468 INFO [train.py:903] (1/4) Epoch 11, batch 750, loss[loss=0.2455, simple_loss=0.3138, pruned_loss=0.08857, over 19681.00 frames. ], tot_loss[loss=0.2358, simple_loss=0.3087, pruned_loss=0.08148, over 3743355.18 frames. ], batch size: 53, lr: 7.72e-03, grad_scale: 4.0 2023-04-01 18:37:44,705 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.8919, 4.9016, 5.6717, 5.6450, 2.1201, 5.3142, 4.5113, 5.2284], device='cuda:1'), covar=tensor([0.1173, 0.0789, 0.0477, 0.0447, 0.4815, 0.0513, 0.0512, 0.1004], device='cuda:1'), in_proj_covar=tensor([0.0659, 0.0596, 0.0782, 0.0667, 0.0720, 0.0542, 0.0482, 0.0726], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1') 2023-04-01 18:37:52,521 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=69047.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 18:38:01,922 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69055.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 18:38:09,126 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69061.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 18:38:20,853 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.322e+02 5.289e+02 6.423e+02 8.063e+02 1.861e+03, threshold=1.285e+03, percent-clipped=2.0 2023-04-01 18:38:33,459 INFO [train.py:903] (1/4) Epoch 11, batch 800, loss[loss=0.243, simple_loss=0.3169, pruned_loss=0.08455, over 19695.00 frames. ], tot_loss[loss=0.2374, simple_loss=0.3104, pruned_loss=0.08224, over 3757231.99 frames. ], batch size: 60, lr: 7.72e-03, grad_scale: 8.0 2023-04-01 18:38:35,044 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.7555, 4.3072, 2.4708, 3.8279, 1.1878, 4.1463, 4.0659, 4.1746], device='cuda:1'), covar=tensor([0.0597, 0.1105, 0.2084, 0.0714, 0.3740, 0.0723, 0.0742, 0.0928], device='cuda:1'), in_proj_covar=tensor([0.0424, 0.0362, 0.0430, 0.0316, 0.0373, 0.0361, 0.0349, 0.0382], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-01 18:38:40,023 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 2023-04-01 18:38:49,985 WARNING [train.py:1073] (1/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 2023-04-01 18:39:34,228 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.5446, 4.0820, 2.5231, 3.6492, 0.9404, 3.8655, 3.8398, 3.9564], device='cuda:1'), covar=tensor([0.0588, 0.0993, 0.1958, 0.0732, 0.3918, 0.0779, 0.0760, 0.0993], device='cuda:1'), in_proj_covar=tensor([0.0421, 0.0362, 0.0429, 0.0315, 0.0373, 0.0361, 0.0349, 0.0382], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-01 18:39:35,201 INFO [train.py:903] (1/4) Epoch 11, batch 850, loss[loss=0.2493, simple_loss=0.3271, pruned_loss=0.08574, over 19754.00 frames. ], tot_loss[loss=0.2363, simple_loss=0.3091, pruned_loss=0.08173, over 3781662.22 frames. ], batch size: 63, lr: 7.72e-03, grad_scale: 8.0 2023-04-01 18:39:39,344 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69132.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 18:39:41,071 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.51 vs. limit=2.0 2023-04-01 18:40:04,893 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.7844, 1.8069, 1.9079, 2.4475, 1.8253, 2.3534, 2.1549, 1.8413], device='cuda:1'), covar=tensor([0.2783, 0.2287, 0.1134, 0.1261, 0.2414, 0.1053, 0.2382, 0.1921], device='cuda:1'), in_proj_covar=tensor([0.0779, 0.0795, 0.0641, 0.0886, 0.0771, 0.0698, 0.0780, 0.0704], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 18:40:09,397 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.7490, 4.0675, 4.4652, 4.5525, 1.6752, 4.2595, 3.7440, 3.8857], device='cuda:1'), covar=tensor([0.1662, 0.1185, 0.0706, 0.0820, 0.5648, 0.1102, 0.0896, 0.1476], device='cuda:1'), in_proj_covar=tensor([0.0651, 0.0589, 0.0772, 0.0660, 0.0711, 0.0531, 0.0476, 0.0716], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1') 2023-04-01 18:40:12,592 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69157.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 18:40:18,314 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69162.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 18:40:26,035 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.965e+02 5.808e+02 7.359e+02 9.451e+02 2.011e+03, threshold=1.472e+03, percent-clipped=12.0 2023-04-01 18:40:32,120 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9_rvb from training. Duration: 25.061125 2023-04-01 18:40:38,097 INFO [train.py:903] (1/4) Epoch 11, batch 900, loss[loss=0.2132, simple_loss=0.2868, pruned_loss=0.06982, over 19729.00 frames. ], tot_loss[loss=0.2378, simple_loss=0.3102, pruned_loss=0.08265, over 3781804.70 frames. ], batch size: 51, lr: 7.72e-03, grad_scale: 8.0 2023-04-01 18:40:39,676 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69180.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 18:40:48,350 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69186.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 18:41:28,503 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.42 vs. limit=2.0 2023-04-01 18:41:42,668 INFO [train.py:903] (1/4) Epoch 11, batch 950, loss[loss=0.2448, simple_loss=0.315, pruned_loss=0.08727, over 19541.00 frames. ], tot_loss[loss=0.2372, simple_loss=0.3096, pruned_loss=0.08236, over 3801448.83 frames. ], batch size: 56, lr: 7.71e-03, grad_scale: 8.0 2023-04-01 18:41:49,497 WARNING [train.py:1073] (1/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9_rvb from training. Duration: 26.32775 2023-04-01 18:42:01,240 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69243.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 18:42:11,247 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 2023-04-01 18:42:24,842 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69263.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 18:42:33,740 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.446e+02 5.204e+02 6.315e+02 7.601e+02 2.315e+03, threshold=1.263e+03, percent-clipped=1.0 2023-04-01 18:42:47,232 INFO [train.py:903] (1/4) Epoch 11, batch 1000, loss[loss=0.2578, simple_loss=0.3233, pruned_loss=0.09616, over 19762.00 frames. ], tot_loss[loss=0.2383, simple_loss=0.3104, pruned_loss=0.08306, over 3808288.75 frames. ], batch size: 54, lr: 7.71e-03, grad_scale: 8.0 2023-04-01 18:43:01,211 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69291.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 18:43:27,050 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69311.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 18:43:43,501 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 2023-04-01 18:43:48,901 INFO [train.py:903] (1/4) Epoch 11, batch 1050, loss[loss=0.2874, simple_loss=0.3451, pruned_loss=0.1149, over 14070.00 frames. ], tot_loss[loss=0.2378, simple_loss=0.3104, pruned_loss=0.08263, over 3818820.47 frames. ], batch size: 136, lr: 7.71e-03, grad_scale: 8.0 2023-04-01 18:43:57,031 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69336.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 18:43:59,186 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.7873, 4.3390, 2.5803, 3.8088, 1.0127, 4.0958, 4.0754, 4.1353], device='cuda:1'), covar=tensor([0.0554, 0.0933, 0.1907, 0.0728, 0.3993, 0.0715, 0.0773, 0.0909], device='cuda:1'), in_proj_covar=tensor([0.0423, 0.0361, 0.0428, 0.0312, 0.0375, 0.0361, 0.0350, 0.0383], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-01 18:43:59,276 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69338.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 18:43:59,411 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.1114, 2.0717, 1.7739, 1.5469, 1.3441, 1.6054, 0.4090, 1.0809], device='cuda:1'), covar=tensor([0.0604, 0.0606, 0.0472, 0.0821, 0.1276, 0.0858, 0.1110, 0.1001], device='cuda:1'), in_proj_covar=tensor([0.0336, 0.0329, 0.0331, 0.0352, 0.0425, 0.0348, 0.0309, 0.0324], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 18:44:02,784 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3764, 2.1870, 1.5695, 1.4458, 2.1114, 1.2013, 1.3216, 1.7447], device='cuda:1'), covar=tensor([0.0968, 0.0647, 0.0996, 0.0715, 0.0452, 0.1148, 0.0706, 0.0475], device='cuda:1'), in_proj_covar=tensor([0.0285, 0.0301, 0.0328, 0.0246, 0.0237, 0.0318, 0.0289, 0.0258], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 18:44:24,560 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 2023-04-01 18:44:38,147 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.334e+02 5.222e+02 6.475e+02 7.672e+02 1.315e+03, threshold=1.295e+03, percent-clipped=1.0 2023-04-01 18:44:49,369 INFO [train.py:903] (1/4) Epoch 11, batch 1100, loss[loss=0.2306, simple_loss=0.3007, pruned_loss=0.0803, over 19480.00 frames. ], tot_loss[loss=0.2382, simple_loss=0.3105, pruned_loss=0.08294, over 3809629.14 frames. ], batch size: 49, lr: 7.70e-03, grad_scale: 8.0 2023-04-01 18:44:57,869 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69386.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 18:45:24,098 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=69405.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 18:45:32,712 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69412.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 18:45:39,607 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69418.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 18:45:53,089 INFO [train.py:903] (1/4) Epoch 11, batch 1150, loss[loss=0.2042, simple_loss=0.2948, pruned_loss=0.0568, over 19543.00 frames. ], tot_loss[loss=0.2376, simple_loss=0.3098, pruned_loss=0.08264, over 3800208.70 frames. ], batch size: 54, lr: 7.70e-03, grad_scale: 8.0 2023-04-01 18:45:59,910 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0 2023-04-01 18:46:11,275 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69443.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 18:46:41,981 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.975e+02 5.885e+02 7.181e+02 8.495e+02 1.651e+03, threshold=1.436e+03, percent-clipped=3.0 2023-04-01 18:46:55,765 INFO [train.py:903] (1/4) Epoch 11, batch 1200, loss[loss=0.2798, simple_loss=0.3474, pruned_loss=0.1061, over 19587.00 frames. ], tot_loss[loss=0.2375, simple_loss=0.3097, pruned_loss=0.08267, over 3809477.25 frames. ], batch size: 61, lr: 7.70e-03, grad_scale: 8.0 2023-04-01 18:46:59,310 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.7010, 1.4412, 1.4974, 2.2671, 1.7592, 2.0409, 2.1653, 1.9054], device='cuda:1'), covar=tensor([0.0737, 0.0962, 0.0969, 0.0747, 0.0812, 0.0670, 0.0765, 0.0635], device='cuda:1'), in_proj_covar=tensor([0.0211, 0.0227, 0.0224, 0.0250, 0.0238, 0.0213, 0.0199, 0.0203], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-01 18:47:27,288 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 2023-04-01 18:47:48,985 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69520.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 18:47:53,392 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=69524.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 18:47:58,998 INFO [train.py:903] (1/4) Epoch 11, batch 1250, loss[loss=0.2228, simple_loss=0.2902, pruned_loss=0.07766, over 19775.00 frames. ], tot_loss[loss=0.238, simple_loss=0.31, pruned_loss=0.08296, over 3804429.76 frames. ], batch size: 54, lr: 7.70e-03, grad_scale: 8.0 2023-04-01 18:48:00,172 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=69530.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 18:48:49,513 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.132e+02 5.575e+02 6.899e+02 8.428e+02 1.860e+03, threshold=1.380e+03, percent-clipped=3.0 2023-04-01 18:49:00,823 INFO [train.py:903] (1/4) Epoch 11, batch 1300, loss[loss=0.2337, simple_loss=0.3168, pruned_loss=0.07526, over 19343.00 frames. ], tot_loss[loss=0.2385, simple_loss=0.3107, pruned_loss=0.08317, over 3809141.11 frames. ], batch size: 70, lr: 7.69e-03, grad_scale: 8.0 2023-04-01 18:49:10,185 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=69587.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 18:49:38,652 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=69607.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 18:50:04,325 INFO [train.py:903] (1/4) Epoch 11, batch 1350, loss[loss=0.2684, simple_loss=0.3443, pruned_loss=0.09628, over 18784.00 frames. ], tot_loss[loss=0.2376, simple_loss=0.3099, pruned_loss=0.08264, over 3805202.19 frames. ], batch size: 74, lr: 7.69e-03, grad_scale: 8.0 2023-04-01 18:50:13,536 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=69635.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 18:50:19,142 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69639.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 18:50:26,795 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69645.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 18:50:54,484 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.470e+02 5.709e+02 6.895e+02 8.476e+02 1.895e+03, threshold=1.379e+03, percent-clipped=5.0 2023-04-01 18:51:08,229 INFO [train.py:903] (1/4) Epoch 11, batch 1400, loss[loss=0.2278, simple_loss=0.2998, pruned_loss=0.07788, over 19592.00 frames. ], tot_loss[loss=0.2368, simple_loss=0.3092, pruned_loss=0.08218, over 3807401.13 frames. ], batch size: 50, lr: 7.69e-03, grad_scale: 8.0 2023-04-01 18:51:11,993 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=69682.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 18:51:25,017 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69692.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 18:51:36,494 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69702.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 18:52:03,576 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69722.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 18:52:09,961 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_rvb from training. Duration: 25.85 2023-04-01 18:52:10,950 INFO [train.py:903] (1/4) Epoch 11, batch 1450, loss[loss=0.2441, simple_loss=0.3308, pruned_loss=0.07871, over 19584.00 frames. ], tot_loss[loss=0.2379, simple_loss=0.3105, pruned_loss=0.08264, over 3810432.99 frames. ], batch size: 61, lr: 7.69e-03, grad_scale: 8.0 2023-04-01 18:52:12,377 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=69730.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 18:52:36,767 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69750.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 18:52:44,800 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=69756.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 18:53:01,501 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.562e+02 5.344e+02 6.531e+02 8.357e+02 2.062e+03, threshold=1.306e+03, percent-clipped=2.0 2023-04-01 18:53:10,378 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69776.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 18:53:13,431 INFO [train.py:903] (1/4) Epoch 11, batch 1500, loss[loss=0.1822, simple_loss=0.2578, pruned_loss=0.05328, over 19709.00 frames. ], tot_loss[loss=0.2377, simple_loss=0.3102, pruned_loss=0.0826, over 3808568.61 frames. ], batch size: 46, lr: 7.68e-03, grad_scale: 8.0 2023-04-01 18:53:34,495 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5081, 1.3188, 1.3096, 1.9066, 1.4469, 1.9410, 1.8742, 1.6501], device='cuda:1'), covar=tensor([0.0869, 0.0975, 0.1035, 0.0766, 0.0873, 0.0597, 0.0797, 0.0691], device='cuda:1'), in_proj_covar=tensor([0.0212, 0.0226, 0.0222, 0.0250, 0.0237, 0.0214, 0.0200, 0.0205], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-01 18:53:36,846 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69797.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 18:53:43,097 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69801.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 18:54:15,942 INFO [train.py:903] (1/4) Epoch 11, batch 1550, loss[loss=0.2378, simple_loss=0.3214, pruned_loss=0.07708, over 19532.00 frames. ], tot_loss[loss=0.2366, simple_loss=0.3096, pruned_loss=0.08182, over 3800210.21 frames. ], batch size: 54, lr: 7.68e-03, grad_scale: 8.0 2023-04-01 18:54:38,367 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69845.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 18:55:07,632 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.582e+02 5.814e+02 6.999e+02 8.951e+02 2.972e+03, threshold=1.400e+03, percent-clipped=5.0 2023-04-01 18:55:09,171 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69871.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 18:55:20,439 INFO [train.py:903] (1/4) Epoch 11, batch 1600, loss[loss=0.1826, simple_loss=0.2611, pruned_loss=0.05208, over 19743.00 frames. ], tot_loss[loss=0.2367, simple_loss=0.3098, pruned_loss=0.08179, over 3806717.63 frames. ], batch size: 47, lr: 7.68e-03, grad_scale: 8.0 2023-04-01 18:55:41,208 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69895.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 18:55:46,726 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9_rvb from training. Duration: 30.1555625 2023-04-01 18:55:48,168 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69901.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 18:56:12,825 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69920.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 18:56:20,542 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69926.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 18:56:23,825 INFO [train.py:903] (1/4) Epoch 11, batch 1650, loss[loss=0.235, simple_loss=0.312, pruned_loss=0.07894, over 19775.00 frames. ], tot_loss[loss=0.2374, simple_loss=0.3105, pruned_loss=0.08217, over 3808277.15 frames. ], batch size: 54, lr: 7.67e-03, grad_scale: 8.0 2023-04-01 18:56:48,514 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.9108, 4.3816, 2.7346, 3.9366, 0.8567, 4.2313, 4.2750, 4.2707], device='cuda:1'), covar=tensor([0.0525, 0.0938, 0.1723, 0.0684, 0.4042, 0.0637, 0.0702, 0.0877], device='cuda:1'), in_proj_covar=tensor([0.0428, 0.0365, 0.0430, 0.0315, 0.0376, 0.0364, 0.0356, 0.0385], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 18:57:00,763 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69958.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 18:57:16,651 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.105e+02 5.294e+02 6.504e+02 8.314e+02 1.576e+03, threshold=1.301e+03, percent-clipped=2.0 2023-04-01 18:57:21,095 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 2023-04-01 18:57:26,304 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69978.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 18:57:27,006 INFO [train.py:903] (1/4) Epoch 11, batch 1700, loss[loss=0.2305, simple_loss=0.3119, pruned_loss=0.07458, over 19565.00 frames. ], tot_loss[loss=0.2362, simple_loss=0.3092, pruned_loss=0.08156, over 3817249.15 frames. ], batch size: 56, lr: 7.67e-03, grad_scale: 8.0 2023-04-01 18:57:32,069 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69983.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 18:57:59,070 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=70003.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 18:58:03,166 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=70006.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 18:58:09,523 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 2023-04-01 18:58:29,317 INFO [train.py:903] (1/4) Epoch 11, batch 1750, loss[loss=0.2586, simple_loss=0.3281, pruned_loss=0.09458, over 19610.00 frames. ], tot_loss[loss=0.2371, simple_loss=0.3094, pruned_loss=0.08234, over 3820346.09 frames. ], batch size: 57, lr: 7.67e-03, grad_scale: 8.0 2023-04-01 18:58:32,022 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=70031.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 18:58:39,012 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=70036.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 18:59:02,463 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=70053.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 18:59:11,018 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 2023-04-01 18:59:22,250 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.188e+02 6.480e+02 8.127e+02 9.904e+02 1.581e+03, threshold=1.625e+03, percent-clipped=5.0 2023-04-01 18:59:34,161 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=70078.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 18:59:34,850 INFO [train.py:903] (1/4) Epoch 11, batch 1800, loss[loss=0.2632, simple_loss=0.3415, pruned_loss=0.09245, over 19693.00 frames. ], tot_loss[loss=0.2362, simple_loss=0.3086, pruned_loss=0.08191, over 3822125.53 frames. ], batch size: 59, lr: 7.67e-03, grad_scale: 8.0 2023-04-01 19:00:02,675 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=70101.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 19:00:08,719 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.63 vs. limit=2.0 2023-04-01 19:00:34,727 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=70126.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 19:00:35,522 WARNING [train.py:1073] (1/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9_rvb from training. Duration: 25.0944375 2023-04-01 19:00:35,947 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=70127.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 19:00:37,728 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=70128.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 19:00:38,700 INFO [train.py:903] (1/4) Epoch 11, batch 1850, loss[loss=0.2551, simple_loss=0.317, pruned_loss=0.09661, over 13665.00 frames. ], tot_loss[loss=0.2356, simple_loss=0.3082, pruned_loss=0.0815, over 3821073.81 frames. ], batch size: 136, lr: 7.66e-03, grad_scale: 8.0 2023-04-01 19:01:05,176 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=70151.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 19:01:06,351 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=70152.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 19:01:13,659 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9_rvb from training. Duration: 27.8166875 2023-04-01 19:01:30,924 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.636e+02 5.171e+02 6.551e+02 7.968e+02 1.780e+03, threshold=1.310e+03, percent-clipped=1.0 2023-04-01 19:01:41,531 INFO [train.py:903] (1/4) Epoch 11, batch 1900, loss[loss=0.2106, simple_loss=0.2905, pruned_loss=0.06533, over 19781.00 frames. ], tot_loss[loss=0.2357, simple_loss=0.3086, pruned_loss=0.08139, over 3827669.31 frames. ], batch size: 56, lr: 7.66e-03, grad_scale: 8.0 2023-04-01 19:01:58,102 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 2023-04-01 19:02:04,767 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9_rvb from training. Duration: 27.02225 2023-04-01 19:02:28,868 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 2023-04-01 19:02:42,956 INFO [train.py:903] (1/4) Epoch 11, batch 1950, loss[loss=0.2664, simple_loss=0.3393, pruned_loss=0.0968, over 19047.00 frames. ], tot_loss[loss=0.2371, simple_loss=0.3095, pruned_loss=0.08236, over 3821572.35 frames. ], batch size: 69, lr: 7.66e-03, grad_scale: 8.0 2023-04-01 19:03:35,384 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.353e+02 4.987e+02 6.114e+02 7.804e+02 3.131e+03, threshold=1.223e+03, percent-clipped=9.0 2023-04-01 19:03:37,258 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 2023-04-01 19:03:43,625 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.2251, 3.7582, 2.3756, 2.2584, 3.3801, 1.8708, 1.4326, 2.1750], device='cuda:1'), covar=tensor([0.0982, 0.0341, 0.0765, 0.0675, 0.0372, 0.0965, 0.0823, 0.0523], device='cuda:1'), in_proj_covar=tensor([0.0282, 0.0293, 0.0321, 0.0242, 0.0231, 0.0314, 0.0284, 0.0255], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 19:03:45,801 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=70278.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 19:03:46,665 INFO [train.py:903] (1/4) Epoch 11, batch 2000, loss[loss=0.2265, simple_loss=0.2942, pruned_loss=0.07939, over 19081.00 frames. ], tot_loss[loss=0.2354, simple_loss=0.3086, pruned_loss=0.08113, over 3830217.20 frames. ], batch size: 42, lr: 7.66e-03, grad_scale: 8.0 2023-04-01 19:04:48,750 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_rvb from training. Duration: 26.8349375 2023-04-01 19:04:51,060 INFO [train.py:903] (1/4) Epoch 11, batch 2050, loss[loss=0.251, simple_loss=0.331, pruned_loss=0.08553, over 19518.00 frames. ], tot_loss[loss=0.2345, simple_loss=0.3079, pruned_loss=0.08055, over 3821401.61 frames. ], batch size: 64, lr: 7.65e-03, grad_scale: 8.0 2023-04-01 19:05:07,061 WARNING [train.py:1073] (1/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 2023-04-01 19:05:08,146 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 2023-04-01 19:05:27,814 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 2023-04-01 19:05:43,672 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.125e+02 5.481e+02 7.126e+02 9.531e+02 2.002e+03, threshold=1.425e+03, percent-clipped=10.0 2023-04-01 19:05:54,409 INFO [train.py:903] (1/4) Epoch 11, batch 2100, loss[loss=0.3308, simple_loss=0.3696, pruned_loss=0.146, over 13327.00 frames. ], tot_loss[loss=0.2359, simple_loss=0.3088, pruned_loss=0.0815, over 3807403.18 frames. ], batch size: 136, lr: 7.65e-03, grad_scale: 8.0 2023-04-01 19:06:24,769 WARNING [train.py:1073] (1/4) Exclude cut with ID 453-131332-0000-131866_sp0.9_rvb from training. Duration: 25.3333125 2023-04-01 19:06:30,839 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=70407.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 19:06:48,496 WARNING [train.py:1073] (1/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9_rvb from training. Duration: 26.6166875 2023-04-01 19:06:57,558 INFO [train.py:903] (1/4) Epoch 11, batch 2150, loss[loss=0.2249, simple_loss=0.3004, pruned_loss=0.07468, over 19580.00 frames. ], tot_loss[loss=0.236, simple_loss=0.309, pruned_loss=0.08156, over 3812430.23 frames. ], batch size: 52, lr: 7.65e-03, grad_scale: 8.0 2023-04-01 19:07:01,569 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=70432.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 19:07:51,371 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.544e+02 5.585e+02 6.733e+02 8.759e+02 1.859e+03, threshold=1.347e+03, percent-clipped=4.0 2023-04-01 19:07:52,830 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=70472.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 19:08:02,115 INFO [train.py:903] (1/4) Epoch 11, batch 2200, loss[loss=0.2586, simple_loss=0.3224, pruned_loss=0.09737, over 19632.00 frames. ], tot_loss[loss=0.2371, simple_loss=0.3099, pruned_loss=0.08213, over 3811397.30 frames. ], batch size: 50, lr: 7.64e-03, grad_scale: 4.0 2023-04-01 19:09:06,387 INFO [train.py:903] (1/4) Epoch 11, batch 2250, loss[loss=0.2361, simple_loss=0.3129, pruned_loss=0.07966, over 17127.00 frames. ], tot_loss[loss=0.2373, simple_loss=0.31, pruned_loss=0.08232, over 3802947.78 frames. ], batch size: 101, lr: 7.64e-03, grad_scale: 4.0 2023-04-01 19:10:01,004 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.382e+02 4.990e+02 6.839e+02 8.349e+02 1.575e+03, threshold=1.368e+03, percent-clipped=2.0 2023-04-01 19:10:10,422 INFO [train.py:903] (1/4) Epoch 11, batch 2300, loss[loss=0.2155, simple_loss=0.2887, pruned_loss=0.07112, over 19662.00 frames. ], tot_loss[loss=0.2374, simple_loss=0.3099, pruned_loss=0.08244, over 3804771.20 frames. ], batch size: 53, lr: 7.64e-03, grad_scale: 4.0 2023-04-01 19:10:19,777 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=70587.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 19:10:23,136 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_rvb from training. Duration: 26.205 2023-04-01 19:11:05,023 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=70622.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 19:11:13,191 INFO [train.py:903] (1/4) Epoch 11, batch 2350, loss[loss=0.2805, simple_loss=0.3413, pruned_loss=0.1098, over 17438.00 frames. ], tot_loss[loss=0.2379, simple_loss=0.3106, pruned_loss=0.0826, over 3798901.14 frames. ], batch size: 101, lr: 7.64e-03, grad_scale: 4.0 2023-04-01 19:11:26,861 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5773, 1.6760, 1.8227, 2.1630, 1.5131, 1.9118, 2.0269, 1.7264], device='cuda:1'), covar=tensor([0.3335, 0.2629, 0.1392, 0.1470, 0.2836, 0.1362, 0.3439, 0.2627], device='cuda:1'), in_proj_covar=tensor([0.0788, 0.0799, 0.0644, 0.0887, 0.0776, 0.0694, 0.0779, 0.0706], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 19:11:56,882 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_rvb from training. Duration: 25.775 2023-04-01 19:12:06,973 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.408e+02 5.270e+02 6.524e+02 8.351e+02 2.247e+03, threshold=1.305e+03, percent-clipped=5.0 2023-04-01 19:12:12,559 WARNING [train.py:1073] (1/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9_rvb from training. Duration: 25.45 2023-04-01 19:12:16,796 INFO [train.py:903] (1/4) Epoch 11, batch 2400, loss[loss=0.2282, simple_loss=0.3085, pruned_loss=0.07389, over 19665.00 frames. ], tot_loss[loss=0.2374, simple_loss=0.31, pruned_loss=0.08237, over 3806098.81 frames. ], batch size: 55, lr: 7.63e-03, grad_scale: 8.0 2023-04-01 19:12:58,799 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1354, 1.2160, 1.7978, 1.1333, 2.5733, 3.5981, 3.2613, 3.7200], device='cuda:1'), covar=tensor([0.1598, 0.3452, 0.2865, 0.2159, 0.0501, 0.0141, 0.0192, 0.0174], device='cuda:1'), in_proj_covar=tensor([0.0253, 0.0295, 0.0320, 0.0251, 0.0215, 0.0155, 0.0205, 0.0196], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 19:13:20,240 INFO [train.py:903] (1/4) Epoch 11, batch 2450, loss[loss=0.2194, simple_loss=0.2977, pruned_loss=0.0706, over 19586.00 frames. ], tot_loss[loss=0.2371, simple_loss=0.3098, pruned_loss=0.08221, over 3821737.81 frames. ], batch size: 52, lr: 7.63e-03, grad_scale: 8.0 2023-04-01 19:13:32,435 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=70737.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 19:13:44,523 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.1060, 2.2941, 2.4638, 3.0068, 2.9341, 2.4546, 2.2090, 2.9369], device='cuda:1'), covar=tensor([0.0614, 0.1541, 0.1204, 0.0926, 0.1023, 0.0424, 0.1119, 0.0484], device='cuda:1'), in_proj_covar=tensor([0.0253, 0.0347, 0.0287, 0.0238, 0.0296, 0.0242, 0.0276, 0.0239], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 19:14:14,278 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.733e+02 5.717e+02 6.752e+02 8.721e+02 4.215e+03, threshold=1.350e+03, percent-clipped=8.0 2023-04-01 19:14:25,717 INFO [train.py:903] (1/4) Epoch 11, batch 2500, loss[loss=0.2643, simple_loss=0.3339, pruned_loss=0.09739, over 19308.00 frames. ], tot_loss[loss=0.2354, simple_loss=0.3084, pruned_loss=0.0812, over 3831745.96 frames. ], batch size: 66, lr: 7.63e-03, grad_scale: 8.0 2023-04-01 19:15:27,565 INFO [train.py:903] (1/4) Epoch 11, batch 2550, loss[loss=0.2312, simple_loss=0.3128, pruned_loss=0.07484, over 19391.00 frames. ], tot_loss[loss=0.2352, simple_loss=0.3084, pruned_loss=0.08095, over 3848588.84 frames. ], batch size: 70, lr: 7.63e-03, grad_scale: 8.0 2023-04-01 19:15:45,706 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=70843.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 19:16:17,533 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=70868.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 19:16:20,506 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.586e+02 5.428e+02 6.511e+02 7.969e+02 1.423e+03, threshold=1.302e+03, percent-clipped=4.0 2023-04-01 19:16:26,480 WARNING [train.py:1073] (1/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 2023-04-01 19:16:30,147 INFO [train.py:903] (1/4) Epoch 11, batch 2600, loss[loss=0.2578, simple_loss=0.3306, pruned_loss=0.09255, over 19676.00 frames. ], tot_loss[loss=0.2338, simple_loss=0.3073, pruned_loss=0.08011, over 3851689.68 frames. ], batch size: 58, lr: 7.62e-03, grad_scale: 8.0 2023-04-01 19:17:06,533 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.13 vs. limit=2.0 2023-04-01 19:17:32,112 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.88 vs. limit=5.0 2023-04-01 19:17:34,834 INFO [train.py:903] (1/4) Epoch 11, batch 2650, loss[loss=0.2528, simple_loss=0.3214, pruned_loss=0.09213, over 19539.00 frames. ], tot_loss[loss=0.2349, simple_loss=0.3081, pruned_loss=0.08087, over 3837608.71 frames. ], batch size: 54, lr: 7.62e-03, grad_scale: 8.0 2023-04-01 19:17:56,630 WARNING [train.py:1073] (1/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9_rvb from training. Duration: 27.25 2023-04-01 19:18:20,080 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.1350, 2.1910, 2.2467, 3.2417, 2.2117, 3.1382, 2.6752, 2.0650], device='cuda:1'), covar=tensor([0.3750, 0.3356, 0.1518, 0.1992, 0.3800, 0.1508, 0.3533, 0.2835], device='cuda:1'), in_proj_covar=tensor([0.0792, 0.0804, 0.0650, 0.0894, 0.0778, 0.0698, 0.0787, 0.0709], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 19:18:28,055 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.326e+02 5.322e+02 6.264e+02 9.085e+02 2.401e+03, threshold=1.253e+03, percent-clipped=10.0 2023-04-01 19:18:39,383 INFO [train.py:903] (1/4) Epoch 11, batch 2700, loss[loss=0.1947, simple_loss=0.2601, pruned_loss=0.06464, over 19704.00 frames. ], tot_loss[loss=0.2356, simple_loss=0.3086, pruned_loss=0.08124, over 3825618.71 frames. ], batch size: 45, lr: 7.62e-03, grad_scale: 8.0 2023-04-01 19:18:56,202 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=70993.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 19:19:18,181 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=71010.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 19:19:30,024 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=71018.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 19:19:43,442 INFO [train.py:903] (1/4) Epoch 11, batch 2750, loss[loss=0.2465, simple_loss=0.3181, pruned_loss=0.08742, over 13890.00 frames. ], tot_loss[loss=0.2354, simple_loss=0.3087, pruned_loss=0.08102, over 3811558.67 frames. ], batch size: 135, lr: 7.62e-03, grad_scale: 8.0 2023-04-01 19:20:08,222 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=71048.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 19:20:37,008 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.532e+02 5.301e+02 6.565e+02 7.951e+02 1.458e+03, threshold=1.313e+03, percent-clipped=6.0 2023-04-01 19:20:45,877 INFO [train.py:903] (1/4) Epoch 11, batch 2800, loss[loss=0.2375, simple_loss=0.3135, pruned_loss=0.08076, over 19331.00 frames. ], tot_loss[loss=0.2372, simple_loss=0.3104, pruned_loss=0.08197, over 3822482.16 frames. ], batch size: 66, lr: 7.61e-03, grad_scale: 8.0 2023-04-01 19:21:18,606 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.0507, 1.3252, 1.4827, 1.4584, 2.6217, 0.9482, 1.9151, 2.9579], device='cuda:1'), covar=tensor([0.0522, 0.2585, 0.2528, 0.1570, 0.0738, 0.2405, 0.1179, 0.0331], device='cuda:1'), in_proj_covar=tensor([0.0352, 0.0336, 0.0351, 0.0318, 0.0342, 0.0332, 0.0327, 0.0347], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 19:21:51,269 INFO [train.py:903] (1/4) Epoch 11, batch 2850, loss[loss=0.2159, simple_loss=0.2917, pruned_loss=0.07011, over 19673.00 frames. ], tot_loss[loss=0.2364, simple_loss=0.3099, pruned_loss=0.08146, over 3817960.31 frames. ], batch size: 53, lr: 7.61e-03, grad_scale: 8.0 2023-04-01 19:22:41,893 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6654, 1.3816, 1.3302, 1.8700, 1.4581, 1.9108, 1.8806, 1.6629], device='cuda:1'), covar=tensor([0.0760, 0.0945, 0.1034, 0.0797, 0.0872, 0.0663, 0.0855, 0.0650], device='cuda:1'), in_proj_covar=tensor([0.0211, 0.0226, 0.0222, 0.0249, 0.0236, 0.0213, 0.0196, 0.0203], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-01 19:22:45,077 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.239e+02 5.466e+02 6.713e+02 8.640e+02 2.236e+03, threshold=1.343e+03, percent-clipped=7.0 2023-04-01 19:22:49,017 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([5.9935, 5.3585, 2.8410, 4.7294, 1.1876, 5.3494, 5.2685, 5.4938], device='cuda:1'), covar=tensor([0.0402, 0.0807, 0.1834, 0.0648, 0.3916, 0.0583, 0.0627, 0.0861], device='cuda:1'), in_proj_covar=tensor([0.0432, 0.0367, 0.0440, 0.0322, 0.0383, 0.0369, 0.0357, 0.0393], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 19:22:54,329 WARNING [train.py:1073] (1/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 2023-04-01 19:22:55,473 INFO [train.py:903] (1/4) Epoch 11, batch 2900, loss[loss=0.2025, simple_loss=0.2876, pruned_loss=0.05873, over 19590.00 frames. ], tot_loss[loss=0.2368, simple_loss=0.31, pruned_loss=0.08177, over 3797286.36 frames. ], batch size: 57, lr: 7.61e-03, grad_scale: 8.0 2023-04-01 19:24:00,113 INFO [train.py:903] (1/4) Epoch 11, batch 2950, loss[loss=0.3307, simple_loss=0.3748, pruned_loss=0.1433, over 18336.00 frames. ], tot_loss[loss=0.2357, simple_loss=0.309, pruned_loss=0.08124, over 3805564.37 frames. ], batch size: 83, lr: 7.60e-03, grad_scale: 8.0 2023-04-01 19:24:25,803 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 2023-04-01 19:24:53,527 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.649e+02 5.471e+02 6.960e+02 8.532e+02 1.699e+03, threshold=1.392e+03, percent-clipped=7.0 2023-04-01 19:25:02,842 INFO [train.py:903] (1/4) Epoch 11, batch 3000, loss[loss=0.2127, simple_loss=0.2958, pruned_loss=0.06484, over 19614.00 frames. ], tot_loss[loss=0.2352, simple_loss=0.3085, pruned_loss=0.08091, over 3816174.90 frames. ], batch size: 57, lr: 7.60e-03, grad_scale: 8.0 2023-04-01 19:25:02,842 INFO [train.py:928] (1/4) Computing validation loss 2023-04-01 19:25:16,077 INFO [train.py:937] (1/4) Epoch 11, validation: loss=0.1785, simple_loss=0.2793, pruned_loss=0.0389, over 944034.00 frames. 2023-04-01 19:25:16,079 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18649MB 2023-04-01 19:25:20,541 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_rvb from training. Duration: 29.735 2023-04-01 19:26:20,104 INFO [train.py:903] (1/4) Epoch 11, batch 3050, loss[loss=0.2461, simple_loss=0.3206, pruned_loss=0.08576, over 19761.00 frames. ], tot_loss[loss=0.2356, simple_loss=0.3089, pruned_loss=0.08115, over 3828494.09 frames. ], batch size: 63, lr: 7.60e-03, grad_scale: 8.0 2023-04-01 19:26:51,672 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=71354.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 19:27:13,587 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.278e+02 5.484e+02 7.005e+02 9.170e+02 2.820e+03, threshold=1.401e+03, percent-clipped=8.0 2023-04-01 19:27:22,999 INFO [train.py:903] (1/4) Epoch 11, batch 3100, loss[loss=0.1994, simple_loss=0.2693, pruned_loss=0.06474, over 19768.00 frames. ], tot_loss[loss=0.2357, simple_loss=0.3091, pruned_loss=0.08121, over 3837255.83 frames. ], batch size: 46, lr: 7.60e-03, grad_scale: 8.0 2023-04-01 19:27:27,743 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.3421, 1.2648, 1.4030, 1.5438, 2.9045, 0.8899, 2.0812, 3.2734], device='cuda:1'), covar=tensor([0.0460, 0.2654, 0.2746, 0.1648, 0.0752, 0.2646, 0.1201, 0.0297], device='cuda:1'), in_proj_covar=tensor([0.0353, 0.0337, 0.0352, 0.0318, 0.0347, 0.0336, 0.0330, 0.0350], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 19:27:36,322 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.62 vs. limit=2.0 2023-04-01 19:27:39,069 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=71392.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 19:28:24,231 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.3747, 1.6214, 1.9566, 2.6799, 2.1230, 2.6211, 2.8283, 2.3557], device='cuda:1'), covar=tensor([0.0645, 0.0931, 0.0888, 0.0864, 0.0855, 0.0624, 0.0761, 0.0570], device='cuda:1'), in_proj_covar=tensor([0.0212, 0.0226, 0.0223, 0.0250, 0.0236, 0.0214, 0.0196, 0.0203], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-01 19:28:26,275 INFO [train.py:903] (1/4) Epoch 11, batch 3150, loss[loss=0.2242, simple_loss=0.3037, pruned_loss=0.07231, over 17986.00 frames. ], tot_loss[loss=0.2356, simple_loss=0.3088, pruned_loss=0.08113, over 3839046.55 frames. ], batch size: 83, lr: 7.59e-03, grad_scale: 8.0 2023-04-01 19:28:55,930 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 2023-04-01 19:29:17,605 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=71469.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 19:29:19,600 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.562e+02 5.450e+02 6.508e+02 8.826e+02 1.534e+03, threshold=1.302e+03, percent-clipped=1.0 2023-04-01 19:29:31,369 INFO [train.py:903] (1/4) Epoch 11, batch 3200, loss[loss=0.1654, simple_loss=0.2431, pruned_loss=0.04389, over 19047.00 frames. ], tot_loss[loss=0.2337, simple_loss=0.3075, pruned_loss=0.07997, over 3842902.78 frames. ], batch size: 42, lr: 7.59e-03, grad_scale: 8.0 2023-04-01 19:30:03,852 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6051, 1.2716, 1.2394, 1.4758, 1.1024, 1.3699, 1.2147, 1.4488], device='cuda:1'), covar=tensor([0.0911, 0.1070, 0.1362, 0.0847, 0.1115, 0.0560, 0.1201, 0.0752], device='cuda:1'), in_proj_covar=tensor([0.0253, 0.0348, 0.0288, 0.0236, 0.0296, 0.0244, 0.0277, 0.0238], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 19:30:05,862 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=71507.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 19:30:32,440 INFO [train.py:903] (1/4) Epoch 11, batch 3250, loss[loss=0.2845, simple_loss=0.3446, pruned_loss=0.1122, over 19770.00 frames. ], tot_loss[loss=0.2361, simple_loss=0.3093, pruned_loss=0.08139, over 3831059.70 frames. ], batch size: 54, lr: 7.59e-03, grad_scale: 8.0 2023-04-01 19:31:26,700 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.701e+02 5.157e+02 6.415e+02 8.641e+02 1.397e+03, threshold=1.283e+03, percent-clipped=1.0 2023-04-01 19:31:36,062 INFO [train.py:903] (1/4) Epoch 11, batch 3300, loss[loss=0.225, simple_loss=0.3045, pruned_loss=0.07278, over 19661.00 frames. ], tot_loss[loss=0.2367, simple_loss=0.3097, pruned_loss=0.08182, over 3833603.36 frames. ], batch size: 55, lr: 7.59e-03, grad_scale: 8.0 2023-04-01 19:31:41,720 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 2023-04-01 19:32:39,937 INFO [train.py:903] (1/4) Epoch 11, batch 3350, loss[loss=0.3015, simple_loss=0.3605, pruned_loss=0.1213, over 18236.00 frames. ], tot_loss[loss=0.2368, simple_loss=0.31, pruned_loss=0.08177, over 3825473.68 frames. ], batch size: 83, lr: 7.58e-03, grad_scale: 8.0 2023-04-01 19:32:47,338 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=71634.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 19:33:13,349 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2309, 1.9794, 1.5543, 1.3658, 1.8911, 1.1623, 1.2733, 1.7985], device='cuda:1'), covar=tensor([0.0807, 0.0688, 0.1008, 0.0729, 0.0435, 0.1102, 0.0593, 0.0342], device='cuda:1'), in_proj_covar=tensor([0.0289, 0.0300, 0.0329, 0.0250, 0.0233, 0.0320, 0.0287, 0.0256], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 19:33:32,601 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 2023-04-01 19:33:34,249 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.542e+02 5.202e+02 6.815e+02 8.173e+02 2.322e+03, threshold=1.363e+03, percent-clipped=5.0 2023-04-01 19:33:44,446 INFO [train.py:903] (1/4) Epoch 11, batch 3400, loss[loss=0.2345, simple_loss=0.3179, pruned_loss=0.07559, over 19693.00 frames. ], tot_loss[loss=0.2366, simple_loss=0.3098, pruned_loss=0.08168, over 3828247.15 frames. ], batch size: 53, lr: 7.58e-03, grad_scale: 8.0 2023-04-01 19:33:45,319 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.72 vs. limit=5.0 2023-04-01 19:34:08,997 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2236, 1.2537, 1.1955, 0.9338, 0.9100, 0.9791, 0.0786, 0.2892], device='cuda:1'), covar=tensor([0.0624, 0.0606, 0.0362, 0.0437, 0.1159, 0.0555, 0.1031, 0.1019], device='cuda:1'), in_proj_covar=tensor([0.0336, 0.0332, 0.0330, 0.0353, 0.0425, 0.0353, 0.0311, 0.0320], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 19:34:44,147 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=71725.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 19:34:48,499 INFO [train.py:903] (1/4) Epoch 11, batch 3450, loss[loss=0.1844, simple_loss=0.2632, pruned_loss=0.0528, over 19351.00 frames. ], tot_loss[loss=0.2374, simple_loss=0.3102, pruned_loss=0.0823, over 3810367.17 frames. ], batch size: 47, lr: 7.58e-03, grad_scale: 8.0 2023-04-01 19:34:52,164 WARNING [train.py:1073] (1/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9_rvb from training. Duration: 25.2444375 2023-04-01 19:35:15,972 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=71750.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 19:35:33,066 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=71763.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 19:35:35,393 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1976, 1.0688, 1.1110, 1.2834, 1.0567, 1.3616, 1.4098, 1.2176], device='cuda:1'), covar=tensor([0.0865, 0.1025, 0.1074, 0.0744, 0.0791, 0.0743, 0.0767, 0.0742], device='cuda:1'), in_proj_covar=tensor([0.0211, 0.0223, 0.0221, 0.0247, 0.0233, 0.0211, 0.0194, 0.0201], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1') 2023-04-01 19:35:42,108 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.543e+02 5.649e+02 6.835e+02 8.667e+02 1.565e+03, threshold=1.367e+03, percent-clipped=4.0 2023-04-01 19:35:52,285 INFO [train.py:903] (1/4) Epoch 11, batch 3500, loss[loss=0.2529, simple_loss=0.3374, pruned_loss=0.08419, over 19669.00 frames. ], tot_loss[loss=0.2353, simple_loss=0.3083, pruned_loss=0.08114, over 3831791.37 frames. ], batch size: 59, lr: 7.58e-03, grad_scale: 8.0 2023-04-01 19:36:04,521 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=71788.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 19:36:56,412 INFO [train.py:903] (1/4) Epoch 11, batch 3550, loss[loss=0.2469, simple_loss=0.3191, pruned_loss=0.08739, over 19557.00 frames. ], tot_loss[loss=0.2343, simple_loss=0.3074, pruned_loss=0.08062, over 3829780.21 frames. ], batch size: 61, lr: 7.57e-03, grad_scale: 8.0 2023-04-01 19:37:12,229 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 2023-04-01 19:37:49,024 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.173e+02 5.129e+02 6.236e+02 8.069e+02 1.994e+03, threshold=1.247e+03, percent-clipped=3.0 2023-04-01 19:37:58,908 INFO [train.py:903] (1/4) Epoch 11, batch 3600, loss[loss=0.2366, simple_loss=0.2875, pruned_loss=0.09284, over 19331.00 frames. ], tot_loss[loss=0.2341, simple_loss=0.3071, pruned_loss=0.08049, over 3840934.62 frames. ], batch size: 44, lr: 7.57e-03, grad_scale: 8.0 2023-04-01 19:38:47,074 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.73 vs. limit=5.0 2023-04-01 19:39:03,005 INFO [train.py:903] (1/4) Epoch 11, batch 3650, loss[loss=0.2354, simple_loss=0.3201, pruned_loss=0.07533, over 19510.00 frames. ], tot_loss[loss=0.2336, simple_loss=0.3068, pruned_loss=0.08023, over 3846949.53 frames. ], batch size: 64, lr: 7.57e-03, grad_scale: 8.0 2023-04-01 19:39:13,505 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=71937.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 19:39:55,948 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.339e+02 5.422e+02 6.726e+02 7.997e+02 1.955e+03, threshold=1.345e+03, percent-clipped=5.0 2023-04-01 19:40:05,242 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=71978.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 19:40:06,261 INFO [train.py:903] (1/4) Epoch 11, batch 3700, loss[loss=0.2015, simple_loss=0.2744, pruned_loss=0.06428, over 19407.00 frames. ], tot_loss[loss=0.2343, simple_loss=0.3074, pruned_loss=0.08059, over 3846801.28 frames. ], batch size: 48, lr: 7.57e-03, grad_scale: 8.0 2023-04-01 19:40:10,654 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=4.07 vs. limit=5.0 2023-04-01 19:40:39,343 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.3167, 2.0575, 1.9871, 2.5045, 2.1688, 2.1107, 2.0716, 2.3824], device='cuda:1'), covar=tensor([0.0877, 0.1621, 0.1297, 0.0923, 0.1287, 0.0471, 0.1045, 0.0605], device='cuda:1'), in_proj_covar=tensor([0.0255, 0.0355, 0.0292, 0.0239, 0.0299, 0.0247, 0.0278, 0.0241], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 19:41:11,920 INFO [train.py:903] (1/4) Epoch 11, batch 3750, loss[loss=0.1959, simple_loss=0.2741, pruned_loss=0.05882, over 19584.00 frames. ], tot_loss[loss=0.235, simple_loss=0.3079, pruned_loss=0.08106, over 3822822.10 frames. ], batch size: 52, lr: 7.56e-03, grad_scale: 8.0 2023-04-01 19:41:40,799 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.3402, 2.4005, 2.6731, 3.2338, 2.2299, 3.1817, 2.8484, 2.4168], device='cuda:1'), covar=tensor([0.3517, 0.3052, 0.1265, 0.1940, 0.3689, 0.1470, 0.3183, 0.2466], device='cuda:1'), in_proj_covar=tensor([0.0788, 0.0804, 0.0647, 0.0892, 0.0778, 0.0705, 0.0782, 0.0706], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 19:42:06,335 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.116e+02 5.141e+02 6.053e+02 7.322e+02 1.150e+03, threshold=1.211e+03, percent-clipped=0.0 2023-04-01 19:42:17,090 INFO [train.py:903] (1/4) Epoch 11, batch 3800, loss[loss=0.2683, simple_loss=0.3381, pruned_loss=0.09924, over 17525.00 frames. ], tot_loss[loss=0.2343, simple_loss=0.3071, pruned_loss=0.08071, over 3818878.25 frames. ], batch size: 101, lr: 7.56e-03, grad_scale: 8.0 2023-04-01 19:42:35,675 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=72093.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 19:42:46,989 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9_rvb from training. Duration: 29.1166875 2023-04-01 19:43:06,527 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.71 vs. limit=5.0 2023-04-01 19:43:21,151 INFO [train.py:903] (1/4) Epoch 11, batch 3850, loss[loss=0.2268, simple_loss=0.3041, pruned_loss=0.07477, over 19655.00 frames. ], tot_loss[loss=0.2337, simple_loss=0.3065, pruned_loss=0.08038, over 3821441.52 frames. ], batch size: 53, lr: 7.56e-03, grad_scale: 8.0 2023-04-01 19:43:45,207 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=72147.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 19:44:16,371 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.216e+02 5.600e+02 6.929e+02 8.545e+02 2.074e+03, threshold=1.386e+03, percent-clipped=5.0 2023-04-01 19:44:25,955 INFO [train.py:903] (1/4) Epoch 11, batch 3900, loss[loss=0.2361, simple_loss=0.3145, pruned_loss=0.07887, over 19484.00 frames. ], tot_loss[loss=0.2347, simple_loss=0.3074, pruned_loss=0.08101, over 3812235.63 frames. ], batch size: 64, lr: 7.55e-03, grad_scale: 8.0 2023-04-01 19:45:30,660 INFO [train.py:903] (1/4) Epoch 11, batch 3950, loss[loss=0.2584, simple_loss=0.309, pruned_loss=0.1039, over 19716.00 frames. ], tot_loss[loss=0.2328, simple_loss=0.3058, pruned_loss=0.07993, over 3826035.36 frames. ], batch size: 46, lr: 7.55e-03, grad_scale: 8.0 2023-04-01 19:45:31,934 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 2023-04-01 19:46:23,815 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.514e+02 5.760e+02 7.487e+02 9.416e+02 2.541e+03, threshold=1.497e+03, percent-clipped=9.0 2023-04-01 19:46:34,092 INFO [train.py:903] (1/4) Epoch 11, batch 4000, loss[loss=0.2722, simple_loss=0.3413, pruned_loss=0.1015, over 19695.00 frames. ], tot_loss[loss=0.2338, simple_loss=0.3067, pruned_loss=0.08042, over 3837754.86 frames. ], batch size: 59, lr: 7.55e-03, grad_scale: 8.0 2023-04-01 19:46:36,459 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=72281.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 19:47:05,063 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.8228, 1.7333, 1.5693, 1.4435, 1.3972, 1.4413, 0.3700, 0.7966], device='cuda:1'), covar=tensor([0.0323, 0.0355, 0.0242, 0.0351, 0.0665, 0.0456, 0.0699, 0.0638], device='cuda:1'), in_proj_covar=tensor([0.0338, 0.0333, 0.0334, 0.0356, 0.0429, 0.0357, 0.0313, 0.0326], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 19:47:22,618 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1_rvb from training. Duration: 0.7545625 2023-04-01 19:47:37,790 INFO [train.py:903] (1/4) Epoch 11, batch 4050, loss[loss=0.198, simple_loss=0.2736, pruned_loss=0.06119, over 19392.00 frames. ], tot_loss[loss=0.2316, simple_loss=0.3049, pruned_loss=0.07916, over 3844164.35 frames. ], batch size: 48, lr: 7.55e-03, grad_scale: 8.0 2023-04-01 19:47:57,889 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=72344.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 19:48:05,237 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=72349.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 19:48:33,049 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.561e+02 4.924e+02 6.616e+02 8.581e+02 2.031e+03, threshold=1.323e+03, percent-clipped=3.0 2023-04-01 19:48:37,036 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=72374.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 19:48:43,655 INFO [train.py:903] (1/4) Epoch 11, batch 4100, loss[loss=0.2506, simple_loss=0.3213, pruned_loss=0.08997, over 19622.00 frames. ], tot_loss[loss=0.2327, simple_loss=0.3058, pruned_loss=0.07975, over 3845271.93 frames. ], batch size: 50, lr: 7.54e-03, grad_scale: 8.0 2023-04-01 19:49:05,249 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=72396.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 19:49:18,929 WARNING [train.py:1073] (1/4) Exclude cut with ID 497-129325-0061-9566_sp1.1_rvb from training. Duration: 0.97725 2023-04-01 19:49:26,315 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.2965, 2.9521, 2.2284, 2.7446, 1.0309, 2.9349, 2.7884, 2.9293], device='cuda:1'), covar=tensor([0.1104, 0.1439, 0.1966, 0.0915, 0.3471, 0.0986, 0.0982, 0.1326], device='cuda:1'), in_proj_covar=tensor([0.0429, 0.0368, 0.0438, 0.0318, 0.0378, 0.0370, 0.0355, 0.0392], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 19:49:48,312 INFO [train.py:903] (1/4) Epoch 11, batch 4150, loss[loss=0.2073, simple_loss=0.2938, pruned_loss=0.06042, over 19785.00 frames. ], tot_loss[loss=0.2333, simple_loss=0.3067, pruned_loss=0.0799, over 3853356.61 frames. ], batch size: 56, lr: 7.54e-03, grad_scale: 8.0 2023-04-01 19:50:42,225 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.080e+02 4.902e+02 6.015e+02 8.099e+02 1.569e+03, threshold=1.203e+03, percent-clipped=3.0 2023-04-01 19:50:51,449 INFO [train.py:903] (1/4) Epoch 11, batch 4200, loss[loss=0.2268, simple_loss=0.3009, pruned_loss=0.0764, over 19623.00 frames. ], tot_loss[loss=0.2342, simple_loss=0.3077, pruned_loss=0.08028, over 3847113.17 frames. ], batch size: 50, lr: 7.54e-03, grad_scale: 16.0 2023-04-01 19:50:57,218 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 2023-04-01 19:51:07,239 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=72491.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 19:51:38,227 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([0.9609, 1.4727, 1.9121, 1.4849, 3.0014, 4.6878, 4.4851, 4.8894], device='cuda:1'), covar=tensor([0.1702, 0.3301, 0.3016, 0.2075, 0.0528, 0.0136, 0.0149, 0.0115], device='cuda:1'), in_proj_covar=tensor([0.0257, 0.0295, 0.0323, 0.0252, 0.0215, 0.0155, 0.0207, 0.0200], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 19:51:56,568 INFO [train.py:903] (1/4) Epoch 11, batch 4250, loss[loss=0.2571, simple_loss=0.3176, pruned_loss=0.09835, over 19849.00 frames. ], tot_loss[loss=0.2341, simple_loss=0.3075, pruned_loss=0.08037, over 3851420.73 frames. ], batch size: 52, lr: 7.54e-03, grad_scale: 8.0 2023-04-01 19:52:17,361 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9_rvb from training. Duration: 29.816625 2023-04-01 19:52:28,262 WARNING [train.py:1073] (1/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 2023-04-01 19:52:28,527 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.9733, 1.1110, 1.2941, 1.1469, 2.4610, 0.8373, 1.8493, 2.7734], device='cuda:1'), covar=tensor([0.0733, 0.3037, 0.3039, 0.2018, 0.1049, 0.2754, 0.1501, 0.0470], device='cuda:1'), in_proj_covar=tensor([0.0355, 0.0339, 0.0353, 0.0322, 0.0348, 0.0334, 0.0337, 0.0354], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 19:52:51,992 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.976e+02 5.433e+02 6.415e+02 7.675e+02 1.468e+03, threshold=1.283e+03, percent-clipped=2.0 2023-04-01 19:53:01,657 INFO [train.py:903] (1/4) Epoch 11, batch 4300, loss[loss=0.208, simple_loss=0.2916, pruned_loss=0.06217, over 19389.00 frames. ], tot_loss[loss=0.2336, simple_loss=0.3072, pruned_loss=0.08002, over 3855917.61 frames. ], batch size: 48, lr: 7.53e-03, grad_scale: 8.0 2023-04-01 19:53:36,395 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=72606.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 19:54:00,212 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 2023-04-01 19:54:04,825 INFO [train.py:903] (1/4) Epoch 11, batch 4350, loss[loss=0.2666, simple_loss=0.3361, pruned_loss=0.09853, over 19766.00 frames. ], tot_loss[loss=0.2339, simple_loss=0.3074, pruned_loss=0.08016, over 3843272.60 frames. ], batch size: 56, lr: 7.53e-03, grad_scale: 8.0 2023-04-01 19:54:13,340 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.1462, 3.8036, 2.3391, 2.1733, 3.3462, 1.8550, 1.3862, 2.1353], device='cuda:1'), covar=tensor([0.1299, 0.0420, 0.0870, 0.0807, 0.0391, 0.1051, 0.1051, 0.0622], device='cuda:1'), in_proj_covar=tensor([0.0292, 0.0304, 0.0329, 0.0252, 0.0236, 0.0320, 0.0291, 0.0261], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 19:54:15,575 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3420, 1.4142, 2.1726, 1.7360, 2.9811, 2.5668, 3.0726, 1.5215], device='cuda:1'), covar=tensor([0.2410, 0.4179, 0.2244, 0.1835, 0.1678, 0.2007, 0.1993, 0.3852], device='cuda:1'), in_proj_covar=tensor([0.0489, 0.0575, 0.0601, 0.0438, 0.0592, 0.0492, 0.0646, 0.0496], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 19:54:33,935 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2307, 1.3187, 1.9185, 1.6692, 2.8647, 4.7035, 4.6612, 4.8705], device='cuda:1'), covar=tensor([0.1497, 0.3311, 0.2780, 0.1839, 0.0515, 0.0122, 0.0129, 0.0120], device='cuda:1'), in_proj_covar=tensor([0.0257, 0.0295, 0.0323, 0.0252, 0.0214, 0.0155, 0.0207, 0.0199], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 19:54:34,018 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=72652.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 19:54:55,755 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.0424, 1.2109, 1.6739, 0.9333, 2.4067, 3.3575, 3.0653, 3.5089], device='cuda:1'), covar=tensor([0.1562, 0.3287, 0.3000, 0.2323, 0.0554, 0.0153, 0.0207, 0.0203], device='cuda:1'), in_proj_covar=tensor([0.0258, 0.0297, 0.0325, 0.0253, 0.0215, 0.0156, 0.0208, 0.0200], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 19:54:58,892 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.400e+02 5.206e+02 6.524e+02 8.372e+02 1.509e+03, threshold=1.305e+03, percent-clipped=4.0 2023-04-01 19:55:06,331 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=72677.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 19:55:08,344 INFO [train.py:903] (1/4) Epoch 11, batch 4400, loss[loss=0.2151, simple_loss=0.2996, pruned_loss=0.0653, over 19346.00 frames. ], tot_loss[loss=0.2323, simple_loss=0.3058, pruned_loss=0.07939, over 3844640.95 frames. ], batch size: 66, lr: 7.53e-03, grad_scale: 8.0 2023-04-01 19:55:20,134 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=72688.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 19:55:30,983 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0134, 1.9766, 1.6876, 1.4976, 1.4301, 1.5384, 0.2874, 0.8703], device='cuda:1'), covar=tensor([0.0418, 0.0396, 0.0288, 0.0514, 0.0835, 0.0583, 0.0880, 0.0740], device='cuda:1'), in_proj_covar=tensor([0.0337, 0.0331, 0.0330, 0.0357, 0.0427, 0.0354, 0.0311, 0.0324], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 19:55:37,557 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_rvb from training. Duration: 25.285 2023-04-01 19:55:47,757 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 2023-04-01 19:56:05,992 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.7471, 1.4381, 1.3384, 1.7166, 1.4681, 1.4894, 1.4226, 1.5905], device='cuda:1'), covar=tensor([0.0965, 0.1380, 0.1455, 0.0961, 0.1205, 0.0540, 0.1211, 0.0741], device='cuda:1'), in_proj_covar=tensor([0.0253, 0.0352, 0.0289, 0.0237, 0.0294, 0.0244, 0.0275, 0.0237], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 19:56:12,262 INFO [train.py:903] (1/4) Epoch 11, batch 4450, loss[loss=0.2225, simple_loss=0.2876, pruned_loss=0.0787, over 19783.00 frames. ], tot_loss[loss=0.2312, simple_loss=0.3051, pruned_loss=0.07865, over 3839546.12 frames. ], batch size: 48, lr: 7.53e-03, grad_scale: 8.0 2023-04-01 19:57:06,051 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.681e+02 5.500e+02 6.922e+02 8.863e+02 1.965e+03, threshold=1.384e+03, percent-clipped=10.0 2023-04-01 19:57:14,262 INFO [train.py:903] (1/4) Epoch 11, batch 4500, loss[loss=0.2224, simple_loss=0.2996, pruned_loss=0.07263, over 19762.00 frames. ], tot_loss[loss=0.2313, simple_loss=0.3048, pruned_loss=0.07895, over 3843544.30 frames. ], batch size: 63, lr: 7.52e-03, grad_scale: 8.0 2023-04-01 19:57:46,549 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=72803.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 19:58:19,534 INFO [train.py:903] (1/4) Epoch 11, batch 4550, loss[loss=0.2586, simple_loss=0.3302, pruned_loss=0.09348, over 19564.00 frames. ], tot_loss[loss=0.2321, simple_loss=0.3055, pruned_loss=0.07937, over 3820259.61 frames. ], batch size: 61, lr: 7.52e-03, grad_scale: 8.0 2023-04-01 19:58:28,668 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9_rvb from training. Duration: 28.72225 2023-04-01 19:58:53,286 WARNING [train.py:1073] (1/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 2023-04-01 19:59:02,249 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=72862.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 19:59:11,462 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5868, 1.6482, 1.7755, 2.0885, 1.3762, 1.8795, 1.9374, 1.6992], device='cuda:1'), covar=tensor([0.3321, 0.2937, 0.1523, 0.1552, 0.3055, 0.1460, 0.3705, 0.2717], device='cuda:1'), in_proj_covar=tensor([0.0781, 0.0805, 0.0641, 0.0885, 0.0771, 0.0698, 0.0782, 0.0702], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 19:59:15,579 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.061e+02 5.895e+02 7.318e+02 9.053e+02 1.617e+03, threshold=1.464e+03, percent-clipped=1.0 2023-04-01 19:59:23,903 INFO [train.py:903] (1/4) Epoch 11, batch 4600, loss[loss=0.2574, simple_loss=0.3231, pruned_loss=0.09581, over 18761.00 frames. ], tot_loss[loss=0.2327, simple_loss=0.3061, pruned_loss=0.07964, over 3826321.61 frames. ], batch size: 74, lr: 7.52e-03, grad_scale: 8.0 2023-04-01 19:59:34,629 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=72887.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 19:59:59,198 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.1443, 2.2986, 2.4285, 3.3786, 2.3277, 3.4243, 2.7593, 2.3830], device='cuda:1'), covar=tensor([0.3859, 0.3371, 0.1429, 0.1771, 0.3543, 0.1285, 0.3464, 0.2506], device='cuda:1'), in_proj_covar=tensor([0.0781, 0.0804, 0.0641, 0.0887, 0.0772, 0.0698, 0.0782, 0.0701], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 20:00:06,147 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=72912.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 20:00:27,758 INFO [train.py:903] (1/4) Epoch 11, batch 4650, loss[loss=0.2668, simple_loss=0.3307, pruned_loss=0.1015, over 19666.00 frames. ], tot_loss[loss=0.2332, simple_loss=0.3067, pruned_loss=0.07982, over 3815978.98 frames. ], batch size: 53, lr: 7.52e-03, grad_scale: 8.0 2023-04-01 20:00:41,032 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([5.1675, 5.2344, 6.0827, 6.0483, 1.8051, 5.7360, 4.8444, 5.7260], device='cuda:1'), covar=tensor([0.1319, 0.0667, 0.0486, 0.0511, 0.5491, 0.0410, 0.0498, 0.1014], device='cuda:1'), in_proj_covar=tensor([0.0665, 0.0597, 0.0787, 0.0678, 0.0728, 0.0553, 0.0482, 0.0729], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1') 2023-04-01 20:00:46,733 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_rvb from training. Duration: 0.92 2023-04-01 20:00:57,790 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_rvb from training. Duration: 0.83 2023-04-01 20:01:22,039 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.839e+02 5.234e+02 6.593e+02 8.122e+02 1.324e+03, threshold=1.319e+03, percent-clipped=0.0 2023-04-01 20:01:30,180 INFO [train.py:903] (1/4) Epoch 11, batch 4700, loss[loss=0.2043, simple_loss=0.2741, pruned_loss=0.06726, over 19400.00 frames. ], tot_loss[loss=0.2348, simple_loss=0.3078, pruned_loss=0.08085, over 3802148.01 frames. ], batch size: 48, lr: 7.51e-03, grad_scale: 8.0 2023-04-01 20:01:34,988 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=72982.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 20:01:55,717 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9_rvb from training. Duration: 0.92225 2023-04-01 20:02:34,869 INFO [train.py:903] (1/4) Epoch 11, batch 4750, loss[loss=0.2175, simple_loss=0.2939, pruned_loss=0.07054, over 19487.00 frames. ], tot_loss[loss=0.236, simple_loss=0.309, pruned_loss=0.08147, over 3808863.57 frames. ], batch size: 49, lr: 7.51e-03, grad_scale: 8.0 2023-04-01 20:03:11,999 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=73059.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 20:03:28,794 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.570e+02 5.728e+02 6.958e+02 8.518e+02 2.013e+03, threshold=1.392e+03, percent-clipped=2.0 2023-04-01 20:03:37,831 INFO [train.py:903] (1/4) Epoch 11, batch 4800, loss[loss=0.256, simple_loss=0.3317, pruned_loss=0.09015, over 18092.00 frames. ], tot_loss[loss=0.237, simple_loss=0.31, pruned_loss=0.082, over 3805525.81 frames. ], batch size: 83, lr: 7.51e-03, grad_scale: 8.0 2023-04-01 20:03:38,473 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 2023-04-01 20:03:44,170 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=73084.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 20:04:18,432 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.9728, 3.6437, 2.3601, 3.2593, 0.9793, 3.4642, 3.4139, 3.4426], device='cuda:1'), covar=tensor([0.0786, 0.1074, 0.1995, 0.0754, 0.3653, 0.0761, 0.0865, 0.1168], device='cuda:1'), in_proj_covar=tensor([0.0430, 0.0364, 0.0433, 0.0315, 0.0376, 0.0370, 0.0356, 0.0389], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 20:04:26,338 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5582, 1.6186, 2.3342, 1.8291, 3.1047, 2.6476, 3.2115, 1.6282], device='cuda:1'), covar=tensor([0.2317, 0.4222, 0.2488, 0.1873, 0.1703, 0.2033, 0.1893, 0.3741], device='cuda:1'), in_proj_covar=tensor([0.0490, 0.0577, 0.0601, 0.0438, 0.0591, 0.0493, 0.0645, 0.0497], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 20:04:40,997 INFO [train.py:903] (1/4) Epoch 11, batch 4850, loss[loss=0.2028, simple_loss=0.2819, pruned_loss=0.06183, over 19585.00 frames. ], tot_loss[loss=0.2364, simple_loss=0.3092, pruned_loss=0.0818, over 3803042.81 frames. ], batch size: 52, lr: 7.51e-03, grad_scale: 8.0 2023-04-01 20:05:02,016 WARNING [train.py:1073] (1/4) Exclude cut with ID 774-127930-0014-48411_sp1.1_rvb from training. Duration: 0.95 2023-04-01 20:05:22,856 WARNING [train.py:1073] (1/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1_rvb from training. Duration: 0.9409375 2023-04-01 20:05:29,936 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 2023-04-01 20:05:29,969 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 2023-04-01 20:05:35,782 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.053e+02 5.592e+02 6.791e+02 8.044e+02 1.982e+03, threshold=1.358e+03, percent-clipped=5.0 2023-04-01 20:05:39,486 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1_rvb from training. Duration: 25.3818125 2023-04-01 20:05:44,089 INFO [train.py:903] (1/4) Epoch 11, batch 4900, loss[loss=0.2639, simple_loss=0.3321, pruned_loss=0.09788, over 19390.00 frames. ], tot_loss[loss=0.2362, simple_loss=0.3094, pruned_loss=0.08156, over 3806893.45 frames. ], batch size: 70, lr: 7.50e-03, grad_scale: 8.0 2023-04-01 20:05:57,064 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.9904, 1.6556, 1.9153, 1.8655, 4.3893, 1.0794, 2.5664, 4.7202], device='cuda:1'), covar=tensor([0.0388, 0.2624, 0.2594, 0.1762, 0.0730, 0.2676, 0.1235, 0.0209], device='cuda:1'), in_proj_covar=tensor([0.0355, 0.0337, 0.0352, 0.0321, 0.0344, 0.0332, 0.0331, 0.0351], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 20:06:01,219 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 2023-04-01 20:06:48,396 INFO [train.py:903] (1/4) Epoch 11, batch 4950, loss[loss=0.2067, simple_loss=0.2805, pruned_loss=0.06644, over 19565.00 frames. ], tot_loss[loss=0.236, simple_loss=0.3091, pruned_loss=0.08142, over 3810293.98 frames. ], batch size: 52, lr: 7.50e-03, grad_scale: 8.0 2023-04-01 20:07:00,959 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 2023-04-01 20:07:21,264 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=73256.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 20:07:24,681 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 2023-04-01 20:07:42,672 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.726e+02 5.448e+02 7.218e+02 1.051e+03 2.092e+03, threshold=1.444e+03, percent-clipped=5.0 2023-04-01 20:07:51,007 INFO [train.py:903] (1/4) Epoch 11, batch 5000, loss[loss=0.2048, simple_loss=0.2683, pruned_loss=0.07062, over 19388.00 frames. ], tot_loss[loss=0.2358, simple_loss=0.3088, pruned_loss=0.08142, over 3821995.75 frames. ], batch size: 47, lr: 7.50e-03, grad_scale: 8.0 2023-04-01 20:07:55,895 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_rvb from training. Duration: 27.14 2023-04-01 20:08:08,611 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 2023-04-01 20:08:13,759 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9847, 1.5738, 1.9887, 2.7346, 1.9473, 2.2631, 2.5739, 2.2542], device='cuda:1'), covar=tensor([0.0871, 0.1051, 0.0986, 0.0903, 0.0956, 0.0764, 0.0910, 0.0626], device='cuda:1'), in_proj_covar=tensor([0.0211, 0.0224, 0.0223, 0.0250, 0.0235, 0.0213, 0.0194, 0.0201], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-01 20:08:32,066 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=73311.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 20:08:50,651 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=73326.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 20:08:54,847 INFO [train.py:903] (1/4) Epoch 11, batch 5050, loss[loss=0.1788, simple_loss=0.257, pruned_loss=0.05034, over 19325.00 frames. ], tot_loss[loss=0.235, simple_loss=0.3081, pruned_loss=0.08095, over 3821211.77 frames. ], batch size: 44, lr: 7.50e-03, grad_scale: 8.0 2023-04-01 20:09:28,973 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9_rvb from training. Duration: 26.62775 2023-04-01 20:09:47,572 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=73371.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 20:09:48,289 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.996e+02 5.337e+02 6.487e+02 8.820e+02 1.986e+03, threshold=1.297e+03, percent-clipped=3.0 2023-04-01 20:09:56,476 INFO [train.py:903] (1/4) Epoch 11, batch 5100, loss[loss=0.2409, simple_loss=0.3163, pruned_loss=0.08277, over 19309.00 frames. ], tot_loss[loss=0.235, simple_loss=0.3076, pruned_loss=0.0812, over 3822192.79 frames. ], batch size: 66, lr: 7.49e-03, grad_scale: 8.0 2023-04-01 20:10:04,903 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 2023-04-01 20:10:10,258 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_rvb from training. Duration: 25.035 2023-04-01 20:10:13,763 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_rvb from training. Duration: 27.92 2023-04-01 20:10:22,130 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.8889, 4.4531, 2.7375, 3.9421, 0.9799, 4.2367, 4.1730, 4.3505], device='cuda:1'), covar=tensor([0.0540, 0.0843, 0.1782, 0.0621, 0.3836, 0.0657, 0.0707, 0.0808], device='cuda:1'), in_proj_covar=tensor([0.0432, 0.0363, 0.0437, 0.0317, 0.0379, 0.0373, 0.0357, 0.0390], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 20:11:00,238 INFO [train.py:903] (1/4) Epoch 11, batch 5150, loss[loss=0.2244, simple_loss=0.3054, pruned_loss=0.07175, over 19673.00 frames. ], tot_loss[loss=0.2355, simple_loss=0.3082, pruned_loss=0.08142, over 3818617.84 frames. ], batch size: 60, lr: 7.49e-03, grad_scale: 8.0 2023-04-01 20:11:09,658 WARNING [train.py:1073] (1/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 2023-04-01 20:11:16,047 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=73441.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 20:11:35,124 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.4595, 1.0707, 1.3141, 1.2981, 2.1188, 0.9833, 1.8337, 2.2861], device='cuda:1'), covar=tensor([0.0686, 0.2644, 0.2575, 0.1458, 0.0890, 0.1971, 0.1039, 0.0516], device='cuda:1'), in_proj_covar=tensor([0.0353, 0.0334, 0.0348, 0.0320, 0.0344, 0.0331, 0.0328, 0.0350], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 20:11:43,059 WARNING [train.py:1073] (1/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1_rvb from training. Duration: 0.9681875 2023-04-01 20:11:50,316 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1508, 1.0064, 1.0854, 1.3304, 1.0322, 1.2048, 1.3194, 1.1896], device='cuda:1'), covar=tensor([0.0899, 0.1086, 0.1117, 0.0690, 0.0917, 0.0856, 0.0831, 0.0774], device='cuda:1'), in_proj_covar=tensor([0.0211, 0.0226, 0.0225, 0.0250, 0.0237, 0.0214, 0.0194, 0.0202], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-01 20:11:54,731 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.282e+02 5.910e+02 7.077e+02 9.438e+02 2.777e+03, threshold=1.415e+03, percent-clipped=4.0 2023-04-01 20:12:04,016 INFO [train.py:903] (1/4) Epoch 11, batch 5200, loss[loss=0.225, simple_loss=0.2938, pruned_loss=0.07806, over 19851.00 frames. ], tot_loss[loss=0.2349, simple_loss=0.3079, pruned_loss=0.08099, over 3817401.96 frames. ], batch size: 52, lr: 7.49e-03, grad_scale: 8.0 2023-04-01 20:12:16,632 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 2023-04-01 20:13:00,064 WARNING [train.py:1073] (1/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 2023-04-01 20:13:07,090 INFO [train.py:903] (1/4) Epoch 11, batch 5250, loss[loss=0.2193, simple_loss=0.3003, pruned_loss=0.06917, over 19533.00 frames. ], tot_loss[loss=0.2353, simple_loss=0.308, pruned_loss=0.08127, over 3811090.40 frames. ], batch size: 56, lr: 7.49e-03, grad_scale: 8.0 2023-04-01 20:13:16,323 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 2023-04-01 20:13:40,014 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 2023-04-01 20:14:02,587 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.872e+02 5.338e+02 6.658e+02 8.757e+02 1.866e+03, threshold=1.332e+03, percent-clipped=3.0 2023-04-01 20:14:11,100 INFO [train.py:903] (1/4) Epoch 11, batch 5300, loss[loss=0.2458, simple_loss=0.3219, pruned_loss=0.08489, over 19663.00 frames. ], tot_loss[loss=0.2344, simple_loss=0.3071, pruned_loss=0.08087, over 3819937.79 frames. ], batch size: 60, lr: 7.48e-03, grad_scale: 8.0 2023-04-01 20:14:14,915 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.7295, 1.8005, 2.0034, 2.4602, 1.6271, 2.2031, 2.2495, 1.9211], device='cuda:1'), covar=tensor([0.3411, 0.2894, 0.1400, 0.1480, 0.3168, 0.1473, 0.3406, 0.2592], device='cuda:1'), in_proj_covar=tensor([0.0788, 0.0808, 0.0646, 0.0889, 0.0777, 0.0702, 0.0786, 0.0710], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 20:14:26,093 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 2023-04-01 20:14:29,974 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.2669, 2.9467, 2.1314, 2.1529, 1.7771, 2.3311, 0.7757, 2.1388], device='cuda:1'), covar=tensor([0.0444, 0.0420, 0.0565, 0.0756, 0.0888, 0.0806, 0.0965, 0.0780], device='cuda:1'), in_proj_covar=tensor([0.0334, 0.0326, 0.0329, 0.0348, 0.0424, 0.0349, 0.0304, 0.0318], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 20:14:43,174 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.2881, 3.7651, 3.8613, 3.8910, 1.4307, 3.6647, 3.1857, 3.5687], device='cuda:1'), covar=tensor([0.1417, 0.0820, 0.0659, 0.0658, 0.5054, 0.0688, 0.0700, 0.1173], device='cuda:1'), in_proj_covar=tensor([0.0666, 0.0600, 0.0794, 0.0678, 0.0724, 0.0553, 0.0485, 0.0730], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1') 2023-04-01 20:14:49,183 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=73608.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 20:15:13,628 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=73627.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 20:15:15,331 INFO [train.py:903] (1/4) Epoch 11, batch 5350, loss[loss=0.2223, simple_loss=0.2957, pruned_loss=0.07443, over 19865.00 frames. ], tot_loss[loss=0.2354, simple_loss=0.3084, pruned_loss=0.08115, over 3808814.84 frames. ], batch size: 52, lr: 7.48e-03, grad_scale: 8.0 2023-04-01 20:15:46,116 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=73652.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 20:15:46,889 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9_rvb from training. Duration: 26.438875 2023-04-01 20:15:49,354 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=73655.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 20:16:09,901 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.174e+02 5.285e+02 6.573e+02 8.256e+02 1.333e+03, threshold=1.315e+03, percent-clipped=1.0 2023-04-01 20:16:19,918 INFO [train.py:903] (1/4) Epoch 11, batch 5400, loss[loss=0.2765, simple_loss=0.3439, pruned_loss=0.1046, over 19733.00 frames. ], tot_loss[loss=0.2359, simple_loss=0.3087, pruned_loss=0.0816, over 3802420.02 frames. ], batch size: 63, lr: 7.48e-03, grad_scale: 8.0 2023-04-01 20:16:41,358 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=73697.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 20:17:12,958 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=73722.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 20:17:20,595 INFO [train.py:903] (1/4) Epoch 11, batch 5450, loss[loss=0.2492, simple_loss=0.3232, pruned_loss=0.08765, over 19772.00 frames. ], tot_loss[loss=0.236, simple_loss=0.3086, pruned_loss=0.08171, over 3815431.83 frames. ], batch size: 56, lr: 7.48e-03, grad_scale: 8.0 2023-04-01 20:17:45,851 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=73750.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 20:18:12,062 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=73770.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 20:18:14,011 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.246e+02 5.707e+02 7.323e+02 9.356e+02 2.024e+03, threshold=1.465e+03, percent-clipped=8.0 2023-04-01 20:18:23,354 INFO [train.py:903] (1/4) Epoch 11, batch 5500, loss[loss=0.2122, simple_loss=0.2736, pruned_loss=0.07546, over 19756.00 frames. ], tot_loss[loss=0.2371, simple_loss=0.3095, pruned_loss=0.08236, over 3820381.50 frames. ], batch size: 47, lr: 7.47e-03, grad_scale: 8.0 2023-04-01 20:18:50,424 WARNING [train.py:1073] (1/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9_rvb from training. Duration: 27.47775 2023-04-01 20:19:25,479 INFO [train.py:903] (1/4) Epoch 11, batch 5550, loss[loss=0.2167, simple_loss=0.2978, pruned_loss=0.06774, over 19560.00 frames. ], tot_loss[loss=0.2372, simple_loss=0.3097, pruned_loss=0.08235, over 3832126.66 frames. ], batch size: 61, lr: 7.47e-03, grad_scale: 8.0 2023-04-01 20:19:35,123 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 2023-04-01 20:20:04,780 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2573, 1.3375, 1.7899, 1.4219, 2.7364, 3.4745, 3.3302, 3.7235], device='cuda:1'), covar=tensor([0.1560, 0.3306, 0.2851, 0.2047, 0.0498, 0.0284, 0.0196, 0.0192], device='cuda:1'), in_proj_covar=tensor([0.0255, 0.0292, 0.0321, 0.0248, 0.0212, 0.0156, 0.0205, 0.0200], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 20:20:20,659 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.349e+02 5.761e+02 6.752e+02 8.536e+02 1.570e+03, threshold=1.350e+03, percent-clipped=1.0 2023-04-01 20:20:26,343 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 2023-04-01 20:20:29,856 INFO [train.py:903] (1/4) Epoch 11, batch 5600, loss[loss=0.254, simple_loss=0.3314, pruned_loss=0.08834, over 19692.00 frames. ], tot_loss[loss=0.2376, simple_loss=0.31, pruned_loss=0.08263, over 3827148.23 frames. ], batch size: 53, lr: 7.47e-03, grad_scale: 8.0 2023-04-01 20:21:24,330 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.8807, 1.7907, 1.8688, 1.4781, 4.3230, 0.8714, 2.4397, 4.6986], device='cuda:1'), covar=tensor([0.0330, 0.2410, 0.2509, 0.1995, 0.0701, 0.2837, 0.1384, 0.0189], device='cuda:1'), in_proj_covar=tensor([0.0353, 0.0338, 0.0349, 0.0322, 0.0347, 0.0331, 0.0332, 0.0353], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 20:21:34,094 INFO [train.py:903] (1/4) Epoch 11, batch 5650, loss[loss=0.1797, simple_loss=0.2473, pruned_loss=0.05602, over 19723.00 frames. ], tot_loss[loss=0.2352, simple_loss=0.308, pruned_loss=0.08117, over 3830317.33 frames. ], batch size: 45, lr: 7.47e-03, grad_scale: 8.0 2023-04-01 20:22:03,021 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=73952.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 20:22:23,877 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9_rvb from training. Duration: 28.0944375 2023-04-01 20:22:28,536 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.512e+02 5.880e+02 7.006e+02 8.632e+02 1.687e+03, threshold=1.401e+03, percent-clipped=2.0 2023-04-01 20:22:37,801 INFO [train.py:903] (1/4) Epoch 11, batch 5700, loss[loss=0.2421, simple_loss=0.311, pruned_loss=0.08659, over 19619.00 frames. ], tot_loss[loss=0.2338, simple_loss=0.3067, pruned_loss=0.08043, over 3844626.47 frames. ], batch size: 50, lr: 7.46e-03, grad_scale: 8.0 2023-04-01 20:23:13,091 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3839, 1.4602, 1.7144, 1.6099, 2.6131, 2.2100, 2.6981, 1.0831], device='cuda:1'), covar=tensor([0.2204, 0.3691, 0.2262, 0.1649, 0.1353, 0.1810, 0.1331, 0.3591], device='cuda:1'), in_proj_covar=tensor([0.0492, 0.0574, 0.0599, 0.0436, 0.0594, 0.0492, 0.0645, 0.0494], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 20:23:38,892 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=74026.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 20:23:42,027 INFO [train.py:903] (1/4) Epoch 11, batch 5750, loss[loss=0.2955, simple_loss=0.3462, pruned_loss=0.1224, over 12881.00 frames. ], tot_loss[loss=0.2341, simple_loss=0.3068, pruned_loss=0.08072, over 3825114.84 frames. ], batch size: 136, lr: 7.46e-03, grad_scale: 8.0 2023-04-01 20:23:43,285 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9_rvb from training. Duration: 33.038875 2023-04-01 20:23:51,329 WARNING [train.py:1073] (1/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 2023-04-01 20:23:56,779 WARNING [train.py:1073] (1/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 2023-04-01 20:24:11,305 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=74051.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 20:24:30,683 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=74067.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 20:24:36,271 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.134e+02 5.413e+02 6.547e+02 7.617e+02 1.298e+03, threshold=1.309e+03, percent-clipped=0.0 2023-04-01 20:24:45,231 INFO [train.py:903] (1/4) Epoch 11, batch 5800, loss[loss=0.2322, simple_loss=0.303, pruned_loss=0.08071, over 19470.00 frames. ], tot_loss[loss=0.2351, simple_loss=0.3076, pruned_loss=0.08135, over 3835174.14 frames. ], batch size: 49, lr: 7.46e-03, grad_scale: 8.0 2023-04-01 20:24:50,599 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=74082.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 20:25:04,165 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=74094.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 20:25:46,662 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.1720, 2.7733, 2.1572, 2.1309, 1.8035, 2.2724, 0.8620, 2.0022], device='cuda:1'), covar=tensor([0.0459, 0.0463, 0.0491, 0.0770, 0.0848, 0.0836, 0.0913, 0.0765], device='cuda:1'), in_proj_covar=tensor([0.0331, 0.0325, 0.0327, 0.0346, 0.0421, 0.0344, 0.0305, 0.0318], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 20:25:48,478 INFO [train.py:903] (1/4) Epoch 11, batch 5850, loss[loss=0.2265, simple_loss=0.3019, pruned_loss=0.0755, over 19835.00 frames. ], tot_loss[loss=0.2351, simple_loss=0.3078, pruned_loss=0.08122, over 3830003.13 frames. ], batch size: 52, lr: 7.46e-03, grad_scale: 8.0 2023-04-01 20:25:52,588 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.53 vs. limit=2.0 2023-04-01 20:26:04,396 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0 2023-04-01 20:26:42,000 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.778e+02 5.734e+02 7.251e+02 9.100e+02 2.751e+03, threshold=1.450e+03, percent-clipped=7.0 2023-04-01 20:26:51,315 INFO [train.py:903] (1/4) Epoch 11, batch 5900, loss[loss=0.2151, simple_loss=0.2957, pruned_loss=0.06723, over 19773.00 frames. ], tot_loss[loss=0.2359, simple_loss=0.3087, pruned_loss=0.08158, over 3832335.86 frames. ], batch size: 54, lr: 7.45e-03, grad_scale: 8.0 2023-04-01 20:26:53,585 WARNING [train.py:1073] (1/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 2023-04-01 20:27:15,056 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9_rvb from training. Duration: 27.511125 2023-04-01 20:27:30,982 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=74209.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 20:27:55,166 INFO [train.py:903] (1/4) Epoch 11, batch 5950, loss[loss=0.2174, simple_loss=0.2827, pruned_loss=0.07612, over 19783.00 frames. ], tot_loss[loss=0.2345, simple_loss=0.3074, pruned_loss=0.08078, over 3842216.63 frames. ], batch size: 48, lr: 7.45e-03, grad_scale: 8.0 2023-04-01 20:28:33,384 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=74258.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 20:28:49,682 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.166e+02 4.883e+02 6.160e+02 7.607e+02 1.521e+03, threshold=1.232e+03, percent-clipped=1.0 2023-04-01 20:28:59,182 INFO [train.py:903] (1/4) Epoch 11, batch 6000, loss[loss=0.1937, simple_loss=0.2655, pruned_loss=0.0609, over 19732.00 frames. ], tot_loss[loss=0.2359, simple_loss=0.3088, pruned_loss=0.0815, over 3845124.50 frames. ], batch size: 45, lr: 7.45e-03, grad_scale: 8.0 2023-04-01 20:28:59,182 INFO [train.py:928] (1/4) Computing validation loss 2023-04-01 20:29:11,906 INFO [train.py:937] (1/4) Epoch 11, validation: loss=0.1778, simple_loss=0.2787, pruned_loss=0.03847, over 944034.00 frames. 2023-04-01 20:29:11,909 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18649MB 2023-04-01 20:30:09,755 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=74323.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 20:30:18,373 INFO [train.py:903] (1/4) Epoch 11, batch 6050, loss[loss=0.2115, simple_loss=0.295, pruned_loss=0.06398, over 19595.00 frames. ], tot_loss[loss=0.2353, simple_loss=0.3085, pruned_loss=0.08108, over 3829507.08 frames. ], batch size: 52, lr: 7.45e-03, grad_scale: 8.0 2023-04-01 20:30:28,827 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.7886, 4.2245, 4.5171, 4.4875, 1.6450, 4.2265, 3.6146, 4.1792], device='cuda:1'), covar=tensor([0.1489, 0.0955, 0.0534, 0.0603, 0.5301, 0.0645, 0.0642, 0.1069], device='cuda:1'), in_proj_covar=tensor([0.0672, 0.0600, 0.0800, 0.0682, 0.0725, 0.0557, 0.0485, 0.0732], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1') 2023-04-01 20:30:42,862 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=74348.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 20:31:12,598 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.290e+02 5.870e+02 7.295e+02 8.913e+02 1.728e+03, threshold=1.459e+03, percent-clipped=8.0 2023-04-01 20:31:21,969 INFO [train.py:903] (1/4) Epoch 11, batch 6100, loss[loss=0.2313, simple_loss=0.3077, pruned_loss=0.07742, over 19768.00 frames. ], tot_loss[loss=0.2352, simple_loss=0.3085, pruned_loss=0.08098, over 3824583.49 frames. ], batch size: 56, lr: 7.44e-03, grad_scale: 8.0 2023-04-01 20:31:56,338 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.0649, 1.2609, 1.4635, 1.2784, 2.6128, 0.9047, 1.9059, 2.8831], device='cuda:1'), covar=tensor([0.0523, 0.2476, 0.2479, 0.1677, 0.0759, 0.2386, 0.1217, 0.0362], device='cuda:1'), in_proj_covar=tensor([0.0353, 0.0334, 0.0347, 0.0317, 0.0344, 0.0330, 0.0332, 0.0350], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 20:32:22,429 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=74426.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 20:32:25,901 INFO [train.py:903] (1/4) Epoch 11, batch 6150, loss[loss=0.1852, simple_loss=0.262, pruned_loss=0.05418, over 19649.00 frames. ], tot_loss[loss=0.2342, simple_loss=0.3079, pruned_loss=0.08026, over 3836776.55 frames. ], batch size: 50, lr: 7.44e-03, grad_scale: 8.0 2023-04-01 20:32:54,729 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9_rvb from training. Duration: 31.02225 2023-04-01 20:33:12,628 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=74465.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 20:33:20,926 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.809e+02 5.576e+02 6.382e+02 8.608e+02 2.367e+03, threshold=1.276e+03, percent-clipped=4.0 2023-04-01 20:33:29,312 INFO [train.py:903] (1/4) Epoch 11, batch 6200, loss[loss=0.2536, simple_loss=0.3286, pruned_loss=0.0893, over 19660.00 frames. ], tot_loss[loss=0.2341, simple_loss=0.3077, pruned_loss=0.08021, over 3834923.19 frames. ], batch size: 60, lr: 7.44e-03, grad_scale: 8.0 2023-04-01 20:33:43,878 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=74490.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 20:34:31,919 INFO [train.py:903] (1/4) Epoch 11, batch 6250, loss[loss=0.2485, simple_loss=0.3338, pruned_loss=0.08158, over 19555.00 frames. ], tot_loss[loss=0.2354, simple_loss=0.3089, pruned_loss=0.08092, over 3827548.96 frames. ], batch size: 56, lr: 7.44e-03, grad_scale: 8.0 2023-04-01 20:34:39,623 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.79 vs. limit=5.0 2023-04-01 20:34:48,638 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=74541.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 20:35:02,986 WARNING [train.py:1073] (1/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9_rvb from training. Duration: 25.988875 2023-04-01 20:35:09,829 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.58 vs. limit=5.0 2023-04-01 20:35:10,724 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.3536, 2.2004, 1.9055, 1.7331, 1.6867, 1.7844, 0.4114, 1.1501], device='cuda:1'), covar=tensor([0.0403, 0.0415, 0.0340, 0.0548, 0.0875, 0.0618, 0.0903, 0.0786], device='cuda:1'), in_proj_covar=tensor([0.0338, 0.0332, 0.0333, 0.0353, 0.0427, 0.0352, 0.0308, 0.0325], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 20:35:30,740 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.915e+02 5.334e+02 6.699e+02 8.918e+02 1.691e+03, threshold=1.340e+03, percent-clipped=6.0 2023-04-01 20:35:38,658 INFO [train.py:903] (1/4) Epoch 11, batch 6300, loss[loss=0.2868, simple_loss=0.346, pruned_loss=0.1137, over 17502.00 frames. ], tot_loss[loss=0.234, simple_loss=0.3076, pruned_loss=0.0802, over 3826709.76 frames. ], batch size: 101, lr: 7.43e-03, grad_scale: 8.0 2023-04-01 20:36:06,829 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=74602.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 20:36:21,235 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=74613.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 20:36:40,491 INFO [train.py:903] (1/4) Epoch 11, batch 6350, loss[loss=0.2276, simple_loss=0.2918, pruned_loss=0.08172, over 19701.00 frames. ], tot_loss[loss=0.2335, simple_loss=0.3071, pruned_loss=0.07995, over 3823599.97 frames. ], batch size: 45, lr: 7.43e-03, grad_scale: 8.0 2023-04-01 20:37:16,964 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4994, 1.2954, 1.3364, 1.9378, 1.4798, 1.7205, 1.9036, 1.6131], device='cuda:1'), covar=tensor([0.0859, 0.1059, 0.1090, 0.0821, 0.0922, 0.0740, 0.0786, 0.0695], device='cuda:1'), in_proj_covar=tensor([0.0210, 0.0225, 0.0223, 0.0246, 0.0236, 0.0212, 0.0194, 0.0200], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-01 20:37:36,488 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.847e+02 5.622e+02 6.880e+02 9.256e+02 3.776e+03, threshold=1.376e+03, percent-clipped=5.0 2023-04-01 20:37:44,831 INFO [train.py:903] (1/4) Epoch 11, batch 6400, loss[loss=0.2359, simple_loss=0.3163, pruned_loss=0.07779, over 19661.00 frames. ], tot_loss[loss=0.2354, simple_loss=0.3089, pruned_loss=0.08098, over 3804050.28 frames. ], batch size: 55, lr: 7.43e-03, grad_scale: 8.0 2023-04-01 20:38:33,474 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=74717.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 20:38:37,067 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1469, 1.1879, 1.3106, 1.2918, 1.6523, 1.6155, 1.5580, 0.5538], device='cuda:1'), covar=tensor([0.1919, 0.3344, 0.1972, 0.1587, 0.1323, 0.1784, 0.1253, 0.3582], device='cuda:1'), in_proj_covar=tensor([0.0485, 0.0569, 0.0596, 0.0435, 0.0589, 0.0486, 0.0640, 0.0488], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 20:38:44,787 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.5125, 1.3450, 1.4512, 1.6140, 3.0383, 1.1549, 2.3268, 3.4259], device='cuda:1'), covar=tensor([0.0416, 0.2588, 0.2675, 0.1679, 0.0693, 0.2400, 0.1208, 0.0285], device='cuda:1'), in_proj_covar=tensor([0.0358, 0.0338, 0.0351, 0.0320, 0.0348, 0.0335, 0.0337, 0.0355], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 20:38:46,860 INFO [train.py:903] (1/4) Epoch 11, batch 6450, loss[loss=0.1977, simple_loss=0.2708, pruned_loss=0.06229, over 19004.00 frames. ], tot_loss[loss=0.2355, simple_loss=0.3088, pruned_loss=0.08108, over 3814734.15 frames. ], batch size: 42, lr: 7.43e-03, grad_scale: 8.0 2023-04-01 20:39:28,526 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9_rvb from training. Duration: 28.638875 2023-04-01 20:39:44,916 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.542e+02 5.590e+02 7.165e+02 9.175e+02 2.194e+03, threshold=1.433e+03, percent-clipped=7.0 2023-04-01 20:39:53,342 INFO [train.py:903] (1/4) Epoch 11, batch 6500, loss[loss=0.2354, simple_loss=0.2979, pruned_loss=0.08647, over 19465.00 frames. ], tot_loss[loss=0.2348, simple_loss=0.3077, pruned_loss=0.08093, over 3816220.50 frames. ], batch size: 49, lr: 7.42e-03, grad_scale: 8.0 2023-04-01 20:39:54,533 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1_rvb from training. Duration: 0.836375 2023-04-01 20:40:16,077 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=74797.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 20:40:48,432 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=74822.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 20:40:52,517 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.81 vs. limit=5.0 2023-04-01 20:40:54,505 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.8111, 1.9378, 2.0920, 2.6573, 1.7638, 2.3609, 2.3395, 1.9684], device='cuda:1'), covar=tensor([0.3376, 0.2804, 0.1364, 0.1597, 0.3223, 0.1452, 0.3191, 0.2483], device='cuda:1'), in_proj_covar=tensor([0.0787, 0.0810, 0.0641, 0.0890, 0.0778, 0.0702, 0.0782, 0.0705], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 20:40:56,363 INFO [train.py:903] (1/4) Epoch 11, batch 6550, loss[loss=0.2389, simple_loss=0.3176, pruned_loss=0.08014, over 19609.00 frames. ], tot_loss[loss=0.235, simple_loss=0.308, pruned_loss=0.081, over 3801570.44 frames. ], batch size: 61, lr: 7.42e-03, grad_scale: 8.0 2023-04-01 20:41:52,115 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.386e+02 5.036e+02 6.201e+02 8.020e+02 1.527e+03, threshold=1.240e+03, percent-clipped=1.0 2023-04-01 20:41:59,122 INFO [train.py:903] (1/4) Epoch 11, batch 6600, loss[loss=0.2217, simple_loss=0.2988, pruned_loss=0.07235, over 19677.00 frames. ], tot_loss[loss=0.2371, simple_loss=0.3097, pruned_loss=0.08222, over 3795823.55 frames. ], batch size: 58, lr: 7.42e-03, grad_scale: 8.0 2023-04-01 20:42:46,208 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3490, 1.4467, 1.7504, 1.6132, 2.7018, 2.2505, 2.7963, 1.1662], device='cuda:1'), covar=tensor([0.2230, 0.3718, 0.2308, 0.1751, 0.1371, 0.1842, 0.1459, 0.3651], device='cuda:1'), in_proj_covar=tensor([0.0496, 0.0577, 0.0604, 0.0441, 0.0597, 0.0493, 0.0651, 0.0495], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 20:43:01,234 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=74927.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 20:43:03,143 INFO [train.py:903] (1/4) Epoch 11, batch 6650, loss[loss=0.3372, simple_loss=0.3817, pruned_loss=0.1464, over 13293.00 frames. ], tot_loss[loss=0.2348, simple_loss=0.3077, pruned_loss=0.08093, over 3806705.86 frames. ], batch size: 136, lr: 7.42e-03, grad_scale: 8.0 2023-04-01 20:43:32,770 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9422, 1.7564, 1.5898, 2.0534, 1.9304, 1.7712, 1.6526, 1.9556], device='cuda:1'), covar=tensor([0.0928, 0.1482, 0.1318, 0.0836, 0.1093, 0.0491, 0.1126, 0.0662], device='cuda:1'), in_proj_covar=tensor([0.0255, 0.0352, 0.0291, 0.0239, 0.0296, 0.0242, 0.0279, 0.0238], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 20:43:40,227 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=74957.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 20:44:00,890 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.349e+02 5.506e+02 7.046e+02 9.043e+02 1.602e+03, threshold=1.409e+03, percent-clipped=2.0 2023-04-01 20:44:01,337 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=74973.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 20:44:07,814 INFO [train.py:903] (1/4) Epoch 11, batch 6700, loss[loss=0.2007, simple_loss=0.2774, pruned_loss=0.06199, over 19853.00 frames. ], tot_loss[loss=0.2347, simple_loss=0.3076, pruned_loss=0.08089, over 3813551.85 frames. ], batch size: 52, lr: 7.41e-03, grad_scale: 8.0 2023-04-01 20:44:12,436 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=74982.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 20:44:31,776 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=74998.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 20:45:06,535 INFO [train.py:903] (1/4) Epoch 11, batch 6750, loss[loss=0.2389, simple_loss=0.32, pruned_loss=0.07885, over 19666.00 frames. ], tot_loss[loss=0.2354, simple_loss=0.3084, pruned_loss=0.08125, over 3822469.61 frames. ], batch size: 58, lr: 7.41e-03, grad_scale: 8.0 2023-04-01 20:45:30,092 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 2023-04-01 20:45:55,351 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=75072.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 20:45:56,174 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.565e+02 5.649e+02 7.002e+02 8.709e+02 1.691e+03, threshold=1.400e+03, percent-clipped=2.0 2023-04-01 20:46:04,211 INFO [train.py:903] (1/4) Epoch 11, batch 6800, loss[loss=0.2435, simple_loss=0.3237, pruned_loss=0.08164, over 19754.00 frames. ], tot_loss[loss=0.2351, simple_loss=0.3083, pruned_loss=0.08098, over 3833737.23 frames. ], batch size: 54, lr: 7.41e-03, grad_scale: 8.0 2023-04-01 20:46:52,333 WARNING [train.py:1073] (1/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 2023-04-01 20:46:53,786 WARNING [train.py:1073] (1/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 2023-04-01 20:46:56,127 INFO [train.py:903] (1/4) Epoch 12, batch 0, loss[loss=0.259, simple_loss=0.3076, pruned_loss=0.1052, over 19754.00 frames. ], tot_loss[loss=0.259, simple_loss=0.3076, pruned_loss=0.1052, over 19754.00 frames. ], batch size: 46, lr: 7.10e-03, grad_scale: 8.0 2023-04-01 20:46:56,128 INFO [train.py:928] (1/4) Computing validation loss 2023-04-01 20:47:06,643 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.5869, 3.2801, 2.6680, 3.0217, 1.2601, 2.9903, 3.0119, 3.1006], device='cuda:1'), covar=tensor([0.0863, 0.0814, 0.1619, 0.0821, 0.3106, 0.1114, 0.0823, 0.1220], device='cuda:1'), in_proj_covar=tensor([0.0430, 0.0361, 0.0432, 0.0312, 0.0374, 0.0364, 0.0352, 0.0385], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-01 20:47:08,133 INFO [train.py:937] (1/4) Epoch 12, validation: loss=0.1777, simple_loss=0.2788, pruned_loss=0.03825, over 944034.00 frames. 2023-04-01 20:47:08,134 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18649MB 2023-04-01 20:47:20,786 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 2023-04-01 20:48:11,513 INFO [train.py:903] (1/4) Epoch 12, batch 50, loss[loss=0.1975, simple_loss=0.2679, pruned_loss=0.06353, over 19715.00 frames. ], tot_loss[loss=0.2316, simple_loss=0.3051, pruned_loss=0.07898, over 870432.97 frames. ], batch size: 46, lr: 7.09e-03, grad_scale: 8.0 2023-04-01 20:48:20,847 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3850, 1.2088, 0.9842, 1.2445, 1.1102, 1.1649, 0.9843, 1.1791], device='cuda:1'), covar=tensor([0.1134, 0.1224, 0.1764, 0.1072, 0.1315, 0.0974, 0.1657, 0.1046], device='cuda:1'), in_proj_covar=tensor([0.0254, 0.0352, 0.0291, 0.0239, 0.0296, 0.0243, 0.0279, 0.0239], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 20:48:29,477 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.226e+02 5.251e+02 6.823e+02 1.011e+03 3.055e+03, threshold=1.365e+03, percent-clipped=9.0 2023-04-01 20:48:41,946 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1_rvb from training. Duration: 27.0318125 2023-04-01 20:49:13,868 INFO [train.py:903] (1/4) Epoch 12, batch 100, loss[loss=0.231, simple_loss=0.3131, pruned_loss=0.07449, over 19301.00 frames. ], tot_loss[loss=0.2358, simple_loss=0.309, pruned_loss=0.08127, over 1513017.50 frames. ], batch size: 70, lr: 7.09e-03, grad_scale: 8.0 2023-04-01 20:49:22,066 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 2023-04-01 20:50:02,907 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=75246.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 20:50:16,771 INFO [train.py:903] (1/4) Epoch 12, batch 150, loss[loss=0.2305, simple_loss=0.3013, pruned_loss=0.07981, over 19584.00 frames. ], tot_loss[loss=0.2324, simple_loss=0.3066, pruned_loss=0.07909, over 2025783.63 frames. ], batch size: 52, lr: 7.09e-03, grad_scale: 8.0 2023-04-01 20:50:34,012 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=75271.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 20:50:36,118 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.416e+02 5.206e+02 6.373e+02 8.343e+02 1.576e+03, threshold=1.275e+03, percent-clipped=5.0 2023-04-01 20:50:56,161 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=75288.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 20:51:15,326 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 2023-04-01 20:51:18,693 INFO [train.py:903] (1/4) Epoch 12, batch 200, loss[loss=0.2296, simple_loss=0.3059, pruned_loss=0.07663, over 19594.00 frames. ], tot_loss[loss=0.2317, simple_loss=0.3058, pruned_loss=0.07885, over 2429406.21 frames. ], batch size: 61, lr: 7.09e-03, grad_scale: 8.0 2023-04-01 20:51:42,634 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=75326.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 20:51:45,339 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=75328.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 20:51:46,573 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.8244, 1.9223, 2.1090, 2.6438, 1.7848, 2.4406, 2.3355, 1.9488], device='cuda:1'), covar=tensor([0.3640, 0.2924, 0.1443, 0.1652, 0.3326, 0.1470, 0.3478, 0.2689], device='cuda:1'), in_proj_covar=tensor([0.0789, 0.0808, 0.0642, 0.0891, 0.0775, 0.0703, 0.0777, 0.0703], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 20:52:16,212 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=75353.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 20:52:22,163 INFO [train.py:903] (1/4) Epoch 12, batch 250, loss[loss=0.2076, simple_loss=0.2907, pruned_loss=0.06227, over 18656.00 frames. ], tot_loss[loss=0.2318, simple_loss=0.3053, pruned_loss=0.07912, over 2748349.10 frames. ], batch size: 74, lr: 7.08e-03, grad_scale: 8.0 2023-04-01 20:52:41,710 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.835e+02 5.714e+02 6.835e+02 8.470e+02 1.829e+03, threshold=1.367e+03, percent-clipped=2.0 2023-04-01 20:52:56,669 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=75386.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 20:53:25,608 INFO [train.py:903] (1/4) Epoch 12, batch 300, loss[loss=0.1989, simple_loss=0.2812, pruned_loss=0.05833, over 19766.00 frames. ], tot_loss[loss=0.2315, simple_loss=0.3057, pruned_loss=0.07865, over 3002157.74 frames. ], batch size: 54, lr: 7.08e-03, grad_scale: 8.0 2023-04-01 20:53:34,115 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.8468, 1.4059, 1.4940, 1.6061, 3.3298, 0.9466, 2.2490, 3.7900], device='cuda:1'), covar=tensor([0.0403, 0.2634, 0.2638, 0.1723, 0.0690, 0.2705, 0.1394, 0.0231], device='cuda:1'), in_proj_covar=tensor([0.0357, 0.0340, 0.0352, 0.0320, 0.0346, 0.0332, 0.0336, 0.0356], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 20:54:07,328 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=75441.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 20:54:28,461 INFO [train.py:903] (1/4) Epoch 12, batch 350, loss[loss=0.2428, simple_loss=0.317, pruned_loss=0.08425, over 19491.00 frames. ], tot_loss[loss=0.2336, simple_loss=0.3075, pruned_loss=0.07991, over 3190840.48 frames. ], batch size: 64, lr: 7.08e-03, grad_scale: 8.0 2023-04-01 20:54:31,954 WARNING [train.py:1073] (1/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1_rvb from training. Duration: 0.9681875 2023-04-01 20:54:45,895 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.322e+02 5.463e+02 6.814e+02 8.627e+02 1.955e+03, threshold=1.363e+03, percent-clipped=3.0 2023-04-01 20:55:24,620 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 2023-04-01 20:55:30,816 INFO [train.py:903] (1/4) Epoch 12, batch 400, loss[loss=0.2978, simple_loss=0.3509, pruned_loss=0.1224, over 13349.00 frames. ], tot_loss[loss=0.2333, simple_loss=0.3072, pruned_loss=0.07974, over 3322947.77 frames. ], batch size: 136, lr: 7.08e-03, grad_scale: 8.0 2023-04-01 20:56:16,839 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4334, 1.5304, 1.8748, 1.6444, 3.0557, 2.5082, 3.2716, 1.5593], device='cuda:1'), covar=tensor([0.2190, 0.3688, 0.2203, 0.1686, 0.1379, 0.1764, 0.1491, 0.3399], device='cuda:1'), in_proj_covar=tensor([0.0492, 0.0577, 0.0604, 0.0438, 0.0598, 0.0493, 0.0646, 0.0492], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 20:56:32,366 INFO [train.py:903] (1/4) Epoch 12, batch 450, loss[loss=0.2529, simple_loss=0.3274, pruned_loss=0.08915, over 19665.00 frames. ], tot_loss[loss=0.2334, simple_loss=0.3072, pruned_loss=0.07982, over 3432249.80 frames. ], batch size: 58, lr: 7.07e-03, grad_scale: 8.0 2023-04-01 20:56:51,546 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.480e+02 5.172e+02 6.423e+02 7.976e+02 2.291e+03, threshold=1.285e+03, percent-clipped=4.0 2023-04-01 20:57:09,451 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 2023-04-01 20:57:10,692 WARNING [train.py:1073] (1/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 2023-04-01 20:57:13,224 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=75590.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 20:57:17,040 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.7832, 4.2137, 4.4169, 4.4096, 1.5617, 4.0850, 3.5753, 4.1004], device='cuda:1'), covar=tensor([0.1302, 0.0803, 0.0597, 0.0584, 0.5294, 0.0721, 0.0643, 0.1100], device='cuda:1'), in_proj_covar=tensor([0.0677, 0.0603, 0.0804, 0.0688, 0.0731, 0.0559, 0.0492, 0.0741], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1') 2023-04-01 20:57:36,544 INFO [train.py:903] (1/4) Epoch 12, batch 500, loss[loss=0.1955, simple_loss=0.2861, pruned_loss=0.05248, over 19620.00 frames. ], tot_loss[loss=0.2341, simple_loss=0.3076, pruned_loss=0.08028, over 3519313.37 frames. ], batch size: 57, lr: 7.07e-03, grad_scale: 8.0 2023-04-01 20:58:06,142 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=75632.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 20:58:17,971 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=75642.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 20:58:38,568 INFO [train.py:903] (1/4) Epoch 12, batch 550, loss[loss=0.2473, simple_loss=0.3158, pruned_loss=0.08939, over 13121.00 frames. ], tot_loss[loss=0.236, simple_loss=0.3096, pruned_loss=0.08121, over 3587916.89 frames. ], batch size: 135, lr: 7.07e-03, grad_scale: 8.0 2023-04-01 20:58:50,357 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=75667.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 20:58:56,746 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.239e+02 5.816e+02 6.939e+02 9.327e+02 2.224e+03, threshold=1.388e+03, percent-clipped=13.0 2023-04-01 20:59:06,364 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5616, 1.3789, 1.3890, 1.8494, 1.6198, 1.8934, 1.9341, 1.7212], device='cuda:1'), covar=tensor([0.0862, 0.0981, 0.1040, 0.0870, 0.0838, 0.0699, 0.0822, 0.0666], device='cuda:1'), in_proj_covar=tensor([0.0211, 0.0223, 0.0222, 0.0246, 0.0234, 0.0212, 0.0195, 0.0201], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-01 20:59:26,721 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.8241, 4.3410, 4.6048, 4.5858, 1.5873, 4.2869, 3.7436, 4.2506], device='cuda:1'), covar=tensor([0.1362, 0.0721, 0.0522, 0.0527, 0.5305, 0.0621, 0.0597, 0.1078], device='cuda:1'), in_proj_covar=tensor([0.0675, 0.0599, 0.0799, 0.0686, 0.0725, 0.0558, 0.0490, 0.0741], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1') 2023-04-01 20:59:26,878 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=75697.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 20:59:38,240 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=75705.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 20:59:41,417 INFO [train.py:903] (1/4) Epoch 12, batch 600, loss[loss=0.2344, simple_loss=0.2991, pruned_loss=0.08487, over 19759.00 frames. ], tot_loss[loss=0.2351, simple_loss=0.3091, pruned_loss=0.08052, over 3642722.59 frames. ], batch size: 47, lr: 7.07e-03, grad_scale: 8.0 2023-04-01 20:59:57,550 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=75722.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 20:59:57,728 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.48 vs. limit=5.0 2023-04-01 21:00:22,881 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 2023-04-01 21:00:30,168 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=75747.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 21:00:42,510 INFO [train.py:903] (1/4) Epoch 12, batch 650, loss[loss=0.1981, simple_loss=0.2834, pruned_loss=0.05638, over 19540.00 frames. ], tot_loss[loss=0.2356, simple_loss=0.3095, pruned_loss=0.0809, over 3661401.65 frames. ], batch size: 54, lr: 7.07e-03, grad_scale: 8.0 2023-04-01 21:00:59,398 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2708, 1.3165, 1.7744, 1.5986, 2.4985, 1.9682, 2.4958, 1.1319], device='cuda:1'), covar=tensor([0.2419, 0.4182, 0.2414, 0.1871, 0.1468, 0.2269, 0.1660, 0.3698], device='cuda:1'), in_proj_covar=tensor([0.0495, 0.0582, 0.0607, 0.0438, 0.0598, 0.0495, 0.0648, 0.0492], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 21:01:01,243 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.249e+02 5.224e+02 6.354e+02 7.929e+02 1.382e+03, threshold=1.271e+03, percent-clipped=0.0 2023-04-01 21:01:31,198 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.13 vs. limit=2.0 2023-04-01 21:01:45,565 INFO [train.py:903] (1/4) Epoch 12, batch 700, loss[loss=0.203, simple_loss=0.2782, pruned_loss=0.06396, over 19778.00 frames. ], tot_loss[loss=0.2344, simple_loss=0.3081, pruned_loss=0.08036, over 3695191.15 frames. ], batch size: 48, lr: 7.06e-03, grad_scale: 8.0 2023-04-01 21:02:10,790 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.4948, 1.6336, 1.7899, 1.7778, 3.9859, 1.1684, 2.4759, 4.0530], device='cuda:1'), covar=tensor([0.0394, 0.2562, 0.2634, 0.1841, 0.0728, 0.2722, 0.1469, 0.0277], device='cuda:1'), in_proj_covar=tensor([0.0357, 0.0341, 0.0353, 0.0321, 0.0349, 0.0333, 0.0337, 0.0356], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 21:02:45,696 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.6164, 1.3562, 1.4616, 1.5473, 3.1125, 1.0755, 2.2841, 3.4082], device='cuda:1'), covar=tensor([0.0410, 0.2492, 0.2587, 0.1752, 0.0698, 0.2502, 0.1193, 0.0293], device='cuda:1'), in_proj_covar=tensor([0.0354, 0.0338, 0.0350, 0.0318, 0.0346, 0.0330, 0.0334, 0.0353], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 21:02:46,462 INFO [train.py:903] (1/4) Epoch 12, batch 750, loss[loss=0.2364, simple_loss=0.307, pruned_loss=0.08292, over 19862.00 frames. ], tot_loss[loss=0.2324, simple_loss=0.3064, pruned_loss=0.07921, over 3726787.66 frames. ], batch size: 52, lr: 7.06e-03, grad_scale: 8.0 2023-04-01 21:03:05,207 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.592e+02 5.547e+02 6.814e+02 8.571e+02 2.504e+03, threshold=1.363e+03, percent-clipped=8.0 2023-04-01 21:03:49,642 INFO [train.py:903] (1/4) Epoch 12, batch 800, loss[loss=0.246, simple_loss=0.3213, pruned_loss=0.08541, over 19308.00 frames. ], tot_loss[loss=0.2334, simple_loss=0.307, pruned_loss=0.07989, over 3757542.86 frames. ], batch size: 66, lr: 7.06e-03, grad_scale: 8.0 2023-04-01 21:04:07,045 WARNING [train.py:1073] (1/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 2023-04-01 21:04:50,833 INFO [train.py:903] (1/4) Epoch 12, batch 850, loss[loss=0.226, simple_loss=0.3039, pruned_loss=0.07401, over 19538.00 frames. ], tot_loss[loss=0.2336, simple_loss=0.3071, pruned_loss=0.0801, over 3773780.97 frames. ], batch size: 54, lr: 7.06e-03, grad_scale: 8.0 2023-04-01 21:04:54,860 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=75961.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 21:05:10,028 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.002e+02 5.208e+02 6.361e+02 7.722e+02 1.579e+03, threshold=1.272e+03, percent-clipped=2.0 2023-04-01 21:05:25,769 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=75986.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 21:05:29,792 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1701, 1.1568, 1.6094, 0.8859, 2.3196, 2.9921, 2.6758, 3.1519], device='cuda:1'), covar=tensor([0.1542, 0.3556, 0.3049, 0.2359, 0.0546, 0.0212, 0.0266, 0.0243], device='cuda:1'), in_proj_covar=tensor([0.0257, 0.0294, 0.0324, 0.0251, 0.0214, 0.0157, 0.0206, 0.0205], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 21:05:46,905 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9_rvb from training. Duration: 25.061125 2023-04-01 21:05:48,487 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=76003.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 21:05:53,894 INFO [train.py:903] (1/4) Epoch 12, batch 900, loss[loss=0.222, simple_loss=0.2993, pruned_loss=0.07241, over 19534.00 frames. ], tot_loss[loss=0.232, simple_loss=0.3062, pruned_loss=0.07892, over 3774457.97 frames. ], batch size: 54, lr: 7.05e-03, grad_scale: 8.0 2023-04-01 21:06:13,739 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.3124, 1.6958, 1.8815, 2.5471, 1.9397, 2.5263, 2.4546, 2.4656], device='cuda:1'), covar=tensor([0.0721, 0.0970, 0.1007, 0.0957, 0.0998, 0.0630, 0.0890, 0.0586], device='cuda:1'), in_proj_covar=tensor([0.0211, 0.0224, 0.0222, 0.0244, 0.0235, 0.0212, 0.0195, 0.0202], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-01 21:06:20,385 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=76028.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 21:06:57,938 INFO [train.py:903] (1/4) Epoch 12, batch 950, loss[loss=0.2731, simple_loss=0.3495, pruned_loss=0.09837, over 19353.00 frames. ], tot_loss[loss=0.232, simple_loss=0.3062, pruned_loss=0.07886, over 3780726.45 frames. ], batch size: 70, lr: 7.05e-03, grad_scale: 8.0 2023-04-01 21:07:02,610 WARNING [train.py:1073] (1/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9_rvb from training. Duration: 26.32775 2023-04-01 21:07:17,511 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.890e+02 5.102e+02 6.493e+02 8.387e+02 1.985e+03, threshold=1.299e+03, percent-clipped=2.0 2023-04-01 21:07:59,850 INFO [train.py:903] (1/4) Epoch 12, batch 1000, loss[loss=0.2134, simple_loss=0.2882, pruned_loss=0.06925, over 19611.00 frames. ], tot_loss[loss=0.2319, simple_loss=0.3062, pruned_loss=0.07877, over 3789302.36 frames. ], batch size: 50, lr: 7.05e-03, grad_scale: 8.0 2023-04-01 21:08:42,484 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.1175, 3.5107, 1.9627, 2.1387, 3.2334, 1.6826, 1.3091, 2.0421], device='cuda:1'), covar=tensor([0.1132, 0.0484, 0.0980, 0.0689, 0.0379, 0.1035, 0.0952, 0.0638], device='cuda:1'), in_proj_covar=tensor([0.0293, 0.0303, 0.0326, 0.0247, 0.0233, 0.0314, 0.0289, 0.0263], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 21:08:54,939 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 2023-04-01 21:09:02,674 INFO [train.py:903] (1/4) Epoch 12, batch 1050, loss[loss=0.2259, simple_loss=0.3049, pruned_loss=0.0735, over 19680.00 frames. ], tot_loss[loss=0.2328, simple_loss=0.3068, pruned_loss=0.07938, over 3799503.43 frames. ], batch size: 55, lr: 7.05e-03, grad_scale: 8.0 2023-04-01 21:09:20,681 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.538e+02 5.178e+02 6.428e+02 8.624e+02 1.751e+03, threshold=1.286e+03, percent-clipped=4.0 2023-04-01 21:09:36,014 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 2023-04-01 21:10:05,084 INFO [train.py:903] (1/4) Epoch 12, batch 1100, loss[loss=0.228, simple_loss=0.3026, pruned_loss=0.07666, over 19655.00 frames. ], tot_loss[loss=0.2325, simple_loss=0.3064, pruned_loss=0.0793, over 3804887.93 frames. ], batch size: 58, lr: 7.04e-03, grad_scale: 8.0 2023-04-01 21:10:12,392 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.8156, 1.4161, 1.4120, 1.7705, 1.5337, 1.5950, 1.4621, 1.6640], device='cuda:1'), covar=tensor([0.0878, 0.1319, 0.1309, 0.0809, 0.1072, 0.0505, 0.1175, 0.0654], device='cuda:1'), in_proj_covar=tensor([0.0252, 0.0352, 0.0292, 0.0241, 0.0296, 0.0243, 0.0280, 0.0238], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 21:11:08,893 INFO [train.py:903] (1/4) Epoch 12, batch 1150, loss[loss=0.2632, simple_loss=0.3216, pruned_loss=0.1023, over 19747.00 frames. ], tot_loss[loss=0.2333, simple_loss=0.3072, pruned_loss=0.07967, over 3816870.97 frames. ], batch size: 51, lr: 7.04e-03, grad_scale: 8.0 2023-04-01 21:11:27,475 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.258e+02 5.432e+02 6.664e+02 8.568e+02 1.731e+03, threshold=1.333e+03, percent-clipped=3.0 2023-04-01 21:11:44,833 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=76287.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 21:12:10,717 INFO [train.py:903] (1/4) Epoch 12, batch 1200, loss[loss=0.2096, simple_loss=0.2868, pruned_loss=0.06618, over 19678.00 frames. ], tot_loss[loss=0.2325, simple_loss=0.3063, pruned_loss=0.07931, over 3818263.47 frames. ], batch size: 60, lr: 7.04e-03, grad_scale: 8.0 2023-04-01 21:12:15,533 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=76312.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 21:12:45,755 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 2023-04-01 21:12:52,904 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=76341.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 21:13:14,063 INFO [train.py:903] (1/4) Epoch 12, batch 1250, loss[loss=0.2374, simple_loss=0.3115, pruned_loss=0.08169, over 19500.00 frames. ], tot_loss[loss=0.2339, simple_loss=0.3075, pruned_loss=0.08014, over 3816416.78 frames. ], batch size: 64, lr: 7.04e-03, grad_scale: 8.0 2023-04-01 21:13:31,217 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.467e+02 5.242e+02 6.279e+02 7.669e+02 1.575e+03, threshold=1.256e+03, percent-clipped=2.0 2023-04-01 21:14:15,615 INFO [train.py:903] (1/4) Epoch 12, batch 1300, loss[loss=0.2437, simple_loss=0.3198, pruned_loss=0.08378, over 19661.00 frames. ], tot_loss[loss=0.233, simple_loss=0.3065, pruned_loss=0.07969, over 3825027.43 frames. ], batch size: 58, lr: 7.04e-03, grad_scale: 8.0 2023-04-01 21:15:18,800 INFO [train.py:903] (1/4) Epoch 12, batch 1350, loss[loss=0.2331, simple_loss=0.3126, pruned_loss=0.0768, over 19175.00 frames. ], tot_loss[loss=0.2317, simple_loss=0.3054, pruned_loss=0.07901, over 3828492.95 frames. ], batch size: 69, lr: 7.03e-03, grad_scale: 8.0 2023-04-01 21:15:37,068 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.121e+02 5.247e+02 6.486e+02 7.910e+02 1.390e+03, threshold=1.297e+03, percent-clipped=4.0 2023-04-01 21:16:20,448 INFO [train.py:903] (1/4) Epoch 12, batch 1400, loss[loss=0.2739, simple_loss=0.3479, pruned_loss=0.0999, over 19483.00 frames. ], tot_loss[loss=0.2317, simple_loss=0.3056, pruned_loss=0.07888, over 3827752.19 frames. ], batch size: 64, lr: 7.03e-03, grad_scale: 16.0 2023-04-01 21:16:50,873 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0273, 1.6931, 1.5085, 1.9315, 1.9655, 1.7616, 1.5002, 1.9106], device='cuda:1'), covar=tensor([0.0857, 0.1535, 0.1448, 0.1002, 0.1077, 0.0505, 0.1316, 0.0660], device='cuda:1'), in_proj_covar=tensor([0.0253, 0.0354, 0.0292, 0.0242, 0.0298, 0.0245, 0.0279, 0.0240], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 21:17:01,834 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.5626, 4.0987, 2.7642, 3.5905, 1.0882, 3.8887, 3.8979, 3.9409], device='cuda:1'), covar=tensor([0.0631, 0.1006, 0.1775, 0.0799, 0.3733, 0.0845, 0.0780, 0.1084], device='cuda:1'), in_proj_covar=tensor([0.0442, 0.0367, 0.0442, 0.0319, 0.0380, 0.0374, 0.0358, 0.0392], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 21:17:23,464 INFO [train.py:903] (1/4) Epoch 12, batch 1450, loss[loss=0.2526, simple_loss=0.3253, pruned_loss=0.08995, over 19647.00 frames. ], tot_loss[loss=0.2318, simple_loss=0.3057, pruned_loss=0.07896, over 3826651.50 frames. ], batch size: 55, lr: 7.03e-03, grad_scale: 16.0 2023-04-01 21:17:25,874 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_rvb from training. Duration: 25.85 2023-04-01 21:17:40,981 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.252e+02 4.935e+02 6.510e+02 8.250e+02 1.774e+03, threshold=1.302e+03, percent-clipped=3.0 2023-04-01 21:18:24,869 INFO [train.py:903] (1/4) Epoch 12, batch 1500, loss[loss=0.2078, simple_loss=0.2919, pruned_loss=0.06181, over 19680.00 frames. ], tot_loss[loss=0.2321, simple_loss=0.3057, pruned_loss=0.07924, over 3837659.37 frames. ], batch size: 53, lr: 7.03e-03, grad_scale: 8.0 2023-04-01 21:18:53,780 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=76631.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 21:19:24,220 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=76656.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 21:19:27,188 INFO [train.py:903] (1/4) Epoch 12, batch 1550, loss[loss=0.2343, simple_loss=0.3049, pruned_loss=0.08189, over 19490.00 frames. ], tot_loss[loss=0.232, simple_loss=0.3055, pruned_loss=0.07921, over 3833595.16 frames. ], batch size: 64, lr: 7.02e-03, grad_scale: 8.0 2023-04-01 21:19:46,335 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.911e+02 5.709e+02 6.905e+02 9.210e+02 1.884e+03, threshold=1.381e+03, percent-clipped=4.0 2023-04-01 21:20:01,238 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=76685.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 21:20:19,367 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 2023-04-01 21:20:23,676 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.59 vs. limit=2.0 2023-04-01 21:20:29,930 INFO [train.py:903] (1/4) Epoch 12, batch 1600, loss[loss=0.2193, simple_loss=0.3013, pruned_loss=0.06868, over 19674.00 frames. ], tot_loss[loss=0.2322, simple_loss=0.3062, pruned_loss=0.07916, over 3823477.81 frames. ], batch size: 58, lr: 7.02e-03, grad_scale: 8.0 2023-04-01 21:20:53,997 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9_rvb from training. Duration: 30.1555625 2023-04-01 21:21:16,436 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=76746.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 21:21:23,393 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2169, 2.0318, 1.5223, 1.2686, 1.8770, 1.0516, 1.2167, 1.8420], device='cuda:1'), covar=tensor([0.0886, 0.0619, 0.0935, 0.0742, 0.0427, 0.1164, 0.0656, 0.0334], device='cuda:1'), in_proj_covar=tensor([0.0289, 0.0300, 0.0321, 0.0243, 0.0231, 0.0316, 0.0289, 0.0260], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 21:21:29,665 INFO [train.py:903] (1/4) Epoch 12, batch 1650, loss[loss=0.2488, simple_loss=0.3283, pruned_loss=0.08462, over 19580.00 frames. ], tot_loss[loss=0.233, simple_loss=0.3069, pruned_loss=0.07956, over 3827884.07 frames. ], batch size: 61, lr: 7.02e-03, grad_scale: 8.0 2023-04-01 21:21:30,361 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.72 vs. limit=2.0 2023-04-01 21:21:46,992 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=76771.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 21:21:49,936 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.672e+02 5.088e+02 6.360e+02 7.707e+02 1.579e+03, threshold=1.272e+03, percent-clipped=3.0 2023-04-01 21:22:06,308 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6956, 1.7811, 1.6827, 2.7454, 1.7648, 2.5248, 1.9678, 1.3949], device='cuda:1'), covar=tensor([0.4250, 0.3568, 0.2197, 0.2073, 0.3865, 0.1693, 0.4714, 0.4092], device='cuda:1'), in_proj_covar=tensor([0.0791, 0.0816, 0.0646, 0.0889, 0.0781, 0.0706, 0.0779, 0.0709], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 21:22:22,654 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=76800.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 21:22:33,393 INFO [train.py:903] (1/4) Epoch 12, batch 1700, loss[loss=0.268, simple_loss=0.33, pruned_loss=0.103, over 19780.00 frames. ], tot_loss[loss=0.2335, simple_loss=0.3068, pruned_loss=0.08008, over 3808245.06 frames. ], batch size: 54, lr: 7.02e-03, grad_scale: 8.0 2023-04-01 21:22:59,166 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5874, 1.3937, 1.3984, 1.8671, 1.5500, 1.8883, 1.8963, 1.6884], device='cuda:1'), covar=tensor([0.0744, 0.0861, 0.0966, 0.0795, 0.0856, 0.0654, 0.0790, 0.0605], device='cuda:1'), in_proj_covar=tensor([0.0212, 0.0225, 0.0224, 0.0246, 0.0236, 0.0213, 0.0195, 0.0201], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-01 21:23:10,306 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 2023-04-01 21:23:16,984 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.1028, 2.0003, 1.8275, 1.6670, 1.4662, 1.6412, 0.3265, 1.0142], device='cuda:1'), covar=tensor([0.0465, 0.0464, 0.0328, 0.0606, 0.0962, 0.0631, 0.1011, 0.0798], device='cuda:1'), in_proj_covar=tensor([0.0339, 0.0329, 0.0332, 0.0356, 0.0427, 0.0354, 0.0309, 0.0322], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 21:23:33,278 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=76857.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 21:23:34,183 INFO [train.py:903] (1/4) Epoch 12, batch 1750, loss[loss=0.1852, simple_loss=0.2563, pruned_loss=0.05706, over 19713.00 frames. ], tot_loss[loss=0.2347, simple_loss=0.3075, pruned_loss=0.08095, over 3791606.37 frames. ], batch size: 46, lr: 7.01e-03, grad_scale: 8.0 2023-04-01 21:23:36,422 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.50 vs. limit=2.0 2023-04-01 21:23:53,640 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.317e+02 5.587e+02 7.065e+02 8.864e+02 2.096e+03, threshold=1.413e+03, percent-clipped=6.0 2023-04-01 21:24:37,132 INFO [train.py:903] (1/4) Epoch 12, batch 1800, loss[loss=0.2321, simple_loss=0.2953, pruned_loss=0.08448, over 19782.00 frames. ], tot_loss[loss=0.233, simple_loss=0.3061, pruned_loss=0.08001, over 3802848.90 frames. ], batch size: 49, lr: 7.01e-03, grad_scale: 8.0 2023-04-01 21:24:44,476 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0835, 3.6261, 2.1608, 2.2173, 3.3335, 1.8662, 1.3115, 2.2553], device='cuda:1'), covar=tensor([0.1311, 0.0515, 0.0996, 0.0724, 0.0429, 0.1049, 0.1012, 0.0580], device='cuda:1'), in_proj_covar=tensor([0.0287, 0.0300, 0.0320, 0.0243, 0.0231, 0.0317, 0.0289, 0.0259], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 21:25:32,729 WARNING [train.py:1073] (1/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9_rvb from training. Duration: 25.0944375 2023-04-01 21:25:38,686 INFO [train.py:903] (1/4) Epoch 12, batch 1850, loss[loss=0.2336, simple_loss=0.3152, pruned_loss=0.07603, over 19603.00 frames. ], tot_loss[loss=0.2333, simple_loss=0.3063, pruned_loss=0.08018, over 3773203.06 frames. ], batch size: 57, lr: 7.01e-03, grad_scale: 8.0 2023-04-01 21:25:45,874 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.06 vs. limit=5.0 2023-04-01 21:25:59,472 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.186e+02 5.640e+02 7.026e+02 8.457e+02 1.689e+03, threshold=1.405e+03, percent-clipped=1.0 2023-04-01 21:26:12,301 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9_rvb from training. Duration: 27.8166875 2023-04-01 21:26:34,388 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=77002.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 21:26:41,890 INFO [train.py:903] (1/4) Epoch 12, batch 1900, loss[loss=0.2049, simple_loss=0.28, pruned_loss=0.06488, over 19835.00 frames. ], tot_loss[loss=0.2345, simple_loss=0.3072, pruned_loss=0.08094, over 3774626.53 frames. ], batch size: 52, lr: 7.01e-03, grad_scale: 8.0 2023-04-01 21:27:00,222 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 2023-04-01 21:27:04,914 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9_rvb from training. Duration: 27.02225 2023-04-01 21:27:06,401 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=77027.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 21:27:06,420 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=77027.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 21:27:28,488 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 2023-04-01 21:27:36,569 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=77052.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 21:27:42,271 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=77056.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 21:27:43,957 INFO [train.py:903] (1/4) Epoch 12, batch 1950, loss[loss=0.2195, simple_loss=0.284, pruned_loss=0.07747, over 19409.00 frames. ], tot_loss[loss=0.2332, simple_loss=0.3063, pruned_loss=0.08006, over 3793178.37 frames. ], batch size: 48, lr: 7.01e-03, grad_scale: 8.0 2023-04-01 21:28:03,223 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.700e+02 5.468e+02 7.092e+02 9.065e+02 1.810e+03, threshold=1.418e+03, percent-clipped=4.0 2023-04-01 21:28:11,819 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=77081.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 21:28:45,470 INFO [train.py:903] (1/4) Epoch 12, batch 2000, loss[loss=0.2151, simple_loss=0.2953, pruned_loss=0.06746, over 19578.00 frames. ], tot_loss[loss=0.2321, simple_loss=0.3055, pruned_loss=0.07937, over 3802527.40 frames. ], batch size: 52, lr: 7.00e-03, grad_scale: 8.0 2023-04-01 21:28:51,484 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=77113.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 21:29:15,583 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.6536, 2.2322, 2.0360, 2.5818, 2.5784, 2.1084, 1.8681, 2.5421], device='cuda:1'), covar=tensor([0.0828, 0.1572, 0.1440, 0.0976, 0.1187, 0.0485, 0.1293, 0.0603], device='cuda:1'), in_proj_covar=tensor([0.0252, 0.0349, 0.0292, 0.0238, 0.0294, 0.0242, 0.0278, 0.0236], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 21:29:43,034 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_rvb from training. Duration: 26.8349375 2023-04-01 21:29:46,534 INFO [train.py:903] (1/4) Epoch 12, batch 2050, loss[loss=0.1989, simple_loss=0.2907, pruned_loss=0.05357, over 19542.00 frames. ], tot_loss[loss=0.231, simple_loss=0.3048, pruned_loss=0.07864, over 3818264.33 frames. ], batch size: 54, lr: 7.00e-03, grad_scale: 8.0 2023-04-01 21:30:02,213 WARNING [train.py:1073] (1/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 2023-04-01 21:30:03,450 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 2023-04-01 21:30:06,829 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.225e+02 5.837e+02 7.308e+02 9.815e+02 2.165e+03, threshold=1.462e+03, percent-clipped=5.0 2023-04-01 21:30:26,558 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 2023-04-01 21:30:40,591 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=77201.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 21:30:49,627 INFO [train.py:903] (1/4) Epoch 12, batch 2100, loss[loss=0.2182, simple_loss=0.284, pruned_loss=0.07616, over 19731.00 frames. ], tot_loss[loss=0.2324, simple_loss=0.3062, pruned_loss=0.07926, over 3826363.73 frames. ], batch size: 46, lr: 7.00e-03, grad_scale: 4.0 2023-04-01 21:31:10,328 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.12 vs. limit=2.0 2023-04-01 21:31:12,359 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.9596, 3.6061, 2.4468, 3.2686, 1.1490, 3.4434, 3.3494, 3.4625], device='cuda:1'), covar=tensor([0.0843, 0.1284, 0.1974, 0.0938, 0.3606, 0.0889, 0.0987, 0.1160], device='cuda:1'), in_proj_covar=tensor([0.0445, 0.0369, 0.0444, 0.0321, 0.0384, 0.0378, 0.0364, 0.0397], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 21:31:17,027 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=77229.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 21:31:20,296 WARNING [train.py:1073] (1/4) Exclude cut with ID 453-131332-0000-131866_sp0.9_rvb from training. Duration: 25.3333125 2023-04-01 21:31:28,740 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1787, 1.1163, 1.1372, 1.3270, 1.0482, 1.3566, 1.2914, 1.2465], device='cuda:1'), covar=tensor([0.0945, 0.1036, 0.1109, 0.0741, 0.0913, 0.0784, 0.0840, 0.0794], device='cuda:1'), in_proj_covar=tensor([0.0214, 0.0226, 0.0226, 0.0249, 0.0236, 0.0215, 0.0197, 0.0204], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-01 21:31:40,676 WARNING [train.py:1073] (1/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9_rvb from training. Duration: 26.6166875 2023-04-01 21:31:52,741 INFO [train.py:903] (1/4) Epoch 12, batch 2150, loss[loss=0.2709, simple_loss=0.3359, pruned_loss=0.103, over 19465.00 frames. ], tot_loss[loss=0.233, simple_loss=0.3067, pruned_loss=0.07962, over 3822835.36 frames. ], batch size: 64, lr: 7.00e-03, grad_scale: 4.0 2023-04-01 21:32:13,146 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.202e+02 5.327e+02 7.168e+02 9.347e+02 2.125e+03, threshold=1.434e+03, percent-clipped=4.0 2023-04-01 21:32:33,277 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=77291.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 21:32:39,473 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=77295.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 21:32:51,916 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.57 vs. limit=5.0 2023-04-01 21:32:53,896 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.0231, 1.2171, 1.7616, 0.9028, 2.3911, 3.0675, 2.7953, 3.2232], device='cuda:1'), covar=tensor([0.1553, 0.3479, 0.2830, 0.2279, 0.0451, 0.0183, 0.0246, 0.0246], device='cuda:1'), in_proj_covar=tensor([0.0254, 0.0295, 0.0324, 0.0250, 0.0215, 0.0158, 0.0205, 0.0205], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 21:32:55,907 INFO [train.py:903] (1/4) Epoch 12, batch 2200, loss[loss=0.2088, simple_loss=0.286, pruned_loss=0.06583, over 19743.00 frames. ], tot_loss[loss=0.232, simple_loss=0.3058, pruned_loss=0.07908, over 3815446.64 frames. ], batch size: 51, lr: 6.99e-03, grad_scale: 4.0 2023-04-01 21:33:05,519 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=77316.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 21:33:25,223 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.13 vs. limit=2.0 2023-04-01 21:33:57,353 INFO [train.py:903] (1/4) Epoch 12, batch 2250, loss[loss=0.2223, simple_loss=0.3065, pruned_loss=0.06903, over 19602.00 frames. ], tot_loss[loss=0.2326, simple_loss=0.3068, pruned_loss=0.07915, over 3807427.23 frames. ], batch size: 57, lr: 6.99e-03, grad_scale: 4.0 2023-04-01 21:34:08,704 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.0990, 1.2988, 1.6383, 1.1776, 2.7245, 3.4799, 3.2007, 3.6744], device='cuda:1'), covar=tensor([0.1555, 0.3326, 0.3088, 0.2141, 0.0477, 0.0154, 0.0215, 0.0202], device='cuda:1'), in_proj_covar=tensor([0.0255, 0.0294, 0.0325, 0.0250, 0.0215, 0.0158, 0.0206, 0.0206], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 21:34:18,143 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.729e+02 5.298e+02 6.341e+02 7.997e+02 1.542e+03, threshold=1.268e+03, percent-clipped=2.0 2023-04-01 21:34:27,405 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.7595, 3.0690, 3.2621, 3.2570, 1.6385, 3.0078, 2.7813, 3.0284], device='cuda:1'), covar=tensor([0.1347, 0.2323, 0.0599, 0.0676, 0.4022, 0.1177, 0.0601, 0.1016], device='cuda:1'), in_proj_covar=tensor([0.0683, 0.0613, 0.0812, 0.0690, 0.0738, 0.0565, 0.0498, 0.0749], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1') 2023-04-01 21:34:35,672 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 2023-04-01 21:34:58,616 INFO [train.py:903] (1/4) Epoch 12, batch 2300, loss[loss=0.2505, simple_loss=0.3242, pruned_loss=0.08843, over 19685.00 frames. ], tot_loss[loss=0.2324, simple_loss=0.3065, pruned_loss=0.07913, over 3813737.46 frames. ], batch size: 59, lr: 6.99e-03, grad_scale: 4.0 2023-04-01 21:35:11,953 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_rvb from training. Duration: 26.205 2023-04-01 21:35:59,772 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=77457.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 21:36:00,826 INFO [train.py:903] (1/4) Epoch 12, batch 2350, loss[loss=0.2026, simple_loss=0.2739, pruned_loss=0.06565, over 19755.00 frames. ], tot_loss[loss=0.2312, simple_loss=0.3055, pruned_loss=0.07844, over 3830733.19 frames. ], batch size: 46, lr: 6.99e-03, grad_scale: 4.0 2023-04-01 21:36:22,276 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.400e+02 5.245e+02 6.457e+02 8.417e+02 4.507e+03, threshold=1.291e+03, percent-clipped=6.0 2023-04-01 21:36:41,243 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_rvb from training. Duration: 25.775 2023-04-01 21:36:59,094 WARNING [train.py:1073] (1/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9_rvb from training. Duration: 25.45 2023-04-01 21:37:02,537 INFO [train.py:903] (1/4) Epoch 12, batch 2400, loss[loss=0.2499, simple_loss=0.3184, pruned_loss=0.09069, over 19794.00 frames. ], tot_loss[loss=0.231, simple_loss=0.3054, pruned_loss=0.07832, over 3819137.41 frames. ], batch size: 56, lr: 6.99e-03, grad_scale: 8.0 2023-04-01 21:37:17,475 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4017, 1.2082, 1.2007, 1.7481, 1.5021, 1.6081, 1.8433, 1.4128], device='cuda:1'), covar=tensor([0.0836, 0.0982, 0.1090, 0.0729, 0.0748, 0.0722, 0.0694, 0.0744], device='cuda:1'), in_proj_covar=tensor([0.0212, 0.0224, 0.0224, 0.0248, 0.0234, 0.0214, 0.0196, 0.0201], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-01 21:37:36,493 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6558, 1.7187, 1.9213, 2.1879, 1.5319, 2.0116, 2.0972, 1.8197], device='cuda:1'), covar=tensor([0.3294, 0.2735, 0.1426, 0.1479, 0.2835, 0.1324, 0.3492, 0.2530], device='cuda:1'), in_proj_covar=tensor([0.0802, 0.0823, 0.0653, 0.0897, 0.0788, 0.0711, 0.0789, 0.0719], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 21:38:04,504 INFO [train.py:903] (1/4) Epoch 12, batch 2450, loss[loss=0.2309, simple_loss=0.2989, pruned_loss=0.08145, over 19833.00 frames. ], tot_loss[loss=0.2301, simple_loss=0.3045, pruned_loss=0.07785, over 3823711.70 frames. ], batch size: 52, lr: 6.98e-03, grad_scale: 8.0 2023-04-01 21:38:21,457 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=77572.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 21:38:21,528 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=77572.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 21:38:22,403 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=77573.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 21:38:24,484 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.532e+02 5.530e+02 6.529e+02 8.263e+02 1.639e+03, threshold=1.306e+03, percent-clipped=2.0 2023-04-01 21:38:52,862 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=77597.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 21:38:56,131 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.8000, 4.3155, 2.7484, 3.9026, 1.3353, 4.1524, 4.1256, 4.2836], device='cuda:1'), covar=tensor([0.0575, 0.0966, 0.1926, 0.0717, 0.3591, 0.0710, 0.0734, 0.0969], device='cuda:1'), in_proj_covar=tensor([0.0440, 0.0365, 0.0440, 0.0319, 0.0380, 0.0373, 0.0360, 0.0392], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 21:39:05,891 INFO [train.py:903] (1/4) Epoch 12, batch 2500, loss[loss=0.2809, simple_loss=0.3458, pruned_loss=0.108, over 19406.00 frames. ], tot_loss[loss=0.2305, simple_loss=0.3049, pruned_loss=0.07805, over 3826195.98 frames. ], batch size: 70, lr: 6.98e-03, grad_scale: 8.0 2023-04-01 21:39:27,226 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6497, 1.3944, 1.4283, 2.0206, 1.6064, 1.9382, 1.9679, 1.8066], device='cuda:1'), covar=tensor([0.0724, 0.0888, 0.0967, 0.0746, 0.0866, 0.0641, 0.0789, 0.0579], device='cuda:1'), in_proj_covar=tensor([0.0210, 0.0221, 0.0222, 0.0244, 0.0232, 0.0211, 0.0193, 0.0199], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-01 21:39:39,437 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=77634.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 21:39:40,540 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=77635.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 21:39:40,770 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.7773, 1.5922, 1.4177, 1.8280, 1.6819, 1.3442, 1.4714, 1.6822], device='cuda:1'), covar=tensor([0.1141, 0.1812, 0.1742, 0.1190, 0.1448, 0.1011, 0.1600, 0.0916], device='cuda:1'), in_proj_covar=tensor([0.0250, 0.0347, 0.0290, 0.0237, 0.0292, 0.0240, 0.0277, 0.0236], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 21:39:45,170 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=77639.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 21:40:08,015 INFO [train.py:903] (1/4) Epoch 12, batch 2550, loss[loss=0.2431, simple_loss=0.3122, pruned_loss=0.08699, over 19489.00 frames. ], tot_loss[loss=0.231, simple_loss=0.3054, pruned_loss=0.07836, over 3812580.24 frames. ], batch size: 64, lr: 6.98e-03, grad_scale: 8.0 2023-04-01 21:40:30,542 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.096e+02 5.340e+02 6.709e+02 8.508e+02 1.809e+03, threshold=1.342e+03, percent-clipped=3.0 2023-04-01 21:40:46,038 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=77688.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 21:40:54,809 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=77695.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 21:41:06,542 WARNING [train.py:1073] (1/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 2023-04-01 21:41:10,775 INFO [train.py:903] (1/4) Epoch 12, batch 2600, loss[loss=0.261, simple_loss=0.3312, pruned_loss=0.09542, over 19672.00 frames. ], tot_loss[loss=0.2315, simple_loss=0.3055, pruned_loss=0.07874, over 3815638.17 frames. ], batch size: 58, lr: 6.98e-03, grad_scale: 8.0 2023-04-01 21:41:54,847 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 2023-04-01 21:42:02,057 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=77750.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 21:42:07,417 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=77754.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 21:42:11,597 INFO [train.py:903] (1/4) Epoch 12, batch 2650, loss[loss=0.242, simple_loss=0.3217, pruned_loss=0.08115, over 18159.00 frames. ], tot_loss[loss=0.232, simple_loss=0.3061, pruned_loss=0.07894, over 3811045.78 frames. ], batch size: 83, lr: 6.97e-03, grad_scale: 8.0 2023-04-01 21:42:32,180 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.366e+02 5.775e+02 6.860e+02 8.613e+02 1.817e+03, threshold=1.372e+03, percent-clipped=5.0 2023-04-01 21:42:33,293 WARNING [train.py:1073] (1/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9_rvb from training. Duration: 27.25 2023-04-01 21:43:06,944 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.6084, 4.2189, 2.6515, 3.7982, 1.2836, 3.9022, 3.9580, 4.0165], device='cuda:1'), covar=tensor([0.0623, 0.0978, 0.2011, 0.0689, 0.3672, 0.0785, 0.0763, 0.1079], device='cuda:1'), in_proj_covar=tensor([0.0449, 0.0374, 0.0450, 0.0326, 0.0386, 0.0379, 0.0367, 0.0400], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 21:43:12,308 INFO [train.py:903] (1/4) Epoch 12, batch 2700, loss[loss=0.2294, simple_loss=0.2941, pruned_loss=0.08238, over 19385.00 frames. ], tot_loss[loss=0.2317, simple_loss=0.3055, pruned_loss=0.07894, over 3813728.96 frames. ], batch size: 48, lr: 6.97e-03, grad_scale: 8.0 2023-04-01 21:43:38,351 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=77828.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 21:44:08,780 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=77853.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 21:44:14,414 INFO [train.py:903] (1/4) Epoch 12, batch 2750, loss[loss=0.2115, simple_loss=0.2848, pruned_loss=0.06912, over 19807.00 frames. ], tot_loss[loss=0.2322, simple_loss=0.306, pruned_loss=0.07923, over 3824501.08 frames. ], batch size: 49, lr: 6.97e-03, grad_scale: 8.0 2023-04-01 21:44:36,445 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.412e+02 5.785e+02 6.935e+02 8.560e+02 1.739e+03, threshold=1.387e+03, percent-clipped=4.0 2023-04-01 21:45:15,196 INFO [train.py:903] (1/4) Epoch 12, batch 2800, loss[loss=0.1736, simple_loss=0.2554, pruned_loss=0.04591, over 19313.00 frames. ], tot_loss[loss=0.2307, simple_loss=0.3046, pruned_loss=0.07839, over 3829890.85 frames. ], batch size: 44, lr: 6.97e-03, grad_scale: 8.0 2023-04-01 21:46:01,236 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=77944.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 21:46:19,705 INFO [train.py:903] (1/4) Epoch 12, batch 2850, loss[loss=0.2383, simple_loss=0.3152, pruned_loss=0.0807, over 19372.00 frames. ], tot_loss[loss=0.2312, simple_loss=0.3051, pruned_loss=0.07862, over 3820199.65 frames. ], batch size: 66, lr: 6.97e-03, grad_scale: 4.0 2023-04-01 21:46:22,373 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.6954, 1.3213, 1.4614, 1.6485, 3.2214, 0.9973, 2.2443, 3.5793], device='cuda:1'), covar=tensor([0.0456, 0.2615, 0.2714, 0.1679, 0.0771, 0.2650, 0.1172, 0.0273], device='cuda:1'), in_proj_covar=tensor([0.0354, 0.0333, 0.0345, 0.0315, 0.0340, 0.0330, 0.0331, 0.0351], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 21:46:32,709 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=77969.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 21:46:41,209 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.493e+02 5.663e+02 6.634e+02 8.773e+02 3.814e+03, threshold=1.327e+03, percent-clipped=6.0 2023-04-01 21:46:43,718 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=77978.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 21:47:20,404 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=78006.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 21:47:22,313 INFO [train.py:903] (1/4) Epoch 12, batch 2900, loss[loss=0.1924, simple_loss=0.2726, pruned_loss=0.05606, over 19478.00 frames. ], tot_loss[loss=0.23, simple_loss=0.3042, pruned_loss=0.0779, over 3817598.55 frames. ], batch size: 49, lr: 6.96e-03, grad_scale: 4.0 2023-04-01 21:47:22,357 WARNING [train.py:1073] (1/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 2023-04-01 21:47:25,113 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=78010.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 21:47:39,742 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3159, 1.5043, 1.9357, 1.6230, 3.1917, 2.6688, 3.5073, 1.4655], device='cuda:1'), covar=tensor([0.2078, 0.3548, 0.2179, 0.1625, 0.1268, 0.1677, 0.1419, 0.3398], device='cuda:1'), in_proj_covar=tensor([0.0489, 0.0578, 0.0608, 0.0437, 0.0594, 0.0489, 0.0644, 0.0495], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 21:47:52,624 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=78031.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 21:47:57,903 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=78035.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 21:48:02,341 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=78039.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 21:48:25,027 INFO [train.py:903] (1/4) Epoch 12, batch 2950, loss[loss=0.2723, simple_loss=0.3419, pruned_loss=0.1013, over 19253.00 frames. ], tot_loss[loss=0.2302, simple_loss=0.3046, pruned_loss=0.07796, over 3821989.75 frames. ], batch size: 66, lr: 6.96e-03, grad_scale: 4.0 2023-04-01 21:48:48,725 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.235e+02 5.583e+02 6.866e+02 9.102e+02 1.641e+03, threshold=1.373e+03, percent-clipped=7.0 2023-04-01 21:49:10,010 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=78093.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 21:49:28,175 INFO [train.py:903] (1/4) Epoch 12, batch 3000, loss[loss=0.2616, simple_loss=0.3457, pruned_loss=0.08871, over 19664.00 frames. ], tot_loss[loss=0.2308, simple_loss=0.305, pruned_loss=0.07832, over 3826602.52 frames. ], batch size: 58, lr: 6.96e-03, grad_scale: 4.0 2023-04-01 21:49:28,175 INFO [train.py:928] (1/4) Computing validation loss 2023-04-01 21:49:40,670 INFO [train.py:937] (1/4) Epoch 12, validation: loss=0.1772, simple_loss=0.2779, pruned_loss=0.0383, over 944034.00 frames. 2023-04-01 21:49:40,671 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18649MB 2023-04-01 21:49:45,519 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_rvb from training. Duration: 29.735 2023-04-01 21:50:38,086 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=78154.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 21:50:42,290 INFO [train.py:903] (1/4) Epoch 12, batch 3050, loss[loss=0.1845, simple_loss=0.2647, pruned_loss=0.05221, over 19743.00 frames. ], tot_loss[loss=0.2318, simple_loss=0.3059, pruned_loss=0.07889, over 3833725.64 frames. ], batch size: 46, lr: 6.96e-03, grad_scale: 4.0 2023-04-01 21:51:04,780 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.897e+02 5.238e+02 6.566e+02 8.426e+02 1.854e+03, threshold=1.313e+03, percent-clipped=6.0 2023-04-01 21:51:20,725 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.8836, 1.9529, 2.1031, 2.5784, 1.7025, 2.4298, 2.3292, 2.0211], device='cuda:1'), covar=tensor([0.3328, 0.3100, 0.1415, 0.1722, 0.3349, 0.1528, 0.3426, 0.2618], device='cuda:1'), in_proj_covar=tensor([0.0795, 0.0822, 0.0648, 0.0891, 0.0783, 0.0708, 0.0785, 0.0715], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 21:51:43,592 INFO [train.py:903] (1/4) Epoch 12, batch 3100, loss[loss=0.2138, simple_loss=0.2909, pruned_loss=0.06835, over 19581.00 frames. ], tot_loss[loss=0.2327, simple_loss=0.3063, pruned_loss=0.07953, over 3815555.84 frames. ], batch size: 52, lr: 6.95e-03, grad_scale: 4.0 2023-04-01 21:52:14,949 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4874, 2.3315, 1.6592, 1.5640, 2.1889, 1.2618, 1.4253, 1.9331], device='cuda:1'), covar=tensor([0.1007, 0.0681, 0.1057, 0.0731, 0.0438, 0.1179, 0.0672, 0.0402], device='cuda:1'), in_proj_covar=tensor([0.0291, 0.0301, 0.0323, 0.0244, 0.0234, 0.0322, 0.0286, 0.0261], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 21:52:18,451 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.69 vs. limit=2.0 2023-04-01 21:52:36,405 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=78250.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 21:52:45,572 INFO [train.py:903] (1/4) Epoch 12, batch 3150, loss[loss=0.2782, simple_loss=0.3479, pruned_loss=0.1043, over 19666.00 frames. ], tot_loss[loss=0.2324, simple_loss=0.3064, pruned_loss=0.07923, over 3831550.74 frames. ], batch size: 53, lr: 6.95e-03, grad_scale: 4.0 2023-04-01 21:52:53,234 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.20 vs. limit=5.0 2023-04-01 21:53:07,631 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.745e+02 5.622e+02 7.364e+02 8.683e+02 1.879e+03, threshold=1.473e+03, percent-clipped=3.0 2023-04-01 21:53:15,174 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 2023-04-01 21:53:32,886 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5452, 1.6921, 2.0968, 1.6930, 2.5594, 2.9289, 2.9202, 3.1285], device='cuda:1'), covar=tensor([0.1318, 0.2711, 0.2333, 0.2047, 0.0868, 0.0382, 0.0200, 0.0280], device='cuda:1'), in_proj_covar=tensor([0.0256, 0.0298, 0.0326, 0.0252, 0.0216, 0.0159, 0.0205, 0.0207], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 21:53:47,084 INFO [train.py:903] (1/4) Epoch 12, batch 3200, loss[loss=0.1952, simple_loss=0.2666, pruned_loss=0.06187, over 19718.00 frames. ], tot_loss[loss=0.2327, simple_loss=0.3062, pruned_loss=0.07963, over 3828539.39 frames. ], batch size: 46, lr: 6.95e-03, grad_scale: 8.0 2023-04-01 21:54:39,758 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=78349.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 21:54:50,766 INFO [train.py:903] (1/4) Epoch 12, batch 3250, loss[loss=0.2207, simple_loss=0.2915, pruned_loss=0.07494, over 19467.00 frames. ], tot_loss[loss=0.2322, simple_loss=0.3058, pruned_loss=0.0793, over 3829758.46 frames. ], batch size: 49, lr: 6.95e-03, grad_scale: 8.0 2023-04-01 21:55:11,650 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=78374.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 21:55:13,652 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.865e+02 5.227e+02 6.164e+02 7.465e+02 1.234e+03, threshold=1.233e+03, percent-clipped=0.0 2023-04-01 21:55:54,296 INFO [train.py:903] (1/4) Epoch 12, batch 3300, loss[loss=0.2397, simple_loss=0.305, pruned_loss=0.08716, over 19844.00 frames. ], tot_loss[loss=0.2333, simple_loss=0.3069, pruned_loss=0.07985, over 3818748.69 frames. ], batch size: 52, lr: 6.95e-03, grad_scale: 8.0 2023-04-01 21:55:57,162 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=78410.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 21:55:57,924 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 2023-04-01 21:56:26,557 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=78435.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 21:56:55,619 INFO [train.py:903] (1/4) Epoch 12, batch 3350, loss[loss=0.2417, simple_loss=0.3145, pruned_loss=0.08449, over 19606.00 frames. ], tot_loss[loss=0.2328, simple_loss=0.3067, pruned_loss=0.07943, over 3814359.84 frames. ], batch size: 61, lr: 6.94e-03, grad_scale: 8.0 2023-04-01 21:57:18,019 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.479e+02 5.251e+02 6.145e+02 6.896e+02 1.617e+03, threshold=1.229e+03, percent-clipped=1.0 2023-04-01 21:57:20,556 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=78478.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 21:57:36,597 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.2991, 2.0666, 1.9702, 2.4447, 2.0639, 1.7392, 1.7153, 2.2719], device='cuda:1'), covar=tensor([0.0961, 0.1660, 0.1501, 0.0897, 0.1417, 0.0717, 0.1424, 0.0726], device='cuda:1'), in_proj_covar=tensor([0.0253, 0.0348, 0.0291, 0.0236, 0.0293, 0.0240, 0.0276, 0.0237], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 21:57:57,201 INFO [train.py:903] (1/4) Epoch 12, batch 3400, loss[loss=0.2199, simple_loss=0.2941, pruned_loss=0.07284, over 19743.00 frames. ], tot_loss[loss=0.2317, simple_loss=0.306, pruned_loss=0.07868, over 3820475.33 frames. ], batch size: 51, lr: 6.94e-03, grad_scale: 8.0 2023-04-01 21:59:00,755 INFO [train.py:903] (1/4) Epoch 12, batch 3450, loss[loss=0.2179, simple_loss=0.2887, pruned_loss=0.0735, over 19396.00 frames. ], tot_loss[loss=0.2319, simple_loss=0.3061, pruned_loss=0.07889, over 3828795.39 frames. ], batch size: 48, lr: 6.94e-03, grad_scale: 8.0 2023-04-01 21:59:06,189 WARNING [train.py:1073] (1/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9_rvb from training. Duration: 25.2444375 2023-04-01 21:59:21,439 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=78574.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 21:59:23,280 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.785e+02 5.692e+02 7.472e+02 9.495e+02 2.057e+03, threshold=1.494e+03, percent-clipped=9.0 2023-04-01 21:59:44,667 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=78594.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 22:00:03,311 INFO [train.py:903] (1/4) Epoch 12, batch 3500, loss[loss=0.3044, simple_loss=0.3572, pruned_loss=0.1258, over 13044.00 frames. ], tot_loss[loss=0.2331, simple_loss=0.3071, pruned_loss=0.07953, over 3827784.49 frames. ], batch size: 135, lr: 6.94e-03, grad_scale: 8.0 2023-04-01 22:00:34,886 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.72 vs. limit=5.0 2023-04-01 22:00:55,214 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.1538, 1.9333, 1.8003, 2.1601, 1.8204, 1.8672, 1.7584, 2.1097], device='cuda:1'), covar=tensor([0.0818, 0.1298, 0.1254, 0.0850, 0.1211, 0.0457, 0.1144, 0.0592], device='cuda:1'), in_proj_covar=tensor([0.0254, 0.0349, 0.0293, 0.0239, 0.0294, 0.0241, 0.0278, 0.0239], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 22:01:05,329 INFO [train.py:903] (1/4) Epoch 12, batch 3550, loss[loss=0.2251, simple_loss=0.2955, pruned_loss=0.07731, over 19458.00 frames. ], tot_loss[loss=0.2329, simple_loss=0.3067, pruned_loss=0.0795, over 3817828.04 frames. ], batch size: 49, lr: 6.93e-03, grad_scale: 8.0 2023-04-01 22:01:26,754 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.897e+02 5.737e+02 7.042e+02 1.025e+03 1.962e+03, threshold=1.408e+03, percent-clipped=6.0 2023-04-01 22:02:07,311 INFO [train.py:903] (1/4) Epoch 12, batch 3600, loss[loss=0.2115, simple_loss=0.2959, pruned_loss=0.06357, over 19623.00 frames. ], tot_loss[loss=0.2319, simple_loss=0.3058, pruned_loss=0.07904, over 3835847.88 frames. ], batch size: 57, lr: 6.93e-03, grad_scale: 8.0 2023-04-01 22:02:07,819 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.2766, 2.1692, 1.9610, 1.7087, 1.6428, 1.8307, 0.4641, 1.0615], device='cuda:1'), covar=tensor([0.0496, 0.0466, 0.0359, 0.0601, 0.0942, 0.0665, 0.1002, 0.0856], device='cuda:1'), in_proj_covar=tensor([0.0342, 0.0336, 0.0334, 0.0360, 0.0432, 0.0359, 0.0317, 0.0328], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 22:02:08,941 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=78709.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 22:03:09,176 INFO [train.py:903] (1/4) Epoch 12, batch 3650, loss[loss=0.3104, simple_loss=0.3606, pruned_loss=0.1301, over 13349.00 frames. ], tot_loss[loss=0.2321, simple_loss=0.306, pruned_loss=0.07905, over 3836917.10 frames. ], batch size: 136, lr: 6.93e-03, grad_scale: 8.0 2023-04-01 22:03:33,806 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.478e+02 5.318e+02 6.562e+02 8.219e+02 2.478e+03, threshold=1.312e+03, percent-clipped=4.0 2023-04-01 22:03:37,859 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2845, 1.3335, 1.5844, 1.4950, 2.1782, 2.0022, 2.2114, 0.7927], device='cuda:1'), covar=tensor([0.2195, 0.3715, 0.2207, 0.1739, 0.1351, 0.1848, 0.1325, 0.3681], device='cuda:1'), in_proj_covar=tensor([0.0488, 0.0577, 0.0606, 0.0437, 0.0592, 0.0488, 0.0643, 0.0492], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 22:04:14,202 INFO [train.py:903] (1/4) Epoch 12, batch 3700, loss[loss=0.2143, simple_loss=0.2966, pruned_loss=0.06603, over 19663.00 frames. ], tot_loss[loss=0.2318, simple_loss=0.306, pruned_loss=0.07878, over 3836741.89 frames. ], batch size: 55, lr: 6.93e-03, grad_scale: 8.0 2023-04-01 22:04:31,407 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=78822.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 22:05:15,856 INFO [train.py:903] (1/4) Epoch 12, batch 3750, loss[loss=0.1827, simple_loss=0.2656, pruned_loss=0.0499, over 19746.00 frames. ], tot_loss[loss=0.232, simple_loss=0.3059, pruned_loss=0.07909, over 3832863.10 frames. ], batch size: 51, lr: 6.93e-03, grad_scale: 8.0 2023-04-01 22:05:18,442 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.3684, 3.9431, 2.5389, 3.5533, 0.9995, 3.7280, 3.7210, 3.8359], device='cuda:1'), covar=tensor([0.0650, 0.1012, 0.2031, 0.0739, 0.3870, 0.0740, 0.0812, 0.0947], device='cuda:1'), in_proj_covar=tensor([0.0444, 0.0369, 0.0445, 0.0321, 0.0381, 0.0376, 0.0364, 0.0399], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 22:05:37,742 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.608e+02 5.233e+02 6.179e+02 8.242e+02 1.500e+03, threshold=1.236e+03, percent-clipped=2.0 2023-04-01 22:06:16,426 INFO [train.py:903] (1/4) Epoch 12, batch 3800, loss[loss=0.2767, simple_loss=0.354, pruned_loss=0.09975, over 19336.00 frames. ], tot_loss[loss=0.2307, simple_loss=0.3049, pruned_loss=0.07823, over 3843437.01 frames. ], batch size: 66, lr: 6.92e-03, grad_scale: 8.0 2023-04-01 22:06:29,218 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=78918.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 22:06:53,675 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9_rvb from training. Duration: 29.1166875 2023-04-01 22:06:54,063 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=78937.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 22:07:17,950 INFO [train.py:903] (1/4) Epoch 12, batch 3850, loss[loss=0.206, simple_loss=0.2772, pruned_loss=0.0674, over 19348.00 frames. ], tot_loss[loss=0.232, simple_loss=0.3058, pruned_loss=0.07914, over 3823085.79 frames. ], batch size: 47, lr: 6.92e-03, grad_scale: 8.0 2023-04-01 22:07:19,949 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 2023-04-01 22:07:27,616 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.44 vs. limit=2.0 2023-04-01 22:07:28,345 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=78965.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 22:07:40,748 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.425e+02 5.982e+02 6.977e+02 9.373e+02 2.137e+03, threshold=1.395e+03, percent-clipped=8.0 2023-04-01 22:07:49,892 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=78983.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 22:07:58,116 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=78990.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 22:08:21,052 INFO [train.py:903] (1/4) Epoch 12, batch 3900, loss[loss=0.2339, simple_loss=0.3114, pruned_loss=0.07824, over 19608.00 frames. ], tot_loss[loss=0.2313, simple_loss=0.3055, pruned_loss=0.07852, over 3824457.24 frames. ], batch size: 57, lr: 6.92e-03, grad_scale: 8.0 2023-04-01 22:08:51,055 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=79033.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 22:09:22,296 INFO [train.py:903] (1/4) Epoch 12, batch 3950, loss[loss=0.2093, simple_loss=0.29, pruned_loss=0.06431, over 19596.00 frames. ], tot_loss[loss=0.2302, simple_loss=0.3047, pruned_loss=0.07792, over 3831057.52 frames. ], batch size: 57, lr: 6.92e-03, grad_scale: 8.0 2023-04-01 22:09:29,052 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 2023-04-01 22:09:40,885 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2283, 2.0240, 1.4462, 1.2943, 1.8473, 1.0539, 1.2407, 1.6617], device='cuda:1'), covar=tensor([0.0891, 0.0657, 0.1086, 0.0730, 0.0437, 0.1257, 0.0663, 0.0403], device='cuda:1'), in_proj_covar=tensor([0.0291, 0.0300, 0.0326, 0.0244, 0.0235, 0.0318, 0.0287, 0.0260], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 22:09:43,634 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.615e+02 6.132e+02 6.993e+02 8.356e+02 2.478e+03, threshold=1.399e+03, percent-clipped=5.0 2023-04-01 22:10:04,420 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.44 vs. limit=5.0 2023-04-01 22:10:18,481 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3708, 1.2029, 1.2099, 1.7510, 1.5307, 1.6003, 1.7525, 1.4262], device='cuda:1'), covar=tensor([0.0951, 0.1053, 0.1182, 0.0753, 0.0823, 0.0770, 0.0799, 0.0773], device='cuda:1'), in_proj_covar=tensor([0.0210, 0.0225, 0.0223, 0.0247, 0.0236, 0.0211, 0.0195, 0.0202], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-01 22:10:22,896 INFO [train.py:903] (1/4) Epoch 12, batch 4000, loss[loss=0.29, simple_loss=0.3584, pruned_loss=0.1108, over 19575.00 frames. ], tot_loss[loss=0.2317, simple_loss=0.3056, pruned_loss=0.07886, over 3829065.82 frames. ], batch size: 61, lr: 6.91e-03, grad_scale: 8.0 2023-04-01 22:11:13,132 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1_rvb from training. Duration: 0.7545625 2023-04-01 22:11:24,603 INFO [train.py:903] (1/4) Epoch 12, batch 4050, loss[loss=0.2489, simple_loss=0.3205, pruned_loss=0.0886, over 19088.00 frames. ], tot_loss[loss=0.2315, simple_loss=0.3053, pruned_loss=0.07882, over 3829147.26 frames. ], batch size: 69, lr: 6.91e-03, grad_scale: 8.0 2023-04-01 22:11:47,128 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.257e+02 5.067e+02 6.190e+02 7.758e+02 2.001e+03, threshold=1.238e+03, percent-clipped=2.0 2023-04-01 22:12:07,821 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=79193.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 22:12:26,661 INFO [train.py:903] (1/4) Epoch 12, batch 4100, loss[loss=0.2383, simple_loss=0.3073, pruned_loss=0.08468, over 19481.00 frames. ], tot_loss[loss=0.2319, simple_loss=0.3058, pruned_loss=0.079, over 3827337.22 frames. ], batch size: 64, lr: 6.91e-03, grad_scale: 8.0 2023-04-01 22:12:39,310 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=79218.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 22:13:03,856 WARNING [train.py:1073] (1/4) Exclude cut with ID 497-129325-0061-9566_sp1.1_rvb from training. Duration: 0.97725 2023-04-01 22:13:27,004 INFO [train.py:903] (1/4) Epoch 12, batch 4150, loss[loss=0.2919, simple_loss=0.3489, pruned_loss=0.1175, over 13230.00 frames. ], tot_loss[loss=0.2319, simple_loss=0.3058, pruned_loss=0.079, over 3816107.97 frames. ], batch size: 135, lr: 6.91e-03, grad_scale: 8.0 2023-04-01 22:13:45,807 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.3211, 2.2779, 2.4155, 3.5116, 2.3207, 3.1529, 2.9168, 2.2624], device='cuda:1'), covar=tensor([0.3844, 0.3362, 0.1441, 0.1738, 0.3751, 0.1452, 0.3382, 0.2737], device='cuda:1'), in_proj_covar=tensor([0.0801, 0.0826, 0.0654, 0.0897, 0.0788, 0.0714, 0.0793, 0.0715], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 22:13:49,453 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.496e+02 5.552e+02 6.928e+02 9.304e+02 2.111e+03, threshold=1.386e+03, percent-clipped=6.0 2023-04-01 22:14:06,867 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=79289.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 22:14:23,433 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=79303.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 22:14:28,763 INFO [train.py:903] (1/4) Epoch 12, batch 4200, loss[loss=0.2943, simple_loss=0.3473, pruned_loss=0.1206, over 14022.00 frames. ], tot_loss[loss=0.2311, simple_loss=0.3053, pruned_loss=0.07844, over 3819886.47 frames. ], batch size: 137, lr: 6.91e-03, grad_scale: 8.0 2023-04-01 22:14:33,318 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 2023-04-01 22:14:35,749 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=79314.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 22:14:53,795 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=79327.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 22:15:22,528 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=79351.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 22:15:31,219 INFO [train.py:903] (1/4) Epoch 12, batch 4250, loss[loss=0.2615, simple_loss=0.3354, pruned_loss=0.09381, over 17816.00 frames. ], tot_loss[loss=0.2313, simple_loss=0.3054, pruned_loss=0.0786, over 3797523.76 frames. ], batch size: 83, lr: 6.90e-03, grad_scale: 8.0 2023-04-01 22:15:43,246 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9_rvb from training. Duration: 29.816625 2023-04-01 22:15:52,357 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.517e+02 4.925e+02 6.557e+02 8.003e+02 1.515e+03, threshold=1.311e+03, percent-clipped=3.0 2023-04-01 22:15:54,756 WARNING [train.py:1073] (1/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 2023-04-01 22:16:00,672 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=79382.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 22:16:13,131 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=79393.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 22:16:33,046 INFO [train.py:903] (1/4) Epoch 12, batch 4300, loss[loss=0.2581, simple_loss=0.3235, pruned_loss=0.09636, over 19577.00 frames. ], tot_loss[loss=0.232, simple_loss=0.3064, pruned_loss=0.07885, over 3817236.95 frames. ], batch size: 61, lr: 6.90e-03, grad_scale: 8.0 2023-04-01 22:17:14,372 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=79442.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 22:17:24,324 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 2023-04-01 22:17:29,465 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.1053, 1.9808, 1.7834, 2.2172, 2.0302, 1.8148, 1.9839, 2.0000], device='cuda:1'), covar=tensor([0.0825, 0.1372, 0.1231, 0.0849, 0.1123, 0.0451, 0.1038, 0.0615], device='cuda:1'), in_proj_covar=tensor([0.0254, 0.0352, 0.0292, 0.0240, 0.0297, 0.0242, 0.0281, 0.0240], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 22:17:32,601 INFO [train.py:903] (1/4) Epoch 12, batch 4350, loss[loss=0.1941, simple_loss=0.2699, pruned_loss=0.05908, over 19745.00 frames. ], tot_loss[loss=0.2318, simple_loss=0.3062, pruned_loss=0.0787, over 3805609.28 frames. ], batch size: 45, lr: 6.90e-03, grad_scale: 8.0 2023-04-01 22:17:52,358 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.5135, 4.0750, 4.2359, 4.1901, 1.5141, 3.9736, 3.4997, 3.9130], device='cuda:1'), covar=tensor([0.1396, 0.0773, 0.0516, 0.0585, 0.5312, 0.0710, 0.0595, 0.1024], device='cuda:1'), in_proj_covar=tensor([0.0682, 0.0618, 0.0812, 0.0691, 0.0740, 0.0567, 0.0496, 0.0749], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1') 2023-04-01 22:17:54,382 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.719e+02 5.544e+02 6.866e+02 8.754e+02 2.036e+03, threshold=1.373e+03, percent-clipped=4.0 2023-04-01 22:18:34,726 INFO [train.py:903] (1/4) Epoch 12, batch 4400, loss[loss=0.2087, simple_loss=0.2802, pruned_loss=0.06861, over 19776.00 frames. ], tot_loss[loss=0.2303, simple_loss=0.3047, pruned_loss=0.078, over 3816277.06 frames. ], batch size: 48, lr: 6.90e-03, grad_scale: 8.0 2023-04-01 22:18:58,958 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_rvb from training. Duration: 25.285 2023-04-01 22:19:07,331 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 2023-04-01 22:19:38,009 INFO [train.py:903] (1/4) Epoch 12, batch 4450, loss[loss=0.2233, simple_loss=0.3052, pruned_loss=0.07071, over 19531.00 frames. ], tot_loss[loss=0.2317, simple_loss=0.3057, pruned_loss=0.07886, over 3801591.39 frames. ], batch size: 56, lr: 6.90e-03, grad_scale: 8.0 2023-04-01 22:20:00,036 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.129e+02 5.261e+02 6.909e+02 8.531e+02 1.990e+03, threshold=1.382e+03, percent-clipped=5.0 2023-04-01 22:20:00,417 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([0.9956, 1.2114, 1.5593, 0.7850, 2.3930, 3.0436, 2.7267, 3.2303], device='cuda:1'), covar=tensor([0.1637, 0.3656, 0.3140, 0.2448, 0.0514, 0.0174, 0.0259, 0.0226], device='cuda:1'), in_proj_covar=tensor([0.0258, 0.0300, 0.0329, 0.0252, 0.0219, 0.0161, 0.0207, 0.0207], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 22:20:33,637 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=79603.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 22:20:41,079 INFO [train.py:903] (1/4) Epoch 12, batch 4500, loss[loss=0.201, simple_loss=0.2813, pruned_loss=0.06031, over 19842.00 frames. ], tot_loss[loss=0.2309, simple_loss=0.3051, pruned_loss=0.07833, over 3805068.10 frames. ], batch size: 52, lr: 6.89e-03, grad_scale: 8.0 2023-04-01 22:20:50,886 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.4106, 2.0667, 1.9383, 2.9793, 2.3230, 2.6391, 2.7347, 2.6322], device='cuda:1'), covar=tensor([0.0709, 0.0806, 0.0947, 0.0790, 0.0790, 0.0624, 0.0780, 0.0543], device='cuda:1'), in_proj_covar=tensor([0.0209, 0.0223, 0.0222, 0.0245, 0.0234, 0.0211, 0.0193, 0.0201], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-01 22:21:29,751 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=79647.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 22:21:43,340 INFO [train.py:903] (1/4) Epoch 12, batch 4550, loss[loss=0.2228, simple_loss=0.2848, pruned_loss=0.08035, over 19778.00 frames. ], tot_loss[loss=0.2309, simple_loss=0.3054, pruned_loss=0.07817, over 3819698.78 frames. ], batch size: 47, lr: 6.89e-03, grad_scale: 8.0 2023-04-01 22:21:47,118 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.0495, 1.2169, 1.5709, 0.6281, 2.0231, 2.4622, 2.1757, 2.6267], device='cuda:1'), covar=tensor([0.1411, 0.3400, 0.2936, 0.2295, 0.0543, 0.0231, 0.0328, 0.0301], device='cuda:1'), in_proj_covar=tensor([0.0256, 0.0298, 0.0327, 0.0250, 0.0218, 0.0160, 0.0206, 0.0206], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 22:21:52,476 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9_rvb from training. Duration: 28.72225 2023-04-01 22:21:58,518 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.5086, 1.1479, 1.2875, 1.2738, 2.1753, 0.9337, 2.0261, 2.3339], device='cuda:1'), covar=tensor([0.0663, 0.2667, 0.2665, 0.1495, 0.0809, 0.2015, 0.0976, 0.0484], device='cuda:1'), in_proj_covar=tensor([0.0360, 0.0342, 0.0352, 0.0320, 0.0344, 0.0330, 0.0336, 0.0356], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 22:22:04,161 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=79675.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 22:22:04,985 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.248e+02 5.458e+02 6.277e+02 7.572e+02 1.495e+03, threshold=1.255e+03, percent-clipped=2.0 2023-04-01 22:22:15,964 WARNING [train.py:1073] (1/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 2023-04-01 22:22:29,768 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=79695.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 22:22:33,648 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=79698.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 22:22:44,776 INFO [train.py:903] (1/4) Epoch 12, batch 4600, loss[loss=0.2382, simple_loss=0.3174, pruned_loss=0.07954, over 19703.00 frames. ], tot_loss[loss=0.2303, simple_loss=0.305, pruned_loss=0.0778, over 3832414.71 frames. ], batch size: 53, lr: 6.89e-03, grad_scale: 8.0 2023-04-01 22:22:47,085 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.4656, 1.8346, 1.9555, 2.8862, 2.0926, 2.8834, 2.6586, 2.6686], device='cuda:1'), covar=tensor([0.0654, 0.0880, 0.0902, 0.0814, 0.0885, 0.0548, 0.0834, 0.0547], device='cuda:1'), in_proj_covar=tensor([0.0209, 0.0225, 0.0223, 0.0247, 0.0236, 0.0212, 0.0194, 0.0203], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-01 22:22:58,547 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.9457, 3.5714, 2.3734, 3.2723, 0.8456, 3.4142, 3.4103, 3.4990], device='cuda:1'), covar=tensor([0.0827, 0.1259, 0.2022, 0.0779, 0.3750, 0.0932, 0.0842, 0.1177], device='cuda:1'), in_proj_covar=tensor([0.0444, 0.0372, 0.0443, 0.0317, 0.0381, 0.0377, 0.0367, 0.0397], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 22:23:04,285 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=79723.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 22:23:07,643 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=79726.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 22:23:21,779 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=79737.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 22:23:46,649 INFO [train.py:903] (1/4) Epoch 12, batch 4650, loss[loss=0.2428, simple_loss=0.3154, pruned_loss=0.08512, over 19613.00 frames. ], tot_loss[loss=0.2309, simple_loss=0.3053, pruned_loss=0.07827, over 3828991.00 frames. ], batch size: 50, lr: 6.89e-03, grad_scale: 8.0 2023-04-01 22:23:51,762 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=79762.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 22:24:06,326 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_rvb from training. Duration: 0.92 2023-04-01 22:24:09,608 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.145e+02 5.614e+02 7.010e+02 8.934e+02 1.991e+03, threshold=1.402e+03, percent-clipped=7.0 2023-04-01 22:24:13,555 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.1644, 1.9349, 1.9617, 2.3193, 2.1210, 1.8350, 1.8927, 2.1681], device='cuda:1'), covar=tensor([0.0721, 0.1197, 0.1055, 0.0662, 0.0934, 0.0449, 0.0987, 0.0509], device='cuda:1'), in_proj_covar=tensor([0.0257, 0.0354, 0.0294, 0.0241, 0.0299, 0.0244, 0.0283, 0.0241], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 22:24:14,686 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=79780.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 22:24:15,616 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_rvb from training. Duration: 0.83 2023-04-01 22:24:48,917 INFO [train.py:903] (1/4) Epoch 12, batch 4700, loss[loss=0.1913, simple_loss=0.2691, pruned_loss=0.05676, over 19721.00 frames. ], tot_loss[loss=0.2322, simple_loss=0.3062, pruned_loss=0.07913, over 3819190.12 frames. ], batch size: 51, lr: 6.88e-03, grad_scale: 8.0 2023-04-01 22:24:52,458 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=79810.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 22:25:12,487 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9_rvb from training. Duration: 0.92225 2023-04-01 22:25:30,392 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=79841.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 22:25:45,590 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=79852.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 22:25:52,119 INFO [train.py:903] (1/4) Epoch 12, batch 4750, loss[loss=0.2222, simple_loss=0.303, pruned_loss=0.07072, over 19304.00 frames. ], tot_loss[loss=0.2308, simple_loss=0.3052, pruned_loss=0.07822, over 3834859.36 frames. ], batch size: 66, lr: 6.88e-03, grad_scale: 8.0 2023-04-01 22:26:14,211 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.848e+02 5.173e+02 6.366e+02 7.623e+02 1.625e+03, threshold=1.273e+03, percent-clipped=2.0 2023-04-01 22:26:20,953 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.52 vs. limit=2.0 2023-04-01 22:26:54,253 INFO [train.py:903] (1/4) Epoch 12, batch 4800, loss[loss=0.2667, simple_loss=0.335, pruned_loss=0.09918, over 19614.00 frames. ], tot_loss[loss=0.2308, simple_loss=0.3055, pruned_loss=0.07807, over 3827262.93 frames. ], batch size: 57, lr: 6.88e-03, grad_scale: 8.0 2023-04-01 22:27:16,980 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2348, 1.2771, 1.2581, 1.0629, 1.0663, 1.0966, 0.0450, 0.3340], device='cuda:1'), covar=tensor([0.0440, 0.0458, 0.0273, 0.0354, 0.0891, 0.0473, 0.0845, 0.0798], device='cuda:1'), in_proj_covar=tensor([0.0340, 0.0327, 0.0330, 0.0355, 0.0428, 0.0356, 0.0310, 0.0325], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 22:27:42,679 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=79947.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 22:27:45,278 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9416, 1.9633, 2.2060, 1.9746, 3.1569, 2.5957, 3.2322, 2.2231], device='cuda:1'), covar=tensor([0.1718, 0.2974, 0.1900, 0.1505, 0.1107, 0.1561, 0.1140, 0.2528], device='cuda:1'), in_proj_covar=tensor([0.0494, 0.0586, 0.0612, 0.0441, 0.0596, 0.0497, 0.0648, 0.0497], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 22:27:56,278 INFO [train.py:903] (1/4) Epoch 12, batch 4850, loss[loss=0.2679, simple_loss=0.3387, pruned_loss=0.09854, over 19712.00 frames. ], tot_loss[loss=0.2307, simple_loss=0.3052, pruned_loss=0.07805, over 3829882.91 frames. ], batch size: 59, lr: 6.88e-03, grad_scale: 16.0 2023-04-01 22:27:57,043 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.97 vs. limit=5.0 2023-04-01 22:28:19,180 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.933e+02 5.432e+02 6.685e+02 9.186e+02 1.976e+03, threshold=1.337e+03, percent-clipped=11.0 2023-04-01 22:28:23,585 WARNING [train.py:1073] (1/4) Exclude cut with ID 774-127930-0014-48411_sp1.1_rvb from training. Duration: 0.95 2023-04-01 22:28:43,813 WARNING [train.py:1073] (1/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1_rvb from training. Duration: 0.9409375 2023-04-01 22:28:48,471 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 2023-04-01 22:28:51,008 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 2023-04-01 22:28:52,775 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.56 vs. limit=2.0 2023-04-01 22:28:59,979 INFO [train.py:903] (1/4) Epoch 12, batch 4900, loss[loss=0.1996, simple_loss=0.2878, pruned_loss=0.05571, over 19628.00 frames. ], tot_loss[loss=0.2299, simple_loss=0.3042, pruned_loss=0.07778, over 3821949.48 frames. ], batch size: 57, lr: 6.88e-03, grad_scale: 8.0 2023-04-01 22:29:02,313 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1_rvb from training. Duration: 25.3818125 2023-04-01 22:29:13,787 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=80018.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 22:29:14,555 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=80019.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 22:29:21,284 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 2023-04-01 22:29:43,697 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=80043.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 22:30:03,747 INFO [train.py:903] (1/4) Epoch 12, batch 4950, loss[loss=0.2817, simple_loss=0.3468, pruned_loss=0.1083, over 18832.00 frames. ], tot_loss[loss=0.2304, simple_loss=0.3044, pruned_loss=0.07821, over 3804118.91 frames. ], batch size: 74, lr: 6.87e-03, grad_scale: 8.0 2023-04-01 22:30:08,656 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=80062.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 22:30:13,652 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=80066.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 22:30:20,440 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 2023-04-01 22:30:26,823 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.980e+02 5.263e+02 6.788e+02 8.958e+02 2.034e+03, threshold=1.358e+03, percent-clipped=5.0 2023-04-01 22:30:44,596 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=80091.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 22:30:45,327 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 2023-04-01 22:30:52,462 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=80097.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 22:31:04,649 INFO [train.py:903] (1/4) Epoch 12, batch 5000, loss[loss=0.2691, simple_loss=0.3389, pruned_loss=0.09959, over 19384.00 frames. ], tot_loss[loss=0.2322, simple_loss=0.3058, pruned_loss=0.07927, over 3803196.01 frames. ], batch size: 70, lr: 6.87e-03, grad_scale: 8.0 2023-04-01 22:31:05,110 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=80108.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 22:31:13,753 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_rvb from training. Duration: 27.14 2023-04-01 22:31:22,088 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=80122.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 22:31:24,081 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=80124.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 22:31:25,179 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 2023-04-01 22:31:37,147 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=80133.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 22:31:38,231 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=80134.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 22:32:06,839 INFO [train.py:903] (1/4) Epoch 12, batch 5050, loss[loss=0.2131, simple_loss=0.2846, pruned_loss=0.07075, over 19855.00 frames. ], tot_loss[loss=0.2317, simple_loss=0.3053, pruned_loss=0.07899, over 3812362.90 frames. ], batch size: 52, lr: 6.87e-03, grad_scale: 8.0 2023-04-01 22:32:30,894 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.181e+02 5.736e+02 7.301e+02 9.465e+02 2.500e+03, threshold=1.460e+03, percent-clipped=5.0 2023-04-01 22:32:35,788 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6312, 1.4074, 1.4378, 2.0101, 1.5626, 1.9325, 1.8649, 1.6758], device='cuda:1'), covar=tensor([0.0797, 0.0984, 0.1041, 0.0744, 0.0853, 0.0673, 0.0871, 0.0681], device='cuda:1'), in_proj_covar=tensor([0.0212, 0.0226, 0.0225, 0.0246, 0.0235, 0.0212, 0.0196, 0.0204], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-01 22:32:41,375 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9_rvb from training. Duration: 26.62775 2023-04-01 22:33:08,740 INFO [train.py:903] (1/4) Epoch 12, batch 5100, loss[loss=0.1933, simple_loss=0.2751, pruned_loss=0.05581, over 19743.00 frames. ], tot_loss[loss=0.2325, simple_loss=0.3061, pruned_loss=0.07947, over 3815487.37 frames. ], batch size: 51, lr: 6.87e-03, grad_scale: 8.0 2023-04-01 22:33:21,056 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 2023-04-01 22:33:23,168 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_rvb from training. Duration: 25.035 2023-04-01 22:33:26,512 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_rvb from training. Duration: 27.92 2023-04-01 22:33:47,554 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=80239.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 22:34:11,775 INFO [train.py:903] (1/4) Epoch 12, batch 5150, loss[loss=0.2862, simple_loss=0.3487, pruned_loss=0.1119, over 18731.00 frames. ], tot_loss[loss=0.2327, simple_loss=0.3064, pruned_loss=0.07949, over 3822384.33 frames. ], batch size: 74, lr: 6.87e-03, grad_scale: 8.0 2023-04-01 22:34:23,653 WARNING [train.py:1073] (1/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 2023-04-01 22:34:34,583 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.377e+02 5.205e+02 6.062e+02 7.794e+02 1.645e+03, threshold=1.212e+03, percent-clipped=2.0 2023-04-01 22:34:59,538 WARNING [train.py:1073] (1/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1_rvb from training. Duration: 0.9681875 2023-04-01 22:35:02,338 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4681, 1.5322, 1.8244, 1.6724, 2.7410, 2.3172, 2.9596, 1.2981], device='cuda:1'), covar=tensor([0.2177, 0.3869, 0.2320, 0.1756, 0.1391, 0.1866, 0.1384, 0.3645], device='cuda:1'), in_proj_covar=tensor([0.0492, 0.0583, 0.0609, 0.0440, 0.0595, 0.0497, 0.0647, 0.0497], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 22:35:05,765 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=80301.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 22:35:13,722 INFO [train.py:903] (1/4) Epoch 12, batch 5200, loss[loss=0.2065, simple_loss=0.2926, pruned_loss=0.06019, over 19761.00 frames. ], tot_loss[loss=0.2316, simple_loss=0.3057, pruned_loss=0.07874, over 3828484.52 frames. ], batch size: 54, lr: 6.86e-03, grad_scale: 8.0 2023-04-01 22:35:26,531 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=80318.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 22:35:27,324 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 2023-04-01 22:35:58,851 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=80343.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 22:36:03,532 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.9027, 4.2976, 4.6093, 4.5905, 1.4610, 4.2494, 3.6955, 4.3217], device='cuda:1'), covar=tensor([0.1627, 0.0686, 0.0569, 0.0664, 0.6171, 0.0718, 0.0633, 0.1132], device='cuda:1'), in_proj_covar=tensor([0.0682, 0.0615, 0.0812, 0.0691, 0.0731, 0.0562, 0.0490, 0.0736], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1') 2023-04-01 22:36:13,577 WARNING [train.py:1073] (1/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 2023-04-01 22:36:16,614 INFO [train.py:903] (1/4) Epoch 12, batch 5250, loss[loss=0.2665, simple_loss=0.3343, pruned_loss=0.09936, over 19763.00 frames. ], tot_loss[loss=0.231, simple_loss=0.3054, pruned_loss=0.07834, over 3820668.59 frames. ], batch size: 63, lr: 6.86e-03, grad_scale: 8.0 2023-04-01 22:36:41,376 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.664e+02 5.603e+02 6.465e+02 8.351e+02 1.434e+03, threshold=1.293e+03, percent-clipped=3.0 2023-04-01 22:36:57,041 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=80390.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 22:37:18,490 INFO [train.py:903] (1/4) Epoch 12, batch 5300, loss[loss=0.2226, simple_loss=0.3034, pruned_loss=0.07095, over 19687.00 frames. ], tot_loss[loss=0.2286, simple_loss=0.3035, pruned_loss=0.07679, over 3830729.66 frames. ], batch size: 59, lr: 6.86e-03, grad_scale: 8.0 2023-04-01 22:37:22,934 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=80411.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 22:37:28,730 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=80415.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 22:37:42,147 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 2023-04-01 22:38:23,299 INFO [train.py:903] (1/4) Epoch 12, batch 5350, loss[loss=0.1912, simple_loss=0.2669, pruned_loss=0.05771, over 19733.00 frames. ], tot_loss[loss=0.2286, simple_loss=0.3035, pruned_loss=0.07681, over 3834599.00 frames. ], batch size: 51, lr: 6.86e-03, grad_scale: 8.0 2023-04-01 22:38:44,528 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.162e+02 5.085e+02 6.728e+02 8.660e+02 2.071e+03, threshold=1.346e+03, percent-clipped=4.0 2023-04-01 22:38:50,312 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=80481.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 22:38:59,005 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9_rvb from training. Duration: 26.438875 2023-04-01 22:39:08,327 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=80495.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 22:39:23,543 INFO [train.py:903] (1/4) Epoch 12, batch 5400, loss[loss=0.2403, simple_loss=0.3124, pruned_loss=0.08411, over 17350.00 frames. ], tot_loss[loss=0.23, simple_loss=0.3045, pruned_loss=0.07773, over 3815170.07 frames. ], batch size: 101, lr: 6.85e-03, grad_scale: 8.0 2023-04-01 22:39:28,312 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=80512.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 22:39:38,431 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=80520.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 22:40:24,807 INFO [train.py:903] (1/4) Epoch 12, batch 5450, loss[loss=0.2189, simple_loss=0.3023, pruned_loss=0.06773, over 19780.00 frames. ], tot_loss[loss=0.2316, simple_loss=0.3059, pruned_loss=0.07867, over 3817627.39 frames. ], batch size: 56, lr: 6.85e-03, grad_scale: 8.0 2023-04-01 22:40:49,073 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.130e+02 5.165e+02 6.319e+02 8.444e+02 1.726e+03, threshold=1.264e+03, percent-clipped=5.0 2023-04-01 22:41:26,465 INFO [train.py:903] (1/4) Epoch 12, batch 5500, loss[loss=0.1994, simple_loss=0.2811, pruned_loss=0.05884, over 19600.00 frames. ], tot_loss[loss=0.2304, simple_loss=0.3047, pruned_loss=0.07802, over 3821329.40 frames. ], batch size: 50, lr: 6.85e-03, grad_scale: 8.0 2023-04-01 22:41:53,692 WARNING [train.py:1073] (1/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9_rvb from training. Duration: 27.47775 2023-04-01 22:42:12,372 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=80645.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 22:42:29,084 INFO [train.py:903] (1/4) Epoch 12, batch 5550, loss[loss=0.2327, simple_loss=0.3125, pruned_loss=0.07644, over 19673.00 frames. ], tot_loss[loss=0.23, simple_loss=0.3042, pruned_loss=0.07786, over 3815369.60 frames. ], batch size: 59, lr: 6.85e-03, grad_scale: 8.0 2023-04-01 22:42:38,027 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 2023-04-01 22:42:51,885 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.910e+02 5.291e+02 6.725e+02 8.423e+02 1.958e+03, threshold=1.345e+03, percent-clipped=4.0 2023-04-01 22:42:53,856 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 2023-04-01 22:43:28,262 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 2023-04-01 22:43:31,868 INFO [train.py:903] (1/4) Epoch 12, batch 5600, loss[loss=0.2052, simple_loss=0.273, pruned_loss=0.06874, over 19711.00 frames. ], tot_loss[loss=0.2303, simple_loss=0.3044, pruned_loss=0.07814, over 3832713.18 frames. ], batch size: 46, lr: 6.85e-03, grad_scale: 8.0 2023-04-01 22:44:12,809 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.6706, 2.0266, 2.3524, 2.8672, 2.2502, 2.2502, 2.1116, 2.7683], device='cuda:1'), covar=tensor([0.0761, 0.1715, 0.1170, 0.0812, 0.1238, 0.0435, 0.1110, 0.0525], device='cuda:1'), in_proj_covar=tensor([0.0254, 0.0351, 0.0293, 0.0238, 0.0299, 0.0240, 0.0281, 0.0241], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 22:44:29,071 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=80754.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 22:44:30,112 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=80755.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 22:44:33,522 INFO [train.py:903] (1/4) Epoch 12, batch 5650, loss[loss=0.223, simple_loss=0.3047, pruned_loss=0.0706, over 19676.00 frames. ], tot_loss[loss=0.2307, simple_loss=0.3043, pruned_loss=0.0785, over 3816762.32 frames. ], batch size: 58, lr: 6.84e-03, grad_scale: 8.0 2023-04-01 22:44:36,175 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=80760.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 22:44:57,739 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.312e+02 5.237e+02 6.303e+02 7.862e+02 2.175e+03, threshold=1.261e+03, percent-clipped=3.0 2023-04-01 22:45:15,155 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=80791.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 22:45:20,703 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9_rvb from training. Duration: 28.0944375 2023-04-01 22:45:35,188 INFO [train.py:903] (1/4) Epoch 12, batch 5700, loss[loss=0.2217, simple_loss=0.3054, pruned_loss=0.069, over 19678.00 frames. ], tot_loss[loss=0.2308, simple_loss=0.3044, pruned_loss=0.07861, over 3816330.68 frames. ], batch size: 53, lr: 6.84e-03, grad_scale: 8.0 2023-04-01 22:45:57,689 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=80825.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 22:46:35,804 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=80856.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 22:46:38,764 INFO [train.py:903] (1/4) Epoch 12, batch 5750, loss[loss=0.1812, simple_loss=0.2571, pruned_loss=0.05259, over 19761.00 frames. ], tot_loss[loss=0.2303, simple_loss=0.3043, pruned_loss=0.07816, over 3830626.59 frames. ], batch size: 47, lr: 6.84e-03, grad_scale: 8.0 2023-04-01 22:46:39,963 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9_rvb from training. Duration: 33.038875 2023-04-01 22:46:47,871 WARNING [train.py:1073] (1/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 2023-04-01 22:46:52,465 WARNING [train.py:1073] (1/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 2023-04-01 22:46:52,775 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=80870.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 22:47:00,404 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.450e+02 5.415e+02 6.686e+02 8.336e+02 1.819e+03, threshold=1.337e+03, percent-clipped=1.0 2023-04-01 22:47:40,324 INFO [train.py:903] (1/4) Epoch 12, batch 5800, loss[loss=0.2064, simple_loss=0.2744, pruned_loss=0.06925, over 19429.00 frames. ], tot_loss[loss=0.2295, simple_loss=0.3038, pruned_loss=0.0776, over 3831646.65 frames. ], batch size: 48, lr: 6.84e-03, grad_scale: 8.0 2023-04-01 22:48:16,662 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0 2023-04-01 22:48:21,655 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=80940.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 22:48:34,497 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.8454, 2.0666, 1.7105, 3.1079, 2.3203, 3.1364, 2.0269, 1.4425], device='cuda:1'), covar=tensor([0.5207, 0.4204, 0.2665, 0.2671, 0.4352, 0.1959, 0.5718, 0.4883], device='cuda:1'), in_proj_covar=tensor([0.0803, 0.0829, 0.0654, 0.0891, 0.0793, 0.0721, 0.0789, 0.0718], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 22:48:41,888 INFO [train.py:903] (1/4) Epoch 12, batch 5850, loss[loss=0.2142, simple_loss=0.2983, pruned_loss=0.06508, over 19416.00 frames. ], tot_loss[loss=0.2297, simple_loss=0.3038, pruned_loss=0.07783, over 3829330.97 frames. ], batch size: 70, lr: 6.84e-03, grad_scale: 8.0 2023-04-01 22:48:58,000 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=80971.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 22:49:06,076 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.947e+02 5.387e+02 6.409e+02 7.183e+02 1.679e+03, threshold=1.282e+03, percent-clipped=1.0 2023-04-01 22:49:10,746 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.5568, 4.1289, 2.4895, 3.7057, 0.8701, 3.8554, 3.8887, 3.9673], device='cuda:1'), covar=tensor([0.0691, 0.1158, 0.2159, 0.0779, 0.4456, 0.0826, 0.0828, 0.1183], device='cuda:1'), in_proj_covar=tensor([0.0443, 0.0372, 0.0442, 0.0320, 0.0379, 0.0375, 0.0366, 0.0397], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 22:49:18,788 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.5796, 4.1277, 2.6353, 3.7135, 0.8912, 3.8616, 3.8535, 3.9607], device='cuda:1'), covar=tensor([0.0633, 0.1152, 0.1963, 0.0773, 0.4245, 0.0813, 0.0846, 0.1178], device='cuda:1'), in_proj_covar=tensor([0.0443, 0.0372, 0.0442, 0.0320, 0.0379, 0.0375, 0.0366, 0.0397], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 22:49:43,641 INFO [train.py:903] (1/4) Epoch 12, batch 5900, loss[loss=0.2231, simple_loss=0.3152, pruned_loss=0.06557, over 18787.00 frames. ], tot_loss[loss=0.2311, simple_loss=0.3049, pruned_loss=0.07866, over 3823708.74 frames. ], batch size: 74, lr: 6.83e-03, grad_scale: 8.0 2023-04-01 22:49:47,128 WARNING [train.py:1073] (1/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 2023-04-01 22:49:55,125 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5323, 2.4701, 1.7628, 1.6531, 2.2305, 1.4036, 1.3519, 1.9374], device='cuda:1'), covar=tensor([0.0996, 0.0628, 0.0883, 0.0665, 0.0438, 0.1064, 0.0669, 0.0434], device='cuda:1'), in_proj_covar=tensor([0.0293, 0.0307, 0.0328, 0.0249, 0.0242, 0.0323, 0.0287, 0.0266], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 22:49:55,145 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=81016.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 22:50:09,739 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9_rvb from training. Duration: 27.511125 2023-04-01 22:50:25,095 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=81041.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 22:50:35,645 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.8880, 1.5499, 1.4134, 1.8224, 1.5501, 1.6007, 1.5140, 1.6463], device='cuda:1'), covar=tensor([0.0846, 0.1193, 0.1397, 0.0822, 0.1095, 0.0486, 0.1153, 0.0707], device='cuda:1'), in_proj_covar=tensor([0.0257, 0.0358, 0.0298, 0.0242, 0.0304, 0.0246, 0.0284, 0.0246], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 22:50:47,179 INFO [train.py:903] (1/4) Epoch 12, batch 5950, loss[loss=0.1979, simple_loss=0.2773, pruned_loss=0.05927, over 19688.00 frames. ], tot_loss[loss=0.2309, simple_loss=0.3048, pruned_loss=0.07853, over 3834965.43 frames. ], batch size: 53, lr: 6.83e-03, grad_scale: 8.0 2023-04-01 22:51:10,058 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.575e+02 5.377e+02 6.760e+02 8.757e+02 1.989e+03, threshold=1.352e+03, percent-clipped=8.0 2023-04-01 22:51:36,989 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=81098.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 22:51:49,674 INFO [train.py:903] (1/4) Epoch 12, batch 6000, loss[loss=0.239, simple_loss=0.3133, pruned_loss=0.08238, over 18882.00 frames. ], tot_loss[loss=0.2293, simple_loss=0.3035, pruned_loss=0.0776, over 3831752.11 frames. ], batch size: 74, lr: 6.83e-03, grad_scale: 8.0 2023-04-01 22:51:49,674 INFO [train.py:928] (1/4) Computing validation loss 2023-04-01 22:52:03,359 INFO [train.py:937] (1/4) Epoch 12, validation: loss=0.1765, simple_loss=0.2774, pruned_loss=0.03779, over 944034.00 frames. 2023-04-01 22:52:03,361 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18649MB 2023-04-01 22:52:25,573 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=81126.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 22:52:34,826 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=81134.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 22:52:35,857 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=81135.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 22:52:57,875 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=81151.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 22:53:05,460 INFO [train.py:903] (1/4) Epoch 12, batch 6050, loss[loss=0.2319, simple_loss=0.3069, pruned_loss=0.07844, over 19651.00 frames. ], tot_loss[loss=0.2294, simple_loss=0.3031, pruned_loss=0.07785, over 3832494.50 frames. ], batch size: 58, lr: 6.83e-03, grad_scale: 8.0 2023-04-01 22:53:27,718 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.001e+02 5.041e+02 6.677e+02 8.260e+02 1.738e+03, threshold=1.335e+03, percent-clipped=2.0 2023-04-01 22:53:52,488 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=81196.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 22:53:57,464 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.41 vs. limit=2.0 2023-04-01 22:54:05,934 INFO [train.py:903] (1/4) Epoch 12, batch 6100, loss[loss=0.2562, simple_loss=0.3296, pruned_loss=0.09138, over 19656.00 frames. ], tot_loss[loss=0.2303, simple_loss=0.3043, pruned_loss=0.0782, over 3830869.48 frames. ], batch size: 55, lr: 6.83e-03, grad_scale: 8.0 2023-04-01 22:54:11,849 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=81213.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 22:54:20,970 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=81221.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 22:54:28,710 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=81227.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 22:54:57,891 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=81250.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 22:55:00,248 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=81252.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 22:55:06,761 INFO [train.py:903] (1/4) Epoch 12, batch 6150, loss[loss=0.2292, simple_loss=0.3098, pruned_loss=0.07427, over 18116.00 frames. ], tot_loss[loss=0.2293, simple_loss=0.3032, pruned_loss=0.07776, over 3842807.25 frames. ], batch size: 83, lr: 6.82e-03, grad_scale: 8.0 2023-04-01 22:55:15,126 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.88 vs. limit=5.0 2023-04-01 22:55:33,027 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.393e+02 5.279e+02 6.402e+02 8.020e+02 2.167e+03, threshold=1.280e+03, percent-clipped=2.0 2023-04-01 22:55:39,197 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9_rvb from training. Duration: 31.02225 2023-04-01 22:55:40,515 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.5696, 4.1181, 2.5793, 3.6461, 1.0901, 3.9728, 3.9651, 3.9403], device='cuda:1'), covar=tensor([0.0639, 0.1038, 0.2024, 0.0828, 0.3763, 0.0720, 0.0762, 0.1191], device='cuda:1'), in_proj_covar=tensor([0.0441, 0.0370, 0.0440, 0.0318, 0.0377, 0.0374, 0.0362, 0.0394], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 22:56:12,045 INFO [train.py:903] (1/4) Epoch 12, batch 6200, loss[loss=0.2483, simple_loss=0.3133, pruned_loss=0.09164, over 19527.00 frames. ], tot_loss[loss=0.2302, simple_loss=0.304, pruned_loss=0.07823, over 3822015.89 frames. ], batch size: 54, lr: 6.82e-03, grad_scale: 8.0 2023-04-01 22:56:22,853 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=81317.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 22:56:25,200 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=81319.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 22:57:11,775 INFO [train.py:903] (1/4) Epoch 12, batch 6250, loss[loss=0.2733, simple_loss=0.3362, pruned_loss=0.1052, over 19668.00 frames. ], tot_loss[loss=0.2301, simple_loss=0.3041, pruned_loss=0.078, over 3829196.50 frames. ], batch size: 59, lr: 6.82e-03, grad_scale: 8.0 2023-04-01 22:57:23,878 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3271, 1.4030, 1.7104, 1.6280, 2.7568, 2.2695, 2.8844, 1.1709], device='cuda:1'), covar=tensor([0.2373, 0.4072, 0.2526, 0.1835, 0.1470, 0.2036, 0.1499, 0.4033], device='cuda:1'), in_proj_covar=tensor([0.0495, 0.0589, 0.0619, 0.0442, 0.0598, 0.0505, 0.0652, 0.0502], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 22:57:33,898 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.420e+02 5.051e+02 6.163e+02 7.297e+02 1.401e+03, threshold=1.233e+03, percent-clipped=3.0 2023-04-01 22:57:42,927 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 2023-04-01 22:57:43,141 WARNING [train.py:1073] (1/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9_rvb from training. Duration: 25.988875 2023-04-01 22:58:13,176 INFO [train.py:903] (1/4) Epoch 12, batch 6300, loss[loss=0.2139, simple_loss=0.2777, pruned_loss=0.07507, over 16570.00 frames. ], tot_loss[loss=0.2298, simple_loss=0.3041, pruned_loss=0.07772, over 3840038.14 frames. ], batch size: 36, lr: 6.82e-03, grad_scale: 8.0 2023-04-01 22:59:14,516 INFO [train.py:903] (1/4) Epoch 12, batch 6350, loss[loss=0.2563, simple_loss=0.3197, pruned_loss=0.09644, over 13265.00 frames. ], tot_loss[loss=0.2302, simple_loss=0.3042, pruned_loss=0.07807, over 3835636.30 frames. ], batch size: 137, lr: 6.81e-03, grad_scale: 8.0 2023-04-01 22:59:28,528 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=81469.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 22:59:38,668 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3949, 1.5006, 1.7595, 1.5901, 2.6326, 2.2289, 2.7023, 1.1696], device='cuda:1'), covar=tensor([0.2145, 0.3701, 0.2278, 0.1794, 0.1313, 0.1874, 0.1320, 0.3568], device='cuda:1'), in_proj_covar=tensor([0.0489, 0.0582, 0.0613, 0.0437, 0.0591, 0.0497, 0.0645, 0.0497], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 22:59:39,338 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.788e+02 5.412e+02 6.997e+02 8.497e+02 1.750e+03, threshold=1.399e+03, percent-clipped=2.0 2023-04-01 22:59:40,730 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=81478.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 22:59:58,963 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 2023-04-01 22:59:59,487 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=81494.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 23:00:14,192 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=81506.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 23:00:16,036 INFO [train.py:903] (1/4) Epoch 12, batch 6400, loss[loss=0.2366, simple_loss=0.3095, pruned_loss=0.08187, over 19469.00 frames. ], tot_loss[loss=0.2313, simple_loss=0.3051, pruned_loss=0.07871, over 3821518.68 frames. ], batch size: 64, lr: 6.81e-03, grad_scale: 8.0 2023-04-01 23:00:45,268 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=81531.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 23:01:19,151 INFO [train.py:903] (1/4) Epoch 12, batch 6450, loss[loss=0.2182, simple_loss=0.295, pruned_loss=0.07066, over 19756.00 frames. ], tot_loss[loss=0.2318, simple_loss=0.3058, pruned_loss=0.07886, over 3819213.52 frames. ], batch size: 54, lr: 6.81e-03, grad_scale: 8.0 2023-04-01 23:01:41,360 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.161e+02 5.839e+02 6.972e+02 8.326e+02 2.886e+03, threshold=1.394e+03, percent-clipped=3.0 2023-04-01 23:02:03,363 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=81593.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 23:02:06,518 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9_rvb from training. Duration: 28.638875 2023-04-01 23:02:20,381 INFO [train.py:903] (1/4) Epoch 12, batch 6500, loss[loss=0.2877, simple_loss=0.3451, pruned_loss=0.1151, over 13936.00 frames. ], tot_loss[loss=0.2303, simple_loss=0.3047, pruned_loss=0.07795, over 3821886.25 frames. ], batch size: 136, lr: 6.81e-03, grad_scale: 8.0 2023-04-01 23:02:27,391 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1_rvb from training. Duration: 0.836375 2023-04-01 23:03:22,584 INFO [train.py:903] (1/4) Epoch 12, batch 6550, loss[loss=0.2338, simple_loss=0.3083, pruned_loss=0.07967, over 19863.00 frames. ], tot_loss[loss=0.2291, simple_loss=0.3038, pruned_loss=0.07714, over 3816919.60 frames. ], batch size: 52, lr: 6.81e-03, grad_scale: 8.0 2023-04-01 23:03:26,197 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=81661.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 23:03:28,586 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=81663.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 23:03:47,197 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.342e+02 5.338e+02 6.617e+02 7.892e+02 1.534e+03, threshold=1.323e+03, percent-clipped=1.0 2023-04-01 23:04:00,911 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=81688.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 23:04:08,934 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.0642, 2.7966, 2.1638, 2.5270, 0.8621, 2.6843, 2.5751, 2.7078], device='cuda:1'), covar=tensor([0.1224, 0.1332, 0.1986, 0.0998, 0.3429, 0.1027, 0.1147, 0.1326], device='cuda:1'), in_proj_covar=tensor([0.0441, 0.0372, 0.0440, 0.0317, 0.0378, 0.0372, 0.0366, 0.0397], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 23:04:24,231 INFO [train.py:903] (1/4) Epoch 12, batch 6600, loss[loss=0.1825, simple_loss=0.2588, pruned_loss=0.05316, over 19734.00 frames. ], tot_loss[loss=0.2297, simple_loss=0.3043, pruned_loss=0.07754, over 3814408.18 frames. ], batch size: 46, lr: 6.80e-03, grad_scale: 8.0 2023-04-01 23:04:56,151 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0 2023-04-01 23:05:15,796 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1410, 1.2106, 1.9009, 1.5822, 3.1188, 4.4040, 4.3073, 4.8031], device='cuda:1'), covar=tensor([0.1694, 0.3733, 0.3047, 0.2047, 0.0503, 0.0190, 0.0161, 0.0124], device='cuda:1'), in_proj_covar=tensor([0.0257, 0.0299, 0.0328, 0.0252, 0.0220, 0.0163, 0.0206, 0.0208], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 23:05:26,106 INFO [train.py:903] (1/4) Epoch 12, batch 6650, loss[loss=0.2168, simple_loss=0.302, pruned_loss=0.06578, over 19477.00 frames. ], tot_loss[loss=0.2299, simple_loss=0.3048, pruned_loss=0.07749, over 3809521.74 frames. ], batch size: 64, lr: 6.80e-03, grad_scale: 8.0 2023-04-01 23:05:46,860 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=81776.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 23:05:47,589 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.542e+02 5.145e+02 6.461e+02 8.134e+02 1.737e+03, threshold=1.292e+03, percent-clipped=3.0 2023-04-01 23:05:49,140 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=81778.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 23:06:26,738 INFO [train.py:903] (1/4) Epoch 12, batch 6700, loss[loss=0.2643, simple_loss=0.3337, pruned_loss=0.09745, over 19740.00 frames. ], tot_loss[loss=0.2309, simple_loss=0.3055, pruned_loss=0.07811, over 3809573.80 frames. ], batch size: 63, lr: 6.80e-03, grad_scale: 8.0 2023-04-01 23:07:17,011 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=81849.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 23:07:26,538 INFO [train.py:903] (1/4) Epoch 12, batch 6750, loss[loss=0.2661, simple_loss=0.3398, pruned_loss=0.09623, over 19270.00 frames. ], tot_loss[loss=0.2327, simple_loss=0.3069, pruned_loss=0.07926, over 3816407.27 frames. ], batch size: 66, lr: 6.80e-03, grad_scale: 4.0 2023-04-01 23:07:37,176 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.2047, 3.6347, 2.2200, 2.2193, 3.2953, 1.8661, 1.6733, 2.2491], device='cuda:1'), covar=tensor([0.1338, 0.0525, 0.0988, 0.0754, 0.0447, 0.1107, 0.0874, 0.0666], device='cuda:1'), in_proj_covar=tensor([0.0293, 0.0306, 0.0325, 0.0246, 0.0240, 0.0325, 0.0288, 0.0265], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 23:07:45,324 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=81874.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 23:07:48,743 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.8937, 1.5062, 1.4784, 1.7751, 1.6193, 1.6357, 1.4456, 1.7315], device='cuda:1'), covar=tensor([0.0894, 0.1227, 0.1356, 0.0868, 0.1038, 0.0519, 0.1250, 0.0698], device='cuda:1'), in_proj_covar=tensor([0.0253, 0.0349, 0.0293, 0.0237, 0.0293, 0.0240, 0.0278, 0.0240], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 23:07:49,446 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.155e+02 6.276e+02 7.333e+02 1.082e+03 2.540e+03, threshold=1.467e+03, percent-clipped=11.0 2023-04-01 23:08:23,232 INFO [train.py:903] (1/4) Epoch 12, batch 6800, loss[loss=0.2507, simple_loss=0.3291, pruned_loss=0.08617, over 18810.00 frames. ], tot_loss[loss=0.2312, simple_loss=0.3058, pruned_loss=0.07832, over 3825968.36 frames. ], batch size: 74, lr: 6.80e-03, grad_scale: 8.0 2023-04-01 23:09:09,106 WARNING [train.py:1073] (1/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 2023-04-01 23:09:10,252 WARNING [train.py:1073] (1/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 2023-04-01 23:09:12,679 INFO [train.py:903] (1/4) Epoch 13, batch 0, loss[loss=0.2187, simple_loss=0.3003, pruned_loss=0.06854, over 19638.00 frames. ], tot_loss[loss=0.2187, simple_loss=0.3003, pruned_loss=0.06854, over 19638.00 frames. ], batch size: 61, lr: 6.53e-03, grad_scale: 8.0 2023-04-01 23:09:12,679 INFO [train.py:928] (1/4) Computing validation loss 2023-04-01 23:09:23,578 INFO [train.py:937] (1/4) Epoch 13, validation: loss=0.176, simple_loss=0.2772, pruned_loss=0.03738, over 944034.00 frames. 2023-04-01 23:09:23,579 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18649MB 2023-04-01 23:09:35,402 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 2023-04-01 23:09:41,795 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.6143, 2.2008, 2.1414, 2.6273, 2.4662, 2.0262, 1.7976, 2.6830], device='cuda:1'), covar=tensor([0.0848, 0.1884, 0.1489, 0.0988, 0.1400, 0.0688, 0.1478, 0.0662], device='cuda:1'), in_proj_covar=tensor([0.0254, 0.0352, 0.0294, 0.0239, 0.0295, 0.0242, 0.0279, 0.0242], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 23:09:47,676 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.44 vs. limit=5.0 2023-04-01 23:10:14,565 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.117e+02 5.222e+02 6.740e+02 8.452e+02 3.268e+03, threshold=1.348e+03, percent-clipped=4.0 2023-04-01 23:10:17,105 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.2944, 3.0082, 2.1547, 2.7310, 0.7927, 2.9003, 2.8211, 2.9120], device='cuda:1'), covar=tensor([0.1143, 0.1490, 0.2189, 0.1096, 0.3856, 0.1078, 0.1090, 0.1450], device='cuda:1'), in_proj_covar=tensor([0.0441, 0.0371, 0.0441, 0.0316, 0.0377, 0.0370, 0.0364, 0.0398], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 23:10:23,830 INFO [train.py:903] (1/4) Epoch 13, batch 50, loss[loss=0.2059, simple_loss=0.2761, pruned_loss=0.0678, over 19416.00 frames. ], tot_loss[loss=0.2244, simple_loss=0.3002, pruned_loss=0.07428, over 859330.05 frames. ], batch size: 48, lr: 6.53e-03, grad_scale: 8.0 2023-04-01 23:10:46,563 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=82005.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 23:10:58,445 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.3800, 3.0251, 2.3363, 2.2841, 2.1597, 2.3507, 0.9865, 2.1820], device='cuda:1'), covar=tensor([0.0543, 0.0432, 0.0525, 0.0913, 0.0908, 0.0948, 0.1050, 0.0832], device='cuda:1'), in_proj_covar=tensor([0.0341, 0.0330, 0.0330, 0.0357, 0.0428, 0.0356, 0.0312, 0.0325], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 23:10:59,203 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1_rvb from training. Duration: 27.0318125 2023-04-01 23:11:20,796 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=82032.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 23:11:21,013 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=82032.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 23:11:23,300 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=82034.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 23:11:25,072 INFO [train.py:903] (1/4) Epoch 13, batch 100, loss[loss=0.2277, simple_loss=0.3025, pruned_loss=0.07648, over 19096.00 frames. ], tot_loss[loss=0.2242, simple_loss=0.3002, pruned_loss=0.07405, over 1522794.91 frames. ], batch size: 69, lr: 6.53e-03, grad_scale: 8.0 2023-04-01 23:11:36,614 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 2023-04-01 23:11:52,552 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=82057.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 23:11:55,566 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=82059.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 23:12:11,460 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=82073.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 23:12:16,819 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.549e+02 4.805e+02 6.218e+02 7.513e+02 1.266e+03, threshold=1.244e+03, percent-clipped=0.0 2023-04-01 23:12:25,778 INFO [train.py:903] (1/4) Epoch 13, batch 150, loss[loss=0.2505, simple_loss=0.3191, pruned_loss=0.09094, over 19623.00 frames. ], tot_loss[loss=0.2286, simple_loss=0.3045, pruned_loss=0.07638, over 2025701.06 frames. ], batch size: 57, lr: 6.52e-03, grad_scale: 8.0 2023-04-01 23:13:10,427 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.3453, 3.9392, 2.6426, 3.5281, 1.0996, 3.7487, 3.7702, 3.7926], device='cuda:1'), covar=tensor([0.0695, 0.1074, 0.1905, 0.0781, 0.3804, 0.0772, 0.0804, 0.1104], device='cuda:1'), in_proj_covar=tensor([0.0443, 0.0370, 0.0444, 0.0316, 0.0380, 0.0374, 0.0367, 0.0399], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 23:13:23,612 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 2023-04-01 23:13:24,756 INFO [train.py:903] (1/4) Epoch 13, batch 200, loss[loss=0.2114, simple_loss=0.3023, pruned_loss=0.06029, over 19775.00 frames. ], tot_loss[loss=0.2297, simple_loss=0.3054, pruned_loss=0.07696, over 2442785.17 frames. ], batch size: 56, lr: 6.52e-03, grad_scale: 8.0 2023-04-01 23:13:40,458 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=82147.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 23:13:42,567 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0729, 1.7677, 1.6239, 2.0952, 1.9407, 1.7793, 1.7000, 1.9641], device='cuda:1'), covar=tensor([0.0879, 0.1531, 0.1417, 0.0946, 0.1168, 0.0492, 0.1164, 0.0690], device='cuda:1'), in_proj_covar=tensor([0.0252, 0.0348, 0.0292, 0.0238, 0.0292, 0.0240, 0.0278, 0.0240], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 23:14:04,570 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.1912, 3.7171, 2.3597, 2.2327, 3.3299, 2.0154, 1.4951, 2.2103], device='cuda:1'), covar=tensor([0.1204, 0.0483, 0.0887, 0.0722, 0.0449, 0.0994, 0.0882, 0.0630], device='cuda:1'), in_proj_covar=tensor([0.0294, 0.0306, 0.0326, 0.0246, 0.0241, 0.0321, 0.0288, 0.0265], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 23:14:14,407 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.039e+02 5.002e+02 5.972e+02 7.403e+02 2.257e+03, threshold=1.194e+03, percent-clipped=4.0 2023-04-01 23:14:26,895 INFO [train.py:903] (1/4) Epoch 13, batch 250, loss[loss=0.2356, simple_loss=0.3095, pruned_loss=0.08087, over 19601.00 frames. ], tot_loss[loss=0.2287, simple_loss=0.3039, pruned_loss=0.07674, over 2751552.45 frames. ], batch size: 57, lr: 6.52e-03, grad_scale: 8.0 2023-04-01 23:15:26,891 INFO [train.py:903] (1/4) Epoch 13, batch 300, loss[loss=0.2079, simple_loss=0.2741, pruned_loss=0.07081, over 19733.00 frames. ], tot_loss[loss=0.2286, simple_loss=0.3035, pruned_loss=0.07686, over 2966225.66 frames. ], batch size: 46, lr: 6.52e-03, grad_scale: 8.0 2023-04-01 23:16:18,765 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.276e+02 5.753e+02 6.815e+02 9.164e+02 1.837e+03, threshold=1.363e+03, percent-clipped=5.0 2023-04-01 23:16:28,128 INFO [train.py:903] (1/4) Epoch 13, batch 350, loss[loss=0.2535, simple_loss=0.3284, pruned_loss=0.08933, over 19615.00 frames. ], tot_loss[loss=0.2287, simple_loss=0.3036, pruned_loss=0.07692, over 3174445.25 frames. ], batch size: 57, lr: 6.52e-03, grad_scale: 8.0 2023-04-01 23:16:30,470 WARNING [train.py:1073] (1/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1_rvb from training. Duration: 0.9681875 2023-04-01 23:17:28,528 INFO [train.py:903] (1/4) Epoch 13, batch 400, loss[loss=0.2257, simple_loss=0.2989, pruned_loss=0.07622, over 19545.00 frames. ], tot_loss[loss=0.229, simple_loss=0.3037, pruned_loss=0.0771, over 3324029.04 frames. ], batch size: 54, lr: 6.51e-03, grad_scale: 8.0 2023-04-01 23:17:47,112 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=82349.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 23:18:04,893 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=82364.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 23:18:10,379 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.5631, 4.1418, 2.6504, 3.6496, 0.8706, 3.9457, 3.9215, 4.0621], device='cuda:1'), covar=tensor([0.0643, 0.1035, 0.1822, 0.0779, 0.3928, 0.0708, 0.0775, 0.0983], device='cuda:1'), in_proj_covar=tensor([0.0440, 0.0370, 0.0442, 0.0318, 0.0378, 0.0371, 0.0365, 0.0398], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 23:18:21,966 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.254e+02 5.325e+02 6.166e+02 7.720e+02 2.046e+03, threshold=1.233e+03, percent-clipped=4.0 2023-04-01 23:18:28,267 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.2360, 1.7880, 1.9373, 2.6583, 2.0679, 2.4776, 2.6591, 2.5353], device='cuda:1'), covar=tensor([0.0755, 0.0931, 0.0977, 0.0899, 0.0911, 0.0741, 0.0839, 0.0573], device='cuda:1'), in_proj_covar=tensor([0.0209, 0.0226, 0.0226, 0.0245, 0.0234, 0.0211, 0.0193, 0.0201], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-01 23:18:31,268 INFO [train.py:903] (1/4) Epoch 13, batch 450, loss[loss=0.2453, simple_loss=0.323, pruned_loss=0.08378, over 19743.00 frames. ], tot_loss[loss=0.2267, simple_loss=0.3016, pruned_loss=0.07585, over 3432298.09 frames. ], batch size: 51, lr: 6.51e-03, grad_scale: 8.0 2023-04-01 23:18:53,540 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=82403.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 23:19:04,678 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 2023-04-01 23:19:04,711 WARNING [train.py:1073] (1/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 2023-04-01 23:19:09,596 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=82417.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 23:19:23,759 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=82428.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 23:19:25,737 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1375, 1.0904, 1.4354, 1.1633, 2.3852, 3.1839, 2.9772, 3.5257], device='cuda:1'), covar=tensor([0.1794, 0.4754, 0.4254, 0.2272, 0.0670, 0.0247, 0.0292, 0.0241], device='cuda:1'), in_proj_covar=tensor([0.0256, 0.0299, 0.0327, 0.0252, 0.0221, 0.0161, 0.0206, 0.0209], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 23:19:33,807 INFO [train.py:903] (1/4) Epoch 13, batch 500, loss[loss=0.2147, simple_loss=0.2779, pruned_loss=0.07572, over 19742.00 frames. ], tot_loss[loss=0.2272, simple_loss=0.3017, pruned_loss=0.07635, over 3520311.68 frames. ], batch size: 47, lr: 6.51e-03, grad_scale: 8.0 2023-04-01 23:20:06,477 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=82464.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 23:20:27,378 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.531e+02 5.153e+02 6.569e+02 8.401e+02 1.477e+03, threshold=1.314e+03, percent-clipped=3.0 2023-04-01 23:20:35,263 INFO [train.py:903] (1/4) Epoch 13, batch 550, loss[loss=0.245, simple_loss=0.3235, pruned_loss=0.08328, over 19502.00 frames. ], tot_loss[loss=0.2273, simple_loss=0.302, pruned_loss=0.07632, over 3590772.18 frames. ], batch size: 64, lr: 6.51e-03, grad_scale: 8.0 2023-04-01 23:21:05,900 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=82511.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 23:21:30,868 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=82532.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 23:21:35,044 INFO [train.py:903] (1/4) Epoch 13, batch 600, loss[loss=0.1926, simple_loss=0.2686, pruned_loss=0.05825, over 19847.00 frames. ], tot_loss[loss=0.2286, simple_loss=0.303, pruned_loss=0.07712, over 3635233.90 frames. ], batch size: 52, lr: 6.51e-03, grad_scale: 8.0 2023-04-01 23:22:17,360 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 2023-04-01 23:22:20,684 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.9061, 4.4273, 2.8733, 3.8769, 1.0826, 4.3040, 4.2699, 4.3306], device='cuda:1'), covar=tensor([0.0533, 0.0982, 0.1767, 0.0751, 0.3725, 0.0629, 0.0706, 0.0907], device='cuda:1'), in_proj_covar=tensor([0.0442, 0.0372, 0.0442, 0.0319, 0.0381, 0.0374, 0.0367, 0.0401], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 23:22:28,776 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.675e+02 5.313e+02 6.751e+02 8.249e+02 1.619e+03, threshold=1.350e+03, percent-clipped=3.0 2023-04-01 23:22:36,736 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.62 vs. limit=5.0 2023-04-01 23:22:36,937 INFO [train.py:903] (1/4) Epoch 13, batch 650, loss[loss=0.2257, simple_loss=0.3009, pruned_loss=0.07529, over 19736.00 frames. ], tot_loss[loss=0.2275, simple_loss=0.3018, pruned_loss=0.07656, over 3683730.17 frames. ], batch size: 51, lr: 6.50e-03, grad_scale: 8.0 2023-04-01 23:22:38,408 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=82587.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 23:23:40,974 INFO [train.py:903] (1/4) Epoch 13, batch 700, loss[loss=0.2062, simple_loss=0.2761, pruned_loss=0.06815, over 19741.00 frames. ], tot_loss[loss=0.2282, simple_loss=0.3029, pruned_loss=0.07677, over 3706759.43 frames. ], batch size: 51, lr: 6.50e-03, grad_scale: 8.0 2023-04-01 23:24:21,263 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9793, 3.4095, 1.9730, 1.8817, 3.0118, 1.5729, 1.3471, 2.2120], device='cuda:1'), covar=tensor([0.1252, 0.0499, 0.0952, 0.0851, 0.0411, 0.1142, 0.0873, 0.0586], device='cuda:1'), in_proj_covar=tensor([0.0295, 0.0307, 0.0329, 0.0248, 0.0242, 0.0323, 0.0289, 0.0267], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 23:24:36,357 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.507e+02 5.358e+02 6.603e+02 8.553e+02 2.977e+03, threshold=1.321e+03, percent-clipped=4.0 2023-04-01 23:24:44,577 INFO [train.py:903] (1/4) Epoch 13, batch 750, loss[loss=0.2773, simple_loss=0.3379, pruned_loss=0.1083, over 13357.00 frames. ], tot_loss[loss=0.2297, simple_loss=0.3041, pruned_loss=0.0776, over 3718742.48 frames. ], batch size: 138, lr: 6.50e-03, grad_scale: 8.0 2023-04-01 23:25:10,146 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=82708.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 23:25:28,783 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=82720.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 23:25:47,330 INFO [train.py:903] (1/4) Epoch 13, batch 800, loss[loss=0.2403, simple_loss=0.3199, pruned_loss=0.0804, over 19533.00 frames. ], tot_loss[loss=0.2296, simple_loss=0.3044, pruned_loss=0.07743, over 3741918.89 frames. ], batch size: 54, lr: 6.50e-03, grad_scale: 8.0 2023-04-01 23:25:58,375 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=82745.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 23:26:01,431 WARNING [train.py:1073] (1/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 2023-04-01 23:26:42,349 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.821e+02 5.442e+02 6.436e+02 7.821e+02 1.140e+03, threshold=1.287e+03, percent-clipped=0.0 2023-04-01 23:26:50,564 INFO [train.py:903] (1/4) Epoch 13, batch 850, loss[loss=0.2143, simple_loss=0.2993, pruned_loss=0.06461, over 19718.00 frames. ], tot_loss[loss=0.229, simple_loss=0.3039, pruned_loss=0.07702, over 3769884.04 frames. ], batch size: 59, lr: 6.50e-03, grad_scale: 8.0 2023-04-01 23:26:53,211 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=82788.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 23:27:27,000 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=82813.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 23:27:38,355 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=82823.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 23:27:44,767 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9_rvb from training. Duration: 25.061125 2023-04-01 23:27:52,721 INFO [train.py:903] (1/4) Epoch 13, batch 900, loss[loss=0.275, simple_loss=0.3229, pruned_loss=0.1135, over 19758.00 frames. ], tot_loss[loss=0.2292, simple_loss=0.3037, pruned_loss=0.07735, over 3792663.55 frames. ], batch size: 46, lr: 6.49e-03, grad_scale: 8.0 2023-04-01 23:27:59,015 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 2023-04-01 23:28:09,491 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 2023-04-01 23:28:19,295 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=82855.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 23:28:47,646 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.460e+02 5.906e+02 6.958e+02 9.103e+02 2.196e+03, threshold=1.392e+03, percent-clipped=5.0 2023-04-01 23:28:59,864 INFO [train.py:903] (1/4) Epoch 13, batch 950, loss[loss=0.1982, simple_loss=0.2803, pruned_loss=0.058, over 19840.00 frames. ], tot_loss[loss=0.229, simple_loss=0.3034, pruned_loss=0.07732, over 3808641.02 frames. ], batch size: 52, lr: 6.49e-03, grad_scale: 8.0 2023-04-01 23:29:04,321 WARNING [train.py:1073] (1/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9_rvb from training. Duration: 26.32775 2023-04-01 23:29:42,097 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.8582, 3.9792, 4.3531, 4.3671, 2.6029, 4.0150, 3.7266, 4.0759], device='cuda:1'), covar=tensor([0.1138, 0.2127, 0.0573, 0.0538, 0.3866, 0.0933, 0.0545, 0.0906], device='cuda:1'), in_proj_covar=tensor([0.0707, 0.0623, 0.0837, 0.0714, 0.0754, 0.0581, 0.0506, 0.0761], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-01 23:29:54,815 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.0320, 3.7072, 2.7911, 3.2357, 1.7832, 3.4549, 3.4288, 3.5715], device='cuda:1'), covar=tensor([0.0813, 0.1043, 0.1983, 0.0931, 0.2782, 0.0897, 0.0938, 0.1302], device='cuda:1'), in_proj_covar=tensor([0.0444, 0.0371, 0.0444, 0.0321, 0.0382, 0.0376, 0.0369, 0.0401], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 23:29:55,944 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=82931.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 23:30:01,534 INFO [train.py:903] (1/4) Epoch 13, batch 1000, loss[loss=0.1918, simple_loss=0.2686, pruned_loss=0.05748, over 19780.00 frames. ], tot_loss[loss=0.2271, simple_loss=0.3017, pruned_loss=0.0762, over 3802419.41 frames. ], batch size: 47, lr: 6.49e-03, grad_scale: 8.0 2023-04-01 23:30:39,444 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 2023-04-01 23:30:44,771 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=82970.0, num_to_drop=1, layers_to_drop={1} 2023-04-01 23:30:50,683 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.3169, 2.0495, 2.0556, 2.7980, 2.3086, 2.2077, 2.2350, 2.6033], device='cuda:1'), covar=tensor([0.0879, 0.1665, 0.1333, 0.0795, 0.1254, 0.0469, 0.1067, 0.0584], device='cuda:1'), in_proj_covar=tensor([0.0255, 0.0350, 0.0296, 0.0238, 0.0296, 0.0241, 0.0280, 0.0242], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 23:30:52,614 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 2023-04-01 23:30:54,571 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.083e+02 5.161e+02 6.395e+02 8.326e+02 2.115e+03, threshold=1.279e+03, percent-clipped=2.0 2023-04-01 23:30:57,047 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=82981.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 23:31:02,718 INFO [train.py:903] (1/4) Epoch 13, batch 1050, loss[loss=0.2264, simple_loss=0.3057, pruned_loss=0.07361, over 19608.00 frames. ], tot_loss[loss=0.2288, simple_loss=0.3032, pruned_loss=0.07715, over 3811262.38 frames. ], batch size: 61, lr: 6.49e-03, grad_scale: 8.0 2023-04-01 23:31:25,460 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.3276, 2.2079, 1.9187, 1.7893, 1.7417, 1.8136, 0.5417, 1.1182], device='cuda:1'), covar=tensor([0.0470, 0.0444, 0.0348, 0.0555, 0.0853, 0.0629, 0.1011, 0.0844], device='cuda:1'), in_proj_covar=tensor([0.0341, 0.0333, 0.0334, 0.0360, 0.0429, 0.0356, 0.0316, 0.0325], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 23:31:34,337 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 2023-04-01 23:32:04,667 INFO [train.py:903] (1/4) Epoch 13, batch 1100, loss[loss=0.304, simple_loss=0.3613, pruned_loss=0.1233, over 13913.00 frames. ], tot_loss[loss=0.229, simple_loss=0.3035, pruned_loss=0.07729, over 3815092.94 frames. ], batch size: 136, lr: 6.49e-03, grad_scale: 8.0 2023-04-01 23:32:19,542 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=83046.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 23:32:57,834 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.212e+02 5.014e+02 6.117e+02 7.878e+02 1.226e+03, threshold=1.223e+03, percent-clipped=0.0 2023-04-01 23:32:58,284 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=83079.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 23:33:08,489 INFO [train.py:903] (1/4) Epoch 13, batch 1150, loss[loss=0.2485, simple_loss=0.3202, pruned_loss=0.08845, over 19778.00 frames. ], tot_loss[loss=0.2284, simple_loss=0.3028, pruned_loss=0.07699, over 3825096.22 frames. ], batch size: 56, lr: 6.48e-03, grad_scale: 8.0 2023-04-01 23:33:30,502 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=83104.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 23:33:40,733 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2573, 1.3761, 1.6824, 1.4674, 2.5175, 2.1442, 2.6514, 1.0473], device='cuda:1'), covar=tensor([0.2260, 0.3847, 0.2329, 0.1773, 0.1455, 0.1900, 0.1396, 0.3779], device='cuda:1'), in_proj_covar=tensor([0.0494, 0.0583, 0.0618, 0.0442, 0.0598, 0.0498, 0.0646, 0.0499], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 23:34:10,933 INFO [train.py:903] (1/4) Epoch 13, batch 1200, loss[loss=0.2406, simple_loss=0.3129, pruned_loss=0.08414, over 18278.00 frames. ], tot_loss[loss=0.2281, simple_loss=0.3025, pruned_loss=0.07681, over 3824180.69 frames. ], batch size: 83, lr: 6.48e-03, grad_scale: 8.0 2023-04-01 23:34:40,376 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 2023-04-01 23:35:06,465 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.490e+02 6.101e+02 7.395e+02 1.032e+03 1.939e+03, threshold=1.479e+03, percent-clipped=13.0 2023-04-01 23:35:12,380 INFO [train.py:903] (1/4) Epoch 13, batch 1250, loss[loss=0.2146, simple_loss=0.2985, pruned_loss=0.06533, over 19527.00 frames. ], tot_loss[loss=0.2277, simple_loss=0.3019, pruned_loss=0.07672, over 3833631.43 frames. ], batch size: 54, lr: 6.48e-03, grad_scale: 4.0 2023-04-01 23:35:19,062 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 2023-04-01 23:35:48,382 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=83214.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 23:36:02,574 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=83226.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 23:36:13,266 INFO [train.py:903] (1/4) Epoch 13, batch 1300, loss[loss=0.2207, simple_loss=0.2979, pruned_loss=0.07179, over 19656.00 frames. ], tot_loss[loss=0.2283, simple_loss=0.3025, pruned_loss=0.07705, over 3836498.18 frames. ], batch size: 53, lr: 6.48e-03, grad_scale: 4.0 2023-04-01 23:36:33,767 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=83251.0, num_to_drop=1, layers_to_drop={0} 2023-04-01 23:37:08,171 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.019e+02 5.042e+02 6.195e+02 7.677e+02 1.204e+03, threshold=1.239e+03, percent-clipped=0.0 2023-04-01 23:37:17,158 INFO [train.py:903] (1/4) Epoch 13, batch 1350, loss[loss=0.2823, simple_loss=0.3417, pruned_loss=0.1115, over 19739.00 frames. ], tot_loss[loss=0.2282, simple_loss=0.3023, pruned_loss=0.07702, over 3845181.95 frames. ], batch size: 63, lr: 6.48e-03, grad_scale: 4.0 2023-04-01 23:37:37,997 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=83302.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 23:37:38,131 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=83302.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 23:38:07,354 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=83325.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 23:38:09,913 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=83327.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 23:38:21,071 INFO [train.py:903] (1/4) Epoch 13, batch 1400, loss[loss=0.2056, simple_loss=0.2795, pruned_loss=0.06587, over 19378.00 frames. ], tot_loss[loss=0.2275, simple_loss=0.3019, pruned_loss=0.07657, over 3845993.21 frames. ], batch size: 47, lr: 6.47e-03, grad_scale: 4.0 2023-04-01 23:38:30,598 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.7746, 4.4010, 2.6210, 3.8352, 1.2006, 4.1235, 4.1356, 4.2190], device='cuda:1'), covar=tensor([0.0551, 0.0919, 0.1963, 0.0711, 0.3574, 0.0712, 0.0793, 0.0940], device='cuda:1'), in_proj_covar=tensor([0.0442, 0.0370, 0.0444, 0.0322, 0.0381, 0.0378, 0.0369, 0.0402], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 23:39:16,727 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.257e+02 5.533e+02 6.494e+02 7.603e+02 1.656e+03, threshold=1.299e+03, percent-clipped=2.0 2023-04-01 23:39:20,296 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_rvb from training. Duration: 25.85 2023-04-01 23:39:22,624 INFO [train.py:903] (1/4) Epoch 13, batch 1450, loss[loss=0.2466, simple_loss=0.3019, pruned_loss=0.09563, over 19796.00 frames. ], tot_loss[loss=0.2282, simple_loss=0.3026, pruned_loss=0.07694, over 3847953.62 frames. ], batch size: 49, lr: 6.47e-03, grad_scale: 4.0 2023-04-01 23:40:24,403 INFO [train.py:903] (1/4) Epoch 13, batch 1500, loss[loss=0.2078, simple_loss=0.2747, pruned_loss=0.07044, over 19360.00 frames. ], tot_loss[loss=0.228, simple_loss=0.3025, pruned_loss=0.07675, over 3849919.43 frames. ], batch size: 48, lr: 6.47e-03, grad_scale: 4.0 2023-04-01 23:40:29,249 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=83440.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 23:41:19,836 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.552e+02 5.496e+02 6.437e+02 7.955e+02 2.023e+03, threshold=1.287e+03, percent-clipped=5.0 2023-04-01 23:41:26,532 INFO [train.py:903] (1/4) Epoch 13, batch 1550, loss[loss=0.1964, simple_loss=0.2762, pruned_loss=0.05824, over 19406.00 frames. ], tot_loss[loss=0.2279, simple_loss=0.3025, pruned_loss=0.07668, over 3826207.91 frames. ], batch size: 48, lr: 6.47e-03, grad_scale: 4.0 2023-04-01 23:41:32,716 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0 2023-04-01 23:42:30,072 INFO [train.py:903] (1/4) Epoch 13, batch 1600, loss[loss=0.198, simple_loss=0.2689, pruned_loss=0.06352, over 19764.00 frames. ], tot_loss[loss=0.2283, simple_loss=0.3028, pruned_loss=0.07689, over 3833672.46 frames. ], batch size: 47, lr: 6.47e-03, grad_scale: 8.0 2023-04-01 23:42:52,458 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([0.8206, 1.1971, 1.5696, 0.5518, 1.9786, 2.3970, 2.0921, 2.5552], device='cuda:1'), covar=tensor([0.1599, 0.3370, 0.2889, 0.2417, 0.0556, 0.0263, 0.0320, 0.0299], device='cuda:1'), in_proj_covar=tensor([0.0258, 0.0298, 0.0325, 0.0251, 0.0219, 0.0161, 0.0205, 0.0210], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 23:42:53,310 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9_rvb from training. Duration: 30.1555625 2023-04-01 23:42:55,751 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=83558.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 23:43:25,327 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.823e+02 5.445e+02 6.386e+02 7.908e+02 1.256e+03, threshold=1.277e+03, percent-clipped=0.0 2023-04-01 23:43:31,085 INFO [train.py:903] (1/4) Epoch 13, batch 1650, loss[loss=0.2133, simple_loss=0.2932, pruned_loss=0.06667, over 19370.00 frames. ], tot_loss[loss=0.2295, simple_loss=0.3037, pruned_loss=0.07772, over 3829744.32 frames. ], batch size: 70, lr: 6.47e-03, grad_scale: 8.0 2023-04-01 23:44:33,574 INFO [train.py:903] (1/4) Epoch 13, batch 1700, loss[loss=0.1953, simple_loss=0.2719, pruned_loss=0.05935, over 19315.00 frames. ], tot_loss[loss=0.2303, simple_loss=0.3043, pruned_loss=0.07813, over 3823794.33 frames. ], batch size: 44, lr: 6.46e-03, grad_scale: 8.0 2023-04-01 23:44:38,735 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.4515, 1.6126, 1.7408, 1.8023, 4.0672, 1.0609, 2.4890, 4.3061], device='cuda:1'), covar=tensor([0.0468, 0.2566, 0.2577, 0.1782, 0.0744, 0.2750, 0.1456, 0.0216], device='cuda:1'), in_proj_covar=tensor([0.0359, 0.0343, 0.0354, 0.0322, 0.0347, 0.0332, 0.0338, 0.0365], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 23:44:38,836 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.1018, 3.1480, 1.9827, 1.9918, 2.8604, 1.7205, 1.4102, 2.1143], device='cuda:1'), covar=tensor([0.1117, 0.0614, 0.0955, 0.0740, 0.0503, 0.1108, 0.0867, 0.0620], device='cuda:1'), in_proj_covar=tensor([0.0293, 0.0305, 0.0325, 0.0248, 0.0238, 0.0321, 0.0284, 0.0262], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 23:44:42,287 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.0499, 1.2559, 1.6566, 1.3968, 2.8688, 3.8082, 3.5852, 4.0491], device='cuda:1'), covar=tensor([0.1626, 0.3428, 0.3021, 0.2002, 0.0472, 0.0131, 0.0184, 0.0179], device='cuda:1'), in_proj_covar=tensor([0.0257, 0.0297, 0.0325, 0.0250, 0.0218, 0.0160, 0.0204, 0.0209], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 23:44:45,679 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=83646.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 23:45:15,482 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 2023-04-01 23:45:19,389 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=83673.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 23:45:27,806 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.234e+02 5.386e+02 6.785e+02 9.131e+02 1.645e+03, threshold=1.357e+03, percent-clipped=5.0 2023-04-01 23:45:33,647 INFO [train.py:903] (1/4) Epoch 13, batch 1750, loss[loss=0.2118, simple_loss=0.2843, pruned_loss=0.06961, over 19858.00 frames. ], tot_loss[loss=0.2293, simple_loss=0.3033, pruned_loss=0.07767, over 3816833.91 frames. ], batch size: 52, lr: 6.46e-03, grad_scale: 8.0 2023-04-01 23:45:49,000 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=83696.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 23:46:19,392 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=83721.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 23:46:37,939 INFO [train.py:903] (1/4) Epoch 13, batch 1800, loss[loss=0.24, simple_loss=0.2959, pruned_loss=0.092, over 19390.00 frames. ], tot_loss[loss=0.2285, simple_loss=0.3029, pruned_loss=0.07703, over 3806712.54 frames. ], batch size: 48, lr: 6.46e-03, grad_scale: 8.0 2023-04-01 23:47:08,593 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=83761.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 23:47:29,816 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=83778.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 23:47:32,957 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.146e+02 5.542e+02 6.939e+02 8.095e+02 2.139e+03, threshold=1.388e+03, percent-clipped=3.0 2023-04-01 23:47:35,257 WARNING [train.py:1073] (1/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9_rvb from training. Duration: 25.0944375 2023-04-01 23:47:39,807 INFO [train.py:903] (1/4) Epoch 13, batch 1850, loss[loss=0.2306, simple_loss=0.3123, pruned_loss=0.07443, over 19552.00 frames. ], tot_loss[loss=0.2292, simple_loss=0.3039, pruned_loss=0.07727, over 3820874.76 frames. ], batch size: 61, lr: 6.46e-03, grad_scale: 8.0 2023-04-01 23:48:11,512 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9_rvb from training. Duration: 27.8166875 2023-04-01 23:48:38,894 INFO [train.py:903] (1/4) Epoch 13, batch 1900, loss[loss=0.2411, simple_loss=0.3153, pruned_loss=0.08344, over 18699.00 frames. ], tot_loss[loss=0.2297, simple_loss=0.3044, pruned_loss=0.07745, over 3819836.66 frames. ], batch size: 74, lr: 6.46e-03, grad_scale: 8.0 2023-04-01 23:48:56,286 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 2023-04-01 23:49:00,755 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9_rvb from training. Duration: 27.02225 2023-04-01 23:49:23,793 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 2023-04-01 23:49:32,947 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.547e+02 5.453e+02 6.640e+02 7.751e+02 1.927e+03, threshold=1.328e+03, percent-clipped=4.0 2023-04-01 23:49:36,680 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.8242, 4.3600, 2.7502, 3.8486, 1.0354, 4.1616, 4.1930, 4.2358], device='cuda:1'), covar=tensor([0.0543, 0.0937, 0.1817, 0.0720, 0.3687, 0.0718, 0.0691, 0.0806], device='cuda:1'), in_proj_covar=tensor([0.0439, 0.0367, 0.0438, 0.0316, 0.0375, 0.0374, 0.0364, 0.0396], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 23:49:38,631 INFO [train.py:903] (1/4) Epoch 13, batch 1950, loss[loss=0.2472, simple_loss=0.3135, pruned_loss=0.09046, over 19517.00 frames. ], tot_loss[loss=0.23, simple_loss=0.3047, pruned_loss=0.07768, over 3816984.61 frames. ], batch size: 54, lr: 6.45e-03, grad_scale: 8.0 2023-04-01 23:50:31,120 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=83929.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 23:50:40,981 INFO [train.py:903] (1/4) Epoch 13, batch 2000, loss[loss=0.2077, simple_loss=0.2787, pruned_loss=0.06837, over 19724.00 frames. ], tot_loss[loss=0.2291, simple_loss=0.304, pruned_loss=0.07715, over 3820050.09 frames. ], batch size: 51, lr: 6.45e-03, grad_scale: 8.0 2023-04-01 23:51:02,466 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=83954.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 23:51:17,791 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=83967.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 23:51:36,105 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.295e+02 5.067e+02 6.527e+02 8.467e+02 1.955e+03, threshold=1.305e+03, percent-clipped=7.0 2023-04-01 23:51:38,291 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_rvb from training. Duration: 26.8349375 2023-04-01 23:51:38,745 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2359, 1.3055, 1.2597, 1.0717, 1.0927, 1.1654, 0.0385, 0.4096], device='cuda:1'), covar=tensor([0.0584, 0.0537, 0.0343, 0.0438, 0.1064, 0.0455, 0.0980, 0.0830], device='cuda:1'), in_proj_covar=tensor([0.0341, 0.0333, 0.0333, 0.0356, 0.0423, 0.0356, 0.0313, 0.0323], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 23:51:42,654 INFO [train.py:903] (1/4) Epoch 13, batch 2050, loss[loss=0.2413, simple_loss=0.3152, pruned_loss=0.08369, over 18183.00 frames. ], tot_loss[loss=0.2302, simple_loss=0.3045, pruned_loss=0.07794, over 3819714.09 frames. ], batch size: 83, lr: 6.45e-03, grad_scale: 8.0 2023-04-01 23:51:46,683 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2840, 1.2522, 1.6571, 1.3662, 2.8034, 3.7396, 3.5104, 3.9899], device='cuda:1'), covar=tensor([0.1527, 0.3548, 0.3124, 0.2070, 0.0498, 0.0160, 0.0198, 0.0177], device='cuda:1'), in_proj_covar=tensor([0.0258, 0.0299, 0.0328, 0.0252, 0.0219, 0.0162, 0.0207, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 23:51:56,805 WARNING [train.py:1073] (1/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 2023-04-01 23:51:57,792 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 2023-04-01 23:52:01,352 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6202, 1.4209, 1.4775, 2.1128, 1.6541, 1.8912, 2.0242, 1.7188], device='cuda:1'), covar=tensor([0.0793, 0.0965, 0.1018, 0.0787, 0.0871, 0.0726, 0.0866, 0.0703], device='cuda:1'), in_proj_covar=tensor([0.0207, 0.0221, 0.0222, 0.0244, 0.0232, 0.0210, 0.0194, 0.0202], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-01 23:52:03,616 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=84003.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 23:52:21,312 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 2023-04-01 23:52:22,804 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=84017.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 23:52:44,657 INFO [train.py:903] (1/4) Epoch 13, batch 2100, loss[loss=0.2464, simple_loss=0.3192, pruned_loss=0.08677, over 19784.00 frames. ], tot_loss[loss=0.2307, simple_loss=0.3051, pruned_loss=0.07818, over 3819107.82 frames. ], batch size: 56, lr: 6.45e-03, grad_scale: 8.0 2023-04-01 23:52:52,281 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=84042.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 23:52:57,292 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.7202, 1.8191, 2.0522, 2.3741, 1.6455, 2.1515, 2.2591, 1.9422], device='cuda:1'), covar=tensor([0.3522, 0.2833, 0.1393, 0.1726, 0.3181, 0.1592, 0.3527, 0.2637], device='cuda:1'), in_proj_covar=tensor([0.0812, 0.0841, 0.0659, 0.0897, 0.0797, 0.0726, 0.0800, 0.0722], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 23:53:14,858 WARNING [train.py:1073] (1/4) Exclude cut with ID 453-131332-0000-131866_sp0.9_rvb from training. Duration: 25.3333125 2023-04-01 23:53:36,228 WARNING [train.py:1073] (1/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9_rvb from training. Duration: 26.6166875 2023-04-01 23:53:39,575 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.864e+02 5.545e+02 6.946e+02 9.457e+02 3.064e+03, threshold=1.389e+03, percent-clipped=12.0 2023-04-01 23:53:45,268 INFO [train.py:903] (1/4) Epoch 13, batch 2150, loss[loss=0.2314, simple_loss=0.3088, pruned_loss=0.07705, over 19765.00 frames. ], tot_loss[loss=0.2317, simple_loss=0.3061, pruned_loss=0.0787, over 3804672.63 frames. ], batch size: 54, lr: 6.45e-03, grad_scale: 8.0 2023-04-01 23:54:23,413 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=84115.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 23:54:31,389 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=84122.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 23:54:49,420 INFO [train.py:903] (1/4) Epoch 13, batch 2200, loss[loss=0.2854, simple_loss=0.3529, pruned_loss=0.109, over 19718.00 frames. ], tot_loss[loss=0.2308, simple_loss=0.3054, pruned_loss=0.07806, over 3809094.31 frames. ], batch size: 63, lr: 6.44e-03, grad_scale: 8.0 2023-04-01 23:55:44,488 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.166e+02 5.936e+02 7.647e+02 9.699e+02 2.302e+03, threshold=1.529e+03, percent-clipped=8.0 2023-04-01 23:55:50,229 INFO [train.py:903] (1/4) Epoch 13, batch 2250, loss[loss=0.2304, simple_loss=0.3037, pruned_loss=0.07856, over 17497.00 frames. ], tot_loss[loss=0.2315, simple_loss=0.3061, pruned_loss=0.07846, over 3806499.07 frames. ], batch size: 101, lr: 6.44e-03, grad_scale: 8.0 2023-04-01 23:55:56,006 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.5559, 1.1464, 1.3993, 1.1906, 2.2196, 0.9319, 1.9622, 2.3817], device='cuda:1'), covar=tensor([0.0656, 0.2645, 0.2556, 0.1603, 0.0869, 0.2043, 0.0976, 0.0494], device='cuda:1'), in_proj_covar=tensor([0.0362, 0.0345, 0.0356, 0.0324, 0.0349, 0.0334, 0.0344, 0.0366], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 23:56:28,845 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=84217.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 23:56:33,074 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.8508, 2.0293, 2.3861, 2.1938, 3.1369, 3.7328, 3.6049, 3.9620], device='cuda:1'), covar=tensor([0.1293, 0.2628, 0.2340, 0.1776, 0.0935, 0.0277, 0.0165, 0.0194], device='cuda:1'), in_proj_covar=tensor([0.0259, 0.0300, 0.0327, 0.0251, 0.0219, 0.0163, 0.0206, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 23:56:36,138 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([6.1302, 5.5206, 3.1219, 4.8354, 1.1394, 5.5643, 5.4932, 5.6231], device='cuda:1'), covar=tensor([0.0400, 0.0757, 0.1622, 0.0584, 0.3936, 0.0505, 0.0620, 0.0773], device='cuda:1'), in_proj_covar=tensor([0.0441, 0.0369, 0.0445, 0.0318, 0.0380, 0.0374, 0.0368, 0.0400], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-01 23:56:51,981 INFO [train.py:903] (1/4) Epoch 13, batch 2300, loss[loss=0.2286, simple_loss=0.3068, pruned_loss=0.07519, over 19618.00 frames. ], tot_loss[loss=0.2312, simple_loss=0.3056, pruned_loss=0.0784, over 3797473.11 frames. ], batch size: 57, lr: 6.44e-03, grad_scale: 8.0 2023-04-01 23:56:53,453 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=84237.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 23:57:05,854 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_rvb from training. Duration: 26.205 2023-04-01 23:57:15,080 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=84254.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 23:57:47,146 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.862e+02 5.192e+02 6.483e+02 8.696e+02 2.103e+03, threshold=1.297e+03, percent-clipped=4.0 2023-04-01 23:57:52,183 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4267, 1.4186, 1.6962, 1.6092, 2.7267, 2.3642, 2.8391, 1.1612], device='cuda:1'), covar=tensor([0.2257, 0.4094, 0.2581, 0.1817, 0.1434, 0.1869, 0.1423, 0.3892], device='cuda:1'), in_proj_covar=tensor([0.0494, 0.0585, 0.0619, 0.0441, 0.0597, 0.0497, 0.0647, 0.0502], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-01 23:57:52,885 INFO [train.py:903] (1/4) Epoch 13, batch 2350, loss[loss=0.1972, simple_loss=0.2694, pruned_loss=0.06252, over 19487.00 frames. ], tot_loss[loss=0.2297, simple_loss=0.3044, pruned_loss=0.07748, over 3802088.96 frames. ], batch size: 49, lr: 6.44e-03, grad_scale: 8.0 2023-04-01 23:58:25,984 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=84311.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 23:58:37,175 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_rvb from training. Duration: 25.775 2023-04-01 23:58:54,395 WARNING [train.py:1073] (1/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9_rvb from training. Duration: 25.45 2023-04-01 23:58:57,889 INFO [train.py:903] (1/4) Epoch 13, batch 2400, loss[loss=0.2508, simple_loss=0.3224, pruned_loss=0.0896, over 19342.00 frames. ], tot_loss[loss=0.2291, simple_loss=0.3039, pruned_loss=0.07716, over 3804512.02 frames. ], batch size: 66, lr: 6.44e-03, grad_scale: 8.0 2023-04-01 23:59:11,480 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=84347.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 23:59:20,714 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=84354.0, num_to_drop=0, layers_to_drop=set() 2023-04-01 23:59:54,315 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.264e+02 5.138e+02 6.932e+02 8.383e+02 1.660e+03, threshold=1.386e+03, percent-clipped=4.0 2023-04-01 23:59:54,739 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.4434, 2.2486, 2.0602, 1.9865, 1.7072, 1.8127, 0.7390, 1.2267], device='cuda:1'), covar=tensor([0.0417, 0.0440, 0.0325, 0.0517, 0.0825, 0.0690, 0.0911, 0.0783], device='cuda:1'), in_proj_covar=tensor([0.0344, 0.0338, 0.0338, 0.0361, 0.0432, 0.0359, 0.0318, 0.0329], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-01 23:59:59,974 INFO [train.py:903] (1/4) Epoch 13, batch 2450, loss[loss=0.2897, simple_loss=0.3632, pruned_loss=0.1081, over 18229.00 frames. ], tot_loss[loss=0.2289, simple_loss=0.3033, pruned_loss=0.07726, over 3810240.69 frames. ], batch size: 83, lr: 6.43e-03, grad_scale: 8.0 2023-04-02 00:00:05,004 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.7160, 4.0684, 4.3436, 4.3198, 1.9617, 4.0119, 3.6463, 4.0611], device='cuda:1'), covar=tensor([0.1368, 0.1342, 0.0545, 0.0584, 0.4911, 0.0866, 0.0547, 0.0974], device='cuda:1'), in_proj_covar=tensor([0.0696, 0.0624, 0.0830, 0.0707, 0.0750, 0.0574, 0.0499, 0.0758], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1') 2023-04-02 00:00:51,591 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=84426.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:01:03,750 INFO [train.py:903] (1/4) Epoch 13, batch 2500, loss[loss=0.2273, simple_loss=0.3202, pruned_loss=0.06721, over 19654.00 frames. ], tot_loss[loss=0.2287, simple_loss=0.3032, pruned_loss=0.07714, over 3812689.42 frames. ], batch size: 59, lr: 6.43e-03, grad_scale: 8.0 2023-04-02 00:01:33,210 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=84459.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:01:38,066 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=84462.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:02:00,518 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.656e+02 5.326e+02 7.100e+02 9.098e+02 1.657e+03, threshold=1.420e+03, percent-clipped=3.0 2023-04-02 00:02:06,433 INFO [train.py:903] (1/4) Epoch 13, batch 2550, loss[loss=0.2327, simple_loss=0.3157, pruned_loss=0.07484, over 19630.00 frames. ], tot_loss[loss=0.2305, simple_loss=0.3049, pruned_loss=0.07804, over 3808461.88 frames. ], batch size: 57, lr: 6.43e-03, grad_scale: 8.0 2023-04-02 00:02:16,073 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=84493.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:02:30,050 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2538, 1.1574, 1.1921, 1.4004, 1.1291, 1.3036, 1.3685, 1.2553], device='cuda:1'), covar=tensor([0.0852, 0.1013, 0.1093, 0.0652, 0.0805, 0.0839, 0.0831, 0.0783], device='cuda:1'), in_proj_covar=tensor([0.0207, 0.0222, 0.0222, 0.0243, 0.0228, 0.0208, 0.0191, 0.0200], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1') 2023-04-02 00:02:48,814 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=84518.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:03:04,536 WARNING [train.py:1073] (1/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 2023-04-02 00:03:10,396 INFO [train.py:903] (1/4) Epoch 13, batch 2600, loss[loss=0.222, simple_loss=0.2909, pruned_loss=0.07661, over 19410.00 frames. ], tot_loss[loss=0.2288, simple_loss=0.3035, pruned_loss=0.07707, over 3816940.01 frames. ], batch size: 47, lr: 6.43e-03, grad_scale: 8.0 2023-04-02 00:03:21,124 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=84543.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:03:42,742 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=84561.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:03:59,459 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=84574.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:04:09,373 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.575e+02 5.097e+02 6.359e+02 8.045e+02 2.004e+03, threshold=1.272e+03, percent-clipped=4.0 2023-04-02 00:04:13,180 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.3123, 1.8814, 1.9842, 1.9761, 2.9708, 1.6937, 2.5544, 3.2066], device='cuda:1'), covar=tensor([0.0563, 0.2126, 0.2181, 0.1601, 0.0721, 0.1996, 0.1787, 0.0485], device='cuda:1'), in_proj_covar=tensor([0.0363, 0.0344, 0.0355, 0.0324, 0.0347, 0.0332, 0.0342, 0.0366], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 00:04:15,154 INFO [train.py:903] (1/4) Epoch 13, batch 2650, loss[loss=0.1915, simple_loss=0.2681, pruned_loss=0.05741, over 18672.00 frames. ], tot_loss[loss=0.2281, simple_loss=0.303, pruned_loss=0.07662, over 3808146.32 frames. ], batch size: 41, lr: 6.43e-03, grad_scale: 8.0 2023-04-02 00:04:30,342 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=84598.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:04:34,944 WARNING [train.py:1073] (1/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9_rvb from training. Duration: 27.25 2023-04-02 00:05:17,498 INFO [train.py:903] (1/4) Epoch 13, batch 2700, loss[loss=0.179, simple_loss=0.26, pruned_loss=0.04898, over 19840.00 frames. ], tot_loss[loss=0.2277, simple_loss=0.3025, pruned_loss=0.07643, over 3803074.13 frames. ], batch size: 52, lr: 6.43e-03, grad_scale: 8.0 2023-04-02 00:05:36,196 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=84651.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:06:08,572 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=84676.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:06:14,002 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.791e+02 5.394e+02 6.395e+02 8.456e+02 1.799e+03, threshold=1.279e+03, percent-clipped=4.0 2023-04-02 00:06:16,785 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=84682.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:06:21,012 INFO [train.py:903] (1/4) Epoch 13, batch 2750, loss[loss=0.2404, simple_loss=0.3137, pruned_loss=0.08355, over 19576.00 frames. ], tot_loss[loss=0.2284, simple_loss=0.3032, pruned_loss=0.07681, over 3791694.45 frames. ], batch size: 61, lr: 6.42e-03, grad_scale: 8.0 2023-04-02 00:06:37,480 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=84698.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:06:49,124 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=84707.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:06:56,197 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=84713.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:07:01,993 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=84718.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:07:24,552 INFO [train.py:903] (1/4) Epoch 13, batch 2800, loss[loss=0.2646, simple_loss=0.3253, pruned_loss=0.102, over 13867.00 frames. ], tot_loss[loss=0.2286, simple_loss=0.3031, pruned_loss=0.077, over 3811318.81 frames. ], batch size: 135, lr: 6.42e-03, grad_scale: 8.0 2023-04-02 00:07:34,084 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=84743.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:08:22,668 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.267e+02 5.423e+02 6.884e+02 8.957e+02 1.568e+03, threshold=1.377e+03, percent-clipped=4.0 2023-04-02 00:08:29,878 INFO [train.py:903] (1/4) Epoch 13, batch 2850, loss[loss=0.2351, simple_loss=0.3202, pruned_loss=0.07505, over 19651.00 frames. ], tot_loss[loss=0.2294, simple_loss=0.3042, pruned_loss=0.07726, over 3817291.05 frames. ], batch size: 58, lr: 6.42e-03, grad_scale: 8.0 2023-04-02 00:09:04,281 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=84813.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:09:11,184 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=84818.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:09:26,619 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=84830.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:09:33,381 INFO [train.py:903] (1/4) Epoch 13, batch 2900, loss[loss=0.2336, simple_loss=0.3109, pruned_loss=0.07822, over 19613.00 frames. ], tot_loss[loss=0.229, simple_loss=0.3043, pruned_loss=0.07689, over 3816690.50 frames. ], batch size: 57, lr: 6.42e-03, grad_scale: 8.0 2023-04-02 00:09:33,412 WARNING [train.py:1073] (1/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 2023-04-02 00:09:38,690 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.5136, 4.0349, 4.2326, 4.2206, 1.6744, 3.9680, 3.5128, 3.8923], device='cuda:1'), covar=tensor([0.1434, 0.0875, 0.0609, 0.0605, 0.5056, 0.0698, 0.0601, 0.1110], device='cuda:1'), in_proj_covar=tensor([0.0697, 0.0620, 0.0832, 0.0697, 0.0749, 0.0570, 0.0496, 0.0757], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1') 2023-04-02 00:09:58,743 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=84855.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:10:21,107 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.77 vs. limit=2.0 2023-04-02 00:10:31,984 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.833e+02 5.191e+02 6.710e+02 8.572e+02 2.238e+03, threshold=1.342e+03, percent-clipped=4.0 2023-04-02 00:10:38,026 INFO [train.py:903] (1/4) Epoch 13, batch 2950, loss[loss=0.1871, simple_loss=0.2593, pruned_loss=0.05744, over 19709.00 frames. ], tot_loss[loss=0.2282, simple_loss=0.3036, pruned_loss=0.07637, over 3826041.16 frames. ], batch size: 45, lr: 6.42e-03, grad_scale: 8.0 2023-04-02 00:10:40,473 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=84887.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:11:00,938 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.3400, 2.1604, 1.9153, 1.6680, 1.4821, 1.7334, 0.3735, 1.2166], device='cuda:1'), covar=tensor([0.0445, 0.0454, 0.0357, 0.0691, 0.0981, 0.0757, 0.1047, 0.0800], device='cuda:1'), in_proj_covar=tensor([0.0341, 0.0334, 0.0335, 0.0357, 0.0430, 0.0358, 0.0316, 0.0325], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 00:11:39,702 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=84932.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:11:45,163 INFO [train.py:903] (1/4) Epoch 13, batch 3000, loss[loss=0.2146, simple_loss=0.2952, pruned_loss=0.06697, over 19777.00 frames. ], tot_loss[loss=0.2263, simple_loss=0.3021, pruned_loss=0.07527, over 3842542.14 frames. ], batch size: 54, lr: 6.41e-03, grad_scale: 8.0 2023-04-02 00:11:45,163 INFO [train.py:928] (1/4) Computing validation loss 2023-04-02 00:12:00,827 INFO [train.py:937] (1/4) Epoch 13, validation: loss=0.1754, simple_loss=0.276, pruned_loss=0.03742, over 944034.00 frames. 2023-04-02 00:12:00,828 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18821MB 2023-04-02 00:12:05,791 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_rvb from training. Duration: 29.735 2023-04-02 00:12:29,052 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 2023-04-02 00:12:29,873 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=84957.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:12:43,910 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=84969.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:12:44,056 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=84969.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:12:59,474 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.728e+02 4.901e+02 6.331e+02 8.447e+02 1.208e+03, threshold=1.266e+03, percent-clipped=0.0 2023-04-02 00:13:05,445 INFO [train.py:903] (1/4) Epoch 13, batch 3050, loss[loss=0.2262, simple_loss=0.3051, pruned_loss=0.07365, over 18077.00 frames. ], tot_loss[loss=0.226, simple_loss=0.3018, pruned_loss=0.07512, over 3841541.89 frames. ], batch size: 83, lr: 6.41e-03, grad_scale: 8.0 2023-04-02 00:13:17,137 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=84994.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:13:18,168 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=84995.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:13:26,452 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=85002.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:14:09,067 INFO [train.py:903] (1/4) Epoch 13, batch 3100, loss[loss=0.2884, simple_loss=0.3445, pruned_loss=0.1161, over 19156.00 frames. ], tot_loss[loss=0.2278, simple_loss=0.3029, pruned_loss=0.07639, over 3840861.68 frames. ], batch size: 69, lr: 6.41e-03, grad_scale: 8.0 2023-04-02 00:14:50,637 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=85069.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:15:05,349 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.206e+02 5.118e+02 6.727e+02 8.323e+02 1.616e+03, threshold=1.345e+03, percent-clipped=6.0 2023-04-02 00:15:11,227 INFO [train.py:903] (1/4) Epoch 13, batch 3150, loss[loss=0.2308, simple_loss=0.3115, pruned_loss=0.07502, over 19642.00 frames. ], tot_loss[loss=0.2269, simple_loss=0.302, pruned_loss=0.07592, over 3819445.97 frames. ], batch size: 58, lr: 6.41e-03, grad_scale: 8.0 2023-04-02 00:15:20,616 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=85094.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:15:38,368 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 2023-04-02 00:15:41,864 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=85110.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:16:00,119 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=85125.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:16:09,357 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=85133.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:16:12,577 INFO [train.py:903] (1/4) Epoch 13, batch 3200, loss[loss=0.2178, simple_loss=0.3059, pruned_loss=0.06483, over 19537.00 frames. ], tot_loss[loss=0.2276, simple_loss=0.3024, pruned_loss=0.07644, over 3808714.85 frames. ], batch size: 56, lr: 6.41e-03, grad_scale: 8.0 2023-04-02 00:16:46,249 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=85162.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:17:10,847 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.817e+02 5.513e+02 6.647e+02 8.266e+02 1.326e+03, threshold=1.329e+03, percent-clipped=0.0 2023-04-02 00:17:16,618 INFO [train.py:903] (1/4) Epoch 13, batch 3250, loss[loss=0.2329, simple_loss=0.3133, pruned_loss=0.07625, over 18419.00 frames. ], tot_loss[loss=0.2275, simple_loss=0.3022, pruned_loss=0.07639, over 3804554.71 frames. ], batch size: 84, lr: 6.40e-03, grad_scale: 8.0 2023-04-02 00:18:20,663 INFO [train.py:903] (1/4) Epoch 13, batch 3300, loss[loss=0.2169, simple_loss=0.2904, pruned_loss=0.07172, over 19687.00 frames. ], tot_loss[loss=0.2265, simple_loss=0.3015, pruned_loss=0.07575, over 3806166.84 frames. ], batch size: 53, lr: 6.40e-03, grad_scale: 8.0 2023-04-02 00:18:21,894 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 2023-04-02 00:18:47,654 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=85258.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:19:12,446 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=85277.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:19:17,833 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.152e+02 5.269e+02 6.447e+02 8.165e+02 1.494e+03, threshold=1.289e+03, percent-clipped=3.0 2023-04-02 00:19:19,464 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=85283.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:19:22,455 INFO [train.py:903] (1/4) Epoch 13, batch 3350, loss[loss=0.2231, simple_loss=0.3074, pruned_loss=0.06938, over 19661.00 frames. ], tot_loss[loss=0.2264, simple_loss=0.3014, pruned_loss=0.07572, over 3802565.74 frames. ], batch size: 58, lr: 6.40e-03, grad_scale: 8.0 2023-04-02 00:19:57,318 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=85313.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:20:24,670 INFO [train.py:903] (1/4) Epoch 13, batch 3400, loss[loss=0.2147, simple_loss=0.2874, pruned_loss=0.07103, over 19586.00 frames. ], tot_loss[loss=0.226, simple_loss=0.3011, pruned_loss=0.07549, over 3816165.85 frames. ], batch size: 52, lr: 6.40e-03, grad_scale: 8.0 2023-04-02 00:20:42,512 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.1215, 3.3276, 2.0278, 2.1433, 2.9510, 1.8731, 1.6321, 2.0610], device='cuda:1'), covar=tensor([0.1055, 0.0495, 0.0876, 0.0732, 0.0456, 0.1004, 0.0776, 0.0608], device='cuda:1'), in_proj_covar=tensor([0.0290, 0.0301, 0.0323, 0.0247, 0.0236, 0.0317, 0.0283, 0.0260], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 00:20:42,590 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.2163, 2.2461, 2.3238, 3.1099, 2.2162, 3.0771, 2.5877, 2.1060], device='cuda:1'), covar=tensor([0.3760, 0.3370, 0.1609, 0.2049, 0.3929, 0.1544, 0.3870, 0.2892], device='cuda:1'), in_proj_covar=tensor([0.0806, 0.0835, 0.0658, 0.0890, 0.0793, 0.0718, 0.0794, 0.0718], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 00:21:03,104 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=85366.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:21:22,545 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.171e+02 5.371e+02 6.826e+02 8.500e+02 2.504e+03, threshold=1.365e+03, percent-clipped=9.0 2023-04-02 00:21:27,160 INFO [train.py:903] (1/4) Epoch 13, batch 3450, loss[loss=0.2439, simple_loss=0.3164, pruned_loss=0.08574, over 19777.00 frames. ], tot_loss[loss=0.2273, simple_loss=0.3025, pruned_loss=0.07607, over 3828909.83 frames. ], batch size: 56, lr: 6.40e-03, grad_scale: 8.0 2023-04-02 00:21:30,637 WARNING [train.py:1073] (1/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9_rvb from training. Duration: 25.2444375 2023-04-02 00:21:34,245 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=85391.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:21:47,609 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5765, 1.4442, 1.4196, 1.8296, 1.6046, 1.8943, 1.8150, 1.7046], device='cuda:1'), covar=tensor([0.0765, 0.0936, 0.0994, 0.0832, 0.0800, 0.0662, 0.0877, 0.0667], device='cuda:1'), in_proj_covar=tensor([0.0209, 0.0225, 0.0225, 0.0245, 0.0231, 0.0212, 0.0193, 0.0203], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-02 00:22:19,367 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=85428.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:22:28,797 INFO [train.py:903] (1/4) Epoch 13, batch 3500, loss[loss=0.1955, simple_loss=0.2777, pruned_loss=0.05664, over 19619.00 frames. ], tot_loss[loss=0.2289, simple_loss=0.3041, pruned_loss=0.07688, over 3827311.31 frames. ], batch size: 50, lr: 6.40e-03, grad_scale: 8.0 2023-04-02 00:22:42,093 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=85446.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:23:10,409 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=85469.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:23:19,014 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.31 vs. limit=2.0 2023-04-02 00:23:20,667 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=85477.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:23:27,197 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.349e+02 5.339e+02 6.675e+02 8.042e+02 1.644e+03, threshold=1.335e+03, percent-clipped=3.0 2023-04-02 00:23:31,545 INFO [train.py:903] (1/4) Epoch 13, batch 3550, loss[loss=0.2509, simple_loss=0.3099, pruned_loss=0.09591, over 19785.00 frames. ], tot_loss[loss=0.2273, simple_loss=0.3027, pruned_loss=0.07599, over 3842950.03 frames. ], batch size: 48, lr: 6.39e-03, grad_scale: 8.0 2023-04-02 00:23:50,557 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=85501.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:24:31,228 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=85533.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:24:34,229 INFO [train.py:903] (1/4) Epoch 13, batch 3600, loss[loss=0.2492, simple_loss=0.3213, pruned_loss=0.08861, over 19683.00 frames. ], tot_loss[loss=0.228, simple_loss=0.3033, pruned_loss=0.07638, over 3841485.07 frames. ], batch size: 59, lr: 6.39e-03, grad_scale: 8.0 2023-04-02 00:24:50,350 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=85548.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:25:03,563 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=85558.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:25:33,196 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.573e+02 5.969e+02 7.367e+02 9.197e+02 1.857e+03, threshold=1.473e+03, percent-clipped=2.0 2023-04-02 00:25:36,786 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=85584.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:25:38,622 INFO [train.py:903] (1/4) Epoch 13, batch 3650, loss[loss=0.2201, simple_loss=0.3074, pruned_loss=0.06645, over 19610.00 frames. ], tot_loss[loss=0.2275, simple_loss=0.3026, pruned_loss=0.07623, over 3833057.57 frames. ], batch size: 57, lr: 6.39e-03, grad_scale: 8.0 2023-04-02 00:25:46,148 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=85592.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:26:41,870 INFO [train.py:903] (1/4) Epoch 13, batch 3700, loss[loss=0.2235, simple_loss=0.2898, pruned_loss=0.07856, over 19271.00 frames. ], tot_loss[loss=0.227, simple_loss=0.3022, pruned_loss=0.07588, over 3849449.76 frames. ], batch size: 44, lr: 6.39e-03, grad_scale: 8.0 2023-04-02 00:26:51,237 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=85643.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:27:41,800 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.698e+02 5.337e+02 6.385e+02 1.025e+03 1.989e+03, threshold=1.277e+03, percent-clipped=4.0 2023-04-02 00:27:44,633 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=85684.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:27:46,548 INFO [train.py:903] (1/4) Epoch 13, batch 3750, loss[loss=0.2452, simple_loss=0.3299, pruned_loss=0.08027, over 19659.00 frames. ], tot_loss[loss=0.2268, simple_loss=0.3023, pruned_loss=0.07562, over 3849511.33 frames. ], batch size: 60, lr: 6.39e-03, grad_scale: 8.0 2023-04-02 00:28:14,946 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=85709.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:28:48,157 INFO [train.py:903] (1/4) Epoch 13, batch 3800, loss[loss=0.2142, simple_loss=0.2943, pruned_loss=0.06706, over 19666.00 frames. ], tot_loss[loss=0.2272, simple_loss=0.3025, pruned_loss=0.07589, over 3846811.78 frames. ], batch size: 58, lr: 6.38e-03, grad_scale: 8.0 2023-04-02 00:29:14,351 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.6560, 4.2031, 2.7163, 3.7194, 1.1485, 4.0482, 3.9787, 4.1126], device='cuda:1'), covar=tensor([0.0609, 0.0875, 0.1950, 0.0760, 0.3739, 0.0706, 0.0849, 0.1026], device='cuda:1'), in_proj_covar=tensor([0.0446, 0.0375, 0.0449, 0.0323, 0.0388, 0.0383, 0.0373, 0.0406], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 00:29:20,860 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9_rvb from training. Duration: 29.1166875 2023-04-02 00:29:44,889 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.310e+02 5.587e+02 7.205e+02 9.194e+02 2.888e+03, threshold=1.441e+03, percent-clipped=8.0 2023-04-02 00:29:50,625 INFO [train.py:903] (1/4) Epoch 13, batch 3850, loss[loss=0.2275, simple_loss=0.3094, pruned_loss=0.07281, over 19698.00 frames. ], tot_loss[loss=0.2262, simple_loss=0.3018, pruned_loss=0.07532, over 3846257.72 frames. ], batch size: 59, lr: 6.38e-03, grad_scale: 8.0 2023-04-02 00:29:56,228 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=85790.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:29:56,576 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9040, 1.9853, 2.1725, 2.6052, 1.8771, 2.4744, 2.3144, 2.0375], device='cuda:1'), covar=tensor([0.3518, 0.3168, 0.1442, 0.1781, 0.3330, 0.1506, 0.3534, 0.2616], device='cuda:1'), in_proj_covar=tensor([0.0812, 0.0846, 0.0661, 0.0899, 0.0797, 0.0723, 0.0800, 0.0722], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 00:30:53,507 INFO [train.py:903] (1/4) Epoch 13, batch 3900, loss[loss=0.2431, simple_loss=0.3125, pruned_loss=0.08682, over 19740.00 frames. ], tot_loss[loss=0.2264, simple_loss=0.3021, pruned_loss=0.07537, over 3838383.86 frames. ], batch size: 63, lr: 6.38e-03, grad_scale: 8.0 2023-04-02 00:31:00,072 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=85840.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:31:06,824 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=85845.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:31:10,504 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=85848.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:31:30,213 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=85865.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:31:40,468 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=85873.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:31:49,440 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.3319, 3.0161, 2.2289, 2.7220, 0.9941, 2.9748, 2.8527, 2.9408], device='cuda:1'), covar=tensor([0.1065, 0.1441, 0.2008, 0.1034, 0.3538, 0.0949, 0.1006, 0.1300], device='cuda:1'), in_proj_covar=tensor([0.0446, 0.0372, 0.0449, 0.0323, 0.0388, 0.0381, 0.0373, 0.0407], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 00:31:51,443 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.894e+02 5.450e+02 6.738e+02 8.252e+02 2.044e+03, threshold=1.348e+03, percent-clipped=4.0 2023-04-02 00:31:57,034 INFO [train.py:903] (1/4) Epoch 13, batch 3950, loss[loss=0.2973, simple_loss=0.3531, pruned_loss=0.1207, over 17487.00 frames. ], tot_loss[loss=0.2287, simple_loss=0.304, pruned_loss=0.07672, over 3829774.18 frames. ], batch size: 101, lr: 6.38e-03, grad_scale: 8.0 2023-04-02 00:32:00,540 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 2023-04-02 00:32:04,165 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=85892.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:32:05,566 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=85893.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:32:20,802 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=85905.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:32:40,635 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.58 vs. limit=2.0 2023-04-02 00:32:59,798 INFO [train.py:903] (1/4) Epoch 13, batch 4000, loss[loss=0.1785, simple_loss=0.2536, pruned_loss=0.0517, over 19731.00 frames. ], tot_loss[loss=0.2284, simple_loss=0.3033, pruned_loss=0.07676, over 3814577.59 frames. ], batch size: 45, lr: 6.38e-03, grad_scale: 8.0 2023-04-02 00:33:14,755 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=85948.0, num_to_drop=1, layers_to_drop={0} 2023-04-02 00:33:30,816 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=85960.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:33:46,142 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1_rvb from training. Duration: 0.7545625 2023-04-02 00:33:48,603 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=85975.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:33:56,481 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.510e+02 5.280e+02 6.258e+02 8.022e+02 1.377e+03, threshold=1.252e+03, percent-clipped=2.0 2023-04-02 00:34:02,168 INFO [train.py:903] (1/4) Epoch 13, batch 4050, loss[loss=0.256, simple_loss=0.336, pruned_loss=0.08801, over 19582.00 frames. ], tot_loss[loss=0.2287, simple_loss=0.3037, pruned_loss=0.07682, over 3824314.16 frames. ], batch size: 61, lr: 6.37e-03, grad_scale: 8.0 2023-04-02 00:34:03,494 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=85987.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:34:06,944 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6883, 1.2354, 1.7057, 1.2676, 2.6162, 3.4814, 3.2419, 3.6783], device='cuda:1'), covar=tensor([0.1249, 0.3579, 0.3178, 0.2192, 0.0533, 0.0193, 0.0212, 0.0212], device='cuda:1'), in_proj_covar=tensor([0.0259, 0.0300, 0.0331, 0.0251, 0.0219, 0.0163, 0.0206, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 00:34:30,787 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86007.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:34:46,116 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86020.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:34:46,324 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.8704, 1.9455, 2.1508, 2.4932, 1.6835, 2.3771, 2.3061, 2.1084], device='cuda:1'), covar=tensor([0.3624, 0.3388, 0.1635, 0.2102, 0.3693, 0.1787, 0.3969, 0.2788], device='cuda:1'), in_proj_covar=tensor([0.0813, 0.0845, 0.0660, 0.0899, 0.0798, 0.0725, 0.0800, 0.0723], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 00:35:05,743 INFO [train.py:903] (1/4) Epoch 13, batch 4100, loss[loss=0.1957, simple_loss=0.2717, pruned_loss=0.0599, over 19477.00 frames. ], tot_loss[loss=0.2286, simple_loss=0.3032, pruned_loss=0.07702, over 3816636.92 frames. ], batch size: 49, lr: 6.37e-03, grad_scale: 8.0 2023-04-02 00:35:07,466 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5811, 1.2933, 1.2998, 2.0322, 1.5798, 1.9345, 1.9731, 1.6102], device='cuda:1'), covar=tensor([0.0794, 0.1008, 0.1108, 0.0761, 0.0857, 0.0674, 0.0810, 0.0712], device='cuda:1'), in_proj_covar=tensor([0.0209, 0.0224, 0.0225, 0.0243, 0.0232, 0.0210, 0.0191, 0.0201], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1') 2023-04-02 00:35:41,500 WARNING [train.py:1073] (1/4) Exclude cut with ID 497-129325-0061-9566_sp1.1_rvb from training. Duration: 0.97725 2023-04-02 00:35:48,369 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86070.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:36:04,017 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.212e+02 4.943e+02 6.094e+02 7.890e+02 1.986e+03, threshold=1.219e+03, percent-clipped=5.0 2023-04-02 00:36:08,782 INFO [train.py:903] (1/4) Epoch 13, batch 4150, loss[loss=0.1932, simple_loss=0.2751, pruned_loss=0.05566, over 19417.00 frames. ], tot_loss[loss=0.226, simple_loss=0.3012, pruned_loss=0.07533, over 3831855.10 frames. ], batch size: 48, lr: 6.37e-03, grad_scale: 8.0 2023-04-02 00:36:11,047 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86087.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:36:24,550 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 2023-04-02 00:36:28,552 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86102.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:37:11,334 INFO [train.py:903] (1/4) Epoch 13, batch 4200, loss[loss=0.2009, simple_loss=0.281, pruned_loss=0.06047, over 19750.00 frames. ], tot_loss[loss=0.2257, simple_loss=0.3009, pruned_loss=0.07526, over 3830281.39 frames. ], batch size: 51, lr: 6.37e-03, grad_scale: 8.0 2023-04-02 00:37:14,932 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 2023-04-02 00:37:42,922 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86161.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:38:08,166 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.547e+02 5.506e+02 6.651e+02 9.015e+02 1.898e+03, threshold=1.330e+03, percent-clipped=8.0 2023-04-02 00:38:12,743 INFO [train.py:903] (1/4) Epoch 13, batch 4250, loss[loss=0.3048, simple_loss=0.3569, pruned_loss=0.1263, over 18061.00 frames. ], tot_loss[loss=0.2268, simple_loss=0.3016, pruned_loss=0.07596, over 3820433.85 frames. ], batch size: 83, lr: 6.37e-03, grad_scale: 8.0 2023-04-02 00:38:13,162 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86186.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:38:31,027 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9_rvb from training. Duration: 29.816625 2023-04-02 00:38:43,582 WARNING [train.py:1073] (1/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 2023-04-02 00:38:52,364 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86216.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:39:15,648 INFO [train.py:903] (1/4) Epoch 13, batch 4300, loss[loss=0.3294, simple_loss=0.376, pruned_loss=0.1414, over 13193.00 frames. ], tot_loss[loss=0.2278, simple_loss=0.3024, pruned_loss=0.07658, over 3811624.66 frames. ], batch size: 135, lr: 6.37e-03, grad_scale: 8.0 2023-04-02 00:39:17,000 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=86237.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:39:17,303 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6049, 2.5263, 1.8724, 1.6768, 2.3215, 1.5355, 1.5711, 2.0411], device='cuda:1'), covar=tensor([0.1065, 0.0602, 0.0954, 0.0760, 0.0432, 0.1065, 0.0603, 0.0432], device='cuda:1'), in_proj_covar=tensor([0.0293, 0.0304, 0.0327, 0.0251, 0.0237, 0.0320, 0.0287, 0.0263], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 00:39:22,965 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86241.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:39:50,480 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86263.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:39:51,570 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86264.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:40:13,250 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.647e+02 5.244e+02 6.314e+02 8.271e+02 2.210e+03, threshold=1.263e+03, percent-clipped=7.0 2023-04-02 00:40:15,212 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 2023-04-02 00:40:19,672 INFO [train.py:903] (1/4) Epoch 13, batch 4350, loss[loss=0.2612, simple_loss=0.3442, pruned_loss=0.08912, over 19609.00 frames. ], tot_loss[loss=0.2267, simple_loss=0.3014, pruned_loss=0.07605, over 3828011.29 frames. ], batch size: 61, lr: 6.36e-03, grad_scale: 8.0 2023-04-02 00:40:22,476 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86288.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:40:27,958 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=86292.0, num_to_drop=1, layers_to_drop={0} 2023-04-02 00:41:00,080 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=86319.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:41:22,570 INFO [train.py:903] (1/4) Epoch 13, batch 4400, loss[loss=0.3222, simple_loss=0.3702, pruned_loss=0.1372, over 17321.00 frames. ], tot_loss[loss=0.227, simple_loss=0.3017, pruned_loss=0.0762, over 3805673.88 frames. ], batch size: 101, lr: 6.36e-03, grad_scale: 8.0 2023-04-02 00:41:42,060 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86352.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:41:49,347 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86358.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:41:50,086 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_rvb from training. Duration: 25.285 2023-04-02 00:41:58,576 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=86364.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:42:00,827 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 2023-04-02 00:42:20,364 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.626e+02 5.303e+02 6.614e+02 7.903e+02 1.976e+03, threshold=1.323e+03, percent-clipped=7.0 2023-04-02 00:42:21,956 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86383.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:42:25,157 INFO [train.py:903] (1/4) Epoch 13, batch 4450, loss[loss=0.2312, simple_loss=0.2978, pruned_loss=0.08227, over 19604.00 frames. ], tot_loss[loss=0.2276, simple_loss=0.3022, pruned_loss=0.07651, over 3817983.18 frames. ], batch size: 50, lr: 6.36e-03, grad_scale: 8.0 2023-04-02 00:42:52,213 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86407.0, num_to_drop=1, layers_to_drop={1} 2023-04-02 00:43:00,999 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=86414.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:43:21,647 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=86431.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:43:25,451 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86434.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:43:27,361 INFO [train.py:903] (1/4) Epoch 13, batch 4500, loss[loss=0.1857, simple_loss=0.2609, pruned_loss=0.05521, over 19334.00 frames. ], tot_loss[loss=0.2268, simple_loss=0.3014, pruned_loss=0.0761, over 3813009.71 frames. ], batch size: 47, lr: 6.36e-03, grad_scale: 8.0 2023-04-02 00:43:52,077 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.7889, 1.7707, 1.6547, 1.4052, 1.3705, 1.4154, 0.2246, 0.7078], device='cuda:1'), covar=tensor([0.0421, 0.0437, 0.0261, 0.0412, 0.0847, 0.0459, 0.0811, 0.0715], device='cuda:1'), in_proj_covar=tensor([0.0343, 0.0337, 0.0334, 0.0360, 0.0433, 0.0360, 0.0316, 0.0327], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 00:43:55,635 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86458.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:44:01,557 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86463.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:44:21,244 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86479.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:44:24,463 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.347e+02 5.430e+02 6.459e+02 7.870e+02 1.871e+03, threshold=1.292e+03, percent-clipped=3.0 2023-04-02 00:44:29,968 INFO [train.py:903] (1/4) Epoch 13, batch 4550, loss[loss=0.2718, simple_loss=0.3413, pruned_loss=0.1011, over 19467.00 frames. ], tot_loss[loss=0.2273, simple_loss=0.3018, pruned_loss=0.07646, over 3798672.18 frames. ], batch size: 64, lr: 6.36e-03, grad_scale: 8.0 2023-04-02 00:44:39,230 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9_rvb from training. Duration: 28.72225 2023-04-02 00:45:02,116 WARNING [train.py:1073] (1/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 2023-04-02 00:45:07,113 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.5131, 4.0831, 2.5100, 3.6109, 0.9933, 3.8786, 3.8947, 4.0373], device='cuda:1'), covar=tensor([0.0701, 0.1163, 0.2223, 0.0819, 0.4215, 0.0855, 0.0777, 0.1234], device='cuda:1'), in_proj_covar=tensor([0.0448, 0.0374, 0.0450, 0.0323, 0.0388, 0.0382, 0.0373, 0.0408], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 00:45:16,170 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.55 vs. limit=2.0 2023-04-02 00:45:25,007 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86529.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:45:33,588 INFO [train.py:903] (1/4) Epoch 13, batch 4600, loss[loss=0.2498, simple_loss=0.3199, pruned_loss=0.0898, over 19715.00 frames. ], tot_loss[loss=0.2272, simple_loss=0.3017, pruned_loss=0.07636, over 3790326.25 frames. ], batch size: 63, lr: 6.35e-03, grad_scale: 8.0 2023-04-02 00:45:45,517 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86546.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:45:50,082 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.8542, 1.5530, 1.2856, 1.6531, 1.6310, 1.4344, 1.3216, 1.6446], device='cuda:1'), covar=tensor([0.0995, 0.1460, 0.1757, 0.1080, 0.1297, 0.0941, 0.1593, 0.0913], device='cuda:1'), in_proj_covar=tensor([0.0255, 0.0346, 0.0292, 0.0240, 0.0292, 0.0239, 0.0281, 0.0239], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 00:46:31,574 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.083e+02 5.452e+02 6.634e+02 8.020e+02 1.693e+03, threshold=1.327e+03, percent-clipped=1.0 2023-04-02 00:46:35,925 INFO [train.py:903] (1/4) Epoch 13, batch 4650, loss[loss=0.2577, simple_loss=0.3265, pruned_loss=0.09441, over 19756.00 frames. ], tot_loss[loss=0.226, simple_loss=0.301, pruned_loss=0.07545, over 3815193.18 frames. ], batch size: 54, lr: 6.35e-03, grad_scale: 8.0 2023-04-02 00:46:51,396 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86598.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:46:52,260 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_rvb from training. Duration: 0.92 2023-04-02 00:46:57,295 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.13 vs. limit=2.0 2023-04-02 00:47:03,884 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_rvb from training. Duration: 0.83 2023-04-02 00:47:04,001 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=86608.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:47:04,266 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86608.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:47:35,054 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86633.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:47:35,329 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 2023-04-02 00:47:38,166 INFO [train.py:903] (1/4) Epoch 13, batch 4700, loss[loss=0.1787, simple_loss=0.2684, pruned_loss=0.0445, over 19787.00 frames. ], tot_loss[loss=0.2249, simple_loss=0.3002, pruned_loss=0.07477, over 3821127.32 frames. ], batch size: 56, lr: 6.35e-03, grad_scale: 8.0 2023-04-02 00:47:53,761 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 2023-04-02 00:48:04,337 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9_rvb from training. Duration: 0.92225 2023-04-02 00:48:05,982 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.7682, 2.6555, 2.1241, 2.0385, 1.8880, 2.3412, 1.0270, 2.0336], device='cuda:1'), covar=tensor([0.0416, 0.0458, 0.0505, 0.0731, 0.0796, 0.0686, 0.0948, 0.0709], device='cuda:1'), in_proj_covar=tensor([0.0341, 0.0335, 0.0332, 0.0357, 0.0432, 0.0357, 0.0314, 0.0325], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 00:48:13,311 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86663.0, num_to_drop=1, layers_to_drop={1} 2023-04-02 00:48:35,954 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.257e+02 5.083e+02 5.953e+02 7.054e+02 1.649e+03, threshold=1.191e+03, percent-clipped=1.0 2023-04-02 00:48:42,062 INFO [train.py:903] (1/4) Epoch 13, batch 4750, loss[loss=0.2204, simple_loss=0.3053, pruned_loss=0.06777, over 18836.00 frames. ], tot_loss[loss=0.2248, simple_loss=0.3004, pruned_loss=0.07462, over 3831203.42 frames. ], batch size: 74, lr: 6.35e-03, grad_scale: 8.0 2023-04-02 00:48:45,761 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86688.0, num_to_drop=1, layers_to_drop={1} 2023-04-02 00:48:48,063 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86690.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:48:58,169 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.9684, 1.6155, 1.7959, 1.6336, 4.4038, 0.9421, 2.4613, 4.7064], device='cuda:1'), covar=tensor([0.0409, 0.2577, 0.2661, 0.1929, 0.0795, 0.2738, 0.1384, 0.0206], device='cuda:1'), in_proj_covar=tensor([0.0365, 0.0345, 0.0356, 0.0325, 0.0352, 0.0332, 0.0346, 0.0365], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 00:49:07,679 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 2023-04-02 00:49:10,081 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 2023-04-02 00:49:15,252 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.10 vs. limit=2.0 2023-04-02 00:49:18,525 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86715.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:49:27,255 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86723.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:49:31,061 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.85 vs. limit=2.0 2023-04-02 00:49:43,088 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86735.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:49:43,756 INFO [train.py:903] (1/4) Epoch 13, batch 4800, loss[loss=0.2133, simple_loss=0.2802, pruned_loss=0.07322, over 17408.00 frames. ], tot_loss[loss=0.2264, simple_loss=0.3015, pruned_loss=0.07564, over 3824164.54 frames. ], batch size: 38, lr: 6.35e-03, grad_scale: 8.0 2023-04-02 00:50:12,865 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9060, 1.6099, 1.4876, 1.8603, 1.7779, 1.6326, 1.4849, 1.7487], device='cuda:1'), covar=tensor([0.0910, 0.1417, 0.1381, 0.0960, 0.1065, 0.0510, 0.1224, 0.0705], device='cuda:1'), in_proj_covar=tensor([0.0256, 0.0348, 0.0293, 0.0241, 0.0294, 0.0240, 0.0282, 0.0239], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 00:50:12,873 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86760.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:50:41,010 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.218e+02 5.270e+02 6.552e+02 8.488e+02 1.715e+03, threshold=1.310e+03, percent-clipped=7.0 2023-04-02 00:50:45,010 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86785.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:50:45,815 INFO [train.py:903] (1/4) Epoch 13, batch 4850, loss[loss=0.27, simple_loss=0.3301, pruned_loss=0.1049, over 13460.00 frames. ], tot_loss[loss=0.2264, simple_loss=0.3015, pruned_loss=0.07567, over 3825860.80 frames. ], batch size: 136, lr: 6.35e-03, grad_scale: 8.0 2023-04-02 00:51:06,440 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=86802.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:51:06,710 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86802.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:51:08,593 WARNING [train.py:1073] (1/4) Exclude cut with ID 774-127930-0014-48411_sp1.1_rvb from training. Duration: 0.95 2023-04-02 00:51:14,209 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=86807.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:51:18,131 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86810.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:51:28,829 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 2023-04-02 00:51:31,701 WARNING [train.py:1073] (1/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1_rvb from training. Duration: 0.9409375 2023-04-02 00:51:37,261 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 2023-04-02 00:51:37,288 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 2023-04-02 00:51:38,813 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86827.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:51:45,264 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86832.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:51:47,470 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1_rvb from training. Duration: 25.3818125 2023-04-02 00:51:49,795 INFO [train.py:903] (1/4) Epoch 13, batch 4900, loss[loss=0.2189, simple_loss=0.299, pruned_loss=0.0694, over 19855.00 frames. ], tot_loss[loss=0.2261, simple_loss=0.3015, pruned_loss=0.07539, over 3831488.45 frames. ], batch size: 52, lr: 6.34e-03, grad_scale: 8.0 2023-04-02 00:51:59,396 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.2338, 2.0696, 1.9069, 2.6670, 1.8651, 2.4477, 2.5413, 2.4608], device='cuda:1'), covar=tensor([0.0709, 0.0808, 0.0961, 0.0886, 0.0967, 0.0709, 0.0831, 0.0563], device='cuda:1'), in_proj_covar=tensor([0.0209, 0.0224, 0.0224, 0.0244, 0.0230, 0.0211, 0.0192, 0.0202], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-02 00:52:07,093 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 2023-04-02 00:52:46,046 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.083e+02 5.393e+02 6.845e+02 8.725e+02 1.892e+03, threshold=1.369e+03, percent-clipped=5.0 2023-04-02 00:52:50,738 INFO [train.py:903] (1/4) Epoch 13, batch 4950, loss[loss=0.2598, simple_loss=0.3355, pruned_loss=0.09207, over 19672.00 frames. ], tot_loss[loss=0.227, simple_loss=0.3024, pruned_loss=0.07583, over 3835549.05 frames. ], batch size: 60, lr: 6.34e-03, grad_scale: 8.0 2023-04-02 00:53:05,802 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86896.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:53:08,080 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 2023-04-02 00:53:31,777 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86917.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:53:32,652 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 2023-04-02 00:53:37,655 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86922.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:53:56,231 INFO [train.py:903] (1/4) Epoch 13, batch 5000, loss[loss=0.3288, simple_loss=0.3714, pruned_loss=0.1431, over 13781.00 frames. ], tot_loss[loss=0.2255, simple_loss=0.3006, pruned_loss=0.0752, over 3826493.19 frames. ], batch size: 136, lr: 6.34e-03, grad_scale: 8.0 2023-04-02 00:54:03,395 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_rvb from training. Duration: 27.14 2023-04-02 00:54:03,562 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=86942.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:54:14,010 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 2023-04-02 00:54:49,281 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86979.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:54:52,319 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.998e+02 5.650e+02 6.679e+02 8.615e+02 1.943e+03, threshold=1.336e+03, percent-clipped=4.0 2023-04-02 00:54:56,905 INFO [train.py:903] (1/4) Epoch 13, batch 5050, loss[loss=0.2205, simple_loss=0.3069, pruned_loss=0.06706, over 19735.00 frames. ], tot_loss[loss=0.226, simple_loss=0.3009, pruned_loss=0.07554, over 3801333.10 frames. ], batch size: 63, lr: 6.34e-03, grad_scale: 8.0 2023-04-02 00:55:19,939 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=87004.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:55:32,447 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9_rvb from training. Duration: 26.62775 2023-04-02 00:55:38,376 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3597, 1.0796, 1.0560, 1.2741, 1.0475, 1.1876, 1.0800, 1.2263], device='cuda:1'), covar=tensor([0.0983, 0.1223, 0.1368, 0.0836, 0.1124, 0.0569, 0.1269, 0.0784], device='cuda:1'), in_proj_covar=tensor([0.0256, 0.0349, 0.0292, 0.0239, 0.0295, 0.0241, 0.0281, 0.0239], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 00:55:41,826 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87020.0, num_to_drop=1, layers_to_drop={0} 2023-04-02 00:56:01,002 INFO [train.py:903] (1/4) Epoch 13, batch 5100, loss[loss=0.2394, simple_loss=0.3022, pruned_loss=0.08832, over 19607.00 frames. ], tot_loss[loss=0.2268, simple_loss=0.3015, pruned_loss=0.076, over 3803435.20 frames. ], batch size: 50, lr: 6.34e-03, grad_scale: 8.0 2023-04-02 00:56:07,916 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 2023-04-02 00:56:11,369 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_rvb from training. Duration: 25.035 2023-04-02 00:56:18,085 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_rvb from training. Duration: 27.92 2023-04-02 00:56:28,876 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=87057.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:56:58,518 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.198e+02 5.089e+02 6.519e+02 8.713e+02 1.677e+03, threshold=1.304e+03, percent-clipped=2.0 2023-04-02 00:57:03,280 INFO [train.py:903] (1/4) Epoch 13, batch 5150, loss[loss=0.2424, simple_loss=0.3185, pruned_loss=0.08311, over 18704.00 frames. ], tot_loss[loss=0.2269, simple_loss=0.3018, pruned_loss=0.07602, over 3799191.91 frames. ], batch size: 74, lr: 6.33e-03, grad_scale: 8.0 2023-04-02 00:57:15,153 WARNING [train.py:1073] (1/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 2023-04-02 00:57:46,754 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5910, 1.6478, 1.8483, 1.9835, 1.3783, 1.8112, 1.9817, 1.7584], device='cuda:1'), covar=tensor([0.3541, 0.2954, 0.1524, 0.1790, 0.3039, 0.1662, 0.3935, 0.2787], device='cuda:1'), in_proj_covar=tensor([0.0815, 0.0844, 0.0661, 0.0894, 0.0795, 0.0728, 0.0796, 0.0728], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 00:57:48,508 WARNING [train.py:1073] (1/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1_rvb from training. Duration: 0.9681875 2023-04-02 00:58:07,739 INFO [train.py:903] (1/4) Epoch 13, batch 5200, loss[loss=0.2184, simple_loss=0.2856, pruned_loss=0.07562, over 19757.00 frames. ], tot_loss[loss=0.2281, simple_loss=0.3028, pruned_loss=0.07668, over 3805551.32 frames. ], batch size: 47, lr: 6.33e-03, grad_scale: 8.0 2023-04-02 00:58:18,410 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 2023-04-02 00:58:36,927 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.2974, 3.8063, 3.9272, 3.9305, 1.5755, 3.7178, 3.2929, 3.6451], device='cuda:1'), covar=tensor([0.1503, 0.0722, 0.0631, 0.0662, 0.5094, 0.0715, 0.0653, 0.1101], device='cuda:1'), in_proj_covar=tensor([0.0705, 0.0622, 0.0829, 0.0704, 0.0748, 0.0578, 0.0499, 0.0768], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1') 2023-04-02 00:58:54,674 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=87173.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:58:57,969 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=87176.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:59:01,460 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=87178.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:59:04,675 WARNING [train.py:1073] (1/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 2023-04-02 00:59:05,687 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.198e+02 5.751e+02 6.794e+02 9.141e+02 2.147e+03, threshold=1.359e+03, percent-clipped=10.0 2023-04-02 00:59:10,442 INFO [train.py:903] (1/4) Epoch 13, batch 5250, loss[loss=0.2143, simple_loss=0.2947, pruned_loss=0.06696, over 19522.00 frames. ], tot_loss[loss=0.2266, simple_loss=0.302, pruned_loss=0.07562, over 3820771.51 frames. ], batch size: 56, lr: 6.33e-03, grad_scale: 16.0 2023-04-02 00:59:25,754 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=87198.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:59:30,575 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6312, 1.7201, 1.8742, 2.0686, 1.4972, 1.9313, 2.0290, 1.8064], device='cuda:1'), covar=tensor([0.3529, 0.2743, 0.1580, 0.1667, 0.2916, 0.1553, 0.3821, 0.2729], device='cuda:1'), in_proj_covar=tensor([0.0814, 0.0843, 0.0661, 0.0896, 0.0797, 0.0728, 0.0796, 0.0726], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 00:59:31,678 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=87203.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 00:59:39,975 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 2023-04-02 01:00:08,225 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.5694, 4.0455, 4.2036, 4.1928, 1.6616, 3.9677, 3.4811, 3.9377], device='cuda:1'), covar=tensor([0.1392, 0.0735, 0.0594, 0.0654, 0.5296, 0.0723, 0.0652, 0.1066], device='cuda:1'), in_proj_covar=tensor([0.0718, 0.0634, 0.0845, 0.0720, 0.0763, 0.0590, 0.0509, 0.0781], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-02 01:00:13,783 INFO [train.py:903] (1/4) Epoch 13, batch 5300, loss[loss=0.2573, simple_loss=0.3269, pruned_loss=0.09387, over 19664.00 frames. ], tot_loss[loss=0.2253, simple_loss=0.3008, pruned_loss=0.0749, over 3813611.88 frames. ], batch size: 58, lr: 6.33e-03, grad_scale: 16.0 2023-04-02 01:00:18,341 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=87240.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 01:00:27,910 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87248.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 01:00:28,740 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 2023-04-02 01:01:10,978 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.140e+02 5.034e+02 6.226e+02 7.840e+02 1.929e+03, threshold=1.245e+03, percent-clipped=3.0 2023-04-02 01:01:15,816 INFO [train.py:903] (1/4) Epoch 13, batch 5350, loss[loss=0.2466, simple_loss=0.3158, pruned_loss=0.08874, over 19679.00 frames. ], tot_loss[loss=0.2259, simple_loss=0.3013, pruned_loss=0.07531, over 3809471.39 frames. ], batch size: 60, lr: 6.33e-03, grad_scale: 16.0 2023-04-02 01:01:24,133 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=87291.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 01:01:49,521 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9_rvb from training. Duration: 26.438875 2023-04-02 01:01:51,107 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=87313.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 01:02:08,434 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.8319, 1.9569, 2.3192, 1.9702, 2.9426, 3.2004, 3.2156, 3.4696], device='cuda:1'), covar=tensor([0.1243, 0.2720, 0.2415, 0.1819, 0.0794, 0.0374, 0.0182, 0.0264], device='cuda:1'), in_proj_covar=tensor([0.0261, 0.0303, 0.0333, 0.0253, 0.0224, 0.0164, 0.0209, 0.0217], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 01:02:21,303 INFO [train.py:903] (1/4) Epoch 13, batch 5400, loss[loss=0.2497, simple_loss=0.3214, pruned_loss=0.089, over 19453.00 frames. ], tot_loss[loss=0.2259, simple_loss=0.3009, pruned_loss=0.07551, over 3797821.33 frames. ], batch size: 64, lr: 6.33e-03, grad_scale: 8.0 2023-04-02 01:02:24,127 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=87338.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 01:02:44,281 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=87355.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 01:02:52,486 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 2023-04-02 01:02:55,503 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=87364.0, num_to_drop=1, layers_to_drop={1} 2023-04-02 01:03:01,391 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.0380, 2.7719, 2.0404, 2.4980, 0.8070, 2.6811, 2.6572, 2.6511], device='cuda:1'), covar=tensor([0.1296, 0.1442, 0.2087, 0.1029, 0.3611, 0.1043, 0.1026, 0.1506], device='cuda:1'), in_proj_covar=tensor([0.0447, 0.0372, 0.0450, 0.0322, 0.0384, 0.0383, 0.0372, 0.0409], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 01:03:16,675 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87380.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 01:03:19,990 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.794e+02 4.952e+02 6.479e+02 7.900e+02 1.578e+03, threshold=1.296e+03, percent-clipped=3.0 2023-04-02 01:03:23,439 INFO [train.py:903] (1/4) Epoch 13, batch 5450, loss[loss=0.1802, simple_loss=0.2646, pruned_loss=0.04793, over 19852.00 frames. ], tot_loss[loss=0.2256, simple_loss=0.3002, pruned_loss=0.07546, over 3795907.61 frames. ], batch size: 52, lr: 6.32e-03, grad_scale: 8.0 2023-04-02 01:04:25,756 INFO [train.py:903] (1/4) Epoch 13, batch 5500, loss[loss=0.2456, simple_loss=0.3206, pruned_loss=0.08532, over 19669.00 frames. ], tot_loss[loss=0.2265, simple_loss=0.301, pruned_loss=0.07599, over 3801982.10 frames. ], batch size: 58, lr: 6.32e-03, grad_scale: 8.0 2023-04-02 01:04:47,876 WARNING [train.py:1073] (1/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9_rvb from training. Duration: 27.47775 2023-04-02 01:05:21,409 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87479.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 01:05:21,482 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=87479.0, num_to_drop=1, layers_to_drop={1} 2023-04-02 01:05:25,841 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.829e+02 5.041e+02 6.443e+02 8.590e+02 2.643e+03, threshold=1.289e+03, percent-clipped=7.0 2023-04-02 01:05:29,325 INFO [train.py:903] (1/4) Epoch 13, batch 5550, loss[loss=0.2518, simple_loss=0.3284, pruned_loss=0.08762, over 19614.00 frames. ], tot_loss[loss=0.2258, simple_loss=0.3004, pruned_loss=0.07561, over 3802772.97 frames. ], batch size: 57, lr: 6.32e-03, grad_scale: 8.0 2023-04-02 01:05:33,840 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 2023-04-02 01:06:16,491 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87523.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 01:06:23,083 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 2023-04-02 01:06:33,799 INFO [train.py:903] (1/4) Epoch 13, batch 5600, loss[loss=0.1996, simple_loss=0.2662, pruned_loss=0.06644, over 18661.00 frames. ], tot_loss[loss=0.2259, simple_loss=0.3006, pruned_loss=0.07562, over 3805260.76 frames. ], batch size: 41, lr: 6.32e-03, grad_scale: 8.0 2023-04-02 01:06:47,647 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=87547.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 01:07:17,518 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=87572.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 01:07:32,977 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.526e+02 5.023e+02 6.222e+02 8.051e+02 1.328e+03, threshold=1.244e+03, percent-clipped=2.0 2023-04-02 01:07:36,359 INFO [train.py:903] (1/4) Epoch 13, batch 5650, loss[loss=0.2364, simple_loss=0.3134, pruned_loss=0.07971, over 19563.00 frames. ], tot_loss[loss=0.2255, simple_loss=0.3004, pruned_loss=0.07531, over 3815079.23 frames. ], batch size: 61, lr: 6.32e-03, grad_scale: 8.0 2023-04-02 01:07:43,445 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=87592.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 01:07:56,303 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87602.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 01:08:01,785 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87607.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 01:08:06,361 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=87611.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 01:08:23,538 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87623.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 01:08:24,450 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9_rvb from training. Duration: 28.0944375 2023-04-02 01:08:38,517 INFO [train.py:903] (1/4) Epoch 13, batch 5700, loss[loss=0.1996, simple_loss=0.2787, pruned_loss=0.06025, over 19765.00 frames. ], tot_loss[loss=0.2253, simple_loss=0.3006, pruned_loss=0.075, over 3819464.42 frames. ], batch size: 54, lr: 6.31e-03, grad_scale: 8.0 2023-04-02 01:08:39,009 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=87636.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 01:09:36,755 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.413e+02 4.997e+02 6.690e+02 8.841e+02 1.921e+03, threshold=1.338e+03, percent-clipped=6.0 2023-04-02 01:09:40,136 INFO [train.py:903] (1/4) Epoch 13, batch 5750, loss[loss=0.2283, simple_loss=0.2994, pruned_loss=0.07861, over 19504.00 frames. ], tot_loss[loss=0.2266, simple_loss=0.3016, pruned_loss=0.07583, over 3818292.91 frames. ], batch size: 49, lr: 6.31e-03, grad_scale: 8.0 2023-04-02 01:09:41,277 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9_rvb from training. Duration: 33.038875 2023-04-02 01:09:49,519 WARNING [train.py:1073] (1/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 2023-04-02 01:09:55,217 WARNING [train.py:1073] (1/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 2023-04-02 01:10:07,765 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=87707.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 01:10:27,831 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=87724.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 01:10:41,827 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=87735.0, num_to_drop=1, layers_to_drop={0} 2023-04-02 01:10:42,503 INFO [train.py:903] (1/4) Epoch 13, batch 5800, loss[loss=0.3317, simple_loss=0.3761, pruned_loss=0.1437, over 13137.00 frames. ], tot_loss[loss=0.2266, simple_loss=0.3016, pruned_loss=0.0758, over 3813038.08 frames. ], batch size: 136, lr: 6.31e-03, grad_scale: 8.0 2023-04-02 01:10:58,848 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.2214, 2.0174, 1.8201, 1.7032, 1.5638, 1.6778, 0.4261, 1.0540], device='cuda:1'), covar=tensor([0.0387, 0.0434, 0.0322, 0.0485, 0.0858, 0.0580, 0.0908, 0.0715], device='cuda:1'), in_proj_covar=tensor([0.0345, 0.0341, 0.0337, 0.0364, 0.0439, 0.0363, 0.0321, 0.0332], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 01:11:13,084 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=87760.0, num_to_drop=1, layers_to_drop={1} 2023-04-02 01:11:43,258 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.294e+02 5.041e+02 6.242e+02 7.928e+02 1.581e+03, threshold=1.248e+03, percent-clipped=4.0 2023-04-02 01:11:46,975 INFO [train.py:903] (1/4) Epoch 13, batch 5850, loss[loss=0.2414, simple_loss=0.3163, pruned_loss=0.08324, over 19606.00 frames. ], tot_loss[loss=0.2269, simple_loss=0.302, pruned_loss=0.07595, over 3833777.23 frames. ], batch size: 57, lr: 6.31e-03, grad_scale: 8.0 2023-04-02 01:12:33,211 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=87823.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 01:12:47,993 INFO [train.py:903] (1/4) Epoch 13, batch 5900, loss[loss=0.2242, simple_loss=0.2967, pruned_loss=0.07587, over 19739.00 frames. ], tot_loss[loss=0.2275, simple_loss=0.3026, pruned_loss=0.07627, over 3840930.52 frames. ], batch size: 51, lr: 6.31e-03, grad_scale: 8.0 2023-04-02 01:12:49,016 WARNING [train.py:1073] (1/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 2023-04-02 01:12:51,614 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=87839.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 01:13:10,535 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9_rvb from training. Duration: 27.511125 2023-04-02 01:13:27,496 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=87867.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 01:13:47,011 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.002e+02 5.447e+02 6.662e+02 8.353e+02 2.279e+03, threshold=1.332e+03, percent-clipped=6.0 2023-04-02 01:13:50,401 INFO [train.py:903] (1/4) Epoch 13, batch 5950, loss[loss=0.2743, simple_loss=0.339, pruned_loss=0.1049, over 18050.00 frames. ], tot_loss[loss=0.2281, simple_loss=0.3028, pruned_loss=0.07666, over 3828478.57 frames. ], batch size: 83, lr: 6.31e-03, grad_scale: 8.0 2023-04-02 01:13:57,836 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.8059, 1.9070, 2.1170, 2.6337, 1.8541, 2.5039, 2.2848, 1.9062], device='cuda:1'), covar=tensor([0.3760, 0.3150, 0.1554, 0.1833, 0.3476, 0.1563, 0.3840, 0.2921], device='cuda:1'), in_proj_covar=tensor([0.0817, 0.0849, 0.0660, 0.0897, 0.0798, 0.0729, 0.0798, 0.0725], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 01:14:53,070 INFO [train.py:903] (1/4) Epoch 13, batch 6000, loss[loss=0.2412, simple_loss=0.317, pruned_loss=0.08276, over 19798.00 frames. ], tot_loss[loss=0.2277, simple_loss=0.3026, pruned_loss=0.07642, over 3829338.31 frames. ], batch size: 56, lr: 6.30e-03, grad_scale: 8.0 2023-04-02 01:14:53,071 INFO [train.py:928] (1/4) Computing validation loss 2023-04-02 01:15:06,412 INFO [train.py:937] (1/4) Epoch 13, validation: loss=0.175, simple_loss=0.2755, pruned_loss=0.03726, over 944034.00 frames. 2023-04-02 01:15:06,415 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18821MB 2023-04-02 01:15:09,194 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=87938.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 01:15:17,514 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3094, 1.2949, 1.4303, 1.4381, 1.7694, 1.8606, 1.7420, 0.5098], device='cuda:1'), covar=tensor([0.2158, 0.3947, 0.2439, 0.1808, 0.1468, 0.2073, 0.1313, 0.4204], device='cuda:1'), in_proj_covar=tensor([0.0500, 0.0591, 0.0632, 0.0447, 0.0603, 0.0498, 0.0649, 0.0509], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 01:15:19,414 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=87946.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 01:15:25,216 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=87951.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 01:15:36,225 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.0744, 1.3859, 1.7422, 1.2532, 2.8966, 3.8901, 3.5963, 4.0691], device='cuda:1'), covar=tensor([0.1572, 0.3343, 0.3002, 0.2068, 0.0463, 0.0134, 0.0177, 0.0172], device='cuda:1'), in_proj_covar=tensor([0.0259, 0.0299, 0.0330, 0.0251, 0.0221, 0.0162, 0.0207, 0.0214], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 01:15:37,486 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2752, 1.6031, 2.0176, 1.5470, 3.1391, 4.8996, 4.7355, 5.0726], device='cuda:1'), covar=tensor([0.1536, 0.3204, 0.2908, 0.1988, 0.0505, 0.0153, 0.0142, 0.0155], device='cuda:1'), in_proj_covar=tensor([0.0259, 0.0299, 0.0330, 0.0251, 0.0220, 0.0162, 0.0207, 0.0214], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 01:15:41,058 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=87963.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 01:15:45,511 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=87967.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 01:15:48,856 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 2023-04-02 01:16:04,014 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=87982.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 01:16:04,764 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.636e+02 5.142e+02 6.649e+02 8.424e+02 2.027e+03, threshold=1.330e+03, percent-clipped=9.0 2023-04-02 01:16:08,330 INFO [train.py:903] (1/4) Epoch 13, batch 6050, loss[loss=0.2509, simple_loss=0.3254, pruned_loss=0.08813, over 19338.00 frames. ], tot_loss[loss=0.228, simple_loss=0.3027, pruned_loss=0.07667, over 3830317.69 frames. ], batch size: 66, lr: 6.30e-03, grad_scale: 8.0 2023-04-02 01:16:12,356 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=87988.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 01:16:34,842 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.8380, 1.6469, 1.9961, 1.7491, 4.3459, 1.0374, 2.5474, 4.7510], device='cuda:1'), covar=tensor([0.0375, 0.2542, 0.2353, 0.1873, 0.0727, 0.2763, 0.1374, 0.0166], device='cuda:1'), in_proj_covar=tensor([0.0365, 0.0345, 0.0356, 0.0325, 0.0352, 0.0332, 0.0343, 0.0367], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 01:17:12,302 INFO [train.py:903] (1/4) Epoch 13, batch 6100, loss[loss=0.2132, simple_loss=0.2846, pruned_loss=0.07092, over 19045.00 frames. ], tot_loss[loss=0.2292, simple_loss=0.3034, pruned_loss=0.07747, over 3821160.52 frames. ], batch size: 42, lr: 6.30e-03, grad_scale: 8.0 2023-04-02 01:17:42,742 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=88061.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 01:17:49,523 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=88066.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 01:18:09,962 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=88082.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 01:18:10,749 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.507e+02 5.311e+02 6.775e+02 8.712e+02 1.953e+03, threshold=1.355e+03, percent-clipped=3.0 2023-04-02 01:18:14,178 INFO [train.py:903] (1/4) Epoch 13, batch 6150, loss[loss=0.2258, simple_loss=0.2981, pruned_loss=0.07674, over 19666.00 frames. ], tot_loss[loss=0.2288, simple_loss=0.3031, pruned_loss=0.07725, over 3823147.82 frames. ], batch size: 53, lr: 6.30e-03, grad_scale: 8.0 2023-04-02 01:18:18,117 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.8590, 1.5150, 1.8511, 1.7263, 4.3453, 0.9795, 2.2684, 4.6117], device='cuda:1'), covar=tensor([0.0384, 0.2738, 0.2845, 0.1918, 0.0757, 0.2729, 0.1498, 0.0217], device='cuda:1'), in_proj_covar=tensor([0.0366, 0.0345, 0.0357, 0.0324, 0.0352, 0.0331, 0.0343, 0.0367], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 01:18:24,944 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=88095.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 01:18:38,419 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9_rvb from training. Duration: 31.02225 2023-04-02 01:18:54,520 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=88118.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 01:18:56,866 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=88120.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 01:19:15,200 INFO [train.py:903] (1/4) Epoch 13, batch 6200, loss[loss=0.2303, simple_loss=0.306, pruned_loss=0.07734, over 19676.00 frames. ], tot_loss[loss=0.2281, simple_loss=0.3029, pruned_loss=0.07667, over 3839636.97 frames. ], batch size: 53, lr: 6.30e-03, grad_scale: 8.0 2023-04-02 01:19:31,807 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.59 vs. limit=2.0 2023-04-02 01:20:13,326 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.409e+02 5.304e+02 6.722e+02 8.627e+02 2.252e+03, threshold=1.344e+03, percent-clipped=3.0 2023-04-02 01:20:16,888 INFO [train.py:903] (1/4) Epoch 13, batch 6250, loss[loss=0.2339, simple_loss=0.3065, pruned_loss=0.08068, over 18720.00 frames. ], tot_loss[loss=0.2278, simple_loss=0.3027, pruned_loss=0.07651, over 3820771.54 frames. ], batch size: 74, lr: 6.29e-03, grad_scale: 8.0 2023-04-02 01:20:29,076 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=88194.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 01:20:44,732 WARNING [train.py:1073] (1/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9_rvb from training. Duration: 25.988875 2023-04-02 01:20:58,217 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=88219.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 01:21:20,745 INFO [train.py:903] (1/4) Epoch 13, batch 6300, loss[loss=0.215, simple_loss=0.2946, pruned_loss=0.06773, over 19664.00 frames. ], tot_loss[loss=0.2284, simple_loss=0.3033, pruned_loss=0.07674, over 3826023.61 frames. ], batch size: 55, lr: 6.29e-03, grad_scale: 8.0 2023-04-02 01:21:23,571 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=88238.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 01:21:32,847 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.45 vs. limit=5.0 2023-04-02 01:21:52,901 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=88263.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 01:22:18,335 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.601e+02 5.052e+02 6.199e+02 7.712e+02 1.989e+03, threshold=1.240e+03, percent-clipped=5.0 2023-04-02 01:22:21,896 INFO [train.py:903] (1/4) Epoch 13, batch 6350, loss[loss=0.2187, simple_loss=0.2861, pruned_loss=0.07559, over 19751.00 frames. ], tot_loss[loss=0.2282, simple_loss=0.303, pruned_loss=0.07667, over 3835895.10 frames. ], batch size: 46, lr: 6.29e-03, grad_scale: 8.0 2023-04-02 01:22:32,648 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([6.2360, 5.5573, 2.8571, 4.8522, 1.4476, 5.5261, 5.5766, 5.7277], device='cuda:1'), covar=tensor([0.0348, 0.0885, 0.1986, 0.0649, 0.3707, 0.0560, 0.0611, 0.0939], device='cuda:1'), in_proj_covar=tensor([0.0444, 0.0374, 0.0451, 0.0321, 0.0386, 0.0380, 0.0373, 0.0408], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 01:22:50,128 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=88309.0, num_to_drop=1, layers_to_drop={1} 2023-04-02 01:23:02,584 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=88317.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 01:23:08,241 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=88322.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 01:23:24,615 INFO [train.py:903] (1/4) Epoch 13, batch 6400, loss[loss=0.2297, simple_loss=0.3004, pruned_loss=0.07949, over 19763.00 frames. ], tot_loss[loss=0.2274, simple_loss=0.3027, pruned_loss=0.07608, over 3828487.18 frames. ], batch size: 54, lr: 6.29e-03, grad_scale: 8.0 2023-04-02 01:23:27,236 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=88338.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 01:23:31,697 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=88342.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 01:23:38,213 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=88347.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 01:23:58,506 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=88363.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 01:23:58,567 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3767, 1.3838, 1.5997, 1.5713, 2.2060, 2.0731, 2.2556, 0.7490], device='cuda:1'), covar=tensor([0.2092, 0.3794, 0.2289, 0.1734, 0.1417, 0.1922, 0.1251, 0.3996], device='cuda:1'), in_proj_covar=tensor([0.0499, 0.0592, 0.0629, 0.0448, 0.0603, 0.0499, 0.0645, 0.0505], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 01:24:16,116 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9134, 2.0409, 2.1761, 2.7091, 1.8589, 2.5844, 2.3074, 2.0339], device='cuda:1'), covar=tensor([0.3845, 0.3140, 0.1591, 0.1905, 0.3588, 0.1579, 0.3873, 0.2799], device='cuda:1'), in_proj_covar=tensor([0.0817, 0.0848, 0.0662, 0.0901, 0.0798, 0.0733, 0.0800, 0.0726], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 01:24:18,746 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.83 vs. limit=2.0 2023-04-02 01:24:22,195 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.590e+02 5.639e+02 6.967e+02 8.699e+02 1.659e+03, threshold=1.393e+03, percent-clipped=7.0 2023-04-02 01:24:25,784 INFO [train.py:903] (1/4) Epoch 13, batch 6450, loss[loss=0.2191, simple_loss=0.2999, pruned_loss=0.06917, over 19787.00 frames. ], tot_loss[loss=0.2278, simple_loss=0.3028, pruned_loss=0.07637, over 3818445.72 frames. ], batch size: 56, lr: 6.29e-03, grad_scale: 8.0 2023-04-02 01:25:05,097 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9_rvb from training. Duration: 28.638875 2023-04-02 01:25:19,074 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.7061, 1.8400, 1.9817, 2.3283, 1.6467, 2.2582, 2.1510, 1.8549], device='cuda:1'), covar=tensor([0.3745, 0.3154, 0.1675, 0.1808, 0.3330, 0.1571, 0.3972, 0.2886], device='cuda:1'), in_proj_covar=tensor([0.0813, 0.0845, 0.0661, 0.0898, 0.0796, 0.0729, 0.0797, 0.0722], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 01:25:28,931 INFO [train.py:903] (1/4) Epoch 13, batch 6500, loss[loss=0.2315, simple_loss=0.3117, pruned_loss=0.07566, over 19733.00 frames. ], tot_loss[loss=0.2272, simple_loss=0.3024, pruned_loss=0.07605, over 3825567.98 frames. ], batch size: 63, lr: 6.29e-03, grad_scale: 8.0 2023-04-02 01:25:30,980 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1_rvb from training. Duration: 0.836375 2023-04-02 01:26:00,488 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=88462.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 01:26:26,785 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.125e+02 5.190e+02 6.145e+02 7.751e+02 1.614e+03, threshold=1.229e+03, percent-clipped=4.0 2023-04-02 01:26:30,115 INFO [train.py:903] (1/4) Epoch 13, batch 6550, loss[loss=0.2743, simple_loss=0.3245, pruned_loss=0.1121, over 19113.00 frames. ], tot_loss[loss=0.227, simple_loss=0.3017, pruned_loss=0.07614, over 3830034.30 frames. ], batch size: 42, lr: 6.28e-03, grad_scale: 8.0 2023-04-02 01:27:11,760 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1279, 1.7958, 1.4562, 1.2326, 1.6542, 1.1471, 1.0882, 1.6482], device='cuda:1'), covar=tensor([0.0716, 0.0745, 0.0940, 0.0656, 0.0455, 0.1085, 0.0571, 0.0360], device='cuda:1'), in_proj_covar=tensor([0.0292, 0.0304, 0.0326, 0.0249, 0.0239, 0.0321, 0.0293, 0.0262], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 01:27:31,536 INFO [train.py:903] (1/4) Epoch 13, batch 6600, loss[loss=0.2674, simple_loss=0.3295, pruned_loss=0.1026, over 19528.00 frames. ], tot_loss[loss=0.2266, simple_loss=0.3015, pruned_loss=0.07584, over 3829925.77 frames. ], batch size: 56, lr: 6.28e-03, grad_scale: 8.0 2023-04-02 01:28:21,382 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=88577.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 01:28:28,437 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.648e+02 5.175e+02 6.632e+02 9.239e+02 1.861e+03, threshold=1.326e+03, percent-clipped=7.0 2023-04-02 01:28:32,108 INFO [train.py:903] (1/4) Epoch 13, batch 6650, loss[loss=0.1941, simple_loss=0.2658, pruned_loss=0.06116, over 19737.00 frames. ], tot_loss[loss=0.2279, simple_loss=0.3026, pruned_loss=0.07662, over 3830728.89 frames. ], batch size: 46, lr: 6.28e-03, grad_scale: 8.0 2023-04-02 01:29:34,640 INFO [train.py:903] (1/4) Epoch 13, batch 6700, loss[loss=0.2261, simple_loss=0.3032, pruned_loss=0.07456, over 19539.00 frames. ], tot_loss[loss=0.2288, simple_loss=0.3035, pruned_loss=0.07706, over 3825503.74 frames. ], batch size: 56, lr: 6.28e-03, grad_scale: 8.0 2023-04-02 01:29:55,526 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=88653.0, num_to_drop=1, layers_to_drop={1} 2023-04-02 01:30:29,681 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.623e+02 5.222e+02 6.682e+02 8.627e+02 1.913e+03, threshold=1.336e+03, percent-clipped=6.0 2023-04-02 01:30:33,147 INFO [train.py:903] (1/4) Epoch 13, batch 6750, loss[loss=0.1952, simple_loss=0.2648, pruned_loss=0.06279, over 19730.00 frames. ], tot_loss[loss=0.2274, simple_loss=0.3026, pruned_loss=0.07607, over 3841142.82 frames. ], batch size: 47, lr: 6.28e-03, grad_scale: 8.0 2023-04-02 01:31:28,207 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.9000, 2.6772, 1.7903, 1.9670, 1.6328, 1.9891, 0.9787, 1.7812], device='cuda:1'), covar=tensor([0.0824, 0.0721, 0.0752, 0.1257, 0.1362, 0.1370, 0.1300, 0.1192], device='cuda:1'), in_proj_covar=tensor([0.0342, 0.0337, 0.0331, 0.0358, 0.0431, 0.0357, 0.0316, 0.0328], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 01:31:31,217 INFO [train.py:903] (1/4) Epoch 13, batch 6800, loss[loss=0.2309, simple_loss=0.3065, pruned_loss=0.07764, over 19756.00 frames. ], tot_loss[loss=0.2259, simple_loss=0.3012, pruned_loss=0.07532, over 3837798.71 frames. ], batch size: 54, lr: 6.28e-03, grad_scale: 8.0 2023-04-02 01:32:17,356 WARNING [train.py:1073] (1/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 2023-04-02 01:32:18,469 WARNING [train.py:1073] (1/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 2023-04-02 01:32:21,842 INFO [train.py:903] (1/4) Epoch 14, batch 0, loss[loss=0.3343, simple_loss=0.3877, pruned_loss=0.1404, over 19677.00 frames. ], tot_loss[loss=0.3343, simple_loss=0.3877, pruned_loss=0.1404, over 19677.00 frames. ], batch size: 60, lr: 6.05e-03, grad_scale: 8.0 2023-04-02 01:32:21,842 INFO [train.py:928] (1/4) Computing validation loss 2023-04-02 01:32:33,653 INFO [train.py:937] (1/4) Epoch 14, validation: loss=0.1763, simple_loss=0.2772, pruned_loss=0.03773, over 944034.00 frames. 2023-04-02 01:32:33,654 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18821MB 2023-04-02 01:32:41,851 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=88768.0, num_to_drop=1, layers_to_drop={1} 2023-04-02 01:32:49,788 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 2023-04-02 01:32:59,168 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.227e+02 4.998e+02 6.377e+02 7.989e+02 1.719e+03, threshold=1.275e+03, percent-clipped=2.0 2023-04-02 01:33:10,456 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.31 vs. limit=2.0 2023-04-02 01:33:40,050 INFO [train.py:903] (1/4) Epoch 14, batch 50, loss[loss=0.1969, simple_loss=0.2845, pruned_loss=0.05464, over 19574.00 frames. ], tot_loss[loss=0.2201, simple_loss=0.2968, pruned_loss=0.07171, over 871526.77 frames. ], batch size: 52, lr: 6.04e-03, grad_scale: 8.0 2023-04-02 01:34:01,775 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=88833.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 01:34:15,362 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1_rvb from training. Duration: 27.0318125 2023-04-02 01:34:34,177 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=88858.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 01:34:40,589 INFO [train.py:903] (1/4) Epoch 14, batch 100, loss[loss=0.2156, simple_loss=0.2954, pruned_loss=0.06794, over 19667.00 frames. ], tot_loss[loss=0.2265, simple_loss=0.3012, pruned_loss=0.07589, over 1517255.67 frames. ], batch size: 60, lr: 6.04e-03, grad_scale: 8.0 2023-04-02 01:34:51,064 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 2023-04-02 01:35:02,778 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.971e+02 5.271e+02 6.621e+02 8.700e+02 2.391e+03, threshold=1.324e+03, percent-clipped=10.0 2023-04-02 01:35:41,058 INFO [train.py:903] (1/4) Epoch 14, batch 150, loss[loss=0.2177, simple_loss=0.2977, pruned_loss=0.06891, over 19740.00 frames. ], tot_loss[loss=0.228, simple_loss=0.3021, pruned_loss=0.07693, over 2018696.70 frames. ], batch size: 63, lr: 6.04e-03, grad_scale: 8.0 2023-04-02 01:35:53,911 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.1461, 1.8825, 1.9801, 2.7963, 1.9596, 2.5410, 2.4743, 2.4701], device='cuda:1'), covar=tensor([0.0742, 0.0845, 0.0959, 0.0827, 0.0860, 0.0668, 0.0918, 0.0555], device='cuda:1'), in_proj_covar=tensor([0.0208, 0.0223, 0.0225, 0.0244, 0.0228, 0.0210, 0.0193, 0.0202], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-02 01:36:00,288 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.45 vs. limit=2.0 2023-04-02 01:36:29,646 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.0276, 1.3025, 1.7388, 0.6088, 2.1086, 2.4804, 2.1636, 2.5308], device='cuda:1'), covar=tensor([0.1386, 0.3244, 0.2650, 0.2298, 0.0489, 0.0247, 0.0342, 0.0327], device='cuda:1'), in_proj_covar=tensor([0.0261, 0.0301, 0.0331, 0.0251, 0.0220, 0.0165, 0.0207, 0.0217], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 01:36:38,770 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 2023-04-02 01:36:39,936 INFO [train.py:903] (1/4) Epoch 14, batch 200, loss[loss=0.2126, simple_loss=0.2986, pruned_loss=0.06336, over 19570.00 frames. ], tot_loss[loss=0.2276, simple_loss=0.3024, pruned_loss=0.07638, over 2418987.98 frames. ], batch size: 61, lr: 6.04e-03, grad_scale: 8.0 2023-04-02 01:37:03,950 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.747e+02 5.170e+02 6.545e+02 8.695e+02 1.666e+03, threshold=1.309e+03, percent-clipped=4.0 2023-04-02 01:37:37,743 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.1744, 3.7528, 3.8681, 3.8327, 1.4536, 3.6174, 3.1392, 3.5686], device='cuda:1'), covar=tensor([0.1650, 0.0966, 0.0672, 0.0770, 0.5595, 0.0853, 0.0726, 0.1268], device='cuda:1'), in_proj_covar=tensor([0.0707, 0.0631, 0.0835, 0.0715, 0.0755, 0.0583, 0.0503, 0.0765], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1') 2023-04-02 01:37:41,109 INFO [train.py:903] (1/4) Epoch 14, batch 250, loss[loss=0.2031, simple_loss=0.2846, pruned_loss=0.06082, over 19689.00 frames. ], tot_loss[loss=0.2262, simple_loss=0.3012, pruned_loss=0.07565, over 2732164.66 frames. ], batch size: 53, lr: 6.04e-03, grad_scale: 8.0 2023-04-02 01:37:53,625 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=89024.0, num_to_drop=1, layers_to_drop={1} 2023-04-02 01:38:17,641 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=89045.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 01:38:22,442 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=89049.0, num_to_drop=1, layers_to_drop={1} 2023-04-02 01:38:43,483 INFO [train.py:903] (1/4) Epoch 14, batch 300, loss[loss=0.2025, simple_loss=0.2781, pruned_loss=0.06341, over 19611.00 frames. ], tot_loss[loss=0.2266, simple_loss=0.3019, pruned_loss=0.07561, over 2960855.87 frames. ], batch size: 50, lr: 6.04e-03, grad_scale: 8.0 2023-04-02 01:38:52,818 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([5.1071, 5.1151, 5.9825, 5.8932, 2.0889, 5.5641, 4.6130, 5.5668], device='cuda:1'), covar=tensor([0.1369, 0.0677, 0.0495, 0.0558, 0.5128, 0.0600, 0.0607, 0.1074], device='cuda:1'), in_proj_covar=tensor([0.0707, 0.0630, 0.0836, 0.0715, 0.0753, 0.0583, 0.0504, 0.0764], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1') 2023-04-02 01:39:05,428 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.258e+02 4.830e+02 5.923e+02 7.544e+02 2.111e+03, threshold=1.185e+03, percent-clipped=1.0 2023-04-02 01:39:45,141 INFO [train.py:903] (1/4) Epoch 14, batch 350, loss[loss=0.2459, simple_loss=0.3196, pruned_loss=0.08611, over 19264.00 frames. ], tot_loss[loss=0.2262, simple_loss=0.3014, pruned_loss=0.07557, over 3153031.94 frames. ], batch size: 66, lr: 6.03e-03, grad_scale: 8.0 2023-04-02 01:39:47,471 WARNING [train.py:1073] (1/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1_rvb from training. Duration: 0.9681875 2023-04-02 01:40:29,861 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.8377, 1.8708, 1.9973, 2.4694, 1.8554, 2.3475, 2.1749, 1.9212], device='cuda:1'), covar=tensor([0.3263, 0.2753, 0.1346, 0.1561, 0.2824, 0.1313, 0.3070, 0.2283], device='cuda:1'), in_proj_covar=tensor([0.0824, 0.0859, 0.0666, 0.0908, 0.0807, 0.0738, 0.0811, 0.0728], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 01:40:46,818 INFO [train.py:903] (1/4) Epoch 14, batch 400, loss[loss=0.1857, simple_loss=0.2512, pruned_loss=0.0601, over 19744.00 frames. ], tot_loss[loss=0.2259, simple_loss=0.3011, pruned_loss=0.07538, over 3297962.65 frames. ], batch size: 46, lr: 6.03e-03, grad_scale: 8.0 2023-04-02 01:41:11,936 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.999e+02 4.893e+02 5.942e+02 7.582e+02 1.529e+03, threshold=1.188e+03, percent-clipped=4.0 2023-04-02 01:41:16,774 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=89187.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 01:41:36,346 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.1215, 1.8563, 1.7278, 2.0325, 1.7865, 1.8730, 1.7854, 2.1138], device='cuda:1'), covar=tensor([0.0886, 0.1510, 0.1450, 0.1002, 0.1382, 0.0488, 0.1206, 0.0623], device='cuda:1'), in_proj_covar=tensor([0.0257, 0.0353, 0.0297, 0.0240, 0.0298, 0.0243, 0.0285, 0.0242], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 01:41:47,614 INFO [train.py:903] (1/4) Epoch 14, batch 450, loss[loss=0.1859, simple_loss=0.2612, pruned_loss=0.05533, over 19738.00 frames. ], tot_loss[loss=0.226, simple_loss=0.3013, pruned_loss=0.0753, over 3422310.84 frames. ], batch size: 47, lr: 6.03e-03, grad_scale: 8.0 2023-04-02 01:42:19,961 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 2023-04-02 01:42:20,947 WARNING [train.py:1073] (1/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 2023-04-02 01:42:51,695 INFO [train.py:903] (1/4) Epoch 14, batch 500, loss[loss=0.2504, simple_loss=0.317, pruned_loss=0.09187, over 19543.00 frames. ], tot_loss[loss=0.2259, simple_loss=0.3011, pruned_loss=0.07534, over 3501569.75 frames. ], batch size: 56, lr: 6.03e-03, grad_scale: 8.0 2023-04-02 01:43:04,791 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6642, 1.7242, 1.3673, 1.7701, 1.7902, 1.3905, 1.4136, 1.6325], device='cuda:1'), covar=tensor([0.1159, 0.1383, 0.1666, 0.1027, 0.1153, 0.0760, 0.1501, 0.0865], device='cuda:1'), in_proj_covar=tensor([0.0256, 0.0350, 0.0295, 0.0239, 0.0295, 0.0241, 0.0285, 0.0242], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 01:43:13,299 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.040e+02 5.086e+02 6.501e+02 8.394e+02 1.936e+03, threshold=1.300e+03, percent-clipped=6.0 2023-04-02 01:43:51,222 INFO [train.py:903] (1/4) Epoch 14, batch 550, loss[loss=0.1956, simple_loss=0.2712, pruned_loss=0.05997, over 19469.00 frames. ], tot_loss[loss=0.2257, simple_loss=0.301, pruned_loss=0.07526, over 3570637.99 frames. ], batch size: 49, lr: 6.03e-03, grad_scale: 8.0 2023-04-02 01:44:05,920 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 2023-04-02 01:44:50,881 INFO [train.py:903] (1/4) Epoch 14, batch 600, loss[loss=0.2453, simple_loss=0.3218, pruned_loss=0.08443, over 18691.00 frames. ], tot_loss[loss=0.2253, simple_loss=0.3007, pruned_loss=0.07498, over 3639262.51 frames. ], batch size: 74, lr: 6.03e-03, grad_scale: 8.0 2023-04-02 01:45:14,771 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.733e+02 5.437e+02 6.463e+02 8.394e+02 1.645e+03, threshold=1.293e+03, percent-clipped=4.0 2023-04-02 01:45:23,619 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=89389.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 01:45:27,423 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=89392.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 01:45:32,878 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 2023-04-02 01:45:52,140 INFO [train.py:903] (1/4) Epoch 14, batch 650, loss[loss=0.2305, simple_loss=0.308, pruned_loss=0.07655, over 19781.00 frames. ], tot_loss[loss=0.2243, simple_loss=0.2997, pruned_loss=0.07445, over 3684945.05 frames. ], batch size: 54, lr: 6.02e-03, grad_scale: 8.0 2023-04-02 01:46:29,612 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=89443.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 01:46:45,753 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=89457.0, num_to_drop=1, layers_to_drop={0} 2023-04-02 01:46:53,905 INFO [train.py:903] (1/4) Epoch 14, batch 700, loss[loss=0.2387, simple_loss=0.3123, pruned_loss=0.08249, over 19496.00 frames. ], tot_loss[loss=0.2249, simple_loss=0.3, pruned_loss=0.07495, over 3722445.43 frames. ], batch size: 64, lr: 6.02e-03, grad_scale: 8.0 2023-04-02 01:47:21,067 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.092e+02 5.132e+02 6.716e+02 8.076e+02 1.293e+03, threshold=1.343e+03, percent-clipped=1.0 2023-04-02 01:47:44,531 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=89504.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 01:47:59,820 INFO [train.py:903] (1/4) Epoch 14, batch 750, loss[loss=0.2227, simple_loss=0.3072, pruned_loss=0.06909, over 19663.00 frames. ], tot_loss[loss=0.2243, simple_loss=0.2995, pruned_loss=0.07457, over 3747475.84 frames. ], batch size: 55, lr: 6.02e-03, grad_scale: 8.0 2023-04-02 01:48:14,759 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=89527.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 01:48:19,167 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=89531.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 01:48:30,280 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5227, 1.2363, 1.1732, 1.4089, 1.1230, 1.3157, 1.1705, 1.3689], device='cuda:1'), covar=tensor([0.1101, 0.1235, 0.1518, 0.0974, 0.1212, 0.0586, 0.1412, 0.0785], device='cuda:1'), in_proj_covar=tensor([0.0256, 0.0350, 0.0294, 0.0239, 0.0294, 0.0241, 0.0284, 0.0240], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 01:48:48,362 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 2023-04-02 01:49:01,549 INFO [train.py:903] (1/4) Epoch 14, batch 800, loss[loss=0.2309, simple_loss=0.303, pruned_loss=0.07943, over 19681.00 frames. ], tot_loss[loss=0.2249, simple_loss=0.3001, pruned_loss=0.07484, over 3763838.50 frames. ], batch size: 53, lr: 6.02e-03, grad_scale: 8.0 2023-04-02 01:49:16,438 WARNING [train.py:1073] (1/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 2023-04-02 01:49:24,440 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.547e+02 5.343e+02 6.523e+02 8.164e+02 1.688e+03, threshold=1.305e+03, percent-clipped=4.0 2023-04-02 01:50:01,790 INFO [train.py:903] (1/4) Epoch 14, batch 850, loss[loss=0.1918, simple_loss=0.2668, pruned_loss=0.05842, over 19476.00 frames. ], tot_loss[loss=0.2248, simple_loss=0.3002, pruned_loss=0.0747, over 3774665.56 frames. ], batch size: 49, lr: 6.02e-03, grad_scale: 8.0 2023-04-02 01:50:38,092 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1731, 1.9843, 1.5589, 1.1688, 1.8088, 1.1210, 1.1237, 1.7740], device='cuda:1'), covar=tensor([0.0949, 0.0745, 0.1079, 0.0963, 0.0535, 0.1295, 0.0748, 0.0440], device='cuda:1'), in_proj_covar=tensor([0.0295, 0.0307, 0.0328, 0.0250, 0.0240, 0.0326, 0.0297, 0.0266], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 01:50:42,887 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=89646.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 01:50:55,556 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9_rvb from training. Duration: 25.061125 2023-04-02 01:51:03,552 INFO [train.py:903] (1/4) Epoch 14, batch 900, loss[loss=0.2259, simple_loss=0.3002, pruned_loss=0.07577, over 19572.00 frames. ], tot_loss[loss=0.2249, simple_loss=0.3003, pruned_loss=0.07478, over 3782097.58 frames. ], batch size: 61, lr: 6.02e-03, grad_scale: 8.0 2023-04-02 01:51:29,677 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.240e+02 4.983e+02 6.131e+02 7.467e+02 1.791e+03, threshold=1.226e+03, percent-clipped=3.0 2023-04-02 01:51:46,421 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=89698.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 01:52:07,432 INFO [train.py:903] (1/4) Epoch 14, batch 950, loss[loss=0.2028, simple_loss=0.2837, pruned_loss=0.06094, over 19738.00 frames. ], tot_loss[loss=0.2253, simple_loss=0.3005, pruned_loss=0.07511, over 3764926.37 frames. ], batch size: 51, lr: 6.01e-03, grad_scale: 8.0 2023-04-02 01:52:07,464 WARNING [train.py:1073] (1/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9_rvb from training. Duration: 26.32775 2023-04-02 01:52:33,958 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=89736.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 01:53:06,417 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=89760.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 01:53:10,800 INFO [train.py:903] (1/4) Epoch 14, batch 1000, loss[loss=0.2233, simple_loss=0.3063, pruned_loss=0.07017, over 19544.00 frames. ], tot_loss[loss=0.226, simple_loss=0.3008, pruned_loss=0.07555, over 3766031.78 frames. ], batch size: 56, lr: 6.01e-03, grad_scale: 8.0 2023-04-02 01:53:34,646 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.523e+02 5.300e+02 6.720e+02 8.420e+02 1.635e+03, threshold=1.344e+03, percent-clipped=5.0 2023-04-02 01:53:36,340 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=89785.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 01:53:38,545 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=89787.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 01:53:59,171 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=89801.0, num_to_drop=1, layers_to_drop={1} 2023-04-02 01:54:02,260 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 2023-04-02 01:54:13,790 INFO [train.py:903] (1/4) Epoch 14, batch 1050, loss[loss=0.2426, simple_loss=0.319, pruned_loss=0.08305, over 17991.00 frames. ], tot_loss[loss=0.2264, simple_loss=0.3014, pruned_loss=0.07572, over 3786867.96 frames. ], batch size: 83, lr: 6.01e-03, grad_scale: 8.0 2023-04-02 01:54:46,483 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 2023-04-02 01:55:02,232 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=89851.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 01:55:16,495 INFO [train.py:903] (1/4) Epoch 14, batch 1100, loss[loss=0.2183, simple_loss=0.2941, pruned_loss=0.07127, over 19541.00 frames. ], tot_loss[loss=0.2267, simple_loss=0.3017, pruned_loss=0.07584, over 3803792.26 frames. ], batch size: 56, lr: 6.01e-03, grad_scale: 8.0 2023-04-02 01:55:24,731 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=89871.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 01:55:43,001 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=89883.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 01:55:43,839 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.219e+02 5.450e+02 6.738e+02 9.006e+02 2.173e+03, threshold=1.348e+03, percent-clipped=6.0 2023-04-02 01:56:05,596 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=89902.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 01:56:05,668 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=89902.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 01:56:19,136 INFO [train.py:903] (1/4) Epoch 14, batch 1150, loss[loss=0.2382, simple_loss=0.321, pruned_loss=0.07772, over 19574.00 frames. ], tot_loss[loss=0.2266, simple_loss=0.3016, pruned_loss=0.07577, over 3814786.83 frames. ], batch size: 61, lr: 6.01e-03, grad_scale: 8.0 2023-04-02 01:56:22,882 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=89916.0, num_to_drop=1, layers_to_drop={1} 2023-04-02 01:56:39,062 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=89927.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 01:56:54,410 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.1865, 3.7870, 2.5207, 3.3927, 1.1987, 3.6155, 3.5929, 3.6594], device='cuda:1'), covar=tensor([0.0864, 0.1115, 0.2185, 0.0844, 0.3957, 0.0865, 0.0949, 0.1256], device='cuda:1'), in_proj_covar=tensor([0.0443, 0.0371, 0.0446, 0.0324, 0.0390, 0.0381, 0.0374, 0.0407], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 01:57:25,200 INFO [train.py:903] (1/4) Epoch 14, batch 1200, loss[loss=0.2365, simple_loss=0.3266, pruned_loss=0.07325, over 19339.00 frames. ], tot_loss[loss=0.2264, simple_loss=0.3017, pruned_loss=0.07556, over 3814496.94 frames. ], batch size: 66, lr: 6.01e-03, grad_scale: 8.0 2023-04-02 01:57:49,394 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.071e+02 4.889e+02 5.798e+02 7.146e+02 1.027e+03, threshold=1.160e+03, percent-clipped=0.0 2023-04-02 01:57:51,931 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=89986.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 01:57:55,090 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 2023-04-02 01:58:28,849 INFO [train.py:903] (1/4) Epoch 14, batch 1250, loss[loss=0.22, simple_loss=0.2955, pruned_loss=0.07221, over 17419.00 frames. ], tot_loss[loss=0.2287, simple_loss=0.3036, pruned_loss=0.07694, over 3793059.70 frames. ], batch size: 101, lr: 6.00e-03, grad_scale: 8.0 2023-04-02 01:59:04,877 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=90042.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 01:59:31,266 INFO [train.py:903] (1/4) Epoch 14, batch 1300, loss[loss=0.2311, simple_loss=0.2969, pruned_loss=0.08267, over 19619.00 frames. ], tot_loss[loss=0.229, simple_loss=0.3036, pruned_loss=0.07723, over 3798047.79 frames. ], batch size: 50, lr: 6.00e-03, grad_scale: 8.0 2023-04-02 01:59:57,906 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.456e+02 5.410e+02 6.769e+02 7.813e+02 2.183e+03, threshold=1.354e+03, percent-clipped=2.0 2023-04-02 02:00:25,742 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=90107.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 02:00:33,598 INFO [train.py:903] (1/4) Epoch 14, batch 1350, loss[loss=0.2364, simple_loss=0.3166, pruned_loss=0.07809, over 19594.00 frames. ], tot_loss[loss=0.2285, simple_loss=0.3033, pruned_loss=0.07683, over 3802511.10 frames. ], batch size: 61, lr: 6.00e-03, grad_scale: 8.0 2023-04-02 02:00:44,611 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.51 vs. limit=2.0 2023-04-02 02:00:58,161 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=90132.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 02:01:16,437 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=90148.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 02:01:16,840 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.65 vs. limit=2.0 2023-04-02 02:01:22,827 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 2023-04-02 02:01:28,085 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=90157.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 02:01:29,296 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=90158.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 02:01:35,838 INFO [train.py:903] (1/4) Epoch 14, batch 1400, loss[loss=0.1953, simple_loss=0.2761, pruned_loss=0.05721, over 19499.00 frames. ], tot_loss[loss=0.2299, simple_loss=0.3046, pruned_loss=0.07761, over 3800585.52 frames. ], batch size: 49, lr: 6.00e-03, grad_scale: 8.0 2023-04-02 02:01:39,491 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6334, 1.6000, 1.2419, 1.5247, 1.5898, 1.0488, 0.9826, 1.4737], device='cuda:1'), covar=tensor([0.1093, 0.1352, 0.1706, 0.1106, 0.1359, 0.1414, 0.1916, 0.0999], device='cuda:1'), in_proj_covar=tensor([0.0258, 0.0356, 0.0297, 0.0243, 0.0298, 0.0242, 0.0288, 0.0244], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 02:01:46,545 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=90172.0, num_to_drop=1, layers_to_drop={0} 2023-04-02 02:01:59,344 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=90183.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 02:02:00,107 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.442e+02 5.450e+02 7.234e+02 1.006e+03 2.886e+03, threshold=1.447e+03, percent-clipped=11.0 2023-04-02 02:02:17,603 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=90197.0, num_to_drop=1, layers_to_drop={1} 2023-04-02 02:02:34,936 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_rvb from training. Duration: 25.85 2023-04-02 02:02:38,344 INFO [train.py:903] (1/4) Epoch 14, batch 1450, loss[loss=0.2566, simple_loss=0.3296, pruned_loss=0.09184, over 17398.00 frames. ], tot_loss[loss=0.2291, simple_loss=0.3041, pruned_loss=0.07711, over 3816977.32 frames. ], batch size: 101, lr: 6.00e-03, grad_scale: 8.0 2023-04-02 02:02:53,328 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=90227.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 02:03:12,051 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=90242.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 02:03:22,147 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=90250.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 02:03:38,187 INFO [train.py:903] (1/4) Epoch 14, batch 1500, loss[loss=0.2696, simple_loss=0.3338, pruned_loss=0.1027, over 19664.00 frames. ], tot_loss[loss=0.2286, simple_loss=0.3037, pruned_loss=0.07674, over 3824777.55 frames. ], batch size: 60, lr: 6.00e-03, grad_scale: 8.0 2023-04-02 02:03:42,138 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=90267.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 02:04:03,121 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.947e+02 5.072e+02 6.110e+02 7.921e+02 2.000e+03, threshold=1.222e+03, percent-clipped=2.0 2023-04-02 02:04:09,166 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.2779, 3.8075, 3.8854, 3.8892, 1.5929, 3.6656, 3.1779, 3.6074], device='cuda:1'), covar=tensor([0.1376, 0.0813, 0.0570, 0.0631, 0.4723, 0.0768, 0.0658, 0.1024], device='cuda:1'), in_proj_covar=tensor([0.0700, 0.0624, 0.0827, 0.0710, 0.0746, 0.0576, 0.0499, 0.0764], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1') 2023-04-02 02:04:23,042 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4983, 2.4172, 1.7129, 1.5193, 2.2076, 1.3633, 1.2174, 1.9418], device='cuda:1'), covar=tensor([0.0992, 0.0600, 0.0937, 0.0772, 0.0455, 0.1160, 0.0792, 0.0473], device='cuda:1'), in_proj_covar=tensor([0.0297, 0.0308, 0.0327, 0.0251, 0.0238, 0.0327, 0.0298, 0.0268], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 02:04:39,089 INFO [train.py:903] (1/4) Epoch 14, batch 1550, loss[loss=0.2303, simple_loss=0.3112, pruned_loss=0.07471, over 19745.00 frames. ], tot_loss[loss=0.228, simple_loss=0.303, pruned_loss=0.07655, over 3805180.75 frames. ], batch size: 63, lr: 5.99e-03, grad_scale: 8.0 2023-04-02 02:05:16,351 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=90342.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 02:05:42,813 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([0.8349, 1.3340, 1.0772, 0.9576, 1.2101, 0.9723, 0.9967, 1.2732], device='cuda:1'), covar=tensor([0.0563, 0.0731, 0.1017, 0.0653, 0.0447, 0.1179, 0.0513, 0.0406], device='cuda:1'), in_proj_covar=tensor([0.0298, 0.0308, 0.0327, 0.0250, 0.0239, 0.0327, 0.0297, 0.0268], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 02:05:44,733 INFO [train.py:903] (1/4) Epoch 14, batch 1600, loss[loss=0.2017, simple_loss=0.2702, pruned_loss=0.06663, over 19751.00 frames. ], tot_loss[loss=0.228, simple_loss=0.303, pruned_loss=0.07647, over 3798015.07 frames. ], batch size: 47, lr: 5.99e-03, grad_scale: 8.0 2023-04-02 02:05:51,698 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=90369.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 02:06:02,629 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9_rvb from training. Duration: 30.1555625 2023-04-02 02:06:08,370 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.055e+02 5.288e+02 6.387e+02 7.705e+02 1.564e+03, threshold=1.277e+03, percent-clipped=2.0 2023-04-02 02:06:46,512 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=90413.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 02:06:47,278 INFO [train.py:903] (1/4) Epoch 14, batch 1650, loss[loss=0.2604, simple_loss=0.3348, pruned_loss=0.09297, over 19746.00 frames. ], tot_loss[loss=0.2281, simple_loss=0.3034, pruned_loss=0.07643, over 3804965.22 frames. ], batch size: 63, lr: 5.99e-03, grad_scale: 8.0 2023-04-02 02:07:15,894 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=90438.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 02:07:47,910 INFO [train.py:903] (1/4) Epoch 14, batch 1700, loss[loss=0.2134, simple_loss=0.2889, pruned_loss=0.0689, over 19682.00 frames. ], tot_loss[loss=0.2271, simple_loss=0.3025, pruned_loss=0.07583, over 3810970.00 frames. ], batch size: 53, lr: 5.99e-03, grad_scale: 8.0 2023-04-02 02:07:50,606 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.2028, 2.2662, 2.3978, 3.1354, 2.2991, 2.9173, 2.6846, 2.2368], device='cuda:1'), covar=tensor([0.3888, 0.3305, 0.1609, 0.2010, 0.3602, 0.1698, 0.3649, 0.2878], device='cuda:1'), in_proj_covar=tensor([0.0817, 0.0852, 0.0665, 0.0893, 0.0797, 0.0737, 0.0800, 0.0729], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 02:08:03,287 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=90477.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 02:08:13,895 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.443e+02 5.354e+02 6.441e+02 8.114e+02 1.316e+03, threshold=1.288e+03, percent-clipped=1.0 2023-04-02 02:08:19,178 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 2023-04-02 02:08:23,205 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=90492.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 02:08:25,327 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 2023-04-02 02:08:49,500 INFO [train.py:903] (1/4) Epoch 14, batch 1750, loss[loss=0.2528, simple_loss=0.3336, pruned_loss=0.08602, over 19266.00 frames. ], tot_loss[loss=0.2262, simple_loss=0.3018, pruned_loss=0.07531, over 3810963.46 frames. ], batch size: 66, lr: 5.99e-03, grad_scale: 8.0 2023-04-02 02:08:49,789 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=90514.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 02:09:54,000 INFO [train.py:903] (1/4) Epoch 14, batch 1800, loss[loss=0.2173, simple_loss=0.3043, pruned_loss=0.06512, over 19688.00 frames. ], tot_loss[loss=0.2255, simple_loss=0.3015, pruned_loss=0.07473, over 3802995.36 frames. ], batch size: 59, lr: 5.99e-03, grad_scale: 8.0 2023-04-02 02:10:17,898 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=4.06 vs. limit=5.0 2023-04-02 02:10:18,177 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.749e+02 5.076e+02 6.157e+02 8.007e+02 1.318e+03, threshold=1.231e+03, percent-clipped=1.0 2023-04-02 02:10:29,852 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=90594.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 02:10:34,872 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=90598.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 02:10:47,912 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=90607.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 02:10:48,789 WARNING [train.py:1073] (1/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9_rvb from training. Duration: 25.0944375 2023-04-02 02:10:55,672 INFO [train.py:903] (1/4) Epoch 14, batch 1850, loss[loss=0.3079, simple_loss=0.3641, pruned_loss=0.1258, over 17620.00 frames. ], tot_loss[loss=0.2256, simple_loss=0.3015, pruned_loss=0.07485, over 3792441.58 frames. ], batch size: 101, lr: 5.98e-03, grad_scale: 8.0 2023-04-02 02:11:07,281 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=90623.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 02:11:24,423 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9_rvb from training. Duration: 27.8166875 2023-04-02 02:11:58,537 INFO [train.py:903] (1/4) Epoch 14, batch 1900, loss[loss=0.1767, simple_loss=0.2559, pruned_loss=0.04873, over 19316.00 frames. ], tot_loss[loss=0.2259, simple_loss=0.3017, pruned_loss=0.07508, over 3789361.01 frames. ], batch size: 44, lr: 5.98e-03, grad_scale: 8.0 2023-04-02 02:12:07,977 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.2333, 3.7556, 3.8761, 3.8874, 1.5064, 3.6459, 3.1969, 3.5982], device='cuda:1'), covar=tensor([0.1551, 0.0878, 0.0592, 0.0670, 0.5163, 0.0813, 0.0705, 0.1089], device='cuda:1'), in_proj_covar=tensor([0.0713, 0.0635, 0.0844, 0.0730, 0.0758, 0.0590, 0.0508, 0.0778], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-02 02:12:12,349 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 2023-04-02 02:12:13,967 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4718, 2.3040, 1.6110, 1.5414, 2.1259, 1.2588, 1.2543, 1.8491], device='cuda:1'), covar=tensor([0.0971, 0.0669, 0.0987, 0.0710, 0.0470, 0.1201, 0.0728, 0.0484], device='cuda:1'), in_proj_covar=tensor([0.0301, 0.0309, 0.0329, 0.0252, 0.0241, 0.0328, 0.0297, 0.0269], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 02:12:17,057 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9_rvb from training. Duration: 27.02225 2023-04-02 02:12:18,378 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.7367, 4.2224, 4.4398, 4.4241, 1.6513, 4.1855, 3.6406, 4.1219], device='cuda:1'), covar=tensor([0.1448, 0.0780, 0.0511, 0.0576, 0.5214, 0.0655, 0.0612, 0.1043], device='cuda:1'), in_proj_covar=tensor([0.0713, 0.0634, 0.0842, 0.0729, 0.0756, 0.0590, 0.0507, 0.0777], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-02 02:12:24,249 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.000e+02 5.065e+02 6.349e+02 7.558e+02 1.663e+03, threshold=1.270e+03, percent-clipped=1.0 2023-04-02 02:12:29,235 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([5.0950, 5.1419, 5.9414, 5.8939, 2.1207, 5.5805, 4.7601, 5.5531], device='cuda:1'), covar=tensor([0.1423, 0.0682, 0.0457, 0.0468, 0.5191, 0.0624, 0.0511, 0.1002], device='cuda:1'), in_proj_covar=tensor([0.0713, 0.0633, 0.0842, 0.0729, 0.0756, 0.0590, 0.0507, 0.0778], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-02 02:12:43,821 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 2023-04-02 02:12:48,698 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4084, 1.2763, 1.3368, 1.6896, 1.3716, 1.6160, 1.7220, 1.5262], device='cuda:1'), covar=tensor([0.0873, 0.0989, 0.1070, 0.0811, 0.0840, 0.0774, 0.0781, 0.0732], device='cuda:1'), in_proj_covar=tensor([0.0210, 0.0222, 0.0223, 0.0243, 0.0230, 0.0209, 0.0192, 0.0203], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-02 02:12:54,512 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=90709.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 02:12:59,237 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=90713.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 02:13:00,240 INFO [train.py:903] (1/4) Epoch 14, batch 1950, loss[loss=0.1654, simple_loss=0.2464, pruned_loss=0.04222, over 19751.00 frames. ], tot_loss[loss=0.2246, simple_loss=0.3008, pruned_loss=0.07423, over 3812812.23 frames. ], batch size: 47, lr: 5.98e-03, grad_scale: 8.0 2023-04-02 02:13:40,006 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9568, 1.9584, 1.7955, 1.6939, 1.6021, 1.7963, 0.8529, 1.3852], device='cuda:1'), covar=tensor([0.0403, 0.0470, 0.0290, 0.0488, 0.0715, 0.0536, 0.0918, 0.0672], device='cuda:1'), in_proj_covar=tensor([0.0339, 0.0336, 0.0331, 0.0359, 0.0434, 0.0358, 0.0315, 0.0324], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 02:14:03,110 INFO [train.py:903] (1/4) Epoch 14, batch 2000, loss[loss=0.18, simple_loss=0.266, pruned_loss=0.04693, over 19769.00 frames. ], tot_loss[loss=0.2237, simple_loss=0.2996, pruned_loss=0.07394, over 3809594.40 frames. ], batch size: 49, lr: 5.98e-03, grad_scale: 8.0 2023-04-02 02:14:22,404 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.5745, 1.0963, 1.3702, 1.2045, 2.2078, 0.9731, 1.9400, 2.4198], device='cuda:1'), covar=tensor([0.0631, 0.2681, 0.2614, 0.1622, 0.0825, 0.2045, 0.1038, 0.0450], device='cuda:1'), in_proj_covar=tensor([0.0370, 0.0347, 0.0360, 0.0329, 0.0351, 0.0335, 0.0346, 0.0368], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 02:14:27,794 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.086e+02 5.207e+02 5.999e+02 7.179e+02 1.269e+03, threshold=1.200e+03, percent-clipped=0.0 2023-04-02 02:14:59,970 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_rvb from training. Duration: 26.8349375 2023-04-02 02:15:05,783 INFO [train.py:903] (1/4) Epoch 14, batch 2050, loss[loss=0.2633, simple_loss=0.3281, pruned_loss=0.09925, over 12550.00 frames. ], tot_loss[loss=0.2244, simple_loss=0.3003, pruned_loss=0.07424, over 3812862.09 frames. ], batch size: 136, lr: 5.98e-03, grad_scale: 8.0 2023-04-02 02:15:14,224 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=90821.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 02:15:18,958 WARNING [train.py:1073] (1/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 2023-04-02 02:15:20,135 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 2023-04-02 02:15:24,012 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=90828.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 02:15:43,046 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 2023-04-02 02:16:01,448 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=90858.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 02:16:07,374 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=90863.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 02:16:08,090 INFO [train.py:903] (1/4) Epoch 14, batch 2100, loss[loss=0.1934, simple_loss=0.2678, pruned_loss=0.05952, over 19749.00 frames. ], tot_loss[loss=0.2256, simple_loss=0.3011, pruned_loss=0.07505, over 3814518.90 frames. ], batch size: 47, lr: 5.98e-03, grad_scale: 8.0 2023-04-02 02:16:12,799 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=90867.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 02:16:32,689 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.435e+02 5.132e+02 6.276e+02 7.348e+02 1.970e+03, threshold=1.255e+03, percent-clipped=2.0 2023-04-02 02:16:37,030 WARNING [train.py:1073] (1/4) Exclude cut with ID 453-131332-0000-131866_sp0.9_rvb from training. Duration: 25.3333125 2023-04-02 02:16:38,585 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=90888.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 02:16:46,388 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.5838, 2.2697, 2.3493, 2.9158, 2.5378, 2.3611, 2.1807, 2.7892], device='cuda:1'), covar=tensor([0.0849, 0.1587, 0.1197, 0.0845, 0.1197, 0.0437, 0.1135, 0.0543], device='cuda:1'), in_proj_covar=tensor([0.0258, 0.0354, 0.0296, 0.0240, 0.0295, 0.0240, 0.0288, 0.0242], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 02:16:58,534 WARNING [train.py:1073] (1/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9_rvb from training. Duration: 26.6166875 2023-04-02 02:17:09,966 INFO [train.py:903] (1/4) Epoch 14, batch 2150, loss[loss=0.2318, simple_loss=0.3096, pruned_loss=0.07698, over 19704.00 frames. ], tot_loss[loss=0.225, simple_loss=0.3005, pruned_loss=0.07477, over 3825996.43 frames. ], batch size: 63, lr: 5.97e-03, grad_scale: 8.0 2023-04-02 02:17:38,836 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=90936.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 02:17:53,046 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.74 vs. limit=2.0 2023-04-02 02:18:10,991 INFO [train.py:903] (1/4) Epoch 14, batch 2200, loss[loss=0.1762, simple_loss=0.2578, pruned_loss=0.04736, over 19467.00 frames. ], tot_loss[loss=0.2247, simple_loss=0.3005, pruned_loss=0.07444, over 3838384.52 frames. ], batch size: 49, lr: 5.97e-03, grad_scale: 8.0 2023-04-02 02:18:12,477 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=90965.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 02:18:18,922 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2672, 1.2358, 1.6266, 0.9943, 2.2314, 3.0353, 2.7491, 3.2445], device='cuda:1'), covar=tensor([0.1432, 0.3562, 0.3117, 0.2350, 0.0576, 0.0208, 0.0246, 0.0229], device='cuda:1'), in_proj_covar=tensor([0.0262, 0.0302, 0.0332, 0.0254, 0.0225, 0.0166, 0.0206, 0.0218], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 02:18:23,353 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=90973.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 02:18:35,015 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.77 vs. limit=5.0 2023-04-02 02:18:35,323 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.217e+02 5.132e+02 5.971e+02 7.684e+02 1.888e+03, threshold=1.194e+03, percent-clipped=2.0 2023-04-02 02:18:43,490 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=90990.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 02:19:13,351 INFO [train.py:903] (1/4) Epoch 14, batch 2250, loss[loss=0.3188, simple_loss=0.3587, pruned_loss=0.1394, over 13641.00 frames. ], tot_loss[loss=0.2244, simple_loss=0.3, pruned_loss=0.07436, over 3830068.74 frames. ], batch size: 136, lr: 5.97e-03, grad_scale: 8.0 2023-04-02 02:20:14,195 INFO [train.py:903] (1/4) Epoch 14, batch 2300, loss[loss=0.1886, simple_loss=0.2694, pruned_loss=0.05389, over 19858.00 frames. ], tot_loss[loss=0.2256, simple_loss=0.3005, pruned_loss=0.07529, over 3834378.08 frames. ], batch size: 52, lr: 5.97e-03, grad_scale: 8.0 2023-04-02 02:20:26,811 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_rvb from training. Duration: 26.205 2023-04-02 02:20:38,185 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.297e+02 5.417e+02 6.829e+02 8.730e+02 1.535e+03, threshold=1.366e+03, percent-clipped=12.0 2023-04-02 02:20:38,659 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=91084.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 02:20:47,645 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0181, 1.7175, 1.6902, 1.9907, 1.8594, 1.7988, 1.6099, 2.0181], device='cuda:1'), covar=tensor([0.0927, 0.1432, 0.1331, 0.0897, 0.1118, 0.0480, 0.1210, 0.0592], device='cuda:1'), in_proj_covar=tensor([0.0256, 0.0350, 0.0294, 0.0239, 0.0293, 0.0239, 0.0285, 0.0239], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 02:21:11,038 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=91109.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 02:21:16,445 INFO [train.py:903] (1/4) Epoch 14, batch 2350, loss[loss=0.2035, simple_loss=0.2713, pruned_loss=0.0678, over 19790.00 frames. ], tot_loss[loss=0.2261, simple_loss=0.3012, pruned_loss=0.07551, over 3830079.37 frames. ], batch size: 47, lr: 5.97e-03, grad_scale: 8.0 2023-04-02 02:21:57,694 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_rvb from training. Duration: 25.775 2023-04-02 02:22:13,113 WARNING [train.py:1073] (1/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9_rvb from training. Duration: 25.45 2023-04-02 02:22:14,983 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.69 vs. limit=5.0 2023-04-02 02:22:17,783 INFO [train.py:903] (1/4) Epoch 14, batch 2400, loss[loss=0.1941, simple_loss=0.2829, pruned_loss=0.05268, over 19665.00 frames. ], tot_loss[loss=0.2259, simple_loss=0.3011, pruned_loss=0.07534, over 3809198.04 frames. ], batch size: 53, lr: 5.97e-03, grad_scale: 8.0 2023-04-02 02:22:42,187 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.550e+02 5.466e+02 6.599e+02 8.174e+02 1.804e+03, threshold=1.320e+03, percent-clipped=5.0 2023-04-02 02:22:52,549 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.9518, 4.9605, 5.6819, 5.6522, 1.8329, 5.2994, 4.5387, 5.2547], device='cuda:1'), covar=tensor([0.1310, 0.0853, 0.0539, 0.0544, 0.5719, 0.0646, 0.0596, 0.1290], device='cuda:1'), in_proj_covar=tensor([0.0708, 0.0636, 0.0839, 0.0724, 0.0756, 0.0586, 0.0504, 0.0777], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1') 2023-04-02 02:22:52,702 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=91192.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 02:23:15,500 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=91211.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 02:23:19,600 INFO [train.py:903] (1/4) Epoch 14, batch 2450, loss[loss=0.1905, simple_loss=0.2651, pruned_loss=0.05792, over 18173.00 frames. ], tot_loss[loss=0.2259, simple_loss=0.3011, pruned_loss=0.0753, over 3821975.52 frames. ], batch size: 40, lr: 5.96e-03, grad_scale: 8.0 2023-04-02 02:23:23,106 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=91217.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 02:23:37,735 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=91229.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 02:24:08,197 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=91254.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 02:24:19,329 INFO [train.py:903] (1/4) Epoch 14, batch 2500, loss[loss=0.2013, simple_loss=0.2851, pruned_loss=0.05881, over 19679.00 frames. ], tot_loss[loss=0.2245, simple_loss=0.3001, pruned_loss=0.07442, over 3815869.66 frames. ], batch size: 53, lr: 5.96e-03, grad_scale: 8.0 2023-04-02 02:24:32,853 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5116, 1.4330, 1.3496, 1.8679, 1.4726, 1.7561, 1.9238, 1.6639], device='cuda:1'), covar=tensor([0.0870, 0.0947, 0.1097, 0.0789, 0.0841, 0.0767, 0.0814, 0.0696], device='cuda:1'), in_proj_covar=tensor([0.0211, 0.0223, 0.0224, 0.0241, 0.0228, 0.0209, 0.0191, 0.0202], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-02 02:24:35,618 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.53 vs. limit=2.0 2023-04-02 02:24:39,626 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.3325, 3.9105, 2.5343, 3.5500, 1.0241, 3.8024, 3.7294, 3.8510], device='cuda:1'), covar=tensor([0.0752, 0.1173, 0.2131, 0.0885, 0.4163, 0.0733, 0.0812, 0.1041], device='cuda:1'), in_proj_covar=tensor([0.0452, 0.0380, 0.0454, 0.0330, 0.0392, 0.0386, 0.0378, 0.0413], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 02:24:42,799 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.074e+02 5.867e+02 6.968e+02 8.618e+02 1.617e+03, threshold=1.394e+03, percent-clipped=2.0 2023-04-02 02:24:45,446 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2320, 1.5211, 2.0116, 1.6788, 3.0878, 4.8063, 4.6801, 5.2275], device='cuda:1'), covar=tensor([0.1588, 0.3403, 0.2944, 0.1920, 0.0503, 0.0156, 0.0143, 0.0126], device='cuda:1'), in_proj_covar=tensor([0.0261, 0.0301, 0.0330, 0.0253, 0.0223, 0.0166, 0.0206, 0.0217], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 02:24:58,191 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.2280, 2.0541, 1.9159, 2.3803, 2.6316, 1.8970, 1.8104, 2.3889], device='cuda:1'), covar=tensor([0.1112, 0.1844, 0.1780, 0.1106, 0.1306, 0.0870, 0.1651, 0.0837], device='cuda:1'), in_proj_covar=tensor([0.0257, 0.0352, 0.0294, 0.0240, 0.0294, 0.0240, 0.0288, 0.0241], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 02:25:06,900 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=91303.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 02:25:19,879 INFO [train.py:903] (1/4) Epoch 14, batch 2550, loss[loss=0.2795, simple_loss=0.3423, pruned_loss=0.1084, over 13164.00 frames. ], tot_loss[loss=0.2236, simple_loss=0.2993, pruned_loss=0.07391, over 3825881.81 frames. ], batch size: 136, lr: 5.96e-03, grad_scale: 8.0 2023-04-02 02:25:33,155 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=91326.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 02:26:12,376 WARNING [train.py:1073] (1/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 2023-04-02 02:26:19,133 INFO [train.py:903] (1/4) Epoch 14, batch 2600, loss[loss=0.182, simple_loss=0.2554, pruned_loss=0.05432, over 19760.00 frames. ], tot_loss[loss=0.2231, simple_loss=0.2988, pruned_loss=0.0737, over 3817609.03 frames. ], batch size: 46, lr: 5.96e-03, grad_scale: 8.0 2023-04-02 02:26:44,915 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.041e+02 5.557e+02 7.203e+02 9.059e+02 1.995e+03, threshold=1.441e+03, percent-clipped=7.0 2023-04-02 02:27:21,521 INFO [train.py:903] (1/4) Epoch 14, batch 2650, loss[loss=0.2554, simple_loss=0.3286, pruned_loss=0.09107, over 19312.00 frames. ], tot_loss[loss=0.2238, simple_loss=0.2991, pruned_loss=0.07424, over 3819504.33 frames. ], batch size: 66, lr: 5.96e-03, grad_scale: 8.0 2023-04-02 02:27:41,110 WARNING [train.py:1073] (1/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9_rvb from training. Duration: 27.25 2023-04-02 02:28:11,797 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.7646, 1.3953, 1.6054, 1.5449, 3.2907, 1.1277, 2.1682, 3.7410], device='cuda:1'), covar=tensor([0.0445, 0.2537, 0.2714, 0.1708, 0.0707, 0.2401, 0.1415, 0.0221], device='cuda:1'), in_proj_covar=tensor([0.0371, 0.0346, 0.0363, 0.0327, 0.0355, 0.0338, 0.0347, 0.0369], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 02:28:18,967 INFO [train.py:903] (1/4) Epoch 14, batch 2700, loss[loss=0.1883, simple_loss=0.2634, pruned_loss=0.05659, over 19379.00 frames. ], tot_loss[loss=0.2254, simple_loss=0.3005, pruned_loss=0.07516, over 3820915.65 frames. ], batch size: 47, lr: 5.96e-03, grad_scale: 8.0 2023-04-02 02:28:43,685 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.192e+02 5.145e+02 6.524e+02 8.606e+02 2.089e+03, threshold=1.305e+03, percent-clipped=4.0 2023-04-02 02:29:13,059 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.41 vs. limit=5.0 2023-04-02 02:29:20,022 INFO [train.py:903] (1/4) Epoch 14, batch 2750, loss[loss=0.1856, simple_loss=0.2756, pruned_loss=0.04785, over 19528.00 frames. ], tot_loss[loss=0.2254, simple_loss=0.3008, pruned_loss=0.07501, over 3831413.16 frames. ], batch size: 54, lr: 5.95e-03, grad_scale: 8.0 2023-04-02 02:29:44,615 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.83 vs. limit=2.0 2023-04-02 02:30:14,288 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=91560.0, num_to_drop=1, layers_to_drop={1} 2023-04-02 02:30:18,387 INFO [train.py:903] (1/4) Epoch 14, batch 2800, loss[loss=0.2509, simple_loss=0.3221, pruned_loss=0.08985, over 19531.00 frames. ], tot_loss[loss=0.2251, simple_loss=0.3005, pruned_loss=0.07489, over 3834328.63 frames. ], batch size: 54, lr: 5.95e-03, grad_scale: 8.0 2023-04-02 02:30:41,211 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=91582.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 02:30:44,142 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.583e+02 4.991e+02 6.176e+02 8.428e+02 2.269e+03, threshold=1.235e+03, percent-clipped=6.0 2023-04-02 02:31:10,021 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=91607.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 02:31:19,255 INFO [train.py:903] (1/4) Epoch 14, batch 2850, loss[loss=0.2336, simple_loss=0.3127, pruned_loss=0.07723, over 19527.00 frames. ], tot_loss[loss=0.227, simple_loss=0.3023, pruned_loss=0.07583, over 3822534.08 frames. ], batch size: 56, lr: 5.95e-03, grad_scale: 8.0 2023-04-02 02:31:58,529 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=91647.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 02:32:20,011 INFO [train.py:903] (1/4) Epoch 14, batch 2900, loss[loss=0.1953, simple_loss=0.2768, pruned_loss=0.05688, over 19748.00 frames. ], tot_loss[loss=0.2249, simple_loss=0.3005, pruned_loss=0.07462, over 3822867.38 frames. ], batch size: 51, lr: 5.95e-03, grad_scale: 8.0 2023-04-02 02:32:20,834 WARNING [train.py:1073] (1/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 2023-04-02 02:32:44,072 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.659e+02 5.126e+02 6.310e+02 7.919e+02 2.445e+03, threshold=1.262e+03, percent-clipped=4.0 2023-04-02 02:33:15,675 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 2023-04-02 02:33:19,571 INFO [train.py:903] (1/4) Epoch 14, batch 2950, loss[loss=0.1947, simple_loss=0.2728, pruned_loss=0.05829, over 19758.00 frames. ], tot_loss[loss=0.2247, simple_loss=0.3005, pruned_loss=0.07439, over 3817837.77 frames. ], batch size: 51, lr: 5.95e-03, grad_scale: 8.0 2023-04-02 02:34:17,799 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=91762.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 02:34:19,630 INFO [train.py:903] (1/4) Epoch 14, batch 3000, loss[loss=0.2457, simple_loss=0.3136, pruned_loss=0.08886, over 19659.00 frames. ], tot_loss[loss=0.226, simple_loss=0.3017, pruned_loss=0.07513, over 3826770.68 frames. ], batch size: 55, lr: 5.95e-03, grad_scale: 8.0 2023-04-02 02:34:19,631 INFO [train.py:928] (1/4) Computing validation loss 2023-04-02 02:34:36,637 INFO [train.py:937] (1/4) Epoch 14, validation: loss=0.1742, simple_loss=0.2751, pruned_loss=0.03671, over 944034.00 frames. 2023-04-02 02:34:36,639 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18821MB 2023-04-02 02:34:42,023 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_rvb from training. Duration: 29.735 2023-04-02 02:35:02,708 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.309e+02 5.261e+02 6.370e+02 8.625e+02 1.479e+03, threshold=1.274e+03, percent-clipped=4.0 2023-04-02 02:35:37,796 INFO [train.py:903] (1/4) Epoch 14, batch 3050, loss[loss=0.1992, simple_loss=0.2793, pruned_loss=0.05955, over 19579.00 frames. ], tot_loss[loss=0.2257, simple_loss=0.3015, pruned_loss=0.07494, over 3826890.55 frames. ], batch size: 52, lr: 5.94e-03, grad_scale: 8.0 2023-04-02 02:36:37,007 INFO [train.py:903] (1/4) Epoch 14, batch 3100, loss[loss=0.2598, simple_loss=0.3269, pruned_loss=0.09633, over 19702.00 frames. ], tot_loss[loss=0.2254, simple_loss=0.3011, pruned_loss=0.0749, over 3814894.68 frames. ], batch size: 59, lr: 5.94e-03, grad_scale: 8.0 2023-04-02 02:37:02,383 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.205e+02 5.361e+02 6.622e+02 8.860e+02 2.580e+03, threshold=1.324e+03, percent-clipped=11.0 2023-04-02 02:37:05,889 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1047, 1.8574, 1.4554, 1.2529, 1.5812, 1.1696, 1.2153, 1.6207], device='cuda:1'), covar=tensor([0.0763, 0.0747, 0.1023, 0.0685, 0.0495, 0.1207, 0.0585, 0.0377], device='cuda:1'), in_proj_covar=tensor([0.0293, 0.0306, 0.0325, 0.0249, 0.0238, 0.0325, 0.0293, 0.0264], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 02:37:24,387 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=91904.0, num_to_drop=1, layers_to_drop={1} 2023-04-02 02:37:31,772 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=91909.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 02:37:37,887 INFO [train.py:903] (1/4) Epoch 14, batch 3150, loss[loss=0.2333, simple_loss=0.3057, pruned_loss=0.08039, over 19684.00 frames. ], tot_loss[loss=0.2259, simple_loss=0.3013, pruned_loss=0.07531, over 3811304.39 frames. ], batch size: 59, lr: 5.94e-03, grad_scale: 8.0 2023-04-02 02:38:04,255 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 2023-04-02 02:38:10,545 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4302, 1.4743, 1.7928, 1.6012, 2.5488, 2.2880, 2.6783, 1.1568], device='cuda:1'), covar=tensor([0.2399, 0.4086, 0.2462, 0.1886, 0.1498, 0.2042, 0.1478, 0.4043], device='cuda:1'), in_proj_covar=tensor([0.0505, 0.0591, 0.0637, 0.0451, 0.0602, 0.0504, 0.0651, 0.0507], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 02:38:37,357 INFO [train.py:903] (1/4) Epoch 14, batch 3200, loss[loss=0.2711, simple_loss=0.3408, pruned_loss=0.1007, over 17569.00 frames. ], tot_loss[loss=0.2257, simple_loss=0.3011, pruned_loss=0.07513, over 3812476.18 frames. ], batch size: 101, lr: 5.94e-03, grad_scale: 8.0 2023-04-02 02:38:48,692 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=91973.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 02:39:02,848 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.458e+02 5.150e+02 6.206e+02 7.874e+02 1.849e+03, threshold=1.241e+03, percent-clipped=5.0 2023-04-02 02:39:20,927 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=91999.0, num_to_drop=1, layers_to_drop={0} 2023-04-02 02:39:39,710 INFO [train.py:903] (1/4) Epoch 14, batch 3250, loss[loss=0.2138, simple_loss=0.2932, pruned_loss=0.06723, over 19854.00 frames. ], tot_loss[loss=0.2249, simple_loss=0.3006, pruned_loss=0.07462, over 3814056.45 frames. ], batch size: 52, lr: 5.94e-03, grad_scale: 8.0 2023-04-02 02:39:44,757 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=92018.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 02:39:45,894 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=92019.0, num_to_drop=1, layers_to_drop={0} 2023-04-02 02:40:15,619 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=92043.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 02:40:39,288 INFO [train.py:903] (1/4) Epoch 14, batch 3300, loss[loss=0.2395, simple_loss=0.3142, pruned_loss=0.08243, over 18792.00 frames. ], tot_loss[loss=0.2234, simple_loss=0.2993, pruned_loss=0.07372, over 3823266.80 frames. ], batch size: 74, lr: 5.94e-03, grad_scale: 8.0 2023-04-02 02:40:44,816 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 2023-04-02 02:40:52,239 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 2023-04-02 02:41:04,997 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.031e+02 4.974e+02 6.176e+02 7.406e+02 2.018e+03, threshold=1.235e+03, percent-clipped=5.0 2023-04-02 02:41:41,635 INFO [train.py:903] (1/4) Epoch 14, batch 3350, loss[loss=0.215, simple_loss=0.2962, pruned_loss=0.06689, over 19683.00 frames. ], tot_loss[loss=0.2237, simple_loss=0.2997, pruned_loss=0.0739, over 3820098.67 frames. ], batch size: 53, lr: 5.94e-03, grad_scale: 8.0 2023-04-02 02:42:40,611 INFO [train.py:903] (1/4) Epoch 14, batch 3400, loss[loss=0.2278, simple_loss=0.309, pruned_loss=0.07325, over 19733.00 frames. ], tot_loss[loss=0.224, simple_loss=0.2995, pruned_loss=0.0742, over 3819520.78 frames. ], batch size: 63, lr: 5.93e-03, grad_scale: 8.0 2023-04-02 02:42:46,780 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.1949, 4.2720, 4.7779, 4.7641, 2.7404, 4.4053, 4.0171, 4.4838], device='cuda:1'), covar=tensor([0.1207, 0.2920, 0.0524, 0.0556, 0.4060, 0.0879, 0.0561, 0.0986], device='cuda:1'), in_proj_covar=tensor([0.0717, 0.0647, 0.0857, 0.0738, 0.0763, 0.0594, 0.0517, 0.0782], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-02 02:43:05,852 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.972e+02 4.934e+02 6.017e+02 7.496e+02 1.650e+03, threshold=1.203e+03, percent-clipped=3.0 2023-04-02 02:43:42,226 INFO [train.py:903] (1/4) Epoch 14, batch 3450, loss[loss=0.2229, simple_loss=0.2842, pruned_loss=0.08086, over 19759.00 frames. ], tot_loss[loss=0.2245, simple_loss=0.2999, pruned_loss=0.0745, over 3792335.17 frames. ], batch size: 45, lr: 5.93e-03, grad_scale: 8.0 2023-04-02 02:43:44,677 WARNING [train.py:1073] (1/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9_rvb from training. Duration: 25.2444375 2023-04-02 02:44:14,167 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 2023-04-02 02:44:28,715 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=92253.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 02:44:40,674 INFO [train.py:903] (1/4) Epoch 14, batch 3500, loss[loss=0.2006, simple_loss=0.28, pruned_loss=0.06055, over 19597.00 frames. ], tot_loss[loss=0.2254, simple_loss=0.3005, pruned_loss=0.07516, over 3801570.60 frames. ], batch size: 52, lr: 5.93e-03, grad_scale: 8.0 2023-04-02 02:44:42,973 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0227, 3.4770, 2.0001, 2.0527, 3.0909, 1.6558, 1.4490, 2.0894], device='cuda:1'), covar=tensor([0.1198, 0.0460, 0.0954, 0.0774, 0.0453, 0.1093, 0.0847, 0.0661], device='cuda:1'), in_proj_covar=tensor([0.0300, 0.0311, 0.0331, 0.0254, 0.0241, 0.0331, 0.0299, 0.0270], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 02:44:43,976 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([6.2031, 5.5365, 3.1418, 4.8138, 1.2654, 5.5618, 5.5573, 5.6844], device='cuda:1'), covar=tensor([0.0360, 0.0887, 0.1842, 0.0688, 0.3966, 0.0591, 0.0655, 0.0720], device='cuda:1'), in_proj_covar=tensor([0.0449, 0.0377, 0.0451, 0.0327, 0.0390, 0.0389, 0.0376, 0.0413], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 02:44:46,017 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.3537, 2.0382, 2.2259, 3.0834, 2.3272, 2.6410, 2.7633, 2.5157], device='cuda:1'), covar=tensor([0.0627, 0.0818, 0.0787, 0.0681, 0.0746, 0.0658, 0.0801, 0.0541], device='cuda:1'), in_proj_covar=tensor([0.0207, 0.0221, 0.0221, 0.0240, 0.0225, 0.0206, 0.0189, 0.0199], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1') 2023-04-02 02:44:54,864 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=92275.0, num_to_drop=1, layers_to_drop={1} 2023-04-02 02:45:05,726 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.029e+02 5.239e+02 6.377e+02 7.871e+02 1.606e+03, threshold=1.275e+03, percent-clipped=5.0 2023-04-02 02:45:24,250 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=92300.0, num_to_drop=1, layers_to_drop={1} 2023-04-02 02:45:34,195 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5520, 2.3180, 1.6887, 1.6325, 2.1594, 1.2784, 1.3999, 1.9151], device='cuda:1'), covar=tensor([0.1016, 0.0643, 0.0944, 0.0692, 0.0584, 0.1164, 0.0701, 0.0478], device='cuda:1'), in_proj_covar=tensor([0.0301, 0.0312, 0.0332, 0.0254, 0.0242, 0.0332, 0.0300, 0.0271], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 02:45:41,110 INFO [train.py:903] (1/4) Epoch 14, batch 3550, loss[loss=0.2206, simple_loss=0.2993, pruned_loss=0.07101, over 19477.00 frames. ], tot_loss[loss=0.2245, simple_loss=0.2997, pruned_loss=0.07472, over 3805087.35 frames. ], batch size: 64, lr: 5.93e-03, grad_scale: 8.0 2023-04-02 02:45:44,521 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=92317.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 02:46:15,216 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=92343.0, num_to_drop=1, layers_to_drop={0} 2023-04-02 02:46:39,900 INFO [train.py:903] (1/4) Epoch 14, batch 3600, loss[loss=0.1948, simple_loss=0.2841, pruned_loss=0.05272, over 19780.00 frames. ], tot_loss[loss=0.2255, simple_loss=0.3009, pruned_loss=0.07505, over 3801449.57 frames. ], batch size: 56, lr: 5.93e-03, grad_scale: 8.0 2023-04-02 02:46:44,958 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=92368.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 02:46:55,932 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.9728, 3.6084, 2.3129, 3.2804, 1.0504, 3.4675, 3.3863, 3.5079], device='cuda:1'), covar=tensor([0.0789, 0.1118, 0.2124, 0.0813, 0.3602, 0.0851, 0.0791, 0.1066], device='cuda:1'), in_proj_covar=tensor([0.0449, 0.0378, 0.0453, 0.0329, 0.0392, 0.0391, 0.0378, 0.0413], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 02:47:04,638 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.193e+02 5.209e+02 6.321e+02 7.842e+02 1.520e+03, threshold=1.264e+03, percent-clipped=2.0 2023-04-02 02:47:40,754 INFO [train.py:903] (1/4) Epoch 14, batch 3650, loss[loss=0.1943, simple_loss=0.2738, pruned_loss=0.0574, over 19737.00 frames. ], tot_loss[loss=0.2246, simple_loss=0.3005, pruned_loss=0.07429, over 3819898.01 frames. ], batch size: 51, lr: 5.93e-03, grad_scale: 8.0 2023-04-02 02:48:02,178 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=92432.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 02:48:34,077 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=92458.0, num_to_drop=1, layers_to_drop={0} 2023-04-02 02:48:40,399 INFO [train.py:903] (1/4) Epoch 14, batch 3700, loss[loss=0.2427, simple_loss=0.3218, pruned_loss=0.08175, over 19789.00 frames. ], tot_loss[loss=0.2253, simple_loss=0.301, pruned_loss=0.0748, over 3812735.77 frames. ], batch size: 56, lr: 5.92e-03, grad_scale: 8.0 2023-04-02 02:48:51,194 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 2023-04-02 02:49:05,824 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.433e+02 4.888e+02 6.023e+02 8.004e+02 1.682e+03, threshold=1.205e+03, percent-clipped=3.0 2023-04-02 02:49:10,965 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.7645, 1.5991, 1.5465, 2.3066, 1.6921, 2.0302, 2.1072, 1.8612], device='cuda:1'), covar=tensor([0.0776, 0.0910, 0.1029, 0.0757, 0.0859, 0.0702, 0.0867, 0.0650], device='cuda:1'), in_proj_covar=tensor([0.0208, 0.0222, 0.0222, 0.0242, 0.0226, 0.0207, 0.0191, 0.0200], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1') 2023-04-02 02:49:11,217 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 2023-04-02 02:49:13,121 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.8495, 1.5949, 1.8160, 1.5723, 4.3002, 1.0206, 2.3412, 4.6906], device='cuda:1'), covar=tensor([0.0340, 0.2524, 0.2688, 0.1947, 0.0749, 0.2658, 0.1490, 0.0166], device='cuda:1'), in_proj_covar=tensor([0.0370, 0.0345, 0.0362, 0.0326, 0.0356, 0.0336, 0.0345, 0.0369], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 02:49:41,646 INFO [train.py:903] (1/4) Epoch 14, batch 3750, loss[loss=0.2149, simple_loss=0.2937, pruned_loss=0.06804, over 19784.00 frames. ], tot_loss[loss=0.2246, simple_loss=0.3004, pruned_loss=0.07439, over 3808300.72 frames. ], batch size: 63, lr: 5.92e-03, grad_scale: 8.0 2023-04-02 02:50:42,053 INFO [train.py:903] (1/4) Epoch 14, batch 3800, loss[loss=0.1902, simple_loss=0.2775, pruned_loss=0.05149, over 19597.00 frames. ], tot_loss[loss=0.2234, simple_loss=0.2994, pruned_loss=0.07368, over 3822600.07 frames. ], batch size: 52, lr: 5.92e-03, grad_scale: 8.0 2023-04-02 02:51:06,367 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.816e+02 4.992e+02 6.384e+02 8.353e+02 1.667e+03, threshold=1.277e+03, percent-clipped=5.0 2023-04-02 02:51:11,002 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9_rvb from training. Duration: 29.1166875 2023-04-02 02:51:40,637 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.85 vs. limit=2.0 2023-04-02 02:51:42,072 INFO [train.py:903] (1/4) Epoch 14, batch 3850, loss[loss=0.2159, simple_loss=0.282, pruned_loss=0.07495, over 19762.00 frames. ], tot_loss[loss=0.2232, simple_loss=0.299, pruned_loss=0.07367, over 3818639.72 frames. ], batch size: 48, lr: 5.92e-03, grad_scale: 8.0 2023-04-02 02:51:54,736 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=92624.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 02:52:19,031 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=92643.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 02:52:26,085 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=92649.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 02:52:43,617 INFO [train.py:903] (1/4) Epoch 14, batch 3900, loss[loss=0.1909, simple_loss=0.2638, pruned_loss=0.05904, over 19382.00 frames. ], tot_loss[loss=0.2248, simple_loss=0.3002, pruned_loss=0.07467, over 3804821.18 frames. ], batch size: 47, lr: 5.92e-03, grad_scale: 4.0 2023-04-02 02:53:10,396 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.910e+02 5.705e+02 7.277e+02 9.203e+02 2.913e+03, threshold=1.455e+03, percent-clipped=9.0 2023-04-02 02:53:13,202 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=92688.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 02:53:43,655 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=92713.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 02:53:44,354 INFO [train.py:903] (1/4) Epoch 14, batch 3950, loss[loss=0.2114, simple_loss=0.2956, pruned_loss=0.06366, over 19719.00 frames. ], tot_loss[loss=0.2248, simple_loss=0.3006, pruned_loss=0.07451, over 3799023.09 frames. ], batch size: 63, lr: 5.92e-03, grad_scale: 4.0 2023-04-02 02:53:44,787 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=92714.0, num_to_drop=1, layers_to_drop={1} 2023-04-02 02:53:48,548 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 2023-04-02 02:54:12,979 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=92738.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 02:54:14,262 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=92739.0, num_to_drop=1, layers_to_drop={0} 2023-04-02 02:54:45,453 INFO [train.py:903] (1/4) Epoch 14, batch 4000, loss[loss=0.1884, simple_loss=0.2729, pruned_loss=0.05192, over 19673.00 frames. ], tot_loss[loss=0.2241, simple_loss=0.3002, pruned_loss=0.07402, over 3803325.13 frames. ], batch size: 53, lr: 5.91e-03, grad_scale: 8.0 2023-04-02 02:55:11,213 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.263e+02 4.983e+02 6.436e+02 8.255e+02 1.908e+03, threshold=1.287e+03, percent-clipped=2.0 2023-04-02 02:55:33,050 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1_rvb from training. Duration: 0.7545625 2023-04-02 02:55:45,562 INFO [train.py:903] (1/4) Epoch 14, batch 4050, loss[loss=0.1775, simple_loss=0.2595, pruned_loss=0.0477, over 19735.00 frames. ], tot_loss[loss=0.2228, simple_loss=0.299, pruned_loss=0.07326, over 3823544.78 frames. ], batch size: 51, lr: 5.91e-03, grad_scale: 4.0 2023-04-02 02:55:52,887 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0907, 1.7028, 1.5857, 1.9348, 1.6602, 1.7329, 1.4801, 1.9691], device='cuda:1'), covar=tensor([0.0910, 0.1381, 0.1449, 0.1027, 0.1325, 0.0512, 0.1367, 0.0691], device='cuda:1'), in_proj_covar=tensor([0.0259, 0.0349, 0.0294, 0.0241, 0.0297, 0.0243, 0.0287, 0.0241], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 02:56:28,995 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=92849.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 02:56:45,576 INFO [train.py:903] (1/4) Epoch 14, batch 4100, loss[loss=0.2441, simple_loss=0.3175, pruned_loss=0.08537, over 19589.00 frames. ], tot_loss[loss=0.2218, simple_loss=0.298, pruned_loss=0.07281, over 3828120.83 frames. ], batch size: 61, lr: 5.91e-03, grad_scale: 4.0 2023-04-02 02:57:12,587 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 2023-04-02 02:57:13,886 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.303e+02 5.414e+02 6.604e+02 8.302e+02 1.654e+03, threshold=1.321e+03, percent-clipped=7.0 2023-04-02 02:57:21,834 WARNING [train.py:1073] (1/4) Exclude cut with ID 497-129325-0061-9566_sp1.1_rvb from training. Duration: 0.97725 2023-04-02 02:57:45,521 INFO [train.py:903] (1/4) Epoch 14, batch 4150, loss[loss=0.2191, simple_loss=0.3021, pruned_loss=0.06804, over 19357.00 frames. ], tot_loss[loss=0.2237, simple_loss=0.2998, pruned_loss=0.07378, over 3830065.76 frames. ], batch size: 66, lr: 5.91e-03, grad_scale: 4.0 2023-04-02 02:57:51,032 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=92917.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 02:58:19,981 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4850, 1.5526, 1.8695, 1.6753, 2.8362, 2.3256, 2.8984, 1.3320], device='cuda:1'), covar=tensor([0.2182, 0.3650, 0.2244, 0.1727, 0.1354, 0.1868, 0.1335, 0.3710], device='cuda:1'), in_proj_covar=tensor([0.0499, 0.0588, 0.0633, 0.0446, 0.0599, 0.0500, 0.0645, 0.0503], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 02:58:28,821 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=92949.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 02:58:48,464 INFO [train.py:903] (1/4) Epoch 14, batch 4200, loss[loss=0.2194, simple_loss=0.3112, pruned_loss=0.06384, over 19691.00 frames. ], tot_loss[loss=0.2232, simple_loss=0.2994, pruned_loss=0.07355, over 3818074.14 frames. ], batch size: 59, lr: 5.91e-03, grad_scale: 4.0 2023-04-02 02:58:51,708 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 2023-04-02 02:59:15,442 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.424e+02 5.036e+02 6.286e+02 7.807e+02 1.684e+03, threshold=1.257e+03, percent-clipped=4.0 2023-04-02 02:59:15,614 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=92987.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 02:59:22,673 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=92993.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 02:59:48,127 INFO [train.py:903] (1/4) Epoch 14, batch 4250, loss[loss=0.1911, simple_loss=0.2731, pruned_loss=0.05454, over 19402.00 frames. ], tot_loss[loss=0.2241, simple_loss=0.2997, pruned_loss=0.07425, over 3817760.81 frames. ], batch size: 48, lr: 5.91e-03, grad_scale: 4.0 2023-04-02 03:00:03,115 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9_rvb from training. Duration: 29.816625 2023-04-02 03:00:05,133 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=93028.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 03:00:15,028 WARNING [train.py:1073] (1/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 2023-04-02 03:00:37,782 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=93054.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 03:00:47,386 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.98 vs. limit=5.0 2023-04-02 03:00:48,956 INFO [train.py:903] (1/4) Epoch 14, batch 4300, loss[loss=0.2013, simple_loss=0.2717, pruned_loss=0.06539, over 19759.00 frames. ], tot_loss[loss=0.2226, simple_loss=0.2983, pruned_loss=0.07339, over 3833207.74 frames. ], batch size: 47, lr: 5.90e-03, grad_scale: 4.0 2023-04-02 03:01:12,680 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=93082.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 03:01:18,339 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.279e+02 5.152e+02 6.308e+02 8.497e+02 2.668e+03, threshold=1.262e+03, percent-clipped=7.0 2023-04-02 03:01:18,720 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.7440, 1.4847, 1.2744, 1.6225, 1.5095, 1.3087, 1.1592, 1.5523], device='cuda:1'), covar=tensor([0.1048, 0.1305, 0.1662, 0.1041, 0.1247, 0.0779, 0.1646, 0.0836], device='cuda:1'), in_proj_covar=tensor([0.0258, 0.0349, 0.0294, 0.0241, 0.0295, 0.0244, 0.0287, 0.0242], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 03:01:36,162 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=93102.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 03:01:42,508 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 2023-04-02 03:01:50,402 INFO [train.py:903] (1/4) Epoch 14, batch 4350, loss[loss=0.1868, simple_loss=0.269, pruned_loss=0.05229, over 19743.00 frames. ], tot_loss[loss=0.2219, simple_loss=0.2977, pruned_loss=0.07307, over 3824050.40 frames. ], batch size: 51, lr: 5.90e-03, grad_scale: 4.0 2023-04-02 03:02:14,541 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=93133.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 03:02:52,476 INFO [train.py:903] (1/4) Epoch 14, batch 4400, loss[loss=0.2471, simple_loss=0.3104, pruned_loss=0.09185, over 19713.00 frames. ], tot_loss[loss=0.2236, simple_loss=0.2991, pruned_loss=0.07409, over 3820935.31 frames. ], batch size: 51, lr: 5.90e-03, grad_scale: 8.0 2023-04-02 03:03:15,301 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_rvb from training. Duration: 25.285 2023-04-02 03:03:18,717 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.086e+02 5.134e+02 6.154e+02 7.916e+02 2.681e+03, threshold=1.231e+03, percent-clipped=4.0 2023-04-02 03:03:25,269 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 2023-04-02 03:03:26,640 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=93193.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 03:03:31,888 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=93197.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 03:03:47,756 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.3767, 2.2035, 2.0149, 1.8985, 1.6769, 1.9037, 0.4909, 1.3458], device='cuda:1'), covar=tensor([0.0441, 0.0482, 0.0380, 0.0646, 0.0966, 0.0745, 0.1048, 0.0795], device='cuda:1'), in_proj_covar=tensor([0.0341, 0.0337, 0.0333, 0.0362, 0.0437, 0.0362, 0.0315, 0.0323], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 03:03:52,930 INFO [train.py:903] (1/4) Epoch 14, batch 4450, loss[loss=0.2382, simple_loss=0.3096, pruned_loss=0.08338, over 19842.00 frames. ], tot_loss[loss=0.2229, simple_loss=0.2985, pruned_loss=0.07362, over 3823525.94 frames. ], batch size: 52, lr: 5.90e-03, grad_scale: 8.0 2023-04-02 03:04:49,432 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=93261.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 03:04:52,743 INFO [train.py:903] (1/4) Epoch 14, batch 4500, loss[loss=0.1775, simple_loss=0.2609, pruned_loss=0.04701, over 19397.00 frames. ], tot_loss[loss=0.2229, simple_loss=0.2984, pruned_loss=0.07365, over 3827490.75 frames. ], batch size: 48, lr: 5.90e-03, grad_scale: 8.0 2023-04-02 03:05:21,651 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.340e+02 5.314e+02 6.344e+02 7.896e+02 1.749e+03, threshold=1.269e+03, percent-clipped=5.0 2023-04-02 03:05:25,895 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 2023-04-02 03:05:28,487 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=93293.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 03:05:45,694 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=93308.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 03:05:52,289 INFO [train.py:903] (1/4) Epoch 14, batch 4550, loss[loss=0.2557, simple_loss=0.3287, pruned_loss=0.09136, over 19509.00 frames. ], tot_loss[loss=0.2222, simple_loss=0.298, pruned_loss=0.07316, over 3819140.50 frames. ], batch size: 64, lr: 5.90e-03, grad_scale: 8.0 2023-04-02 03:06:06,151 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9_rvb from training. Duration: 28.72225 2023-04-02 03:06:21,890 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=93337.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 03:06:27,650 WARNING [train.py:1073] (1/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 2023-04-02 03:06:39,765 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.43 vs. limit=5.0 2023-04-02 03:06:46,896 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=93358.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 03:06:55,967 INFO [train.py:903] (1/4) Epoch 14, batch 4600, loss[loss=0.2069, simple_loss=0.2934, pruned_loss=0.06021, over 19793.00 frames. ], tot_loss[loss=0.2215, simple_loss=0.2975, pruned_loss=0.07279, over 3821435.65 frames. ], batch size: 56, lr: 5.90e-03, grad_scale: 8.0 2023-04-02 03:07:05,026 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=93372.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 03:07:09,477 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=93376.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 03:07:17,775 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=93383.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 03:07:22,006 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.600e+02 5.339e+02 6.388e+02 8.227e+02 2.509e+03, threshold=1.278e+03, percent-clipped=2.0 2023-04-02 03:07:35,566 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=93398.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 03:07:49,443 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=93408.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 03:07:56,013 INFO [train.py:903] (1/4) Epoch 14, batch 4650, loss[loss=0.2931, simple_loss=0.3543, pruned_loss=0.1159, over 13553.00 frames. ], tot_loss[loss=0.2219, simple_loss=0.2978, pruned_loss=0.07298, over 3813779.92 frames. ], batch size: 136, lr: 5.89e-03, grad_scale: 8.0 2023-04-02 03:07:56,725 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 2023-04-02 03:08:12,038 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_rvb from training. Duration: 0.92 2023-04-02 03:08:23,193 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_rvb from training. Duration: 0.83 2023-04-02 03:08:43,050 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=93452.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 03:08:44,284 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=93453.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 03:08:45,671 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.75 vs. limit=5.0 2023-04-02 03:08:56,300 INFO [train.py:903] (1/4) Epoch 14, batch 4700, loss[loss=0.1988, simple_loss=0.2817, pruned_loss=0.05795, over 19598.00 frames. ], tot_loss[loss=0.222, simple_loss=0.2982, pruned_loss=0.07295, over 3825979.98 frames. ], batch size: 52, lr: 5.89e-03, grad_scale: 8.0 2023-04-02 03:09:11,832 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=93477.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 03:09:13,376 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=93478.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 03:09:20,327 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9_rvb from training. Duration: 0.92225 2023-04-02 03:09:25,716 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.143e+02 5.199e+02 6.337e+02 7.857e+02 1.524e+03, threshold=1.267e+03, percent-clipped=2.0 2023-04-02 03:09:26,075 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=93487.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 03:09:55,156 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=93513.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 03:09:55,940 INFO [train.py:903] (1/4) Epoch 14, batch 4750, loss[loss=0.1788, simple_loss=0.2542, pruned_loss=0.05168, over 19328.00 frames. ], tot_loss[loss=0.222, simple_loss=0.2979, pruned_loss=0.07304, over 3825410.55 frames. ], batch size: 44, lr: 5.89e-03, grad_scale: 8.0 2023-04-02 03:10:55,740 INFO [train.py:903] (1/4) Epoch 14, batch 4800, loss[loss=0.2293, simple_loss=0.3069, pruned_loss=0.07585, over 19384.00 frames. ], tot_loss[loss=0.2227, simple_loss=0.2986, pruned_loss=0.07344, over 3825692.28 frames. ], batch size: 70, lr: 5.89e-03, grad_scale: 8.0 2023-04-02 03:10:56,149 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=93564.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 03:11:18,610 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6838, 1.8746, 2.2407, 1.9512, 2.9632, 3.2656, 3.2349, 3.4916], device='cuda:1'), covar=tensor([0.1420, 0.2897, 0.2439, 0.1971, 0.0901, 0.0379, 0.0190, 0.0273], device='cuda:1'), in_proj_covar=tensor([0.0261, 0.0301, 0.0330, 0.0254, 0.0222, 0.0164, 0.0207, 0.0216], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 03:11:22,945 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.531e+02 5.541e+02 6.642e+02 8.296e+02 2.320e+03, threshold=1.328e+03, percent-clipped=4.0 2023-04-02 03:11:25,723 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=93589.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 03:11:29,076 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=93592.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 03:11:39,593 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 2023-04-02 03:11:43,676 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([5.1680, 5.2138, 6.1189, 6.0139, 1.9138, 5.7499, 5.0110, 5.7046], device='cuda:1'), covar=tensor([0.1512, 0.0650, 0.0480, 0.0531, 0.5896, 0.0490, 0.0470, 0.1037], device='cuda:1'), in_proj_covar=tensor([0.0710, 0.0644, 0.0851, 0.0729, 0.0757, 0.0594, 0.0510, 0.0776], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-02 03:11:57,047 INFO [train.py:903] (1/4) Epoch 14, batch 4850, loss[loss=0.2435, simple_loss=0.3074, pruned_loss=0.0898, over 19842.00 frames. ], tot_loss[loss=0.224, simple_loss=0.2997, pruned_loss=0.07409, over 3818370.98 frames. ], batch size: 52, lr: 5.89e-03, grad_scale: 8.0 2023-04-02 03:12:17,835 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=93632.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 03:12:23,245 WARNING [train.py:1073] (1/4) Exclude cut with ID 774-127930-0014-48411_sp1.1_rvb from training. Duration: 0.95 2023-04-02 03:12:42,754 WARNING [train.py:1073] (1/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1_rvb from training. Duration: 0.9409375 2023-04-02 03:12:47,284 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 2023-04-02 03:12:48,583 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 2023-04-02 03:12:48,962 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=93657.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 03:12:56,528 INFO [train.py:903] (1/4) Epoch 14, batch 4900, loss[loss=0.2406, simple_loss=0.3163, pruned_loss=0.08246, over 19602.00 frames. ], tot_loss[loss=0.2243, simple_loss=0.3002, pruned_loss=0.07426, over 3802695.94 frames. ], batch size: 57, lr: 5.89e-03, grad_scale: 8.0 2023-04-02 03:12:56,938 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=93664.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 03:12:57,707 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1_rvb from training. Duration: 25.3818125 2023-04-02 03:13:18,090 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 2023-04-02 03:13:25,555 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.007e+02 4.842e+02 5.818e+02 7.123e+02 1.786e+03, threshold=1.164e+03, percent-clipped=2.0 2023-04-02 03:13:26,357 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 2023-04-02 03:13:26,360 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.59 vs. limit=2.0 2023-04-02 03:13:28,151 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=93689.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 03:13:33,914 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.7131, 1.8349, 1.6480, 2.7201, 1.9802, 2.5372, 1.8996, 1.5499], device='cuda:1'), covar=tensor([0.4667, 0.4150, 0.2479, 0.2514, 0.4068, 0.1939, 0.5382, 0.4388], device='cuda:1'), in_proj_covar=tensor([0.0824, 0.0861, 0.0669, 0.0901, 0.0810, 0.0745, 0.0806, 0.0733], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 03:13:49,664 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=93708.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 03:13:55,991 INFO [train.py:903] (1/4) Epoch 14, batch 4950, loss[loss=0.2377, simple_loss=0.3134, pruned_loss=0.08099, over 19749.00 frames. ], tot_loss[loss=0.2234, simple_loss=0.2997, pruned_loss=0.07355, over 3808116.33 frames. ], batch size: 63, lr: 5.88e-03, grad_scale: 8.0 2023-04-02 03:14:16,294 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 2023-04-02 03:14:21,090 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=93733.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 03:14:32,410 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=93743.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 03:14:36,689 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 2023-04-02 03:14:58,213 INFO [train.py:903] (1/4) Epoch 14, batch 5000, loss[loss=0.253, simple_loss=0.3339, pruned_loss=0.08609, over 18599.00 frames. ], tot_loss[loss=0.2233, simple_loss=0.2995, pruned_loss=0.07358, over 3824084.33 frames. ], batch size: 74, lr: 5.88e-03, grad_scale: 8.0 2023-04-02 03:15:04,082 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=93768.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 03:15:05,337 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=93769.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 03:15:07,254 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_rvb from training. Duration: 27.14 2023-04-02 03:15:17,358 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 2023-04-02 03:15:18,126 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.98 vs. limit=5.0 2023-04-02 03:15:25,225 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.057e+02 5.347e+02 6.962e+02 9.103e+02 2.417e+03, threshold=1.392e+03, percent-clipped=9.0 2023-04-02 03:15:26,646 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=93788.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 03:15:33,775 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=93794.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 03:15:56,382 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3072, 1.9207, 1.5009, 1.3704, 1.8456, 1.2508, 1.3372, 1.7269], device='cuda:1'), covar=tensor([0.0726, 0.0626, 0.0765, 0.0622, 0.0399, 0.0920, 0.0486, 0.0358], device='cuda:1'), in_proj_covar=tensor([0.0294, 0.0303, 0.0324, 0.0246, 0.0236, 0.0323, 0.0288, 0.0263], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 03:15:59,315 INFO [train.py:903] (1/4) Epoch 14, batch 5050, loss[loss=0.2362, simple_loss=0.3135, pruned_loss=0.07949, over 19732.00 frames. ], tot_loss[loss=0.2219, simple_loss=0.2983, pruned_loss=0.07274, over 3831835.33 frames. ], batch size: 63, lr: 5.88e-03, grad_scale: 8.0 2023-04-02 03:16:35,021 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9_rvb from training. Duration: 26.62775 2023-04-02 03:16:40,863 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=93848.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 03:16:59,046 INFO [train.py:903] (1/4) Epoch 14, batch 5100, loss[loss=0.2021, simple_loss=0.2738, pruned_loss=0.06525, over 15571.00 frames. ], tot_loss[loss=0.2216, simple_loss=0.2979, pruned_loss=0.07267, over 3824840.22 frames. ], batch size: 34, lr: 5.88e-03, grad_scale: 8.0 2023-04-02 03:17:09,141 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=93873.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 03:17:09,903 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 2023-04-02 03:17:13,175 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_rvb from training. Duration: 25.035 2023-04-02 03:17:16,699 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_rvb from training. Duration: 27.92 2023-04-02 03:17:26,263 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.008e+02 4.904e+02 5.934e+02 7.645e+02 1.361e+03, threshold=1.187e+03, percent-clipped=0.0 2023-04-02 03:17:56,876 INFO [train.py:903] (1/4) Epoch 14, batch 5150, loss[loss=0.2222, simple_loss=0.3018, pruned_loss=0.07125, over 18311.00 frames. ], tot_loss[loss=0.2227, simple_loss=0.2989, pruned_loss=0.07331, over 3820349.33 frames. ], batch size: 84, lr: 5.88e-03, grad_scale: 8.0 2023-04-02 03:18:09,225 WARNING [train.py:1073] (1/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 2023-04-02 03:18:43,207 WARNING [train.py:1073] (1/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1_rvb from training. Duration: 0.9681875 2023-04-02 03:18:58,008 INFO [train.py:903] (1/4) Epoch 14, batch 5200, loss[loss=0.291, simple_loss=0.3493, pruned_loss=0.1164, over 18797.00 frames. ], tot_loss[loss=0.2224, simple_loss=0.2986, pruned_loss=0.07311, over 3828905.46 frames. ], batch size: 74, lr: 5.88e-03, grad_scale: 8.0 2023-04-02 03:19:13,843 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 2023-04-02 03:19:25,470 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.052e+02 5.268e+02 6.485e+02 8.631e+02 2.638e+03, threshold=1.297e+03, percent-clipped=7.0 2023-04-02 03:19:39,236 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=93999.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 03:19:57,305 WARNING [train.py:1073] (1/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 2023-04-02 03:19:59,390 INFO [train.py:903] (1/4) Epoch 14, batch 5250, loss[loss=0.2381, simple_loss=0.318, pruned_loss=0.07913, over 19850.00 frames. ], tot_loss[loss=0.2215, simple_loss=0.2979, pruned_loss=0.07251, over 3824014.10 frames. ], batch size: 52, lr: 5.88e-03, grad_scale: 8.0 2023-04-02 03:20:59,247 INFO [train.py:903] (1/4) Epoch 14, batch 5300, loss[loss=0.2328, simple_loss=0.3118, pruned_loss=0.07695, over 19598.00 frames. ], tot_loss[loss=0.2209, simple_loss=0.2971, pruned_loss=0.07234, over 3822526.35 frames. ], batch size: 61, lr: 5.87e-03, grad_scale: 8.0 2023-04-02 03:21:16,449 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 2023-04-02 03:21:27,961 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.389e+02 5.368e+02 7.020e+02 9.283e+02 2.840e+03, threshold=1.404e+03, percent-clipped=4.0 2023-04-02 03:21:58,997 INFO [train.py:903] (1/4) Epoch 14, batch 5350, loss[loss=0.1817, simple_loss=0.2707, pruned_loss=0.04639, over 19765.00 frames. ], tot_loss[loss=0.2216, simple_loss=0.2976, pruned_loss=0.07282, over 3816174.51 frames. ], batch size: 54, lr: 5.87e-03, grad_scale: 8.0 2023-04-02 03:22:23,143 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=94132.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 03:22:34,750 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9_rvb from training. Duration: 26.438875 2023-04-02 03:23:01,401 INFO [train.py:903] (1/4) Epoch 14, batch 5400, loss[loss=0.1844, simple_loss=0.255, pruned_loss=0.05693, over 19736.00 frames. ], tot_loss[loss=0.2221, simple_loss=0.2984, pruned_loss=0.07288, over 3809818.33 frames. ], batch size: 45, lr: 5.87e-03, grad_scale: 8.0 2023-04-02 03:23:29,187 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.364e+02 5.537e+02 7.248e+02 8.700e+02 2.021e+03, threshold=1.450e+03, percent-clipped=3.0 2023-04-02 03:24:03,213 INFO [train.py:903] (1/4) Epoch 14, batch 5450, loss[loss=0.2062, simple_loss=0.2829, pruned_loss=0.06478, over 19846.00 frames. ], tot_loss[loss=0.2223, simple_loss=0.2988, pruned_loss=0.07291, over 3812335.63 frames. ], batch size: 52, lr: 5.87e-03, grad_scale: 8.0 2023-04-02 03:24:33,948 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=94241.0, num_to_drop=1, layers_to_drop={0} 2023-04-02 03:24:43,539 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=94247.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 03:24:54,139 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2136, 1.1930, 1.4393, 1.3133, 2.3226, 2.0089, 2.4527, 1.0876], device='cuda:1'), covar=tensor([0.2432, 0.4210, 0.2541, 0.2062, 0.1562, 0.2104, 0.1460, 0.3948], device='cuda:1'), in_proj_covar=tensor([0.0502, 0.0591, 0.0639, 0.0450, 0.0604, 0.0505, 0.0649, 0.0510], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 03:25:02,870 INFO [train.py:903] (1/4) Epoch 14, batch 5500, loss[loss=0.2589, simple_loss=0.3281, pruned_loss=0.0949, over 17548.00 frames. ], tot_loss[loss=0.2229, simple_loss=0.2994, pruned_loss=0.07318, over 3812054.87 frames. ], batch size: 101, lr: 5.87e-03, grad_scale: 8.0 2023-04-02 03:25:24,846 WARNING [train.py:1073] (1/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9_rvb from training. Duration: 27.47775 2023-04-02 03:25:30,870 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.158e+02 4.805e+02 5.794e+02 7.462e+02 1.465e+03, threshold=1.159e+03, percent-clipped=1.0 2023-04-02 03:25:46,411 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.2094, 2.9276, 2.0126, 2.1228, 1.9505, 2.3843, 0.7510, 2.0718], device='cuda:1'), covar=tensor([0.0579, 0.0517, 0.0668, 0.1034, 0.1011, 0.0895, 0.1204, 0.0900], device='cuda:1'), in_proj_covar=tensor([0.0344, 0.0342, 0.0336, 0.0363, 0.0437, 0.0363, 0.0316, 0.0328], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 03:26:01,458 INFO [train.py:903] (1/4) Epoch 14, batch 5550, loss[loss=0.2326, simple_loss=0.31, pruned_loss=0.07757, over 18692.00 frames. ], tot_loss[loss=0.2238, simple_loss=0.3, pruned_loss=0.07386, over 3800093.06 frames. ], batch size: 74, lr: 5.87e-03, grad_scale: 8.0 2023-04-02 03:26:08,355 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 2023-04-02 03:26:23,616 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.14 vs. limit=5.0 2023-04-02 03:26:37,729 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=94343.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 03:26:53,084 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 2023-04-02 03:26:57,912 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 2023-04-02 03:27:01,443 INFO [train.py:903] (1/4) Epoch 14, batch 5600, loss[loss=0.2088, simple_loss=0.2892, pruned_loss=0.06413, over 19667.00 frames. ], tot_loss[loss=0.2242, simple_loss=0.3003, pruned_loss=0.07409, over 3809276.93 frames. ], batch size: 58, lr: 5.86e-03, grad_scale: 8.0 2023-04-02 03:27:05,647 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=94366.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 03:27:30,022 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.083e+02 5.188e+02 6.005e+02 7.911e+02 1.925e+03, threshold=1.201e+03, percent-clipped=8.0 2023-04-02 03:28:03,375 INFO [train.py:903] (1/4) Epoch 14, batch 5650, loss[loss=0.2259, simple_loss=0.3088, pruned_loss=0.07152, over 19723.00 frames. ], tot_loss[loss=0.2242, simple_loss=0.3003, pruned_loss=0.07403, over 3824732.49 frames. ], batch size: 63, lr: 5.86e-03, grad_scale: 8.0 2023-04-02 03:28:49,715 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9_rvb from training. Duration: 28.0944375 2023-04-02 03:28:55,923 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=94458.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 03:29:02,016 INFO [train.py:903] (1/4) Epoch 14, batch 5700, loss[loss=0.2317, simple_loss=0.315, pruned_loss=0.07424, over 18744.00 frames. ], tot_loss[loss=0.2256, simple_loss=0.3012, pruned_loss=0.07495, over 3831051.53 frames. ], batch size: 74, lr: 5.86e-03, grad_scale: 8.0 2023-04-02 03:29:29,828 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.310e+02 4.949e+02 6.008e+02 7.817e+02 2.884e+03, threshold=1.202e+03, percent-clipped=11.0 2023-04-02 03:29:50,189 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=94503.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 03:30:02,383 INFO [train.py:903] (1/4) Epoch 14, batch 5750, loss[loss=0.2026, simple_loss=0.2809, pruned_loss=0.06217, over 19723.00 frames. ], tot_loss[loss=0.2246, simple_loss=0.3004, pruned_loss=0.07436, over 3839061.40 frames. ], batch size: 51, lr: 5.86e-03, grad_scale: 8.0 2023-04-02 03:30:04,717 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9_rvb from training. Duration: 33.038875 2023-04-02 03:30:09,071 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.62 vs. limit=5.0 2023-04-02 03:30:11,537 WARNING [train.py:1073] (1/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 2023-04-02 03:30:17,736 WARNING [train.py:1073] (1/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 2023-04-02 03:30:21,288 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=94528.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 03:31:04,683 INFO [train.py:903] (1/4) Epoch 14, batch 5800, loss[loss=0.2166, simple_loss=0.2958, pruned_loss=0.06867, over 19661.00 frames. ], tot_loss[loss=0.2244, simple_loss=0.3003, pruned_loss=0.07423, over 3826780.43 frames. ], batch size: 55, lr: 5.86e-03, grad_scale: 8.0 2023-04-02 03:31:30,529 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=94585.0, num_to_drop=1, layers_to_drop={0} 2023-04-02 03:31:32,528 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.175e+02 5.430e+02 7.155e+02 9.192e+02 1.752e+03, threshold=1.431e+03, percent-clipped=10.0 2023-04-02 03:31:35,425 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 2023-04-02 03:31:38,986 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.09 vs. limit=2.0 2023-04-02 03:32:06,961 INFO [train.py:903] (1/4) Epoch 14, batch 5850, loss[loss=0.2542, simple_loss=0.3345, pruned_loss=0.08695, over 19704.00 frames. ], tot_loss[loss=0.2251, simple_loss=0.3009, pruned_loss=0.0746, over 3829580.06 frames. ], batch size: 59, lr: 5.86e-03, grad_scale: 8.0 2023-04-02 03:33:06,738 INFO [train.py:903] (1/4) Epoch 14, batch 5900, loss[loss=0.2478, simple_loss=0.3218, pruned_loss=0.0869, over 19696.00 frames. ], tot_loss[loss=0.2262, simple_loss=0.302, pruned_loss=0.07515, over 3807593.94 frames. ], batch size: 59, lr: 5.85e-03, grad_scale: 8.0 2023-04-02 03:33:07,924 WARNING [train.py:1073] (1/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 2023-04-02 03:33:27,825 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9_rvb from training. Duration: 27.511125 2023-04-02 03:33:33,163 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.177e+02 5.122e+02 5.971e+02 8.409e+02 2.018e+03, threshold=1.194e+03, percent-clipped=4.0 2023-04-02 03:33:43,156 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.2421, 2.0681, 2.0139, 2.4789, 2.3528, 1.8927, 1.8627, 2.4353], device='cuda:1'), covar=tensor([0.1023, 0.1785, 0.1453, 0.0950, 0.1369, 0.0654, 0.1460, 0.0681], device='cuda:1'), in_proj_covar=tensor([0.0262, 0.0356, 0.0300, 0.0246, 0.0301, 0.0246, 0.0292, 0.0247], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 03:33:50,048 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=94700.0, num_to_drop=1, layers_to_drop={0} 2023-04-02 03:34:01,315 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=94710.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 03:34:05,812 INFO [train.py:903] (1/4) Epoch 14, batch 5950, loss[loss=0.1958, simple_loss=0.2742, pruned_loss=0.05869, over 19734.00 frames. ], tot_loss[loss=0.2256, simple_loss=0.3015, pruned_loss=0.07486, over 3824189.50 frames. ], batch size: 46, lr: 5.85e-03, grad_scale: 8.0 2023-04-02 03:34:06,225 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=94714.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 03:34:37,319 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=94739.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 03:35:04,629 INFO [train.py:903] (1/4) Epoch 14, batch 6000, loss[loss=0.2892, simple_loss=0.3529, pruned_loss=0.1127, over 18229.00 frames. ], tot_loss[loss=0.2262, simple_loss=0.3021, pruned_loss=0.0751, over 3830036.79 frames. ], batch size: 84, lr: 5.85e-03, grad_scale: 8.0 2023-04-02 03:35:04,629 INFO [train.py:928] (1/4) Computing validation loss 2023-04-02 03:35:17,168 INFO [train.py:937] (1/4) Epoch 14, validation: loss=0.1744, simple_loss=0.2748, pruned_loss=0.03705, over 944034.00 frames. 2023-04-02 03:35:17,169 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18821MB 2023-04-02 03:35:47,201 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.285e+02 5.018e+02 6.191e+02 7.483e+02 1.325e+03, threshold=1.238e+03, percent-clipped=2.0 2023-04-02 03:36:17,853 INFO [train.py:903] (1/4) Epoch 14, batch 6050, loss[loss=0.2517, simple_loss=0.323, pruned_loss=0.09017, over 19506.00 frames. ], tot_loss[loss=0.2243, simple_loss=0.3003, pruned_loss=0.07415, over 3841594.31 frames. ], batch size: 64, lr: 5.85e-03, grad_scale: 8.0 2023-04-02 03:36:33,154 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=94825.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 03:37:20,914 INFO [train.py:903] (1/4) Epoch 14, batch 6100, loss[loss=0.1968, simple_loss=0.267, pruned_loss=0.06328, over 19751.00 frames. ], tot_loss[loss=0.2247, simple_loss=0.3005, pruned_loss=0.07451, over 3836327.13 frames. ], batch size: 46, lr: 5.85e-03, grad_scale: 8.0 2023-04-02 03:37:48,990 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.227e+02 5.260e+02 6.294e+02 8.137e+02 1.551e+03, threshold=1.259e+03, percent-clipped=3.0 2023-04-02 03:38:21,673 INFO [train.py:903] (1/4) Epoch 14, batch 6150, loss[loss=0.2605, simple_loss=0.326, pruned_loss=0.09748, over 17382.00 frames. ], tot_loss[loss=0.2245, simple_loss=0.3001, pruned_loss=0.07443, over 3829246.66 frames. ], batch size: 101, lr: 5.85e-03, grad_scale: 8.0 2023-04-02 03:38:48,820 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9_rvb from training. Duration: 31.02225 2023-04-02 03:39:13,109 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=94956.0, num_to_drop=1, layers_to_drop={1} 2023-04-02 03:39:21,882 INFO [train.py:903] (1/4) Epoch 14, batch 6200, loss[loss=0.1804, simple_loss=0.2588, pruned_loss=0.05097, over 19394.00 frames. ], tot_loss[loss=0.2246, simple_loss=0.3003, pruned_loss=0.07451, over 3823989.79 frames. ], batch size: 48, lr: 5.85e-03, grad_scale: 8.0 2023-04-02 03:39:44,503 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=94981.0, num_to_drop=1, layers_to_drop={1} 2023-04-02 03:39:51,883 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.111e+02 5.470e+02 6.385e+02 8.085e+02 2.296e+03, threshold=1.277e+03, percent-clipped=5.0 2023-04-02 03:40:19,513 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([0.8539, 1.1173, 1.5624, 0.4878, 2.0308, 2.4464, 2.0822, 2.5796], device='cuda:1'), covar=tensor([0.1618, 0.3755, 0.3139, 0.2601, 0.0557, 0.0258, 0.0353, 0.0308], device='cuda:1'), in_proj_covar=tensor([0.0263, 0.0303, 0.0330, 0.0253, 0.0224, 0.0167, 0.0208, 0.0218], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 03:40:22,504 INFO [train.py:903] (1/4) Epoch 14, batch 6250, loss[loss=0.2252, simple_loss=0.3106, pruned_loss=0.06991, over 18862.00 frames. ], tot_loss[loss=0.224, simple_loss=0.3, pruned_loss=0.07401, over 3828714.82 frames. ], batch size: 74, lr: 5.84e-03, grad_scale: 8.0 2023-04-02 03:40:31,405 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.82 vs. limit=2.0 2023-04-02 03:40:55,024 WARNING [train.py:1073] (1/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9_rvb from training. Duration: 25.988875 2023-04-02 03:41:24,163 INFO [train.py:903] (1/4) Epoch 14, batch 6300, loss[loss=0.2038, simple_loss=0.283, pruned_loss=0.06225, over 19380.00 frames. ], tot_loss[loss=0.2247, simple_loss=0.3006, pruned_loss=0.07441, over 3826422.24 frames. ], batch size: 48, lr: 5.84e-03, grad_scale: 8.0 2023-04-02 03:41:44,490 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=95081.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 03:41:51,896 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.537e+02 5.238e+02 6.215e+02 7.195e+02 1.642e+03, threshold=1.243e+03, percent-clipped=4.0 2023-04-02 03:42:15,045 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=95106.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 03:42:24,131 INFO [train.py:903] (1/4) Epoch 14, batch 6350, loss[loss=0.1768, simple_loss=0.2556, pruned_loss=0.04897, over 19747.00 frames. ], tot_loss[loss=0.2246, simple_loss=0.3001, pruned_loss=0.07452, over 3829779.00 frames. ], batch size: 45, lr: 5.84e-03, grad_scale: 8.0 2023-04-02 03:42:34,692 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=95123.0, num_to_drop=1, layers_to_drop={1} 2023-04-02 03:42:50,475 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.1271, 2.2813, 2.3788, 3.0982, 2.2629, 2.9686, 2.6110, 2.1616], device='cuda:1'), covar=tensor([0.3947, 0.3477, 0.1577, 0.2171, 0.3955, 0.1745, 0.3708, 0.2857], device='cuda:1'), in_proj_covar=tensor([0.0826, 0.0864, 0.0667, 0.0903, 0.0811, 0.0752, 0.0809, 0.0735], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 03:43:23,501 INFO [train.py:903] (1/4) Epoch 14, batch 6400, loss[loss=0.2056, simple_loss=0.2865, pruned_loss=0.06238, over 19858.00 frames. ], tot_loss[loss=0.2262, simple_loss=0.3015, pruned_loss=0.07543, over 3829969.78 frames. ], batch size: 52, lr: 5.84e-03, grad_scale: 8.0 2023-04-02 03:43:31,393 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([0.8906, 0.7972, 0.8508, 1.0371, 0.8584, 0.9323, 1.0013, 0.8730], device='cuda:1'), covar=tensor([0.0715, 0.0819, 0.0831, 0.0563, 0.0746, 0.0686, 0.0737, 0.0646], device='cuda:1'), in_proj_covar=tensor([0.0210, 0.0220, 0.0222, 0.0243, 0.0228, 0.0211, 0.0194, 0.0202], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-02 03:43:52,814 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.332e+02 5.689e+02 7.116e+02 8.755e+02 2.889e+03, threshold=1.423e+03, percent-clipped=3.0 2023-04-02 03:44:23,637 INFO [train.py:903] (1/4) Epoch 14, batch 6450, loss[loss=0.2051, simple_loss=0.2976, pruned_loss=0.05635, over 19687.00 frames. ], tot_loss[loss=0.226, simple_loss=0.3015, pruned_loss=0.07521, over 3818357.80 frames. ], batch size: 60, lr: 5.84e-03, grad_scale: 8.0 2023-04-02 03:45:09,466 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9_rvb from training. Duration: 28.638875 2023-04-02 03:45:25,844 INFO [train.py:903] (1/4) Epoch 14, batch 6500, loss[loss=0.183, simple_loss=0.2714, pruned_loss=0.04729, over 19673.00 frames. ], tot_loss[loss=0.2235, simple_loss=0.2995, pruned_loss=0.07374, over 3822232.38 frames. ], batch size: 53, lr: 5.84e-03, grad_scale: 8.0 2023-04-02 03:45:32,076 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.70 vs. limit=2.0 2023-04-02 03:45:32,332 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1_rvb from training. Duration: 0.836375 2023-04-02 03:45:43,522 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=95278.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 03:45:52,733 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=95286.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 03:45:54,573 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.481e+02 5.245e+02 6.559e+02 8.783e+02 2.152e+03, threshold=1.312e+03, percent-clipped=6.0 2023-04-02 03:46:27,867 INFO [train.py:903] (1/4) Epoch 14, batch 6550, loss[loss=0.2107, simple_loss=0.2877, pruned_loss=0.06687, over 19393.00 frames. ], tot_loss[loss=0.2238, simple_loss=0.2999, pruned_loss=0.07386, over 3808924.98 frames. ], batch size: 47, lr: 5.84e-03, grad_scale: 8.0 2023-04-02 03:47:20,329 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=95357.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 03:47:24,832 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1279, 1.4403, 1.8738, 1.6265, 2.9287, 4.3280, 4.3142, 4.8626], device='cuda:1'), covar=tensor([0.1679, 0.3539, 0.2995, 0.1969, 0.0557, 0.0210, 0.0164, 0.0131], device='cuda:1'), in_proj_covar=tensor([0.0262, 0.0304, 0.0330, 0.0253, 0.0225, 0.0167, 0.0207, 0.0219], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 03:47:28,131 INFO [train.py:903] (1/4) Epoch 14, batch 6600, loss[loss=0.2373, simple_loss=0.3223, pruned_loss=0.07615, over 19509.00 frames. ], tot_loss[loss=0.2232, simple_loss=0.2999, pruned_loss=0.07322, over 3822643.23 frames. ], batch size: 64, lr: 5.83e-03, grad_scale: 8.0 2023-04-02 03:47:57,394 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.401e+02 5.166e+02 6.061e+02 7.266e+02 1.890e+03, threshold=1.212e+03, percent-clipped=2.0 2023-04-02 03:48:28,437 INFO [train.py:903] (1/4) Epoch 14, batch 6650, loss[loss=0.2449, simple_loss=0.3234, pruned_loss=0.08323, over 19757.00 frames. ], tot_loss[loss=0.2238, simple_loss=0.3003, pruned_loss=0.07363, over 3821094.01 frames. ], batch size: 54, lr: 5.83e-03, grad_scale: 8.0 2023-04-02 03:49:29,378 INFO [train.py:903] (1/4) Epoch 14, batch 6700, loss[loss=0.1678, simple_loss=0.2463, pruned_loss=0.04467, over 19743.00 frames. ], tot_loss[loss=0.2234, simple_loss=0.2992, pruned_loss=0.07377, over 3813311.33 frames. ], batch size: 47, lr: 5.83e-03, grad_scale: 8.0 2023-04-02 03:49:33,785 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=95467.0, num_to_drop=1, layers_to_drop={0} 2023-04-02 03:49:53,206 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2314, 1.3880, 1.8473, 1.2912, 2.6608, 3.7552, 3.5429, 3.9575], device='cuda:1'), covar=tensor([0.1500, 0.3362, 0.2696, 0.2086, 0.0519, 0.0159, 0.0176, 0.0180], device='cuda:1'), in_proj_covar=tensor([0.0261, 0.0302, 0.0328, 0.0252, 0.0223, 0.0167, 0.0207, 0.0219], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 03:49:55,591 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3894, 1.4999, 1.8454, 1.7044, 2.7313, 2.4024, 2.8243, 1.3270], device='cuda:1'), covar=tensor([0.2268, 0.3941, 0.2390, 0.1704, 0.1451, 0.1805, 0.1482, 0.3707], device='cuda:1'), in_proj_covar=tensor([0.0499, 0.0589, 0.0639, 0.0449, 0.0598, 0.0504, 0.0646, 0.0508], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 03:49:57,461 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.175e+02 5.267e+02 5.901e+02 8.158e+02 1.902e+03, threshold=1.180e+03, percent-clipped=6.0 2023-04-02 03:50:25,036 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.6383, 2.5622, 2.4612, 2.9736, 2.8599, 2.4403, 2.3237, 2.9046], device='cuda:1'), covar=tensor([0.0752, 0.1268, 0.1034, 0.0799, 0.0970, 0.0395, 0.1006, 0.0473], device='cuda:1'), in_proj_covar=tensor([0.0258, 0.0353, 0.0297, 0.0245, 0.0298, 0.0245, 0.0288, 0.0244], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 03:50:25,738 INFO [train.py:903] (1/4) Epoch 14, batch 6750, loss[loss=0.1999, simple_loss=0.2733, pruned_loss=0.06322, over 19482.00 frames. ], tot_loss[loss=0.2244, simple_loss=0.2998, pruned_loss=0.0745, over 3798672.61 frames. ], batch size: 49, lr: 5.83e-03, grad_scale: 8.0 2023-04-02 03:51:21,226 INFO [train.py:903] (1/4) Epoch 14, batch 6800, loss[loss=0.2246, simple_loss=0.3026, pruned_loss=0.0733, over 18667.00 frames. ], tot_loss[loss=0.2258, simple_loss=0.3008, pruned_loss=0.07539, over 3808891.38 frames. ], batch size: 74, lr: 5.83e-03, grad_scale: 8.0 2023-04-02 03:51:41,557 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=95582.0, num_to_drop=1, layers_to_drop={0} 2023-04-02 03:51:46,734 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.931e+02 5.198e+02 6.166e+02 8.008e+02 1.508e+03, threshold=1.233e+03, percent-clipped=6.0 2023-04-02 03:52:06,365 WARNING [train.py:1073] (1/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 2023-04-02 03:52:07,471 WARNING [train.py:1073] (1/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 2023-04-02 03:52:10,204 INFO [train.py:903] (1/4) Epoch 15, batch 0, loss[loss=0.2369, simple_loss=0.3082, pruned_loss=0.08283, over 19584.00 frames. ], tot_loss[loss=0.2369, simple_loss=0.3082, pruned_loss=0.08283, over 19584.00 frames. ], batch size: 52, lr: 5.63e-03, grad_scale: 8.0 2023-04-02 03:52:10,204 INFO [train.py:928] (1/4) Computing validation loss 2023-04-02 03:52:21,744 INFO [train.py:937] (1/4) Epoch 15, validation: loss=0.1744, simple_loss=0.2751, pruned_loss=0.03681, over 944034.00 frames. 2023-04-02 03:52:21,744 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18821MB 2023-04-02 03:52:26,583 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3945, 1.4538, 1.7619, 1.6596, 2.6826, 2.2880, 2.8095, 1.0163], device='cuda:1'), covar=tensor([0.2331, 0.4075, 0.2578, 0.1820, 0.1425, 0.1963, 0.1475, 0.4187], device='cuda:1'), in_proj_covar=tensor([0.0503, 0.0592, 0.0642, 0.0452, 0.0602, 0.0507, 0.0650, 0.0510], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 03:52:33,134 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 2023-04-02 03:52:58,948 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=95622.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 03:53:08,225 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=95630.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 03:53:14,626 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=95635.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 03:53:22,136 INFO [train.py:903] (1/4) Epoch 15, batch 50, loss[loss=0.2313, simple_loss=0.2943, pruned_loss=0.08419, over 19749.00 frames. ], tot_loss[loss=0.2252, simple_loss=0.3028, pruned_loss=0.07383, over 881537.54 frames. ], batch size: 47, lr: 5.63e-03, grad_scale: 8.0 2023-04-02 03:53:57,869 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.5360, 1.1640, 1.3691, 1.2755, 2.1283, 0.9809, 1.9279, 2.4582], device='cuda:1'), covar=tensor([0.0646, 0.2657, 0.2628, 0.1512, 0.0896, 0.1965, 0.1007, 0.0425], device='cuda:1'), in_proj_covar=tensor([0.0374, 0.0347, 0.0367, 0.0327, 0.0357, 0.0336, 0.0344, 0.0371], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 03:53:58,791 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1_rvb from training. Duration: 27.0318125 2023-04-02 03:54:07,437 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 2023-04-02 03:54:20,258 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.439e+02 5.472e+02 6.461e+02 8.212e+02 1.912e+03, threshold=1.292e+03, percent-clipped=7.0 2023-04-02 03:54:26,851 INFO [train.py:903] (1/4) Epoch 15, batch 100, loss[loss=0.1881, simple_loss=0.2595, pruned_loss=0.0584, over 19752.00 frames. ], tot_loss[loss=0.2227, simple_loss=0.2992, pruned_loss=0.07311, over 1543074.68 frames. ], batch size: 46, lr: 5.62e-03, grad_scale: 8.0 2023-04-02 03:54:37,473 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 2023-04-02 03:54:37,589 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=95701.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 03:55:22,804 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=95737.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 03:55:28,252 INFO [train.py:903] (1/4) Epoch 15, batch 150, loss[loss=0.2247, simple_loss=0.3065, pruned_loss=0.07146, over 17194.00 frames. ], tot_loss[loss=0.2224, simple_loss=0.2991, pruned_loss=0.07288, over 2050728.00 frames. ], batch size: 101, lr: 5.62e-03, grad_scale: 8.0 2023-04-02 03:55:32,093 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=95745.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 03:56:23,968 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.923e+02 5.351e+02 6.322e+02 7.452e+02 1.833e+03, threshold=1.264e+03, percent-clipped=1.0 2023-04-02 03:56:27,304 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 2023-04-02 03:56:28,490 INFO [train.py:903] (1/4) Epoch 15, batch 200, loss[loss=0.174, simple_loss=0.2453, pruned_loss=0.0514, over 19101.00 frames. ], tot_loss[loss=0.2233, simple_loss=0.2997, pruned_loss=0.07349, over 2441490.99 frames. ], batch size: 42, lr: 5.62e-03, grad_scale: 8.0 2023-04-02 03:56:45,610 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.4566, 2.2090, 2.1448, 2.5855, 2.4715, 2.1984, 1.8488, 2.6220], device='cuda:1'), covar=tensor([0.0912, 0.1638, 0.1405, 0.1065, 0.1293, 0.0490, 0.1386, 0.0606], device='cuda:1'), in_proj_covar=tensor([0.0260, 0.0357, 0.0301, 0.0247, 0.0301, 0.0247, 0.0291, 0.0247], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 03:56:51,099 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.5636, 1.2010, 1.4227, 1.4997, 2.1877, 1.1606, 1.9962, 2.4872], device='cuda:1'), covar=tensor([0.0676, 0.2538, 0.2529, 0.1268, 0.0870, 0.1750, 0.0986, 0.0435], device='cuda:1'), in_proj_covar=tensor([0.0374, 0.0346, 0.0369, 0.0328, 0.0354, 0.0337, 0.0344, 0.0372], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 03:56:59,688 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=95816.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 03:57:15,604 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=95830.0, num_to_drop=1, layers_to_drop={0} 2023-04-02 03:57:24,888 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=95838.0, num_to_drop=1, layers_to_drop={0} 2023-04-02 03:57:29,639 INFO [train.py:903] (1/4) Epoch 15, batch 250, loss[loss=0.2407, simple_loss=0.308, pruned_loss=0.08674, over 19657.00 frames. ], tot_loss[loss=0.2239, simple_loss=0.3003, pruned_loss=0.07374, over 2753227.61 frames. ], batch size: 60, lr: 5.62e-03, grad_scale: 8.0 2023-04-02 03:57:56,109 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=95863.0, num_to_drop=1, layers_to_drop={0} 2023-04-02 03:57:58,216 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=95865.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 03:58:24,515 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.450e+02 5.277e+02 6.948e+02 9.039e+02 3.101e+03, threshold=1.390e+03, percent-clipped=9.0 2023-04-02 03:58:28,478 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5116, 1.5221, 1.7756, 1.6873, 2.5579, 2.2889, 2.5613, 1.2631], device='cuda:1'), covar=tensor([0.2055, 0.3844, 0.2288, 0.1660, 0.1312, 0.1724, 0.1331, 0.3521], device='cuda:1'), in_proj_covar=tensor([0.0498, 0.0589, 0.0636, 0.0449, 0.0599, 0.0503, 0.0644, 0.0507], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 03:58:30,109 INFO [train.py:903] (1/4) Epoch 15, batch 300, loss[loss=0.1935, simple_loss=0.2671, pruned_loss=0.05989, over 16037.00 frames. ], tot_loss[loss=0.224, simple_loss=0.2999, pruned_loss=0.074, over 2988748.15 frames. ], batch size: 35, lr: 5.62e-03, grad_scale: 8.0 2023-04-02 03:59:32,839 INFO [train.py:903] (1/4) Epoch 15, batch 350, loss[loss=0.1882, simple_loss=0.2752, pruned_loss=0.05057, over 19844.00 frames. ], tot_loss[loss=0.2239, simple_loss=0.3001, pruned_loss=0.07383, over 3165834.78 frames. ], batch size: 52, lr: 5.62e-03, grad_scale: 8.0 2023-04-02 03:59:33,868 WARNING [train.py:1073] (1/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1_rvb from training. Duration: 0.9681875 2023-04-02 03:59:58,802 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.7933, 4.8712, 5.5846, 5.5470, 2.1224, 5.2046, 4.4657, 5.1858], device='cuda:1'), covar=tensor([0.1408, 0.1043, 0.0521, 0.0556, 0.5416, 0.0689, 0.0567, 0.1150], device='cuda:1'), in_proj_covar=tensor([0.0719, 0.0646, 0.0856, 0.0740, 0.0762, 0.0600, 0.0516, 0.0787], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-02 04:00:17,462 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=95979.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:00:28,193 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.384e+02 5.018e+02 5.906e+02 6.897e+02 1.495e+03, threshold=1.181e+03, percent-clipped=1.0 2023-04-02 04:00:32,776 INFO [train.py:903] (1/4) Epoch 15, batch 400, loss[loss=0.2233, simple_loss=0.2997, pruned_loss=0.07343, over 19578.00 frames. ], tot_loss[loss=0.2242, simple_loss=0.2999, pruned_loss=0.07427, over 3320809.67 frames. ], batch size: 52, lr: 5.62e-03, grad_scale: 8.0 2023-04-02 04:00:34,350 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=95993.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:00:39,881 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=95998.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:00:44,452 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=96001.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:01:04,791 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=96018.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:01:12,818 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3514, 1.3484, 1.4415, 1.4873, 1.7744, 1.8971, 1.7467, 0.4449], device='cuda:1'), covar=tensor([0.2250, 0.3887, 0.2478, 0.1911, 0.1565, 0.2157, 0.1361, 0.4193], device='cuda:1'), in_proj_covar=tensor([0.0502, 0.0590, 0.0643, 0.0453, 0.0605, 0.0508, 0.0648, 0.0510], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 04:01:16,058 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=96026.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:01:33,841 INFO [train.py:903] (1/4) Epoch 15, batch 450, loss[loss=0.2467, simple_loss=0.3152, pruned_loss=0.08906, over 13509.00 frames. ], tot_loss[loss=0.2234, simple_loss=0.2993, pruned_loss=0.07376, over 3438283.54 frames. ], batch size: 136, lr: 5.61e-03, grad_scale: 8.0 2023-04-02 04:02:07,790 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 2023-04-02 04:02:07,820 WARNING [train.py:1073] (1/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 2023-04-02 04:02:12,880 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=96072.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:02:31,246 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.196e+02 4.837e+02 5.995e+02 7.453e+02 1.580e+03, threshold=1.199e+03, percent-clipped=6.0 2023-04-02 04:02:36,669 INFO [train.py:903] (1/4) Epoch 15, batch 500, loss[loss=0.2201, simple_loss=0.293, pruned_loss=0.07359, over 19451.00 frames. ], tot_loss[loss=0.2234, simple_loss=0.2993, pruned_loss=0.07378, over 3531510.65 frames. ], batch size: 49, lr: 5.61e-03, grad_scale: 8.0 2023-04-02 04:02:39,309 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=96094.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:02:43,796 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=96097.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:02:44,805 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=96098.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:03:38,906 INFO [train.py:903] (1/4) Epoch 15, batch 550, loss[loss=0.2835, simple_loss=0.3463, pruned_loss=0.1104, over 19286.00 frames. ], tot_loss[loss=0.2241, simple_loss=0.3001, pruned_loss=0.07405, over 3604247.15 frames. ], batch size: 66, lr: 5.61e-03, grad_scale: 8.0 2023-04-02 04:04:18,072 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=96174.0, num_to_drop=1, layers_to_drop={1} 2023-04-02 04:04:29,221 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=96183.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:04:35,629 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.160e+02 5.342e+02 6.491e+02 8.104e+02 1.503e+03, threshold=1.298e+03, percent-clipped=3.0 2023-04-02 04:04:40,051 INFO [train.py:903] (1/4) Epoch 15, batch 600, loss[loss=0.2543, simple_loss=0.3391, pruned_loss=0.08471, over 19669.00 frames. ], tot_loss[loss=0.2233, simple_loss=0.2995, pruned_loss=0.07354, over 3655719.16 frames. ], batch size: 55, lr: 5.61e-03, grad_scale: 8.0 2023-04-02 04:05:00,805 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=96209.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:05:20,412 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 2023-04-02 04:05:35,506 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9963, 2.1162, 2.3023, 2.7434, 2.0473, 2.6306, 2.5013, 2.0608], device='cuda:1'), covar=tensor([0.3755, 0.3140, 0.1496, 0.2083, 0.3523, 0.1682, 0.3668, 0.2827], device='cuda:1'), in_proj_covar=tensor([0.0828, 0.0867, 0.0667, 0.0899, 0.0814, 0.0747, 0.0806, 0.0731], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 04:05:43,298 INFO [train.py:903] (1/4) Epoch 15, batch 650, loss[loss=0.2179, simple_loss=0.3006, pruned_loss=0.06755, over 19289.00 frames. ], tot_loss[loss=0.2232, simple_loss=0.2991, pruned_loss=0.07369, over 3687358.98 frames. ], batch size: 66, lr: 5.61e-03, grad_scale: 8.0 2023-04-02 04:06:41,549 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.029e+02 5.105e+02 6.385e+02 8.770e+02 1.706e+03, threshold=1.277e+03, percent-clipped=3.0 2023-04-02 04:06:42,914 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=96289.0, num_to_drop=1, layers_to_drop={0} 2023-04-02 04:06:46,045 INFO [train.py:903] (1/4) Epoch 15, batch 700, loss[loss=0.2118, simple_loss=0.2903, pruned_loss=0.06662, over 19662.00 frames. ], tot_loss[loss=0.2221, simple_loss=0.2988, pruned_loss=0.07267, over 3720085.16 frames. ], batch size: 53, lr: 5.61e-03, grad_scale: 8.0 2023-04-02 04:07:12,184 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.67 vs. limit=5.0 2023-04-02 04:07:19,714 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.7366, 1.8157, 2.0280, 2.3256, 1.6282, 2.1960, 2.1821, 1.8693], device='cuda:1'), covar=tensor([0.3807, 0.3441, 0.1637, 0.1980, 0.3610, 0.1799, 0.4087, 0.3064], device='cuda:1'), in_proj_covar=tensor([0.0830, 0.0868, 0.0668, 0.0902, 0.0816, 0.0749, 0.0807, 0.0732], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 04:07:26,410 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=96324.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:07:47,390 INFO [train.py:903] (1/4) Epoch 15, batch 750, loss[loss=0.1995, simple_loss=0.2704, pruned_loss=0.06433, over 19745.00 frames. ], tot_loss[loss=0.2244, simple_loss=0.301, pruned_loss=0.07385, over 3734229.52 frames. ], batch size: 47, lr: 5.61e-03, grad_scale: 8.0 2023-04-02 04:07:47,554 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=96342.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:07:47,699 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=96342.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:07:57,753 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=96350.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:08:29,486 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=96375.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:08:44,437 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.337e+02 5.276e+02 6.207e+02 7.536e+02 1.572e+03, threshold=1.241e+03, percent-clipped=2.0 2023-04-02 04:08:49,809 INFO [train.py:903] (1/4) Epoch 15, batch 800, loss[loss=0.2273, simple_loss=0.3057, pruned_loss=0.07442, over 19661.00 frames. ], tot_loss[loss=0.2242, simple_loss=0.3009, pruned_loss=0.07376, over 3755585.26 frames. ], batch size: 55, lr: 5.60e-03, grad_scale: 8.0 2023-04-02 04:08:53,677 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.6392, 1.3358, 1.4623, 1.4394, 3.2232, 0.9623, 2.2594, 3.5874], device='cuda:1'), covar=tensor([0.0423, 0.2659, 0.2784, 0.1784, 0.0736, 0.2528, 0.1226, 0.0247], device='cuda:1'), in_proj_covar=tensor([0.0374, 0.0347, 0.0367, 0.0327, 0.0353, 0.0334, 0.0344, 0.0372], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 04:09:04,710 WARNING [train.py:1073] (1/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 2023-04-02 04:09:09,567 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=96408.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:09:50,620 INFO [train.py:903] (1/4) Epoch 15, batch 850, loss[loss=0.1937, simple_loss=0.2588, pruned_loss=0.06435, over 19741.00 frames. ], tot_loss[loss=0.2244, simple_loss=0.3006, pruned_loss=0.07405, over 3775918.60 frames. ], batch size: 46, lr: 5.60e-03, grad_scale: 8.0 2023-04-02 04:09:51,925 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=96442.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:10:10,272 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=96457.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:10:41,148 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9_rvb from training. Duration: 25.061125 2023-04-02 04:10:47,719 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.001e+02 5.238e+02 6.465e+02 7.879e+02 1.664e+03, threshold=1.293e+03, percent-clipped=4.0 2023-04-02 04:10:52,486 INFO [train.py:903] (1/4) Epoch 15, batch 900, loss[loss=0.2016, simple_loss=0.2882, pruned_loss=0.05752, over 19746.00 frames. ], tot_loss[loss=0.2251, simple_loss=0.3012, pruned_loss=0.07451, over 3791485.20 frames. ], batch size: 54, lr: 5.60e-03, grad_scale: 8.0 2023-04-02 04:11:20,266 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2487, 1.3068, 1.4527, 1.4068, 1.7514, 1.7679, 1.7580, 0.5710], device='cuda:1'), covar=tensor([0.2289, 0.3872, 0.2423, 0.1809, 0.1498, 0.2182, 0.1366, 0.4068], device='cuda:1'), in_proj_covar=tensor([0.0498, 0.0586, 0.0637, 0.0448, 0.0598, 0.0503, 0.0641, 0.0505], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 04:11:36,486 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=96527.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:11:55,050 INFO [train.py:903] (1/4) Epoch 15, batch 950, loss[loss=0.2342, simple_loss=0.3157, pruned_loss=0.07638, over 19681.00 frames. ], tot_loss[loss=0.225, simple_loss=0.3012, pruned_loss=0.0744, over 3801514.16 frames. ], batch size: 63, lr: 5.60e-03, grad_scale: 8.0 2023-04-02 04:11:56,232 WARNING [train.py:1073] (1/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9_rvb from training. Duration: 26.32775 2023-04-02 04:11:59,030 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=96545.0, num_to_drop=1, layers_to_drop={0} 2023-04-02 04:12:14,131 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=96557.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:12:30,393 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=96570.0, num_to_drop=1, layers_to_drop={0} 2023-04-02 04:12:41,739 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=96580.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:12:52,446 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.674e+02 5.225e+02 6.143e+02 7.839e+02 1.754e+03, threshold=1.229e+03, percent-clipped=1.0 2023-04-02 04:12:53,871 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.2659, 3.7246, 3.8797, 3.8765, 1.4392, 3.6538, 3.2122, 3.5754], device='cuda:1'), covar=tensor([0.1592, 0.0917, 0.0677, 0.0715, 0.5730, 0.0862, 0.0696, 0.1148], device='cuda:1'), in_proj_covar=tensor([0.0732, 0.0659, 0.0873, 0.0745, 0.0777, 0.0609, 0.0523, 0.0802], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-02 04:12:57,219 INFO [train.py:903] (1/4) Epoch 15, batch 1000, loss[loss=0.1771, simple_loss=0.2586, pruned_loss=0.04777, over 19729.00 frames. ], tot_loss[loss=0.224, simple_loss=0.3005, pruned_loss=0.07379, over 3824376.28 frames. ], batch size: 47, lr: 5.60e-03, grad_scale: 8.0 2023-04-02 04:13:13,543 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=96605.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:13:51,308 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 2023-04-02 04:13:59,278 INFO [train.py:903] (1/4) Epoch 15, batch 1050, loss[loss=0.179, simple_loss=0.2532, pruned_loss=0.05237, over 19754.00 frames. ], tot_loss[loss=0.2219, simple_loss=0.2986, pruned_loss=0.07263, over 3838885.66 frames. ], batch size: 47, lr: 5.60e-03, grad_scale: 8.0 2023-04-02 04:13:59,635 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=96642.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:14:31,315 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 2023-04-02 04:14:53,730 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=96686.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:14:57,023 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.831e+02 5.170e+02 6.505e+02 8.376e+02 1.590e+03, threshold=1.301e+03, percent-clipped=4.0 2023-04-02 04:15:01,344 INFO [train.py:903] (1/4) Epoch 15, batch 1100, loss[loss=0.2404, simple_loss=0.3158, pruned_loss=0.08247, over 19749.00 frames. ], tot_loss[loss=0.2218, simple_loss=0.2981, pruned_loss=0.07274, over 3833803.56 frames. ], batch size: 63, lr: 5.60e-03, grad_scale: 4.0 2023-04-02 04:15:14,750 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.1194, 3.4191, 1.9125, 2.1677, 3.1050, 1.8434, 1.5150, 2.1763], device='cuda:1'), covar=tensor([0.1226, 0.0534, 0.1024, 0.0704, 0.0482, 0.1089, 0.0919, 0.0699], device='cuda:1'), in_proj_covar=tensor([0.0295, 0.0305, 0.0325, 0.0248, 0.0238, 0.0328, 0.0291, 0.0265], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 04:15:28,053 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=96713.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:15:58,781 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=96738.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:16:02,838 INFO [train.py:903] (1/4) Epoch 15, batch 1150, loss[loss=0.2467, simple_loss=0.3211, pruned_loss=0.08617, over 18140.00 frames. ], tot_loss[loss=0.2206, simple_loss=0.2972, pruned_loss=0.07202, over 3830010.61 frames. ], batch size: 83, lr: 5.59e-03, grad_scale: 4.0 2023-04-02 04:16:15,362 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=96752.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:17:01,609 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.369e+02 5.122e+02 6.460e+02 8.216e+02 1.619e+03, threshold=1.292e+03, percent-clipped=5.0 2023-04-02 04:17:06,202 INFO [train.py:903] (1/4) Epoch 15, batch 1200, loss[loss=0.1886, simple_loss=0.2676, pruned_loss=0.05478, over 19789.00 frames. ], tot_loss[loss=0.222, simple_loss=0.2984, pruned_loss=0.07284, over 3827035.34 frames. ], batch size: 49, lr: 5.59e-03, grad_scale: 8.0 2023-04-02 04:17:08,690 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=96794.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:17:17,127 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=96801.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:17:32,588 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=96813.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:17:39,085 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 2023-04-02 04:18:03,059 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=96838.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:18:07,999 INFO [train.py:903] (1/4) Epoch 15, batch 1250, loss[loss=0.2166, simple_loss=0.2824, pruned_loss=0.07537, over 19618.00 frames. ], tot_loss[loss=0.2219, simple_loss=0.2988, pruned_loss=0.07254, over 3836618.68 frames. ], batch size: 50, lr: 5.59e-03, grad_scale: 8.0 2023-04-02 04:18:38,294 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=96867.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:19:05,685 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.585e+02 5.442e+02 6.897e+02 8.528e+02 1.967e+03, threshold=1.379e+03, percent-clipped=7.0 2023-04-02 04:19:09,075 INFO [train.py:903] (1/4) Epoch 15, batch 1300, loss[loss=0.2584, simple_loss=0.3262, pruned_loss=0.09525, over 19321.00 frames. ], tot_loss[loss=0.2232, simple_loss=0.2997, pruned_loss=0.07338, over 3834673.52 frames. ], batch size: 66, lr: 5.59e-03, grad_scale: 8.0 2023-04-02 04:19:17,614 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=96898.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:19:48,615 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=96923.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:20:12,477 INFO [train.py:903] (1/4) Epoch 15, batch 1350, loss[loss=0.2255, simple_loss=0.3085, pruned_loss=0.07125, over 19313.00 frames. ], tot_loss[loss=0.2233, simple_loss=0.2996, pruned_loss=0.07354, over 3839243.43 frames. ], batch size: 66, lr: 5.59e-03, grad_scale: 8.0 2023-04-02 04:21:11,447 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.829e+02 4.520e+02 5.515e+02 7.272e+02 1.782e+03, threshold=1.103e+03, percent-clipped=1.0 2023-04-02 04:21:15,870 INFO [train.py:903] (1/4) Epoch 15, batch 1400, loss[loss=0.2572, simple_loss=0.3256, pruned_loss=0.09438, over 13040.00 frames. ], tot_loss[loss=0.2219, simple_loss=0.2983, pruned_loss=0.0727, over 3828444.39 frames. ], batch size: 137, lr: 5.59e-03, grad_scale: 8.0 2023-04-02 04:21:49,679 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=97018.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:22:19,067 INFO [train.py:903] (1/4) Epoch 15, batch 1450, loss[loss=0.2384, simple_loss=0.3232, pruned_loss=0.07681, over 19675.00 frames. ], tot_loss[loss=0.2224, simple_loss=0.2989, pruned_loss=0.07297, over 3820371.13 frames. ], batch size: 55, lr: 5.59e-03, grad_scale: 8.0 2023-04-02 04:22:20,263 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_rvb from training. Duration: 25.85 2023-04-02 04:22:38,917 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=97057.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:23:09,811 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=97082.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:23:18,570 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.200e+02 5.198e+02 6.417e+02 9.294e+02 1.968e+03, threshold=1.283e+03, percent-clipped=11.0 2023-04-02 04:23:21,918 INFO [train.py:903] (1/4) Epoch 15, batch 1500, loss[loss=0.2336, simple_loss=0.3138, pruned_loss=0.0767, over 19681.00 frames. ], tot_loss[loss=0.2213, simple_loss=0.2977, pruned_loss=0.07247, over 3832923.42 frames. ], batch size: 59, lr: 5.58e-03, grad_scale: 8.0 2023-04-02 04:23:30,901 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.14 vs. limit=2.0 2023-04-02 04:24:01,256 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=97123.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:24:20,465 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=97138.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:24:24,793 INFO [train.py:903] (1/4) Epoch 15, batch 1550, loss[loss=0.1937, simple_loss=0.2702, pruned_loss=0.05858, over 19290.00 frames. ], tot_loss[loss=0.2225, simple_loss=0.2987, pruned_loss=0.07316, over 3819109.78 frames. ], batch size: 44, lr: 5.58e-03, grad_scale: 8.0 2023-04-02 04:24:31,842 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=97148.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:25:22,551 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.982e+02 5.223e+02 6.490e+02 8.468e+02 1.572e+03, threshold=1.298e+03, percent-clipped=7.0 2023-04-02 04:25:26,786 INFO [train.py:903] (1/4) Epoch 15, batch 1600, loss[loss=0.2236, simple_loss=0.3032, pruned_loss=0.07199, over 17363.00 frames. ], tot_loss[loss=0.223, simple_loss=0.2994, pruned_loss=0.07328, over 3821448.23 frames. ], batch size: 101, lr: 5.58e-03, grad_scale: 8.0 2023-04-02 04:25:51,568 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9_rvb from training. Duration: 30.1555625 2023-04-02 04:26:20,094 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.13 vs. limit=2.0 2023-04-02 04:26:28,830 INFO [train.py:903] (1/4) Epoch 15, batch 1650, loss[loss=0.226, simple_loss=0.317, pruned_loss=0.06751, over 19660.00 frames. ], tot_loss[loss=0.223, simple_loss=0.2993, pruned_loss=0.07337, over 3828699.88 frames. ], batch size: 55, lr: 5.58e-03, grad_scale: 8.0 2023-04-02 04:26:42,598 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=97253.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:27:03,853 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.7646, 2.1556, 2.0032, 2.7082, 2.5079, 2.3063, 2.1396, 2.6570], device='cuda:1'), covar=tensor([0.0774, 0.1668, 0.1416, 0.0913, 0.1153, 0.0469, 0.1215, 0.0576], device='cuda:1'), in_proj_covar=tensor([0.0256, 0.0349, 0.0296, 0.0242, 0.0293, 0.0244, 0.0287, 0.0242], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 04:27:14,810 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=97278.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:27:27,487 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.574e+02 5.168e+02 6.563e+02 8.938e+02 1.305e+03, threshold=1.313e+03, percent-clipped=1.0 2023-04-02 04:27:30,909 INFO [train.py:903] (1/4) Epoch 15, batch 1700, loss[loss=0.2371, simple_loss=0.3068, pruned_loss=0.08369, over 19772.00 frames. ], tot_loss[loss=0.223, simple_loss=0.2996, pruned_loss=0.07321, over 3835869.76 frames. ], batch size: 54, lr: 5.58e-03, grad_scale: 8.0 2023-04-02 04:27:49,503 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.41 vs. limit=2.0 2023-04-02 04:28:02,160 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.3546, 3.9334, 2.6968, 3.4991, 0.8491, 3.8497, 3.7115, 3.9013], device='cuda:1'), covar=tensor([0.0722, 0.1185, 0.1928, 0.0863, 0.4146, 0.0735, 0.0766, 0.0863], device='cuda:1'), in_proj_covar=tensor([0.0457, 0.0384, 0.0458, 0.0326, 0.0389, 0.0391, 0.0382, 0.0414], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 04:28:11,292 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 2023-04-02 04:28:32,695 INFO [train.py:903] (1/4) Epoch 15, batch 1750, loss[loss=0.1915, simple_loss=0.2623, pruned_loss=0.06034, over 19776.00 frames. ], tot_loss[loss=0.222, simple_loss=0.2985, pruned_loss=0.07271, over 3829691.47 frames. ], batch size: 47, lr: 5.58e-03, grad_scale: 8.0 2023-04-02 04:28:55,651 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=97360.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:28:57,846 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=97362.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:29:01,596 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=97365.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:29:30,673 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.259e+02 5.096e+02 6.393e+02 8.037e+02 1.428e+03, threshold=1.279e+03, percent-clipped=5.0 2023-04-02 04:29:33,918 INFO [train.py:903] (1/4) Epoch 15, batch 1800, loss[loss=0.2398, simple_loss=0.3183, pruned_loss=0.08071, over 18204.00 frames. ], tot_loss[loss=0.2223, simple_loss=0.2987, pruned_loss=0.07293, over 3821553.05 frames. ], batch size: 83, lr: 5.58e-03, grad_scale: 8.0 2023-04-02 04:29:34,237 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=97392.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:30:03,917 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.3751, 2.9950, 2.2921, 2.3387, 2.1975, 2.5913, 1.0693, 2.1430], device='cuda:1'), covar=tensor([0.0531, 0.0504, 0.0570, 0.0906, 0.0935, 0.0892, 0.1153, 0.0896], device='cuda:1'), in_proj_covar=tensor([0.0342, 0.0340, 0.0337, 0.0366, 0.0438, 0.0365, 0.0319, 0.0325], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 04:30:32,252 WARNING [train.py:1073] (1/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9_rvb from training. Duration: 25.0944375 2023-04-02 04:30:36,841 INFO [train.py:903] (1/4) Epoch 15, batch 1850, loss[loss=0.2547, simple_loss=0.3199, pruned_loss=0.09479, over 19721.00 frames. ], tot_loss[loss=0.2202, simple_loss=0.2968, pruned_loss=0.07178, over 3816195.16 frames. ], batch size: 63, lr: 5.57e-03, grad_scale: 8.0 2023-04-02 04:31:10,807 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9_rvb from training. Duration: 27.8166875 2023-04-02 04:31:21,349 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=97477.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:31:25,848 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.9168, 4.3658, 4.6697, 4.6547, 1.6822, 4.3668, 3.8600, 4.3539], device='cuda:1'), covar=tensor([0.1538, 0.0708, 0.0563, 0.0591, 0.5537, 0.0693, 0.0609, 0.1073], device='cuda:1'), in_proj_covar=tensor([0.0728, 0.0655, 0.0865, 0.0744, 0.0773, 0.0609, 0.0523, 0.0797], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-02 04:31:35,889 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.266e+02 4.535e+02 5.674e+02 7.772e+02 1.771e+03, threshold=1.135e+03, percent-clipped=3.0 2023-04-02 04:31:39,234 INFO [train.py:903] (1/4) Epoch 15, batch 1900, loss[loss=0.1982, simple_loss=0.2778, pruned_loss=0.05933, over 19725.00 frames. ], tot_loss[loss=0.2214, simple_loss=0.2978, pruned_loss=0.07249, over 3806603.80 frames. ], batch size: 51, lr: 5.57e-03, grad_scale: 8.0 2023-04-02 04:31:47,728 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.5551, 4.0667, 4.2186, 4.2376, 1.6360, 3.9939, 3.4544, 3.9449], device='cuda:1'), covar=tensor([0.1424, 0.0727, 0.0598, 0.0614, 0.5007, 0.0715, 0.0642, 0.1024], device='cuda:1'), in_proj_covar=tensor([0.0725, 0.0654, 0.0863, 0.0743, 0.0771, 0.0607, 0.0522, 0.0794], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-02 04:31:58,013 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 2023-04-02 04:32:00,939 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=97509.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:32:02,867 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9_rvb from training. Duration: 27.02225 2023-04-02 04:32:28,310 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 2023-04-02 04:32:32,289 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=97534.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:32:40,693 INFO [train.py:903] (1/4) Epoch 15, batch 1950, loss[loss=0.2269, simple_loss=0.2911, pruned_loss=0.08138, over 19621.00 frames. ], tot_loss[loss=0.2221, simple_loss=0.2984, pruned_loss=0.07297, over 3807044.37 frames. ], batch size: 50, lr: 5.57e-03, grad_scale: 8.0 2023-04-02 04:33:19,888 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=97573.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:33:39,666 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.621e+02 4.980e+02 6.456e+02 8.636e+02 2.349e+03, threshold=1.291e+03, percent-clipped=8.0 2023-04-02 04:33:43,294 INFO [train.py:903] (1/4) Epoch 15, batch 2000, loss[loss=0.1852, simple_loss=0.2591, pruned_loss=0.05563, over 19755.00 frames. ], tot_loss[loss=0.2211, simple_loss=0.2974, pruned_loss=0.07243, over 3802209.92 frames. ], batch size: 46, lr: 5.57e-03, grad_scale: 8.0 2023-04-02 04:34:06,540 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.57 vs. limit=2.0 2023-04-02 04:34:20,870 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=97622.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:34:35,244 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.81 vs. limit=2.0 2023-04-02 04:34:42,264 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_rvb from training. Duration: 26.8349375 2023-04-02 04:34:47,027 INFO [train.py:903] (1/4) Epoch 15, batch 2050, loss[loss=0.2211, simple_loss=0.3058, pruned_loss=0.06817, over 19596.00 frames. ], tot_loss[loss=0.2207, simple_loss=0.2971, pruned_loss=0.07215, over 3807939.10 frames. ], batch size: 61, lr: 5.57e-03, grad_scale: 8.0 2023-04-02 04:35:01,879 WARNING [train.py:1073] (1/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 2023-04-02 04:35:03,049 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 2023-04-02 04:35:23,949 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 2023-04-02 04:35:47,018 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.214e+02 4.998e+02 6.604e+02 7.928e+02 1.987e+03, threshold=1.321e+03, percent-clipped=5.0 2023-04-02 04:35:50,609 INFO [train.py:903] (1/4) Epoch 15, batch 2100, loss[loss=0.2216, simple_loss=0.3017, pruned_loss=0.07076, over 19498.00 frames. ], tot_loss[loss=0.22, simple_loss=0.2966, pruned_loss=0.07171, over 3810901.61 frames. ], batch size: 64, lr: 5.57e-03, grad_scale: 8.0 2023-04-02 04:35:50,923 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=97692.0, num_to_drop=1, layers_to_drop={0} 2023-04-02 04:35:56,735 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.5214, 4.0201, 4.1847, 4.2032, 1.5741, 3.9460, 3.4894, 3.8778], device='cuda:1'), covar=tensor([0.1465, 0.0925, 0.0641, 0.0604, 0.5689, 0.0935, 0.0645, 0.1132], device='cuda:1'), in_proj_covar=tensor([0.0732, 0.0658, 0.0867, 0.0747, 0.0777, 0.0612, 0.0524, 0.0798], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-02 04:36:04,914 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=97704.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:36:11,603 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=97709.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:36:20,739 WARNING [train.py:1073] (1/4) Exclude cut with ID 453-131332-0000-131866_sp0.9_rvb from training. Duration: 25.3333125 2023-04-02 04:36:41,966 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=97733.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:36:43,971 WARNING [train.py:1073] (1/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9_rvb from training. Duration: 26.6166875 2023-04-02 04:36:45,261 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=97736.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:36:46,649 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=97737.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:36:49,900 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1074, 1.3080, 1.7183, 1.1166, 2.5485, 3.3662, 3.1064, 3.6013], device='cuda:1'), covar=tensor([0.1576, 0.3463, 0.2932, 0.2231, 0.0505, 0.0166, 0.0212, 0.0214], device='cuda:1'), in_proj_covar=tensor([0.0261, 0.0305, 0.0331, 0.0253, 0.0225, 0.0169, 0.0208, 0.0221], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 04:36:51,863 INFO [train.py:903] (1/4) Epoch 15, batch 2150, loss[loss=0.2156, simple_loss=0.2944, pruned_loss=0.06841, over 19854.00 frames. ], tot_loss[loss=0.2187, simple_loss=0.2957, pruned_loss=0.07084, over 3822885.75 frames. ], batch size: 52, lr: 5.57e-03, grad_scale: 8.0 2023-04-02 04:37:12,262 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=97758.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:37:49,741 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.357e+02 4.862e+02 6.186e+02 7.185e+02 1.323e+03, threshold=1.237e+03, percent-clipped=1.0 2023-04-02 04:37:54,000 INFO [train.py:903] (1/4) Epoch 15, batch 2200, loss[loss=0.1949, simple_loss=0.2739, pruned_loss=0.05795, over 19722.00 frames. ], tot_loss[loss=0.2187, simple_loss=0.296, pruned_loss=0.07069, over 3831343.84 frames. ], batch size: 51, lr: 5.56e-03, grad_scale: 8.0 2023-04-02 04:37:55,550 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.1050, 2.7837, 2.0125, 2.0649, 1.8237, 2.4296, 0.9180, 1.9289], device='cuda:1'), covar=tensor([0.0494, 0.0456, 0.0571, 0.0893, 0.0933, 0.0840, 0.1064, 0.0877], device='cuda:1'), in_proj_covar=tensor([0.0342, 0.0340, 0.0336, 0.0365, 0.0438, 0.0364, 0.0318, 0.0324], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 04:38:26,774 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.7255, 1.3340, 1.5359, 1.3784, 3.2116, 0.9509, 2.2860, 3.6688], device='cuda:1'), covar=tensor([0.0406, 0.2742, 0.2730, 0.1943, 0.0737, 0.2678, 0.1310, 0.0243], device='cuda:1'), in_proj_covar=tensor([0.0375, 0.0350, 0.0366, 0.0329, 0.0356, 0.0338, 0.0350, 0.0375], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 04:38:28,035 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=97819.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:38:33,740 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=97824.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:38:57,678 INFO [train.py:903] (1/4) Epoch 15, batch 2250, loss[loss=0.2435, simple_loss=0.3199, pruned_loss=0.0835, over 19673.00 frames. ], tot_loss[loss=0.2195, simple_loss=0.2961, pruned_loss=0.07146, over 3824107.76 frames. ], batch size: 59, lr: 5.56e-03, grad_scale: 8.0 2023-04-02 04:39:09,185 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=97851.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:39:56,865 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.382e+02 4.977e+02 6.292e+02 8.077e+02 1.831e+03, threshold=1.258e+03, percent-clipped=5.0 2023-04-02 04:40:00,318 INFO [train.py:903] (1/4) Epoch 15, batch 2300, loss[loss=0.238, simple_loss=0.3097, pruned_loss=0.08312, over 19699.00 frames. ], tot_loss[loss=0.2191, simple_loss=0.2956, pruned_loss=0.07128, over 3818957.80 frames. ], batch size: 59, lr: 5.56e-03, grad_scale: 8.0 2023-04-02 04:40:12,675 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_rvb from training. Duration: 26.205 2023-04-02 04:40:16,279 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=97906.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:40:29,876 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=97917.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:40:37,935 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.8705, 1.9116, 2.1119, 2.5228, 1.9119, 2.4445, 2.2559, 2.0520], device='cuda:1'), covar=tensor([0.3540, 0.2993, 0.1464, 0.1828, 0.3232, 0.1443, 0.3548, 0.2482], device='cuda:1'), in_proj_covar=tensor([0.0834, 0.0871, 0.0671, 0.0903, 0.0817, 0.0750, 0.0806, 0.0735], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 04:41:01,537 INFO [train.py:903] (1/4) Epoch 15, batch 2350, loss[loss=0.2088, simple_loss=0.2974, pruned_loss=0.06007, over 19763.00 frames. ], tot_loss[loss=0.2204, simple_loss=0.2973, pruned_loss=0.07176, over 3821132.10 frames. ], batch size: 54, lr: 5.56e-03, grad_scale: 8.0 2023-04-02 04:41:44,937 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_rvb from training. Duration: 25.775 2023-04-02 04:41:58,281 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.032e+02 5.416e+02 6.244e+02 7.823e+02 1.587e+03, threshold=1.249e+03, percent-clipped=4.0 2023-04-02 04:41:58,343 WARNING [train.py:1073] (1/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9_rvb from training. Duration: 25.45 2023-04-02 04:42:02,750 INFO [train.py:903] (1/4) Epoch 15, batch 2400, loss[loss=0.1718, simple_loss=0.2507, pruned_loss=0.04649, over 19762.00 frames. ], tot_loss[loss=0.22, simple_loss=0.2968, pruned_loss=0.0716, over 3819364.10 frames. ], batch size: 47, lr: 5.56e-03, grad_scale: 8.0 2023-04-02 04:42:04,391 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=97993.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:42:36,274 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=98018.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:42:53,715 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=98032.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:42:58,101 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=98036.0, num_to_drop=1, layers_to_drop={0} 2023-04-02 04:43:05,611 INFO [train.py:903] (1/4) Epoch 15, batch 2450, loss[loss=0.1831, simple_loss=0.262, pruned_loss=0.05212, over 19443.00 frames. ], tot_loss[loss=0.2191, simple_loss=0.296, pruned_loss=0.07106, over 3822201.12 frames. ], batch size: 48, lr: 5.56e-03, grad_scale: 8.0 2023-04-02 04:43:47,635 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=98075.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:43:54,226 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=98080.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:44:05,272 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.133e+02 5.305e+02 6.204e+02 8.091e+02 1.870e+03, threshold=1.241e+03, percent-clipped=5.0 2023-04-02 04:44:09,724 INFO [train.py:903] (1/4) Epoch 15, batch 2500, loss[loss=0.2567, simple_loss=0.3278, pruned_loss=0.09285, over 19675.00 frames. ], tot_loss[loss=0.2192, simple_loss=0.2964, pruned_loss=0.07101, over 3815479.75 frames. ], batch size: 59, lr: 5.56e-03, grad_scale: 8.0 2023-04-02 04:44:19,640 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=98100.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:44:25,570 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=98105.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:44:27,999 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=98107.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:44:46,010 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.81 vs. limit=2.0 2023-04-02 04:44:51,235 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=98124.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:45:00,593 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=98132.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:45:12,196 INFO [train.py:903] (1/4) Epoch 15, batch 2550, loss[loss=0.2408, simple_loss=0.3217, pruned_loss=0.07993, over 19751.00 frames. ], tot_loss[loss=0.2203, simple_loss=0.2974, pruned_loss=0.07158, over 3813342.50 frames. ], batch size: 51, lr: 5.55e-03, grad_scale: 8.0 2023-04-02 04:45:12,519 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.4681, 3.9594, 4.1569, 4.1338, 1.7480, 3.8807, 3.3788, 3.8378], device='cuda:1'), covar=tensor([0.1673, 0.1028, 0.0673, 0.0727, 0.5165, 0.0911, 0.0734, 0.1191], device='cuda:1'), in_proj_covar=tensor([0.0724, 0.0654, 0.0863, 0.0742, 0.0770, 0.0609, 0.0521, 0.0788], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-02 04:45:23,242 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=98151.0, num_to_drop=1, layers_to_drop={0} 2023-04-02 04:45:52,140 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([5.9262, 5.2894, 2.9925, 4.6140, 1.3308, 5.2699, 5.1904, 5.3731], device='cuda:1'), covar=tensor([0.0358, 0.0867, 0.2016, 0.0720, 0.3892, 0.0683, 0.0781, 0.0953], device='cuda:1'), in_proj_covar=tensor([0.0462, 0.0382, 0.0461, 0.0326, 0.0392, 0.0395, 0.0383, 0.0419], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 04:46:07,120 WARNING [train.py:1073] (1/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 2023-04-02 04:46:10,538 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.981e+02 5.247e+02 6.339e+02 8.638e+02 2.352e+03, threshold=1.268e+03, percent-clipped=5.0 2023-04-02 04:46:14,038 INFO [train.py:903] (1/4) Epoch 15, batch 2600, loss[loss=0.1889, simple_loss=0.2721, pruned_loss=0.05288, over 19809.00 frames. ], tot_loss[loss=0.2195, simple_loss=0.2965, pruned_loss=0.07123, over 3811902.69 frames. ], batch size: 49, lr: 5.55e-03, grad_scale: 8.0 2023-04-02 04:47:18,022 INFO [train.py:903] (1/4) Epoch 15, batch 2650, loss[loss=0.2485, simple_loss=0.3302, pruned_loss=0.08342, over 19543.00 frames. ], tot_loss[loss=0.2204, simple_loss=0.2974, pruned_loss=0.07169, over 3817596.60 frames. ], batch size: 56, lr: 5.55e-03, grad_scale: 8.0 2023-04-02 04:47:28,827 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=98250.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:47:39,871 WARNING [train.py:1073] (1/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9_rvb from training. Duration: 27.25 2023-04-02 04:48:17,340 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=98288.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:48:18,027 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.302e+02 4.990e+02 6.118e+02 7.575e+02 1.335e+03, threshold=1.224e+03, percent-clipped=1.0 2023-04-02 04:48:21,600 INFO [train.py:903] (1/4) Epoch 15, batch 2700, loss[loss=0.2128, simple_loss=0.2927, pruned_loss=0.06642, over 19603.00 frames. ], tot_loss[loss=0.2207, simple_loss=0.2975, pruned_loss=0.07194, over 3815032.47 frames. ], batch size: 57, lr: 5.55e-03, grad_scale: 8.0 2023-04-02 04:48:29,611 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.7689, 2.2534, 2.2491, 2.9111, 2.6243, 2.5165, 2.0007, 3.1030], device='cuda:1'), covar=tensor([0.0789, 0.1635, 0.1285, 0.0906, 0.1278, 0.0421, 0.1308, 0.0490], device='cuda:1'), in_proj_covar=tensor([0.0260, 0.0351, 0.0297, 0.0244, 0.0295, 0.0245, 0.0290, 0.0247], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 04:48:47,474 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=98313.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:49:24,161 INFO [train.py:903] (1/4) Epoch 15, batch 2750, loss[loss=0.1912, simple_loss=0.2789, pruned_loss=0.05175, over 19660.00 frames. ], tot_loss[loss=0.2208, simple_loss=0.2974, pruned_loss=0.07209, over 3814615.14 frames. ], batch size: 53, lr: 5.55e-03, grad_scale: 8.0 2023-04-02 04:49:46,583 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.8724, 1.4165, 1.5488, 1.7252, 3.4068, 1.1887, 2.4354, 3.8203], device='cuda:1'), covar=tensor([0.0402, 0.2825, 0.2862, 0.1738, 0.0709, 0.2433, 0.1219, 0.0249], device='cuda:1'), in_proj_covar=tensor([0.0375, 0.0350, 0.0368, 0.0329, 0.0357, 0.0339, 0.0347, 0.0375], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 04:49:54,681 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=98365.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:50:23,802 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.213e+02 5.173e+02 6.107e+02 7.991e+02 1.431e+03, threshold=1.221e+03, percent-clipped=3.0 2023-04-02 04:50:27,282 INFO [train.py:903] (1/4) Epoch 15, batch 2800, loss[loss=0.2262, simple_loss=0.3123, pruned_loss=0.07007, over 19685.00 frames. ], tot_loss[loss=0.2199, simple_loss=0.2967, pruned_loss=0.07158, over 3822349.63 frames. ], batch size: 59, lr: 5.55e-03, grad_scale: 8.0 2023-04-02 04:50:48,801 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=98407.0, num_to_drop=1, layers_to_drop={0} 2023-04-02 04:51:08,186 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9256, 2.0152, 2.2742, 2.6793, 1.9151, 2.6011, 2.4553, 2.0601], device='cuda:1'), covar=tensor([0.3937, 0.3479, 0.1559, 0.2014, 0.3731, 0.1724, 0.3947, 0.3034], device='cuda:1'), in_proj_covar=tensor([0.0829, 0.0875, 0.0670, 0.0902, 0.0816, 0.0748, 0.0807, 0.0737], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 04:51:13,925 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=98428.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:51:18,527 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=98432.0, num_to_drop=1, layers_to_drop={0} 2023-04-02 04:51:30,816 INFO [train.py:903] (1/4) Epoch 15, batch 2850, loss[loss=0.2102, simple_loss=0.2891, pruned_loss=0.06566, over 19730.00 frames. ], tot_loss[loss=0.2191, simple_loss=0.2957, pruned_loss=0.07126, over 3816869.70 frames. ], batch size: 51, lr: 5.55e-03, grad_scale: 8.0 2023-04-02 04:51:44,200 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.50 vs. limit=2.0 2023-04-02 04:52:03,918 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=98468.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:52:30,661 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.993e+02 5.164e+02 6.295e+02 8.927e+02 2.262e+03, threshold=1.259e+03, percent-clipped=4.0 2023-04-02 04:52:34,167 INFO [train.py:903] (1/4) Epoch 15, batch 2900, loss[loss=0.23, simple_loss=0.3048, pruned_loss=0.07764, over 17394.00 frames. ], tot_loss[loss=0.2203, simple_loss=0.2966, pruned_loss=0.07193, over 3801512.31 frames. ], batch size: 101, lr: 5.54e-03, grad_scale: 8.0 2023-04-02 04:52:35,432 WARNING [train.py:1073] (1/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 2023-04-02 04:52:52,023 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0517, 1.8695, 1.6271, 2.2154, 1.9151, 1.8214, 1.7285, 2.0172], device='cuda:1'), covar=tensor([0.1005, 0.1480, 0.1446, 0.0871, 0.1269, 0.0506, 0.1296, 0.0673], device='cuda:1'), in_proj_covar=tensor([0.0260, 0.0352, 0.0297, 0.0245, 0.0295, 0.0246, 0.0290, 0.0248], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 04:53:36,705 INFO [train.py:903] (1/4) Epoch 15, batch 2950, loss[loss=0.2546, simple_loss=0.3175, pruned_loss=0.09589, over 19671.00 frames. ], tot_loss[loss=0.2211, simple_loss=0.2972, pruned_loss=0.07248, over 3813085.58 frames. ], batch size: 58, lr: 5.54e-03, grad_scale: 8.0 2023-04-02 04:53:47,983 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.2917, 2.1516, 1.8838, 1.7916, 1.6300, 1.8776, 0.4931, 1.1402], device='cuda:1'), covar=tensor([0.0500, 0.0484, 0.0388, 0.0669, 0.0977, 0.0651, 0.1075, 0.0862], device='cuda:1'), in_proj_covar=tensor([0.0344, 0.0342, 0.0340, 0.0369, 0.0442, 0.0368, 0.0320, 0.0328], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 04:54:00,568 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=98561.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:54:29,176 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=98583.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:54:35,382 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.243e+02 5.043e+02 6.244e+02 8.249e+02 2.456e+03, threshold=1.249e+03, percent-clipped=2.0 2023-04-02 04:54:38,823 INFO [train.py:903] (1/4) Epoch 15, batch 3000, loss[loss=0.1872, simple_loss=0.2644, pruned_loss=0.05503, over 19332.00 frames. ], tot_loss[loss=0.2223, simple_loss=0.2979, pruned_loss=0.07335, over 3797184.85 frames. ], batch size: 47, lr: 5.54e-03, grad_scale: 8.0 2023-04-02 04:54:38,824 INFO [train.py:928] (1/4) Computing validation loss 2023-04-02 04:54:51,336 INFO [train.py:937] (1/4) Epoch 15, validation: loss=0.1735, simple_loss=0.2738, pruned_loss=0.0366, over 944034.00 frames. 2023-04-02 04:54:51,337 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18821MB 2023-04-02 04:54:53,545 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_rvb from training. Duration: 29.735 2023-04-02 04:55:28,729 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=98621.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:55:52,975 INFO [train.py:903] (1/4) Epoch 15, batch 3050, loss[loss=0.2047, simple_loss=0.2883, pruned_loss=0.06051, over 18076.00 frames. ], tot_loss[loss=0.2219, simple_loss=0.2977, pruned_loss=0.07301, over 3811254.61 frames. ], batch size: 83, lr: 5.54e-03, grad_scale: 8.0 2023-04-02 04:55:57,893 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=98646.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:56:51,966 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.865e+02 5.607e+02 7.054e+02 9.171e+02 2.046e+03, threshold=1.411e+03, percent-clipped=6.0 2023-04-02 04:56:54,314 INFO [train.py:903] (1/4) Epoch 15, batch 3100, loss[loss=0.264, simple_loss=0.3432, pruned_loss=0.09241, over 19752.00 frames. ], tot_loss[loss=0.2217, simple_loss=0.2978, pruned_loss=0.07279, over 3815367.01 frames. ], batch size: 63, lr: 5.54e-03, grad_scale: 8.0 2023-04-02 04:57:58,288 INFO [train.py:903] (1/4) Epoch 15, batch 3150, loss[loss=0.2076, simple_loss=0.2976, pruned_loss=0.05883, over 19673.00 frames. ], tot_loss[loss=0.2219, simple_loss=0.298, pruned_loss=0.07293, over 3819939.63 frames. ], batch size: 59, lr: 5.54e-03, grad_scale: 8.0 2023-04-02 04:58:26,300 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 2023-04-02 04:58:31,122 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.0644, 4.4610, 4.7872, 4.7850, 1.7988, 4.5064, 4.0050, 4.4813], device='cuda:1'), covar=tensor([0.1486, 0.0711, 0.0535, 0.0558, 0.5329, 0.0638, 0.0581, 0.1019], device='cuda:1'), in_proj_covar=tensor([0.0725, 0.0656, 0.0860, 0.0744, 0.0774, 0.0609, 0.0517, 0.0795], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-02 04:58:34,635 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=98772.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 04:58:58,670 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.173e+02 5.220e+02 6.366e+02 8.908e+02 1.802e+03, threshold=1.273e+03, percent-clipped=4.0 2023-04-02 04:59:01,103 INFO [train.py:903] (1/4) Epoch 15, batch 3200, loss[loss=0.2569, simple_loss=0.323, pruned_loss=0.0954, over 19534.00 frames. ], tot_loss[loss=0.2203, simple_loss=0.2965, pruned_loss=0.07199, over 3831828.53 frames. ], batch size: 64, lr: 5.54e-03, grad_scale: 8.0 2023-04-02 04:59:33,978 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.0066, 1.2959, 1.5664, 1.1795, 2.6199, 3.8054, 3.5146, 3.9561], device='cuda:1'), covar=tensor([0.1766, 0.3579, 0.3297, 0.2295, 0.0613, 0.0165, 0.0195, 0.0194], device='cuda:1'), in_proj_covar=tensor([0.0260, 0.0303, 0.0330, 0.0251, 0.0223, 0.0168, 0.0206, 0.0222], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 04:59:59,407 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=98839.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 05:00:02,401 INFO [train.py:903] (1/4) Epoch 15, batch 3250, loss[loss=0.2417, simple_loss=0.3305, pruned_loss=0.07645, over 18159.00 frames. ], tot_loss[loss=0.2208, simple_loss=0.2972, pruned_loss=0.07216, over 3827285.64 frames. ], batch size: 83, lr: 5.53e-03, grad_scale: 8.0 2023-04-02 05:00:29,951 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=98864.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 05:00:56,791 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=98887.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 05:00:59,973 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.355e+02 4.885e+02 6.090e+02 7.322e+02 1.846e+03, threshold=1.218e+03, percent-clipped=3.0 2023-04-02 05:01:02,394 INFO [train.py:903] (1/4) Epoch 15, batch 3300, loss[loss=0.1756, simple_loss=0.252, pruned_loss=0.04963, over 19362.00 frames. ], tot_loss[loss=0.2214, simple_loss=0.2977, pruned_loss=0.0725, over 3813186.76 frames. ], batch size: 47, lr: 5.53e-03, grad_scale: 8.0 2023-04-02 05:01:08,196 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 2023-04-02 05:01:20,961 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=98905.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 05:01:25,246 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=98908.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 05:01:30,541 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=98913.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 05:01:41,396 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5882, 1.6893, 1.8625, 2.0006, 1.4434, 1.8611, 1.9606, 1.7597], device='cuda:1'), covar=tensor([0.3898, 0.3126, 0.1642, 0.1910, 0.3364, 0.1758, 0.4372, 0.2960], device='cuda:1'), in_proj_covar=tensor([0.0831, 0.0873, 0.0669, 0.0903, 0.0815, 0.0748, 0.0807, 0.0735], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 05:02:07,324 INFO [train.py:903] (1/4) Epoch 15, batch 3350, loss[loss=0.2194, simple_loss=0.3024, pruned_loss=0.0682, over 19597.00 frames. ], tot_loss[loss=0.2217, simple_loss=0.2982, pruned_loss=0.07263, over 3799148.95 frames. ], batch size: 57, lr: 5.53e-03, grad_scale: 8.0 2023-04-02 05:02:09,397 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 2023-04-02 05:03:06,853 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=98989.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 05:03:07,561 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.074e+02 5.628e+02 7.317e+02 9.748e+02 2.071e+03, threshold=1.463e+03, percent-clipped=8.0 2023-04-02 05:03:09,815 INFO [train.py:903] (1/4) Epoch 15, batch 3400, loss[loss=0.211, simple_loss=0.2949, pruned_loss=0.06351, over 19595.00 frames. ], tot_loss[loss=0.2213, simple_loss=0.2976, pruned_loss=0.07246, over 3817805.71 frames. ], batch size: 57, lr: 5.53e-03, grad_scale: 8.0 2023-04-02 05:03:44,058 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=99020.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 05:04:10,160 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5253, 2.2732, 1.5958, 1.6852, 2.0798, 1.3395, 1.4649, 1.8437], device='cuda:1'), covar=tensor([0.0944, 0.0648, 0.0979, 0.0573, 0.0509, 0.1093, 0.0642, 0.0514], device='cuda:1'), in_proj_covar=tensor([0.0296, 0.0307, 0.0324, 0.0251, 0.0239, 0.0327, 0.0293, 0.0267], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 05:04:10,819 INFO [train.py:903] (1/4) Epoch 15, batch 3450, loss[loss=0.2653, simple_loss=0.3312, pruned_loss=0.09974, over 19395.00 frames. ], tot_loss[loss=0.221, simple_loss=0.2975, pruned_loss=0.07227, over 3818908.75 frames. ], batch size: 70, lr: 5.53e-03, grad_scale: 4.0 2023-04-02 05:04:14,058 WARNING [train.py:1073] (1/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9_rvb from training. Duration: 25.2444375 2023-04-02 05:04:34,759 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2482, 1.1395, 1.2069, 1.3884, 1.1001, 1.3436, 1.3407, 1.2916], device='cuda:1'), covar=tensor([0.0917, 0.1061, 0.1099, 0.0672, 0.0846, 0.0831, 0.0873, 0.0755], device='cuda:1'), in_proj_covar=tensor([0.0209, 0.0221, 0.0221, 0.0240, 0.0225, 0.0207, 0.0189, 0.0202], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-02 05:04:45,794 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99069.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 05:05:11,143 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.718e+02 4.946e+02 5.886e+02 7.201e+02 1.354e+03, threshold=1.177e+03, percent-clipped=0.0 2023-04-02 05:05:12,313 INFO [train.py:903] (1/4) Epoch 15, batch 3500, loss[loss=0.2109, simple_loss=0.2872, pruned_loss=0.06731, over 19455.00 frames. ], tot_loss[loss=0.22, simple_loss=0.2966, pruned_loss=0.0717, over 3831630.59 frames. ], batch size: 49, lr: 5.53e-03, grad_scale: 4.0 2023-04-02 05:06:15,625 INFO [train.py:903] (1/4) Epoch 15, batch 3550, loss[loss=0.1931, simple_loss=0.2764, pruned_loss=0.05494, over 19565.00 frames. ], tot_loss[loss=0.2205, simple_loss=0.2966, pruned_loss=0.07224, over 3819255.55 frames. ], batch size: 52, lr: 5.53e-03, grad_scale: 4.0 2023-04-02 05:06:18,419 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=99143.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 05:06:48,304 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=99168.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 05:06:53,014 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6231, 1.4207, 1.4078, 2.0434, 1.6108, 1.9206, 1.9813, 1.6660], device='cuda:1'), covar=tensor([0.0818, 0.0949, 0.1023, 0.0779, 0.0830, 0.0712, 0.0832, 0.0687], device='cuda:1'), in_proj_covar=tensor([0.0209, 0.0221, 0.0221, 0.0240, 0.0226, 0.0208, 0.0189, 0.0202], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-02 05:07:18,042 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.289e+02 4.734e+02 6.110e+02 7.809e+02 1.736e+03, threshold=1.222e+03, percent-clipped=8.0 2023-04-02 05:07:19,077 INFO [train.py:903] (1/4) Epoch 15, batch 3600, loss[loss=0.2188, simple_loss=0.2871, pruned_loss=0.07532, over 19636.00 frames. ], tot_loss[loss=0.2208, simple_loss=0.2972, pruned_loss=0.0722, over 3833959.07 frames. ], batch size: 50, lr: 5.52e-03, grad_scale: 8.0 2023-04-02 05:08:20,334 INFO [train.py:903] (1/4) Epoch 15, batch 3650, loss[loss=0.2126, simple_loss=0.2962, pruned_loss=0.0645, over 18902.00 frames. ], tot_loss[loss=0.2215, simple_loss=0.2978, pruned_loss=0.07257, over 3826165.43 frames. ], batch size: 74, lr: 5.52e-03, grad_scale: 8.0 2023-04-02 05:08:26,432 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99247.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 05:08:32,040 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=99252.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 05:08:37,706 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=99257.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 05:09:03,806 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=99276.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 05:09:15,281 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99286.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 05:09:20,758 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.641e+02 5.562e+02 6.527e+02 8.007e+02 1.277e+03, threshold=1.305e+03, percent-clipped=3.0 2023-04-02 05:09:21,908 INFO [train.py:903] (1/4) Epoch 15, batch 3700, loss[loss=0.2709, simple_loss=0.3391, pruned_loss=0.1014, over 19661.00 frames. ], tot_loss[loss=0.223, simple_loss=0.2992, pruned_loss=0.07341, over 3819606.53 frames. ], batch size: 60, lr: 5.52e-03, grad_scale: 8.0 2023-04-02 05:09:32,857 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=99301.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 05:10:13,724 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=99333.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 05:10:24,822 INFO [train.py:903] (1/4) Epoch 15, batch 3750, loss[loss=0.2138, simple_loss=0.3004, pruned_loss=0.06356, over 19491.00 frames. ], tot_loss[loss=0.2226, simple_loss=0.299, pruned_loss=0.07317, over 3820629.45 frames. ], batch size: 64, lr: 5.52e-03, grad_scale: 8.0 2023-04-02 05:10:25,120 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99342.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 05:10:40,376 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.6296, 2.0624, 2.2692, 2.7443, 2.4581, 2.3200, 2.0275, 2.6875], device='cuda:1'), covar=tensor([0.0899, 0.1833, 0.1294, 0.1030, 0.1333, 0.0487, 0.1257, 0.0621], device='cuda:1'), in_proj_covar=tensor([0.0262, 0.0354, 0.0297, 0.0244, 0.0298, 0.0249, 0.0292, 0.0248], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 05:10:56,150 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=99367.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 05:11:01,984 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=99372.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 05:11:26,374 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.684e+02 5.640e+02 6.861e+02 8.808e+02 1.909e+03, threshold=1.372e+03, percent-clipped=3.0 2023-04-02 05:11:28,560 INFO [train.py:903] (1/4) Epoch 15, batch 3800, loss[loss=0.217, simple_loss=0.2908, pruned_loss=0.0716, over 19569.00 frames. ], tot_loss[loss=0.2235, simple_loss=0.2993, pruned_loss=0.07389, over 3807643.56 frames. ], batch size: 52, lr: 5.52e-03, grad_scale: 8.0 2023-04-02 05:11:41,815 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.3471, 2.1444, 1.9166, 1.8990, 1.6317, 1.8003, 0.4950, 1.2406], device='cuda:1'), covar=tensor([0.0467, 0.0524, 0.0451, 0.0690, 0.1022, 0.0814, 0.1185, 0.0900], device='cuda:1'), in_proj_covar=tensor([0.0346, 0.0346, 0.0341, 0.0373, 0.0444, 0.0369, 0.0322, 0.0330], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 05:11:53,180 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=99413.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 05:11:58,869 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9_rvb from training. Duration: 29.1166875 2023-04-02 05:12:26,312 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99438.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 05:12:30,510 INFO [train.py:903] (1/4) Epoch 15, batch 3850, loss[loss=0.2233, simple_loss=0.3108, pruned_loss=0.06787, over 19614.00 frames. ], tot_loss[loss=0.2231, simple_loss=0.2991, pruned_loss=0.07354, over 3801078.37 frames. ], batch size: 57, lr: 5.52e-03, grad_scale: 4.0 2023-04-02 05:12:37,667 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=99448.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 05:12:44,235 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 2023-04-02 05:13:25,909 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99486.0, num_to_drop=1, layers_to_drop={1} 2023-04-02 05:13:32,483 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.342e+02 4.923e+02 6.054e+02 7.410e+02 1.518e+03, threshold=1.211e+03, percent-clipped=1.0 2023-04-02 05:13:32,501 INFO [train.py:903] (1/4) Epoch 15, batch 3900, loss[loss=0.1861, simple_loss=0.2603, pruned_loss=0.05593, over 19853.00 frames. ], tot_loss[loss=0.222, simple_loss=0.2982, pruned_loss=0.07293, over 3800265.64 frames. ], batch size: 52, lr: 5.52e-03, grad_scale: 4.0 2023-04-02 05:14:18,206 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=99528.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 05:14:33,478 INFO [train.py:903] (1/4) Epoch 15, batch 3950, loss[loss=0.23, simple_loss=0.3077, pruned_loss=0.07612, over 19776.00 frames. ], tot_loss[loss=0.2214, simple_loss=0.2976, pruned_loss=0.07261, over 3815213.45 frames. ], batch size: 56, lr: 5.52e-03, grad_scale: 4.0 2023-04-02 05:14:41,168 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 2023-04-02 05:15:28,304 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99586.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 05:15:35,962 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=99591.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 05:15:36,880 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.680e+02 5.329e+02 6.331e+02 8.679e+02 1.427e+03, threshold=1.266e+03, percent-clipped=6.0 2023-04-02 05:15:36,899 INFO [train.py:903] (1/4) Epoch 15, batch 4000, loss[loss=0.2646, simple_loss=0.3317, pruned_loss=0.09874, over 19088.00 frames. ], tot_loss[loss=0.2229, simple_loss=0.2991, pruned_loss=0.07336, over 3808089.18 frames. ], batch size: 69, lr: 5.51e-03, grad_scale: 8.0 2023-04-02 05:15:54,644 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99607.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 05:16:14,225 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=99623.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 05:16:20,951 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=99628.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 05:16:23,025 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1_rvb from training. Duration: 0.7545625 2023-04-02 05:16:23,146 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=99630.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 05:16:30,447 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0234, 3.1657, 1.8523, 1.8842, 2.7802, 1.5632, 1.4332, 1.9871], device='cuda:1'), covar=tensor([0.1219, 0.0599, 0.0955, 0.0889, 0.0543, 0.1208, 0.0894, 0.0660], device='cuda:1'), in_proj_covar=tensor([0.0292, 0.0305, 0.0321, 0.0250, 0.0239, 0.0325, 0.0290, 0.0266], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 05:16:38,158 INFO [train.py:903] (1/4) Epoch 15, batch 4050, loss[loss=0.2257, simple_loss=0.305, pruned_loss=0.07319, over 19541.00 frames. ], tot_loss[loss=0.2219, simple_loss=0.2984, pruned_loss=0.07269, over 3813663.07 frames. ], batch size: 56, lr: 5.51e-03, grad_scale: 4.0 2023-04-02 05:16:45,324 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=99648.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 05:16:50,828 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=99653.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 05:17:32,691 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=99686.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 05:17:39,405 INFO [train.py:903] (1/4) Epoch 15, batch 4100, loss[loss=0.2636, simple_loss=0.3377, pruned_loss=0.09477, over 19602.00 frames. ], tot_loss[loss=0.2208, simple_loss=0.2971, pruned_loss=0.07222, over 3820075.11 frames. ], batch size: 57, lr: 5.51e-03, grad_scale: 4.0 2023-04-02 05:17:40,546 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.802e+02 5.566e+02 7.429e+02 9.161e+02 2.166e+03, threshold=1.486e+03, percent-clipped=8.0 2023-04-02 05:17:46,992 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99698.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 05:17:54,944 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=99704.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 05:17:58,070 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=99706.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 05:18:14,908 WARNING [train.py:1073] (1/4) Exclude cut with ID 497-129325-0061-9566_sp1.1_rvb from training. Duration: 0.97725 2023-04-02 05:18:25,800 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.5515, 4.0829, 4.2679, 4.2543, 1.5232, 3.9943, 3.4684, 3.9549], device='cuda:1'), covar=tensor([0.1493, 0.0778, 0.0588, 0.0617, 0.5547, 0.0828, 0.0667, 0.1081], device='cuda:1'), in_proj_covar=tensor([0.0725, 0.0657, 0.0860, 0.0736, 0.0771, 0.0609, 0.0519, 0.0797], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-02 05:18:27,081 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=99729.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 05:18:43,271 INFO [train.py:903] (1/4) Epoch 15, batch 4150, loss[loss=0.1948, simple_loss=0.2776, pruned_loss=0.05598, over 19604.00 frames. ], tot_loss[loss=0.2199, simple_loss=0.2969, pruned_loss=0.07142, over 3822839.01 frames. ], batch size: 50, lr: 5.51e-03, grad_scale: 4.0 2023-04-02 05:18:47,063 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=99745.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 05:19:11,154 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.0515, 1.1945, 1.7040, 1.0187, 2.4212, 3.0256, 2.7422, 3.2151], device='cuda:1'), covar=tensor([0.1612, 0.3661, 0.3038, 0.2328, 0.0518, 0.0187, 0.0265, 0.0270], device='cuda:1'), in_proj_covar=tensor([0.0260, 0.0304, 0.0332, 0.0252, 0.0226, 0.0169, 0.0207, 0.0224], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 05:19:13,809 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 2023-04-02 05:19:31,718 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=99782.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 05:19:35,410 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=99784.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 05:19:45,181 INFO [train.py:903] (1/4) Epoch 15, batch 4200, loss[loss=0.1763, simple_loss=0.2529, pruned_loss=0.04987, over 18644.00 frames. ], tot_loss[loss=0.2197, simple_loss=0.2966, pruned_loss=0.07139, over 3826202.59 frames. ], batch size: 41, lr: 5.51e-03, grad_scale: 4.0 2023-04-02 05:19:47,437 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.476e+02 5.549e+02 6.772e+02 8.720e+02 1.402e+03, threshold=1.354e+03, percent-clipped=0.0 2023-04-02 05:19:50,955 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 2023-04-02 05:19:57,147 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=99801.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 05:20:06,407 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=99809.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 05:20:32,496 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=99830.0, num_to_drop=1, layers_to_drop={0} 2023-04-02 05:20:47,373 INFO [train.py:903] (1/4) Epoch 15, batch 4250, loss[loss=0.3191, simple_loss=0.3796, pruned_loss=0.1293, over 19623.00 frames. ], tot_loss[loss=0.2189, simple_loss=0.2957, pruned_loss=0.07107, over 3836732.47 frames. ], batch size: 61, lr: 5.51e-03, grad_scale: 4.0 2023-04-02 05:21:03,965 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9_rvb from training. Duration: 29.816625 2023-04-02 05:21:15,068 WARNING [train.py:1073] (1/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 2023-04-02 05:21:47,822 INFO [train.py:903] (1/4) Epoch 15, batch 4300, loss[loss=0.2401, simple_loss=0.3114, pruned_loss=0.08441, over 19661.00 frames. ], tot_loss[loss=0.2196, simple_loss=0.2966, pruned_loss=0.07132, over 3844971.83 frames. ], batch size: 55, lr: 5.51e-03, grad_scale: 4.0 2023-04-02 05:21:48,970 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.369e+02 5.517e+02 6.494e+02 8.260e+02 1.741e+03, threshold=1.299e+03, percent-clipped=4.0 2023-04-02 05:21:53,749 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=99897.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 05:22:35,370 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=99930.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 05:22:41,699 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 2023-04-02 05:22:48,542 INFO [train.py:903] (1/4) Epoch 15, batch 4350, loss[loss=0.2118, simple_loss=0.2943, pruned_loss=0.06463, over 19795.00 frames. ], tot_loss[loss=0.2209, simple_loss=0.2973, pruned_loss=0.07226, over 3824639.06 frames. ], batch size: 56, lr: 5.50e-03, grad_scale: 4.0 2023-04-02 05:22:52,432 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=99945.0, num_to_drop=1, layers_to_drop={1} 2023-04-02 05:23:01,033 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=99951.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 05:23:15,990 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=99962.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 05:23:45,440 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=99987.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 05:23:51,927 INFO [train.py:903] (1/4) Epoch 15, batch 4400, loss[loss=0.218, simple_loss=0.2978, pruned_loss=0.06904, over 19687.00 frames. ], tot_loss[loss=0.2194, simple_loss=0.2961, pruned_loss=0.07133, over 3836607.88 frames. ], batch size: 59, lr: 5.50e-03, grad_scale: 8.0 2023-04-02 05:23:53,157 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.223e+02 4.844e+02 5.787e+02 7.470e+02 1.170e+03, threshold=1.157e+03, percent-clipped=0.0 2023-04-02 05:23:56,390 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.1794, 1.3015, 1.6187, 1.8074, 3.5452, 1.3719, 2.8197, 3.8014], device='cuda:1'), covar=tensor([0.0626, 0.3488, 0.3273, 0.2218, 0.1046, 0.2971, 0.1484, 0.0429], device='cuda:1'), in_proj_covar=tensor([0.0379, 0.0349, 0.0370, 0.0329, 0.0356, 0.0339, 0.0350, 0.0372], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 05:24:06,750 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=100001.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 05:24:23,724 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_rvb from training. Duration: 25.285 2023-04-02 05:24:31,846 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 2023-04-02 05:24:34,564 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=100025.0, num_to_drop=1, layers_to_drop={1} 2023-04-02 05:24:35,770 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=100026.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 05:24:47,757 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.92 vs. limit=5.0 2023-04-02 05:24:57,046 INFO [train.py:903] (1/4) Epoch 15, batch 4450, loss[loss=0.248, simple_loss=0.3273, pruned_loss=0.08437, over 19299.00 frames. ], tot_loss[loss=0.22, simple_loss=0.2966, pruned_loss=0.07168, over 3808560.98 frames. ], batch size: 66, lr: 5.50e-03, grad_scale: 8.0 2023-04-02 05:24:57,237 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=100042.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 05:25:00,853 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=100045.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 05:25:07,801 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.7162, 1.3638, 1.5688, 1.5270, 3.2482, 1.0694, 2.2457, 3.6325], device='cuda:1'), covar=tensor([0.0438, 0.2683, 0.2687, 0.1774, 0.0710, 0.2603, 0.1316, 0.0266], device='cuda:1'), in_proj_covar=tensor([0.0379, 0.0350, 0.0369, 0.0330, 0.0357, 0.0339, 0.0350, 0.0372], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 05:25:14,926 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=100057.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 05:25:25,300 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=100066.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 05:25:47,697 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=100082.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 05:25:59,166 INFO [train.py:903] (1/4) Epoch 15, batch 4500, loss[loss=0.2333, simple_loss=0.318, pruned_loss=0.07436, over 19618.00 frames. ], tot_loss[loss=0.2199, simple_loss=0.2967, pruned_loss=0.07152, over 3812509.38 frames. ], batch size: 57, lr: 5.50e-03, grad_scale: 8.0 2023-04-02 05:25:59,584 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4248, 1.4659, 1.7378, 1.6248, 2.5596, 2.1670, 2.5601, 1.1080], device='cuda:1'), covar=tensor([0.2387, 0.4257, 0.2496, 0.1899, 0.1597, 0.2178, 0.1653, 0.4221], device='cuda:1'), in_proj_covar=tensor([0.0508, 0.0600, 0.0651, 0.0456, 0.0606, 0.0508, 0.0646, 0.0515], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 05:26:00,176 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.254e+02 5.166e+02 6.659e+02 8.670e+02 1.796e+03, threshold=1.332e+03, percent-clipped=6.0 2023-04-02 05:27:01,359 INFO [train.py:903] (1/4) Epoch 15, batch 4550, loss[loss=0.1892, simple_loss=0.2747, pruned_loss=0.05184, over 19438.00 frames. ], tot_loss[loss=0.2193, simple_loss=0.2965, pruned_loss=0.07104, over 3815925.87 frames. ], batch size: 48, lr: 5.50e-03, grad_scale: 8.0 2023-04-02 05:27:10,388 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9_rvb from training. Duration: 28.72225 2023-04-02 05:27:16,340 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=100153.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 05:27:21,791 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=100157.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 05:27:38,129 WARNING [train.py:1073] (1/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 2023-04-02 05:27:47,833 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=100178.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 05:28:04,926 INFO [train.py:903] (1/4) Epoch 15, batch 4600, loss[loss=0.2127, simple_loss=0.2903, pruned_loss=0.06753, over 19658.00 frames. ], tot_loss[loss=0.2184, simple_loss=0.2959, pruned_loss=0.07048, over 3826516.79 frames. ], batch size: 53, lr: 5.50e-03, grad_scale: 8.0 2023-04-02 05:28:06,058 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.047e+02 4.707e+02 5.654e+02 7.541e+02 1.184e+03, threshold=1.131e+03, percent-clipped=0.0 2023-04-02 05:28:19,063 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=100201.0, num_to_drop=1, layers_to_drop={1} 2023-04-02 05:28:30,974 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([5.1209, 5.2056, 6.0228, 5.9970, 1.8987, 5.6877, 4.7764, 5.6911], device='cuda:1'), covar=tensor([0.1389, 0.0625, 0.0427, 0.0460, 0.5683, 0.0544, 0.0549, 0.0827], device='cuda:1'), in_proj_covar=tensor([0.0728, 0.0664, 0.0865, 0.0744, 0.0771, 0.0612, 0.0522, 0.0799], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-02 05:28:47,089 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=100226.0, num_to_drop=1, layers_to_drop={1} 2023-04-02 05:29:08,985 INFO [train.py:903] (1/4) Epoch 15, batch 4650, loss[loss=0.2132, simple_loss=0.2842, pruned_loss=0.07108, over 19842.00 frames. ], tot_loss[loss=0.2187, simple_loss=0.2961, pruned_loss=0.07063, over 3832931.05 frames. ], batch size: 52, lr: 5.50e-03, grad_scale: 8.0 2023-04-02 05:29:25,484 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_rvb from training. Duration: 0.92 2023-04-02 05:29:25,778 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=100256.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 05:29:35,849 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_rvb from training. Duration: 0.83 2023-04-02 05:30:11,077 INFO [train.py:903] (1/4) Epoch 15, batch 4700, loss[loss=0.1772, simple_loss=0.2472, pruned_loss=0.05362, over 19741.00 frames. ], tot_loss[loss=0.2182, simple_loss=0.2953, pruned_loss=0.07058, over 3835876.28 frames. ], batch size: 46, lr: 5.49e-03, grad_scale: 8.0 2023-04-02 05:30:12,227 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.373e+02 5.601e+02 6.328e+02 8.331e+02 3.311e+03, threshold=1.266e+03, percent-clipped=13.0 2023-04-02 05:30:22,215 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=100301.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 05:30:33,232 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9_rvb from training. Duration: 0.92225 2023-04-02 05:30:50,868 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=100322.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 05:30:55,458 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=100326.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 05:31:13,902 INFO [train.py:903] (1/4) Epoch 15, batch 4750, loss[loss=0.2271, simple_loss=0.3006, pruned_loss=0.07681, over 19770.00 frames. ], tot_loss[loss=0.217, simple_loss=0.2945, pruned_loss=0.06978, over 3837171.04 frames. ], batch size: 54, lr: 5.49e-03, grad_scale: 8.0 2023-04-02 05:31:21,218 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=100347.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 05:31:49,662 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=100369.0, num_to_drop=1, layers_to_drop={1} 2023-04-02 05:32:16,840 INFO [train.py:903] (1/4) Epoch 15, batch 4800, loss[loss=0.2001, simple_loss=0.2703, pruned_loss=0.06495, over 19707.00 frames. ], tot_loss[loss=0.2174, simple_loss=0.2947, pruned_loss=0.07005, over 3841060.04 frames. ], batch size: 46, lr: 5.49e-03, grad_scale: 8.0 2023-04-02 05:32:18,029 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.926e+02 4.713e+02 6.387e+02 8.455e+02 2.006e+03, threshold=1.277e+03, percent-clipped=3.0 2023-04-02 05:32:44,427 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=100413.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 05:32:52,660 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=100420.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 05:33:17,165 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=100438.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 05:33:21,577 INFO [train.py:903] (1/4) Epoch 15, batch 4850, loss[loss=0.1824, simple_loss=0.2596, pruned_loss=0.05263, over 19375.00 frames. ], tot_loss[loss=0.2192, simple_loss=0.2962, pruned_loss=0.07108, over 3840217.87 frames. ], batch size: 47, lr: 5.49e-03, grad_scale: 8.0 2023-04-02 05:33:48,708 WARNING [train.py:1073] (1/4) Exclude cut with ID 774-127930-0014-48411_sp1.1_rvb from training. Duration: 0.95 2023-04-02 05:34:10,983 WARNING [train.py:1073] (1/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1_rvb from training. Duration: 0.9409375 2023-04-02 05:34:14,880 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=100484.0, num_to_drop=1, layers_to_drop={0} 2023-04-02 05:34:15,715 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 2023-04-02 05:34:17,774 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 2023-04-02 05:34:24,793 INFO [train.py:903] (1/4) Epoch 15, batch 4900, loss[loss=0.1606, simple_loss=0.2395, pruned_loss=0.04086, over 19329.00 frames. ], tot_loss[loss=0.2188, simple_loss=0.2954, pruned_loss=0.07108, over 3831161.74 frames. ], batch size: 44, lr: 5.49e-03, grad_scale: 8.0 2023-04-02 05:34:25,927 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.390e+02 5.042e+02 5.846e+02 7.925e+02 1.600e+03, threshold=1.169e+03, percent-clipped=3.0 2023-04-02 05:34:25,983 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1_rvb from training. Duration: 25.3818125 2023-04-02 05:34:46,436 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 2023-04-02 05:35:19,182 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.5926, 2.3593, 2.2980, 2.7712, 2.3652, 2.3655, 2.1278, 2.6625], device='cuda:1'), covar=tensor([0.0872, 0.1565, 0.1222, 0.0904, 0.1303, 0.0439, 0.1204, 0.0587], device='cuda:1'), in_proj_covar=tensor([0.0260, 0.0355, 0.0297, 0.0242, 0.0298, 0.0245, 0.0292, 0.0247], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 05:35:27,073 INFO [train.py:903] (1/4) Epoch 15, batch 4950, loss[loss=0.1769, simple_loss=0.2565, pruned_loss=0.04861, over 19722.00 frames. ], tot_loss[loss=0.2176, simple_loss=0.2945, pruned_loss=0.07035, over 3832683.22 frames. ], batch size: 46, lr: 5.49e-03, grad_scale: 8.0 2023-04-02 05:35:45,772 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 2023-04-02 05:36:05,331 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([5.2047, 5.1277, 6.1067, 6.0598, 2.0177, 5.7120, 4.9163, 5.7390], device='cuda:1'), covar=tensor([0.1450, 0.0706, 0.0517, 0.0514, 0.5664, 0.0483, 0.0534, 0.1040], device='cuda:1'), in_proj_covar=tensor([0.0735, 0.0669, 0.0873, 0.0750, 0.0778, 0.0619, 0.0528, 0.0808], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-02 05:36:09,664 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 2023-04-02 05:36:29,678 INFO [train.py:903] (1/4) Epoch 15, batch 5000, loss[loss=0.213, simple_loss=0.2925, pruned_loss=0.06678, over 19605.00 frames. ], tot_loss[loss=0.219, simple_loss=0.2956, pruned_loss=0.07114, over 3830226.24 frames. ], batch size: 50, lr: 5.49e-03, grad_scale: 4.0 2023-04-02 05:36:31,852 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.571e+02 5.199e+02 6.623e+02 8.418e+02 1.165e+03, threshold=1.325e+03, percent-clipped=0.0 2023-04-02 05:36:40,510 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_rvb from training. Duration: 27.14 2023-04-02 05:36:40,821 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=100600.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 05:36:52,295 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 2023-04-02 05:37:33,609 INFO [train.py:903] (1/4) Epoch 15, batch 5050, loss[loss=0.2241, simple_loss=0.315, pruned_loss=0.06661, over 19762.00 frames. ], tot_loss[loss=0.2177, simple_loss=0.2944, pruned_loss=0.07053, over 3837527.85 frames. ], batch size: 56, lr: 5.49e-03, grad_scale: 4.0 2023-04-02 05:38:09,675 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9_rvb from training. Duration: 26.62775 2023-04-02 05:38:37,492 INFO [train.py:903] (1/4) Epoch 15, batch 5100, loss[loss=0.2298, simple_loss=0.3088, pruned_loss=0.0754, over 19787.00 frames. ], tot_loss[loss=0.2183, simple_loss=0.295, pruned_loss=0.07079, over 3823847.87 frames. ], batch size: 56, lr: 5.48e-03, grad_scale: 4.0 2023-04-02 05:38:39,901 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.188e+02 5.134e+02 6.124e+02 7.830e+02 1.941e+03, threshold=1.225e+03, percent-clipped=4.0 2023-04-02 05:38:49,377 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 2023-04-02 05:38:51,815 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_rvb from training. Duration: 25.035 2023-04-02 05:38:57,354 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_rvb from training. Duration: 27.92 2023-04-02 05:38:57,698 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1211, 1.1990, 1.7393, 1.4528, 2.9249, 4.4969, 4.3763, 4.8593], device='cuda:1'), covar=tensor([0.1713, 0.3870, 0.3386, 0.2184, 0.0585, 0.0191, 0.0178, 0.0142], device='cuda:1'), in_proj_covar=tensor([0.0258, 0.0302, 0.0332, 0.0251, 0.0225, 0.0169, 0.0206, 0.0223], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 05:39:05,807 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=100715.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 05:39:37,104 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=100740.0, num_to_drop=1, layers_to_drop={1} 2023-04-02 05:39:38,888 INFO [train.py:903] (1/4) Epoch 15, batch 5150, loss[loss=0.2281, simple_loss=0.3167, pruned_loss=0.06974, over 18780.00 frames. ], tot_loss[loss=0.2188, simple_loss=0.2957, pruned_loss=0.07094, over 3818831.35 frames. ], batch size: 74, lr: 5.48e-03, grad_scale: 4.0 2023-04-02 05:39:39,296 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.9849, 2.6996, 2.1176, 2.0631, 1.6955, 2.2513, 0.9927, 1.9741], device='cuda:1'), covar=tensor([0.0619, 0.0563, 0.0569, 0.0929, 0.1131, 0.0994, 0.1160, 0.0915], device='cuda:1'), in_proj_covar=tensor([0.0347, 0.0342, 0.0340, 0.0369, 0.0442, 0.0370, 0.0322, 0.0328], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 05:39:47,018 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=100748.0, num_to_drop=1, layers_to_drop={0} 2023-04-02 05:39:51,353 WARNING [train.py:1073] (1/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 2023-04-02 05:40:07,665 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=100764.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 05:40:09,096 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=100765.0, num_to_drop=1, layers_to_drop={1} 2023-04-02 05:40:28,216 WARNING [train.py:1073] (1/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1_rvb from training. Duration: 0.9681875 2023-04-02 05:40:41,942 INFO [train.py:903] (1/4) Epoch 15, batch 5200, loss[loss=0.2142, simple_loss=0.2857, pruned_loss=0.07135, over 19605.00 frames. ], tot_loss[loss=0.2192, simple_loss=0.2959, pruned_loss=0.07128, over 3822896.34 frames. ], batch size: 50, lr: 5.48e-03, grad_scale: 8.0 2023-04-02 05:40:44,524 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.350e+02 5.325e+02 6.515e+02 8.501e+02 1.618e+03, threshold=1.303e+03, percent-clipped=5.0 2023-04-02 05:40:58,645 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 2023-04-02 05:41:43,051 WARNING [train.py:1073] (1/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 2023-04-02 05:41:47,605 INFO [train.py:903] (1/4) Epoch 15, batch 5250, loss[loss=0.2101, simple_loss=0.282, pruned_loss=0.06911, over 19451.00 frames. ], tot_loss[loss=0.2209, simple_loss=0.2975, pruned_loss=0.07211, over 3813257.48 frames. ], batch size: 49, lr: 5.48e-03, grad_scale: 8.0 2023-04-02 05:42:20,902 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.7902, 1.9287, 2.1314, 2.6085, 1.8963, 2.4892, 2.2186, 1.9392], device='cuda:1'), covar=tensor([0.3821, 0.3422, 0.1694, 0.1940, 0.3610, 0.1628, 0.4183, 0.3048], device='cuda:1'), in_proj_covar=tensor([0.0835, 0.0877, 0.0675, 0.0906, 0.0817, 0.0752, 0.0812, 0.0739], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 05:42:25,538 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1404, 1.2125, 1.7069, 1.4012, 2.7180, 3.6836, 3.4693, 4.0128], device='cuda:1'), covar=tensor([0.1746, 0.3797, 0.3225, 0.2292, 0.0566, 0.0212, 0.0204, 0.0197], device='cuda:1'), in_proj_covar=tensor([0.0260, 0.0304, 0.0333, 0.0252, 0.0225, 0.0170, 0.0208, 0.0224], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 05:42:33,438 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=100879.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 05:42:50,408 INFO [train.py:903] (1/4) Epoch 15, batch 5300, loss[loss=0.2015, simple_loss=0.2804, pruned_loss=0.06136, over 19592.00 frames. ], tot_loss[loss=0.2199, simple_loss=0.2966, pruned_loss=0.07157, over 3826355.81 frames. ], batch size: 52, lr: 5.48e-03, grad_scale: 8.0 2023-04-02 05:42:52,714 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.288e+02 5.131e+02 6.120e+02 8.353e+02 1.768e+03, threshold=1.224e+03, percent-clipped=4.0 2023-04-02 05:43:08,018 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 2023-04-02 05:43:52,929 INFO [train.py:903] (1/4) Epoch 15, batch 5350, loss[loss=0.2097, simple_loss=0.2986, pruned_loss=0.06039, over 19670.00 frames. ], tot_loss[loss=0.2192, simple_loss=0.296, pruned_loss=0.07119, over 3821482.32 frames. ], batch size: 55, lr: 5.48e-03, grad_scale: 8.0 2023-04-02 05:44:29,379 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9_rvb from training. Duration: 26.438875 2023-04-02 05:44:30,857 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=100971.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 05:44:50,766 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6925, 1.5844, 1.4474, 2.3618, 1.8359, 2.1251, 2.0455, 1.7741], device='cuda:1'), covar=tensor([0.0815, 0.0915, 0.1045, 0.0702, 0.0776, 0.0681, 0.0848, 0.0694], device='cuda:1'), in_proj_covar=tensor([0.0210, 0.0220, 0.0221, 0.0242, 0.0226, 0.0209, 0.0189, 0.0202], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-02 05:44:56,005 INFO [train.py:903] (1/4) Epoch 15, batch 5400, loss[loss=0.2055, simple_loss=0.2924, pruned_loss=0.05934, over 18256.00 frames. ], tot_loss[loss=0.2207, simple_loss=0.2975, pruned_loss=0.07199, over 3810282.20 frames. ], batch size: 83, lr: 5.48e-03, grad_scale: 8.0 2023-04-02 05:44:58,255 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.408e+02 5.355e+02 6.832e+02 8.412e+02 2.240e+03, threshold=1.366e+03, percent-clipped=7.0 2023-04-02 05:45:01,674 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=100996.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 05:46:00,085 INFO [train.py:903] (1/4) Epoch 15, batch 5450, loss[loss=0.2359, simple_loss=0.3119, pruned_loss=0.07997, over 19682.00 frames. ], tot_loss[loss=0.2209, simple_loss=0.2977, pruned_loss=0.07206, over 3813359.38 frames. ], batch size: 60, lr: 5.47e-03, grad_scale: 8.0 2023-04-02 05:46:08,863 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.9868, 3.5943, 2.6811, 3.2611, 0.8639, 3.4759, 3.3759, 3.5026], device='cuda:1'), covar=tensor([0.0853, 0.1217, 0.1901, 0.0922, 0.4208, 0.0834, 0.0934, 0.1221], device='cuda:1'), in_proj_covar=tensor([0.0462, 0.0379, 0.0458, 0.0326, 0.0390, 0.0393, 0.0383, 0.0420], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 05:46:32,261 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4112, 2.1057, 1.5054, 1.4533, 1.9394, 1.0558, 1.2460, 1.8312], device='cuda:1'), covar=tensor([0.1033, 0.0772, 0.1044, 0.0810, 0.0540, 0.1372, 0.0769, 0.0431], device='cuda:1'), in_proj_covar=tensor([0.0296, 0.0307, 0.0326, 0.0252, 0.0239, 0.0330, 0.0294, 0.0269], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 05:47:04,589 INFO [train.py:903] (1/4) Epoch 15, batch 5500, loss[loss=0.2205, simple_loss=0.3072, pruned_loss=0.06692, over 19700.00 frames. ], tot_loss[loss=0.2197, simple_loss=0.2964, pruned_loss=0.07149, over 3816489.77 frames. ], batch size: 59, lr: 5.47e-03, grad_scale: 8.0 2023-04-02 05:47:04,772 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=101092.0, num_to_drop=1, layers_to_drop={1} 2023-04-02 05:47:05,797 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.5731, 4.1336, 2.6438, 3.6209, 0.7829, 3.9835, 3.8826, 4.0478], device='cuda:1'), covar=tensor([0.0662, 0.1076, 0.2077, 0.0871, 0.4444, 0.0739, 0.0893, 0.1046], device='cuda:1'), in_proj_covar=tensor([0.0466, 0.0381, 0.0462, 0.0329, 0.0393, 0.0395, 0.0386, 0.0423], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 05:47:06,824 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.049e+02 5.715e+02 6.860e+02 8.623e+02 1.531e+03, threshold=1.372e+03, percent-clipped=1.0 2023-04-02 05:47:08,470 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.7445, 1.6425, 1.5146, 2.1868, 1.6242, 2.0705, 2.0891, 1.8578], device='cuda:1'), covar=tensor([0.0778, 0.0906, 0.1017, 0.0801, 0.0904, 0.0707, 0.0869, 0.0669], device='cuda:1'), in_proj_covar=tensor([0.0209, 0.0221, 0.0221, 0.0243, 0.0227, 0.0209, 0.0189, 0.0202], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-02 05:47:27,319 WARNING [train.py:1073] (1/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9_rvb from training. Duration: 27.47775 2023-04-02 05:47:30,839 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.8410, 1.7254, 1.4695, 1.8642, 1.7013, 1.3799, 1.4274, 1.7533], device='cuda:1'), covar=tensor([0.1114, 0.1562, 0.1681, 0.1139, 0.1441, 0.0788, 0.1640, 0.0838], device='cuda:1'), in_proj_covar=tensor([0.0260, 0.0354, 0.0298, 0.0246, 0.0299, 0.0246, 0.0294, 0.0247], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 05:47:45,301 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=101125.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 05:47:58,086 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=101135.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 05:48:05,674 INFO [train.py:903] (1/4) Epoch 15, batch 5550, loss[loss=0.2436, simple_loss=0.3235, pruned_loss=0.08187, over 19676.00 frames. ], tot_loss[loss=0.2203, simple_loss=0.2969, pruned_loss=0.07188, over 3816216.51 frames. ], batch size: 53, lr: 5.47e-03, grad_scale: 8.0 2023-04-02 05:48:12,790 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 2023-04-02 05:48:24,404 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0782, 1.9738, 1.6937, 2.8098, 2.0438, 2.5980, 2.4759, 2.1499], device='cuda:1'), covar=tensor([0.0761, 0.0830, 0.1056, 0.0828, 0.0838, 0.0663, 0.0883, 0.0667], device='cuda:1'), in_proj_covar=tensor([0.0209, 0.0221, 0.0222, 0.0243, 0.0227, 0.0209, 0.0189, 0.0202], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-02 05:48:28,803 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=101160.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 05:49:04,235 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 2023-04-02 05:49:08,929 INFO [train.py:903] (1/4) Epoch 15, batch 5600, loss[loss=0.2393, simple_loss=0.3176, pruned_loss=0.08053, over 19431.00 frames. ], tot_loss[loss=0.22, simple_loss=0.2968, pruned_loss=0.07163, over 3825033.77 frames. ], batch size: 64, lr: 5.47e-03, grad_scale: 8.0 2023-04-02 05:49:11,012 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.323e+02 5.272e+02 6.555e+02 8.017e+02 1.573e+03, threshold=1.311e+03, percent-clipped=2.0 2023-04-02 05:49:28,262 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=101207.0, num_to_drop=1, layers_to_drop={0} 2023-04-02 05:50:11,398 INFO [train.py:903] (1/4) Epoch 15, batch 5650, loss[loss=0.2848, simple_loss=0.3337, pruned_loss=0.118, over 13367.00 frames. ], tot_loss[loss=0.2202, simple_loss=0.2968, pruned_loss=0.07173, over 3823069.63 frames. ], batch size: 137, lr: 5.47e-03, grad_scale: 8.0 2023-04-02 05:51:00,539 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9_rvb from training. Duration: 28.0944375 2023-04-02 05:51:14,089 INFO [train.py:903] (1/4) Epoch 15, batch 5700, loss[loss=0.2212, simple_loss=0.3049, pruned_loss=0.06875, over 19649.00 frames. ], tot_loss[loss=0.2209, simple_loss=0.2973, pruned_loss=0.07229, over 3842373.71 frames. ], batch size: 58, lr: 5.47e-03, grad_scale: 8.0 2023-04-02 05:51:17,701 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.786e+02 5.651e+02 6.610e+02 8.419e+02 1.957e+03, threshold=1.322e+03, percent-clipped=7.0 2023-04-02 05:51:48,345 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.8277, 1.6276, 1.5658, 1.9659, 1.6280, 1.6782, 1.5881, 1.8433], device='cuda:1'), covar=tensor([0.1007, 0.1429, 0.1379, 0.0943, 0.1296, 0.0516, 0.1255, 0.0680], device='cuda:1'), in_proj_covar=tensor([0.0258, 0.0351, 0.0296, 0.0244, 0.0297, 0.0244, 0.0291, 0.0246], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 05:52:17,921 INFO [train.py:903] (1/4) Epoch 15, batch 5750, loss[loss=0.2003, simple_loss=0.286, pruned_loss=0.05733, over 18694.00 frames. ], tot_loss[loss=0.2198, simple_loss=0.2965, pruned_loss=0.07156, over 3852579.95 frames. ], batch size: 74, lr: 5.47e-03, grad_scale: 8.0 2023-04-02 05:52:20,179 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9_rvb from training. Duration: 33.038875 2023-04-02 05:52:28,373 WARNING [train.py:1073] (1/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 2023-04-02 05:52:32,998 WARNING [train.py:1073] (1/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 2023-04-02 05:52:51,640 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2996, 1.2246, 1.5304, 1.1202, 2.4137, 3.3385, 3.0492, 3.5241], device='cuda:1'), covar=tensor([0.1489, 0.3642, 0.3305, 0.2340, 0.0588, 0.0171, 0.0230, 0.0236], device='cuda:1'), in_proj_covar=tensor([0.0259, 0.0304, 0.0334, 0.0253, 0.0225, 0.0170, 0.0208, 0.0224], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 05:53:10,170 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.5414, 1.3733, 1.5917, 1.5605, 3.0867, 1.0721, 2.2726, 3.4468], device='cuda:1'), covar=tensor([0.0444, 0.2665, 0.2617, 0.1826, 0.0682, 0.2573, 0.1341, 0.0277], device='cuda:1'), in_proj_covar=tensor([0.0382, 0.0353, 0.0370, 0.0336, 0.0362, 0.0341, 0.0354, 0.0374], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 05:53:21,267 INFO [train.py:903] (1/4) Epoch 15, batch 5800, loss[loss=0.1928, simple_loss=0.2755, pruned_loss=0.05503, over 19586.00 frames. ], tot_loss[loss=0.2198, simple_loss=0.2967, pruned_loss=0.07146, over 3846084.49 frames. ], batch size: 52, lr: 5.46e-03, grad_scale: 8.0 2023-04-02 05:53:23,495 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.554e+02 5.132e+02 6.075e+02 7.663e+02 2.298e+03, threshold=1.215e+03, percent-clipped=3.0 2023-04-02 05:54:24,295 INFO [train.py:903] (1/4) Epoch 15, batch 5850, loss[loss=0.2076, simple_loss=0.2778, pruned_loss=0.06871, over 19779.00 frames. ], tot_loss[loss=0.2195, simple_loss=0.2964, pruned_loss=0.07135, over 3849795.49 frames. ], batch size: 47, lr: 5.46e-03, grad_scale: 8.0 2023-04-02 05:54:37,742 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0533, 2.0803, 2.2283, 2.6991, 2.0099, 2.6631, 2.2892, 2.0366], device='cuda:1'), covar=tensor([0.4045, 0.3640, 0.1760, 0.2194, 0.3895, 0.1762, 0.4442, 0.3141], device='cuda:1'), in_proj_covar=tensor([0.0836, 0.0882, 0.0679, 0.0910, 0.0823, 0.0756, 0.0815, 0.0744], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 05:54:52,493 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=101463.0, num_to_drop=1, layers_to_drop={1} 2023-04-02 05:54:58,858 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=101469.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 05:55:24,159 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=101488.0, num_to_drop=1, layers_to_drop={1} 2023-04-02 05:55:28,446 INFO [train.py:903] (1/4) Epoch 15, batch 5900, loss[loss=0.2294, simple_loss=0.307, pruned_loss=0.0759, over 18875.00 frames. ], tot_loss[loss=0.2203, simple_loss=0.2973, pruned_loss=0.07171, over 3844510.51 frames. ], batch size: 74, lr: 5.46e-03, grad_scale: 8.0 2023-04-02 05:55:30,755 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.093e+02 5.163e+02 6.557e+02 7.983e+02 1.612e+03, threshold=1.311e+03, percent-clipped=3.0 2023-04-02 05:55:30,812 WARNING [train.py:1073] (1/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 2023-04-02 05:55:51,849 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9_rvb from training. Duration: 27.511125 2023-04-02 05:56:03,826 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9145, 1.5393, 1.4953, 1.8025, 1.5086, 1.6424, 1.4517, 1.8316], device='cuda:1'), covar=tensor([0.0958, 0.1290, 0.1453, 0.0959, 0.1292, 0.0539, 0.1421, 0.0692], device='cuda:1'), in_proj_covar=tensor([0.0259, 0.0352, 0.0296, 0.0245, 0.0297, 0.0244, 0.0291, 0.0246], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 05:56:20,785 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4168, 1.4588, 1.6198, 1.5666, 2.2949, 2.0294, 2.2868, 0.9927], device='cuda:1'), covar=tensor([0.2210, 0.4045, 0.2471, 0.1810, 0.1411, 0.2059, 0.1357, 0.4049], device='cuda:1'), in_proj_covar=tensor([0.0508, 0.0608, 0.0654, 0.0457, 0.0609, 0.0513, 0.0650, 0.0516], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 05:56:30,793 INFO [train.py:903] (1/4) Epoch 15, batch 5950, loss[loss=0.2201, simple_loss=0.2956, pruned_loss=0.0723, over 19575.00 frames. ], tot_loss[loss=0.2195, simple_loss=0.2965, pruned_loss=0.07125, over 3819219.22 frames. ], batch size: 52, lr: 5.46e-03, grad_scale: 8.0 2023-04-02 05:57:24,024 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=101584.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 05:57:34,460 INFO [train.py:903] (1/4) Epoch 15, batch 6000, loss[loss=0.2273, simple_loss=0.3109, pruned_loss=0.07181, over 19741.00 frames. ], tot_loss[loss=0.2206, simple_loss=0.2975, pruned_loss=0.07186, over 3828935.39 frames. ], batch size: 63, lr: 5.46e-03, grad_scale: 8.0 2023-04-02 05:57:34,460 INFO [train.py:928] (1/4) Computing validation loss 2023-04-02 05:57:47,185 INFO [train.py:937] (1/4) Epoch 15, validation: loss=0.1729, simple_loss=0.2735, pruned_loss=0.0362, over 944034.00 frames. 2023-04-02 05:57:47,186 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18821MB 2023-04-02 05:57:49,631 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.269e+02 5.208e+02 6.128e+02 8.316e+02 1.573e+03, threshold=1.226e+03, percent-clipped=3.0 2023-04-02 05:58:26,438 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.3079, 2.0601, 2.0481, 2.8329, 2.0675, 2.6338, 2.3886, 2.4042], device='cuda:1'), covar=tensor([0.0743, 0.0839, 0.0922, 0.0850, 0.0897, 0.0709, 0.0959, 0.0593], device='cuda:1'), in_proj_covar=tensor([0.0210, 0.0221, 0.0221, 0.0241, 0.0227, 0.0207, 0.0187, 0.0201], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-02 05:58:49,959 INFO [train.py:903] (1/4) Epoch 15, batch 6050, loss[loss=0.1823, simple_loss=0.2532, pruned_loss=0.05569, over 18987.00 frames. ], tot_loss[loss=0.2207, simple_loss=0.2973, pruned_loss=0.07207, over 3815023.00 frames. ], batch size: 42, lr: 5.46e-03, grad_scale: 8.0 2023-04-02 05:59:17,804 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4536, 1.4497, 1.8083, 1.4386, 2.3418, 2.6806, 2.5538, 2.8225], device='cuda:1'), covar=tensor([0.1325, 0.3079, 0.2554, 0.2277, 0.1047, 0.0352, 0.0259, 0.0338], device='cuda:1'), in_proj_covar=tensor([0.0260, 0.0305, 0.0334, 0.0253, 0.0225, 0.0169, 0.0208, 0.0224], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 05:59:52,010 INFO [train.py:903] (1/4) Epoch 15, batch 6100, loss[loss=0.2743, simple_loss=0.3412, pruned_loss=0.1037, over 18121.00 frames. ], tot_loss[loss=0.2203, simple_loss=0.297, pruned_loss=0.07174, over 3817493.30 frames. ], batch size: 84, lr: 5.46e-03, grad_scale: 8.0 2023-04-02 05:59:55,079 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.053e+02 5.047e+02 6.326e+02 7.867e+02 1.574e+03, threshold=1.265e+03, percent-clipped=9.0 2023-04-02 06:00:05,531 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=101702.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 06:00:56,520 INFO [train.py:903] (1/4) Epoch 15, batch 6150, loss[loss=0.1802, simple_loss=0.2587, pruned_loss=0.05088, over 19782.00 frames. ], tot_loss[loss=0.2198, simple_loss=0.2966, pruned_loss=0.07151, over 3822347.77 frames. ], batch size: 45, lr: 5.46e-03, grad_scale: 8.0 2023-04-02 06:01:23,273 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.63 vs. limit=5.0 2023-04-02 06:01:23,845 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9_rvb from training. Duration: 31.02225 2023-04-02 06:01:59,351 INFO [train.py:903] (1/4) Epoch 15, batch 6200, loss[loss=0.2171, simple_loss=0.2971, pruned_loss=0.06851, over 19596.00 frames. ], tot_loss[loss=0.2201, simple_loss=0.2966, pruned_loss=0.07177, over 3826958.53 frames. ], batch size: 52, lr: 5.45e-03, grad_scale: 8.0 2023-04-02 06:02:01,561 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.367e+02 4.959e+02 6.303e+02 7.991e+02 1.526e+03, threshold=1.261e+03, percent-clipped=1.0 2023-04-02 06:02:59,281 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=101840.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 06:03:02,065 INFO [train.py:903] (1/4) Epoch 15, batch 6250, loss[loss=0.2075, simple_loss=0.2763, pruned_loss=0.06929, over 19186.00 frames. ], tot_loss[loss=0.2202, simple_loss=0.2967, pruned_loss=0.07183, over 3821257.63 frames. ], batch size: 42, lr: 5.45e-03, grad_scale: 8.0 2023-04-02 06:03:30,467 WARNING [train.py:1073] (1/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9_rvb from training. Duration: 25.988875 2023-04-02 06:03:30,831 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=101865.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 06:04:04,525 INFO [train.py:903] (1/4) Epoch 15, batch 6300, loss[loss=0.2069, simple_loss=0.2847, pruned_loss=0.06456, over 19855.00 frames. ], tot_loss[loss=0.2193, simple_loss=0.2963, pruned_loss=0.0712, over 3832795.65 frames. ], batch size: 52, lr: 5.45e-03, grad_scale: 4.0 2023-04-02 06:04:07,996 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.483e+02 5.666e+02 6.616e+02 7.934e+02 1.912e+03, threshold=1.323e+03, percent-clipped=2.0 2023-04-02 06:05:03,699 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=101938.0, num_to_drop=1, layers_to_drop={0} 2023-04-02 06:05:08,246 INFO [train.py:903] (1/4) Epoch 15, batch 6350, loss[loss=0.2344, simple_loss=0.3162, pruned_loss=0.07627, over 19509.00 frames. ], tot_loss[loss=0.2193, simple_loss=0.2964, pruned_loss=0.07111, over 3828651.75 frames. ], batch size: 64, lr: 5.45e-03, grad_scale: 2.0 2023-04-02 06:05:17,916 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0381, 2.0066, 1.8035, 1.6313, 1.5929, 1.6543, 0.4501, 0.8864], device='cuda:1'), covar=tensor([0.0482, 0.0481, 0.0318, 0.0544, 0.0978, 0.0619, 0.1037, 0.0918], device='cuda:1'), in_proj_covar=tensor([0.0350, 0.0341, 0.0338, 0.0370, 0.0442, 0.0370, 0.0323, 0.0331], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 06:05:55,737 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2072, 1.2788, 1.2171, 1.0250, 1.0746, 1.0989, 0.0707, 0.3537], device='cuda:1'), covar=tensor([0.0568, 0.0545, 0.0330, 0.0471, 0.1049, 0.0477, 0.1066, 0.0904], device='cuda:1'), in_proj_covar=tensor([0.0349, 0.0340, 0.0338, 0.0370, 0.0442, 0.0370, 0.0323, 0.0331], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 06:06:11,821 INFO [train.py:903] (1/4) Epoch 15, batch 6400, loss[loss=0.2409, simple_loss=0.3181, pruned_loss=0.08184, over 17467.00 frames. ], tot_loss[loss=0.2211, simple_loss=0.298, pruned_loss=0.07209, over 3817220.86 frames. ], batch size: 101, lr: 5.45e-03, grad_scale: 4.0 2023-04-02 06:06:16,593 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.303e+02 4.874e+02 5.936e+02 7.490e+02 2.019e+03, threshold=1.187e+03, percent-clipped=4.0 2023-04-02 06:06:30,770 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=102005.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 06:07:05,512 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.8319, 1.6312, 1.8503, 1.8365, 4.3150, 1.1305, 2.4647, 4.6899], device='cuda:1'), covar=tensor([0.0381, 0.2788, 0.2701, 0.1776, 0.0771, 0.2671, 0.1391, 0.0218], device='cuda:1'), in_proj_covar=tensor([0.0377, 0.0348, 0.0366, 0.0331, 0.0357, 0.0337, 0.0349, 0.0372], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 06:07:17,094 INFO [train.py:903] (1/4) Epoch 15, batch 6450, loss[loss=0.2272, simple_loss=0.3094, pruned_loss=0.07251, over 19853.00 frames. ], tot_loss[loss=0.2209, simple_loss=0.2978, pruned_loss=0.07199, over 3812908.16 frames. ], batch size: 52, lr: 5.45e-03, grad_scale: 4.0 2023-04-02 06:07:23,152 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=102046.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 06:08:03,614 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9_rvb from training. Duration: 28.638875 2023-04-02 06:08:08,946 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 2023-04-02 06:08:20,704 INFO [train.py:903] (1/4) Epoch 15, batch 6500, loss[loss=0.2019, simple_loss=0.2719, pruned_loss=0.06599, over 19331.00 frames. ], tot_loss[loss=0.221, simple_loss=0.2981, pruned_loss=0.07196, over 3819894.18 frames. ], batch size: 44, lr: 5.45e-03, grad_scale: 4.0 2023-04-02 06:08:25,529 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.034e+02 4.860e+02 6.100e+02 7.866e+02 2.286e+03, threshold=1.220e+03, percent-clipped=9.0 2023-04-02 06:08:26,642 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1_rvb from training. Duration: 0.836375 2023-04-02 06:08:41,072 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.58 vs. limit=5.0 2023-04-02 06:09:23,516 INFO [train.py:903] (1/4) Epoch 15, batch 6550, loss[loss=0.2426, simple_loss=0.3146, pruned_loss=0.08536, over 19473.00 frames. ], tot_loss[loss=0.221, simple_loss=0.2981, pruned_loss=0.07196, over 3815803.84 frames. ], batch size: 64, lr: 5.44e-03, grad_scale: 4.0 2023-04-02 06:09:47,187 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=102161.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 06:10:26,465 INFO [train.py:903] (1/4) Epoch 15, batch 6600, loss[loss=0.2087, simple_loss=0.2771, pruned_loss=0.07018, over 19676.00 frames. ], tot_loss[loss=0.2204, simple_loss=0.2972, pruned_loss=0.07182, over 3823688.51 frames. ], batch size: 53, lr: 5.44e-03, grad_scale: 4.0 2023-04-02 06:10:31,177 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.095e+02 4.937e+02 6.611e+02 8.175e+02 1.787e+03, threshold=1.322e+03, percent-clipped=6.0 2023-04-02 06:11:10,228 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4288, 1.4785, 1.9986, 1.6113, 3.3090, 2.6415, 3.5291, 1.6252], device='cuda:1'), covar=tensor([0.2378, 0.4151, 0.2553, 0.1807, 0.1386, 0.1927, 0.1539, 0.3780], device='cuda:1'), in_proj_covar=tensor([0.0507, 0.0605, 0.0655, 0.0457, 0.0608, 0.0514, 0.0653, 0.0515], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 06:11:29,855 INFO [train.py:903] (1/4) Epoch 15, batch 6650, loss[loss=0.2299, simple_loss=0.3075, pruned_loss=0.07612, over 17575.00 frames. ], tot_loss[loss=0.2184, simple_loss=0.2955, pruned_loss=0.07071, over 3833414.77 frames. ], batch size: 101, lr: 5.44e-03, grad_scale: 4.0 2023-04-02 06:11:39,331 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2018, 1.8385, 1.4348, 1.2165, 1.6214, 1.1987, 1.1285, 1.6638], device='cuda:1'), covar=tensor([0.0708, 0.0691, 0.1015, 0.0726, 0.0464, 0.1156, 0.0598, 0.0357], device='cuda:1'), in_proj_covar=tensor([0.0294, 0.0309, 0.0329, 0.0254, 0.0241, 0.0329, 0.0293, 0.0269], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 06:12:01,500 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.31 vs. limit=2.0 2023-04-02 06:12:20,466 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=102282.0, num_to_drop=1, layers_to_drop={0} 2023-04-02 06:12:21,725 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4170, 1.4611, 1.7909, 1.5664, 2.7345, 2.2747, 2.8610, 1.2463], device='cuda:1'), covar=tensor([0.2385, 0.4085, 0.2437, 0.1888, 0.1424, 0.1977, 0.1396, 0.3974], device='cuda:1'), in_proj_covar=tensor([0.0506, 0.0603, 0.0654, 0.0456, 0.0607, 0.0512, 0.0650, 0.0514], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 06:12:33,719 INFO [train.py:903] (1/4) Epoch 15, batch 6700, loss[loss=0.2023, simple_loss=0.2705, pruned_loss=0.06709, over 19358.00 frames. ], tot_loss[loss=0.2188, simple_loss=0.2956, pruned_loss=0.07097, over 3834837.86 frames. ], batch size: 47, lr: 5.44e-03, grad_scale: 4.0 2023-04-02 06:12:38,442 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.247e+02 5.015e+02 6.423e+02 7.841e+02 1.581e+03, threshold=1.285e+03, percent-clipped=1.0 2023-04-02 06:13:32,250 INFO [train.py:903] (1/4) Epoch 15, batch 6750, loss[loss=0.1876, simple_loss=0.2643, pruned_loss=0.05543, over 18648.00 frames. ], tot_loss[loss=0.2181, simple_loss=0.2955, pruned_loss=0.07034, over 3833652.30 frames. ], batch size: 41, lr: 5.44e-03, grad_scale: 4.0 2023-04-02 06:13:40,334 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=102349.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 06:13:42,065 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.65 vs. limit=2.0 2023-04-02 06:13:49,042 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.44 vs. limit=2.0 2023-04-02 06:13:57,400 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3264, 1.3378, 1.5370, 1.4783, 2.3310, 2.0993, 2.3192, 0.8829], device='cuda:1'), covar=tensor([0.2114, 0.3757, 0.2363, 0.1700, 0.1332, 0.1867, 0.1283, 0.3788], device='cuda:1'), in_proj_covar=tensor([0.0504, 0.0601, 0.0651, 0.0454, 0.0606, 0.0511, 0.0649, 0.0514], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 06:14:17,262 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9196, 2.0267, 2.2514, 2.6447, 1.9260, 2.4973, 2.3650, 2.1006], device='cuda:1'), covar=tensor([0.3909, 0.3478, 0.1602, 0.1982, 0.3695, 0.1754, 0.4131, 0.2886], device='cuda:1'), in_proj_covar=tensor([0.0832, 0.0879, 0.0675, 0.0903, 0.0819, 0.0754, 0.0810, 0.0739], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 06:14:28,435 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.2035, 2.0033, 2.0127, 2.9990, 2.2006, 2.6103, 2.5519, 2.5201], device='cuda:1'), covar=tensor([0.0816, 0.0919, 0.1000, 0.0755, 0.0866, 0.0745, 0.0928, 0.0604], device='cuda:1'), in_proj_covar=tensor([0.0212, 0.0223, 0.0224, 0.0244, 0.0228, 0.0210, 0.0190, 0.0203], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-02 06:14:30,399 INFO [train.py:903] (1/4) Epoch 15, batch 6800, loss[loss=0.1985, simple_loss=0.2706, pruned_loss=0.0632, over 19391.00 frames. ], tot_loss[loss=0.2201, simple_loss=0.2971, pruned_loss=0.07154, over 3827656.66 frames. ], batch size: 48, lr: 5.44e-03, grad_scale: 8.0 2023-04-02 06:14:35,326 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.506e+02 4.944e+02 6.022e+02 7.665e+02 3.022e+03, threshold=1.204e+03, percent-clipped=5.0 2023-04-02 06:14:36,986 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=102397.0, num_to_drop=1, layers_to_drop={0} 2023-04-02 06:14:57,417 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=102417.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 06:15:15,616 WARNING [train.py:1073] (1/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 2023-04-02 06:15:16,097 WARNING [train.py:1073] (1/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 2023-04-02 06:15:19,104 INFO [train.py:903] (1/4) Epoch 16, batch 0, loss[loss=0.2719, simple_loss=0.3319, pruned_loss=0.106, over 13218.00 frames. ], tot_loss[loss=0.2719, simple_loss=0.3319, pruned_loss=0.106, over 13218.00 frames. ], batch size: 136, lr: 5.26e-03, grad_scale: 8.0 2023-04-02 06:15:19,104 INFO [train.py:928] (1/4) Computing validation loss 2023-04-02 06:15:29,714 INFO [train.py:937] (1/4) Epoch 16, validation: loss=0.1737, simple_loss=0.2745, pruned_loss=0.03646, over 944034.00 frames. 2023-04-02 06:15:29,715 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18821MB 2023-04-02 06:15:45,604 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 2023-04-02 06:15:58,368 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=102442.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 06:16:24,717 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=102464.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 06:16:33,054 INFO [train.py:903] (1/4) Epoch 16, batch 50, loss[loss=0.2048, simple_loss=0.2824, pruned_loss=0.06358, over 19770.00 frames. ], tot_loss[loss=0.2132, simple_loss=0.2888, pruned_loss=0.06884, over 870884.00 frames. ], batch size: 54, lr: 5.26e-03, grad_scale: 8.0 2023-04-02 06:17:04,309 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.709e+02 4.865e+02 6.426e+02 8.395e+02 1.744e+03, threshold=1.285e+03, percent-clipped=5.0 2023-04-02 06:17:08,775 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1_rvb from training. Duration: 27.0318125 2023-04-02 06:17:33,677 INFO [train.py:903] (1/4) Epoch 16, batch 100, loss[loss=0.1942, simple_loss=0.276, pruned_loss=0.05625, over 19576.00 frames. ], tot_loss[loss=0.2143, simple_loss=0.2904, pruned_loss=0.06908, over 1529852.74 frames. ], batch size: 52, lr: 5.26e-03, grad_scale: 8.0 2023-04-02 06:17:47,707 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 2023-04-02 06:18:34,758 INFO [train.py:903] (1/4) Epoch 16, batch 150, loss[loss=0.2001, simple_loss=0.2813, pruned_loss=0.05945, over 19837.00 frames. ], tot_loss[loss=0.2175, simple_loss=0.2937, pruned_loss=0.07067, over 2047842.55 frames. ], batch size: 52, lr: 5.26e-03, grad_scale: 8.0 2023-04-02 06:19:07,433 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.312e+02 5.578e+02 6.638e+02 8.298e+02 1.665e+03, threshold=1.328e+03, percent-clipped=4.0 2023-04-02 06:19:36,772 INFO [train.py:903] (1/4) Epoch 16, batch 200, loss[loss=0.2466, simple_loss=0.3178, pruned_loss=0.08768, over 19676.00 frames. ], tot_loss[loss=0.2205, simple_loss=0.297, pruned_loss=0.07199, over 2446815.40 frames. ], batch size: 60, lr: 5.26e-03, grad_scale: 8.0 2023-04-02 06:19:38,846 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 2023-04-02 06:20:18,503 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=102653.0, num_to_drop=1, layers_to_drop={0} 2023-04-02 06:20:38,298 INFO [train.py:903] (1/4) Epoch 16, batch 250, loss[loss=0.2153, simple_loss=0.2761, pruned_loss=0.07722, over 19752.00 frames. ], tot_loss[loss=0.2204, simple_loss=0.2969, pruned_loss=0.07197, over 2756083.33 frames. ], batch size: 45, lr: 5.26e-03, grad_scale: 8.0 2023-04-02 06:20:50,385 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=102678.0, num_to_drop=1, layers_to_drop={0} 2023-04-02 06:21:12,172 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.486e+02 5.293e+02 5.976e+02 7.347e+02 1.638e+03, threshold=1.195e+03, percent-clipped=3.0 2023-04-02 06:21:43,451 INFO [train.py:903] (1/4) Epoch 16, batch 300, loss[loss=0.1932, simple_loss=0.2616, pruned_loss=0.06239, over 19312.00 frames. ], tot_loss[loss=0.2208, simple_loss=0.2972, pruned_loss=0.07223, over 2990763.79 frames. ], batch size: 44, lr: 5.26e-03, grad_scale: 8.0 2023-04-02 06:21:43,849 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=102720.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 06:22:13,592 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=102745.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 06:22:44,947 INFO [train.py:903] (1/4) Epoch 16, batch 350, loss[loss=0.2305, simple_loss=0.3071, pruned_loss=0.0769, over 19677.00 frames. ], tot_loss[loss=0.2206, simple_loss=0.2969, pruned_loss=0.07214, over 3175432.07 frames. ], batch size: 60, lr: 5.25e-03, grad_scale: 8.0 2023-04-02 06:22:50,836 WARNING [train.py:1073] (1/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1_rvb from training. Duration: 0.9681875 2023-04-02 06:23:16,118 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.122e+02 5.254e+02 6.186e+02 7.503e+02 2.205e+03, threshold=1.237e+03, percent-clipped=4.0 2023-04-02 06:23:47,556 INFO [train.py:903] (1/4) Epoch 16, batch 400, loss[loss=0.1921, simple_loss=0.2603, pruned_loss=0.06197, over 19749.00 frames. ], tot_loss[loss=0.2213, simple_loss=0.2979, pruned_loss=0.07235, over 3324594.30 frames. ], batch size: 46, lr: 5.25e-03, grad_scale: 8.0 2023-04-02 06:24:49,412 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=102869.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 06:24:50,281 INFO [train.py:903] (1/4) Epoch 16, batch 450, loss[loss=0.1701, simple_loss=0.249, pruned_loss=0.04564, over 19393.00 frames. ], tot_loss[loss=0.2206, simple_loss=0.2975, pruned_loss=0.07188, over 3440838.88 frames. ], batch size: 48, lr: 5.25e-03, grad_scale: 8.0 2023-04-02 06:25:22,281 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.385e+02 4.837e+02 5.955e+02 8.003e+02 1.401e+03, threshold=1.191e+03, percent-clipped=4.0 2023-04-02 06:25:24,667 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 2023-04-02 06:25:25,879 WARNING [train.py:1073] (1/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 2023-04-02 06:25:32,995 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=102905.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 06:25:52,008 INFO [train.py:903] (1/4) Epoch 16, batch 500, loss[loss=0.1554, simple_loss=0.2376, pruned_loss=0.03663, over 19390.00 frames. ], tot_loss[loss=0.2194, simple_loss=0.2965, pruned_loss=0.0712, over 3535605.14 frames. ], batch size: 47, lr: 5.25e-03, grad_scale: 8.0 2023-04-02 06:26:37,590 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.3842, 2.4630, 2.5554, 3.2712, 2.3759, 3.2491, 2.7258, 2.4204], device='cuda:1'), covar=tensor([0.4151, 0.3652, 0.1704, 0.2097, 0.3997, 0.1706, 0.4060, 0.2875], device='cuda:1'), in_proj_covar=tensor([0.0837, 0.0886, 0.0679, 0.0906, 0.0824, 0.0758, 0.0814, 0.0743], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 06:26:54,173 INFO [train.py:903] (1/4) Epoch 16, batch 550, loss[loss=0.2072, simple_loss=0.2917, pruned_loss=0.06141, over 19579.00 frames. ], tot_loss[loss=0.22, simple_loss=0.297, pruned_loss=0.07153, over 3599039.92 frames. ], batch size: 52, lr: 5.25e-03, grad_scale: 8.0 2023-04-02 06:27:00,535 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=4.43 vs. limit=5.0 2023-04-02 06:27:18,359 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([0.9092, 1.2519, 1.5348, 0.6318, 2.0311, 2.4492, 2.1501, 2.5955], device='cuda:1'), covar=tensor([0.1557, 0.3513, 0.3139, 0.2506, 0.0548, 0.0272, 0.0343, 0.0331], device='cuda:1'), in_proj_covar=tensor([0.0262, 0.0306, 0.0335, 0.0256, 0.0228, 0.0171, 0.0211, 0.0227], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 06:27:24,958 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.789e+02 5.227e+02 6.443e+02 7.702e+02 1.436e+03, threshold=1.289e+03, percent-clipped=3.0 2023-04-02 06:27:54,425 INFO [train.py:903] (1/4) Epoch 16, batch 600, loss[loss=0.2376, simple_loss=0.3025, pruned_loss=0.08632, over 19583.00 frames. ], tot_loss[loss=0.2196, simple_loss=0.2966, pruned_loss=0.07127, over 3647536.12 frames. ], batch size: 52, lr: 5.25e-03, grad_scale: 8.0 2023-04-02 06:28:37,067 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 2023-04-02 06:28:55,609 INFO [train.py:903] (1/4) Epoch 16, batch 650, loss[loss=0.1946, simple_loss=0.2783, pruned_loss=0.05545, over 19542.00 frames. ], tot_loss[loss=0.2197, simple_loss=0.2968, pruned_loss=0.07129, over 3685268.79 frames. ], batch size: 56, lr: 5.25e-03, grad_scale: 8.0 2023-04-02 06:29:28,782 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.210e+02 4.715e+02 6.008e+02 7.942e+02 1.451e+03, threshold=1.202e+03, percent-clipped=1.0 2023-04-02 06:29:53,906 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=103116.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 06:29:58,298 INFO [train.py:903] (1/4) Epoch 16, batch 700, loss[loss=0.1784, simple_loss=0.2567, pruned_loss=0.05006, over 19754.00 frames. ], tot_loss[loss=0.2197, simple_loss=0.2967, pruned_loss=0.0714, over 3722127.32 frames. ], batch size: 46, lr: 5.25e-03, grad_scale: 8.0 2023-04-02 06:30:34,540 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.58 vs. limit=2.0 2023-04-02 06:30:35,138 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=103150.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 06:31:00,302 INFO [train.py:903] (1/4) Epoch 16, batch 750, loss[loss=0.2738, simple_loss=0.3398, pruned_loss=0.1039, over 19606.00 frames. ], tot_loss[loss=0.2212, simple_loss=0.2977, pruned_loss=0.07238, over 3737046.05 frames. ], batch size: 61, lr: 5.24e-03, grad_scale: 8.0 2023-04-02 06:31:33,686 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.142e+02 5.236e+02 6.262e+02 8.063e+02 1.865e+03, threshold=1.252e+03, percent-clipped=5.0 2023-04-02 06:31:55,090 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=103213.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 06:32:03,182 INFO [train.py:903] (1/4) Epoch 16, batch 800, loss[loss=0.245, simple_loss=0.3182, pruned_loss=0.08588, over 18003.00 frames. ], tot_loss[loss=0.2218, simple_loss=0.2982, pruned_loss=0.07269, over 3744505.68 frames. ], batch size: 83, lr: 5.24e-03, grad_scale: 8.0 2023-04-02 06:32:18,120 WARNING [train.py:1073] (1/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 2023-04-02 06:32:26,524 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1126, 1.7836, 1.4600, 1.1710, 1.6145, 1.2174, 1.0850, 1.6489], device='cuda:1'), covar=tensor([0.0780, 0.0812, 0.1023, 0.0763, 0.0524, 0.1185, 0.0636, 0.0391], device='cuda:1'), in_proj_covar=tensor([0.0295, 0.0309, 0.0330, 0.0254, 0.0242, 0.0330, 0.0292, 0.0267], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 06:32:38,488 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=103249.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 06:32:52,100 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([5.0859, 5.1148, 5.9450, 5.9181, 1.8844, 5.6137, 4.6076, 5.5529], device='cuda:1'), covar=tensor([0.1528, 0.0801, 0.0566, 0.0540, 0.5939, 0.0612, 0.0572, 0.1212], device='cuda:1'), in_proj_covar=tensor([0.0730, 0.0666, 0.0869, 0.0746, 0.0772, 0.0615, 0.0519, 0.0802], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-02 06:33:04,736 INFO [train.py:903] (1/4) Epoch 16, batch 850, loss[loss=0.2144, simple_loss=0.2872, pruned_loss=0.07079, over 19472.00 frames. ], tot_loss[loss=0.2195, simple_loss=0.2962, pruned_loss=0.07141, over 3768176.84 frames. ], batch size: 49, lr: 5.24e-03, grad_scale: 8.0 2023-04-02 06:33:38,414 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.826e+02 4.865e+02 6.263e+02 7.829e+02 1.710e+03, threshold=1.253e+03, percent-clipped=2.0 2023-04-02 06:33:40,861 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.7889, 4.0025, 4.3806, 4.3724, 2.6636, 4.0979, 3.7633, 4.1537], device='cuda:1'), covar=tensor([0.1237, 0.2542, 0.0549, 0.0602, 0.3893, 0.0895, 0.0511, 0.0888], device='cuda:1'), in_proj_covar=tensor([0.0730, 0.0665, 0.0870, 0.0747, 0.0770, 0.0616, 0.0520, 0.0802], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-02 06:33:44,513 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4042, 1.5479, 1.5727, 1.9061, 1.4751, 1.6774, 1.7609, 1.4066], device='cuda:1'), covar=tensor([0.4454, 0.3853, 0.2473, 0.2479, 0.3727, 0.2174, 0.5398, 0.4400], device='cuda:1'), in_proj_covar=tensor([0.0841, 0.0891, 0.0684, 0.0911, 0.0828, 0.0764, 0.0814, 0.0747], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 06:33:57,916 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9_rvb from training. Duration: 25.061125 2023-04-02 06:34:06,719 INFO [train.py:903] (1/4) Epoch 16, batch 900, loss[loss=0.1833, simple_loss=0.2693, pruned_loss=0.04863, over 19609.00 frames. ], tot_loss[loss=0.2199, simple_loss=0.2967, pruned_loss=0.07154, over 3783082.35 frames. ], batch size: 50, lr: 5.24e-03, grad_scale: 8.0 2023-04-02 06:34:17,425 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=103328.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 06:35:01,541 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=103364.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 06:35:06,280 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.7578, 2.6185, 1.9869, 1.8967, 1.8622, 2.1437, 1.2583, 1.9524], device='cuda:1'), covar=tensor([0.0575, 0.0537, 0.0635, 0.0951, 0.0943, 0.0910, 0.1021, 0.0837], device='cuda:1'), in_proj_covar=tensor([0.0346, 0.0338, 0.0336, 0.0364, 0.0437, 0.0367, 0.0320, 0.0327], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 06:35:08,179 INFO [train.py:903] (1/4) Epoch 16, batch 950, loss[loss=0.2125, simple_loss=0.2984, pruned_loss=0.06333, over 19538.00 frames. ], tot_loss[loss=0.219, simple_loss=0.296, pruned_loss=0.07103, over 3808300.05 frames. ], batch size: 56, lr: 5.24e-03, grad_scale: 8.0 2023-04-02 06:35:13,554 WARNING [train.py:1073] (1/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9_rvb from training. Duration: 26.32775 2023-04-02 06:35:19,093 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([6.2362, 5.6318, 2.8743, 4.8766, 1.1389, 5.5887, 5.5376, 5.7012], device='cuda:1'), covar=tensor([0.0377, 0.0789, 0.2034, 0.0674, 0.3997, 0.0558, 0.0735, 0.0763], device='cuda:1'), in_proj_covar=tensor([0.0470, 0.0384, 0.0460, 0.0329, 0.0391, 0.0398, 0.0392, 0.0427], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 06:35:27,489 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=103384.0, num_to_drop=1, layers_to_drop={1} 2023-04-02 06:35:40,817 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.146e+02 5.367e+02 6.234e+02 7.823e+02 2.113e+03, threshold=1.247e+03, percent-clipped=3.0 2023-04-02 06:36:12,129 INFO [train.py:903] (1/4) Epoch 16, batch 1000, loss[loss=0.2795, simple_loss=0.3542, pruned_loss=0.1024, over 19723.00 frames. ], tot_loss[loss=0.22, simple_loss=0.2969, pruned_loss=0.07157, over 3814661.21 frames. ], batch size: 63, lr: 5.24e-03, grad_scale: 8.0 2023-04-02 06:37:03,132 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=103460.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 06:37:07,731 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 2023-04-02 06:37:14,579 INFO [train.py:903] (1/4) Epoch 16, batch 1050, loss[loss=0.2141, simple_loss=0.2831, pruned_loss=0.07255, over 19609.00 frames. ], tot_loss[loss=0.2202, simple_loss=0.2964, pruned_loss=0.07203, over 3830727.94 frames. ], batch size: 50, lr: 5.24e-03, grad_scale: 8.0 2023-04-02 06:37:43,713 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=103494.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 06:37:45,897 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.718e+02 5.497e+02 6.450e+02 8.583e+02 2.663e+03, threshold=1.290e+03, percent-clipped=6.0 2023-04-02 06:37:49,187 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 2023-04-02 06:38:15,970 INFO [train.py:903] (1/4) Epoch 16, batch 1100, loss[loss=0.2543, simple_loss=0.3387, pruned_loss=0.08496, over 19608.00 frames. ], tot_loss[loss=0.2216, simple_loss=0.298, pruned_loss=0.0726, over 3818908.33 frames. ], batch size: 61, lr: 5.24e-03, grad_scale: 8.0 2023-04-02 06:38:21,841 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.7855, 4.2833, 4.4814, 4.4667, 1.7055, 4.2008, 3.6680, 4.1627], device='cuda:1'), covar=tensor([0.1476, 0.0782, 0.0552, 0.0606, 0.5493, 0.0812, 0.0601, 0.1071], device='cuda:1'), in_proj_covar=tensor([0.0741, 0.0676, 0.0879, 0.0754, 0.0780, 0.0623, 0.0524, 0.0810], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-02 06:39:18,053 INFO [train.py:903] (1/4) Epoch 16, batch 1150, loss[loss=0.2175, simple_loss=0.2911, pruned_loss=0.07196, over 19687.00 frames. ], tot_loss[loss=0.2209, simple_loss=0.2977, pruned_loss=0.07209, over 3817463.03 frames. ], batch size: 53, lr: 5.23e-03, grad_scale: 8.0 2023-04-02 06:39:24,748 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=103575.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 06:39:37,382 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=103584.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 06:39:50,620 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.809e+02 5.106e+02 6.190e+02 8.719e+02 1.567e+03, threshold=1.238e+03, percent-clipped=4.0 2023-04-02 06:39:58,913 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4852, 1.5285, 2.0102, 1.6770, 3.0578, 2.5714, 3.3614, 1.4731], device='cuda:1'), covar=tensor([0.2287, 0.4000, 0.2525, 0.1851, 0.1513, 0.1976, 0.1538, 0.4012], device='cuda:1'), in_proj_covar=tensor([0.0501, 0.0599, 0.0653, 0.0453, 0.0604, 0.0509, 0.0647, 0.0514], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 06:40:06,805 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=103609.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 06:40:06,846 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=103609.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 06:40:21,166 INFO [train.py:903] (1/4) Epoch 16, batch 1200, loss[loss=0.1849, simple_loss=0.2642, pruned_loss=0.05278, over 19402.00 frames. ], tot_loss[loss=0.2199, simple_loss=0.2968, pruned_loss=0.07149, over 3827177.65 frames. ], batch size: 48, lr: 5.23e-03, grad_scale: 8.0 2023-04-02 06:40:21,584 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=103620.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 06:40:52,908 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=103645.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 06:40:54,983 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 2023-04-02 06:41:24,587 INFO [train.py:903] (1/4) Epoch 16, batch 1250, loss[loss=0.2166, simple_loss=0.2941, pruned_loss=0.06957, over 19739.00 frames. ], tot_loss[loss=0.2204, simple_loss=0.2975, pruned_loss=0.07168, over 3843113.24 frames. ], batch size: 51, lr: 5.23e-03, grad_scale: 8.0 2023-04-02 06:41:56,584 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.406e+02 4.955e+02 6.144e+02 7.729e+02 1.641e+03, threshold=1.229e+03, percent-clipped=4.0 2023-04-02 06:42:20,680 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4371, 1.4649, 2.1754, 1.8189, 3.1110, 4.7465, 4.6885, 5.2202], device='cuda:1'), covar=tensor([0.1505, 0.3432, 0.2804, 0.1897, 0.0527, 0.0200, 0.0150, 0.0155], device='cuda:1'), in_proj_covar=tensor([0.0265, 0.0307, 0.0336, 0.0257, 0.0228, 0.0172, 0.0211, 0.0227], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 06:42:24,690 INFO [train.py:903] (1/4) Epoch 16, batch 1300, loss[loss=0.229, simple_loss=0.3087, pruned_loss=0.07469, over 18059.00 frames. ], tot_loss[loss=0.22, simple_loss=0.2969, pruned_loss=0.07159, over 3841057.39 frames. ], batch size: 83, lr: 5.23e-03, grad_scale: 8.0 2023-04-02 06:42:35,158 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=103728.0, num_to_drop=1, layers_to_drop={1} 2023-04-02 06:42:53,056 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.8273, 1.9611, 2.2257, 2.6513, 1.8753, 2.4244, 2.3752, 2.0670], device='cuda:1'), covar=tensor([0.3893, 0.3377, 0.1633, 0.1928, 0.3712, 0.1781, 0.3907, 0.2890], device='cuda:1'), in_proj_covar=tensor([0.0838, 0.0885, 0.0682, 0.0910, 0.0826, 0.0763, 0.0811, 0.0744], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 06:43:12,120 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.3652, 3.9660, 2.5938, 3.4712, 0.8649, 3.8700, 3.7573, 3.8157], device='cuda:1'), covar=tensor([0.0675, 0.1091, 0.1911, 0.0911, 0.3891, 0.0718, 0.0845, 0.1075], device='cuda:1'), in_proj_covar=tensor([0.0473, 0.0386, 0.0462, 0.0334, 0.0395, 0.0400, 0.0395, 0.0431], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 06:43:12,210 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=103758.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 06:43:26,084 INFO [train.py:903] (1/4) Epoch 16, batch 1350, loss[loss=0.2014, simple_loss=0.2778, pruned_loss=0.06248, over 19590.00 frames. ], tot_loss[loss=0.22, simple_loss=0.2968, pruned_loss=0.07158, over 3840775.24 frames. ], batch size: 52, lr: 5.23e-03, grad_scale: 8.0 2023-04-02 06:43:43,807 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=103783.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 06:43:47,409 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.01 vs. limit=5.0 2023-04-02 06:43:59,428 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.194e+02 5.214e+02 6.633e+02 9.392e+02 2.118e+03, threshold=1.327e+03, percent-clipped=8.0 2023-04-02 06:44:30,120 INFO [train.py:903] (1/4) Epoch 16, batch 1400, loss[loss=0.2201, simple_loss=0.2984, pruned_loss=0.07087, over 19381.00 frames. ], tot_loss[loss=0.2199, simple_loss=0.2965, pruned_loss=0.07166, over 3825555.37 frames. ], batch size: 70, lr: 5.23e-03, grad_scale: 8.0 2023-04-02 06:44:43,946 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=103831.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 06:44:57,984 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=103843.0, num_to_drop=1, layers_to_drop={0} 2023-04-02 06:45:14,492 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=103856.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 06:45:20,193 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6945, 1.4790, 1.4022, 1.7501, 1.4804, 1.4592, 1.3888, 1.6461], device='cuda:1'), covar=tensor([0.0957, 0.1308, 0.1358, 0.0894, 0.1189, 0.0554, 0.1296, 0.0717], device='cuda:1'), in_proj_covar=tensor([0.0261, 0.0354, 0.0298, 0.0246, 0.0301, 0.0247, 0.0293, 0.0248], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 06:45:26,666 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=103865.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 06:45:32,154 INFO [train.py:903] (1/4) Epoch 16, batch 1450, loss[loss=0.2146, simple_loss=0.2958, pruned_loss=0.0667, over 18788.00 frames. ], tot_loss[loss=0.22, simple_loss=0.2965, pruned_loss=0.07175, over 3830331.82 frames. ], batch size: 74, lr: 5.23e-03, grad_scale: 8.0 2023-04-02 06:45:32,196 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_rvb from training. Duration: 25.85 2023-04-02 06:45:42,907 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.3754, 3.9907, 2.6400, 3.6579, 1.0638, 3.8043, 3.8175, 3.8782], device='cuda:1'), covar=tensor([0.0618, 0.1052, 0.1989, 0.0757, 0.3781, 0.0773, 0.0851, 0.0942], device='cuda:1'), in_proj_covar=tensor([0.0470, 0.0385, 0.0461, 0.0331, 0.0392, 0.0398, 0.0393, 0.0430], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 06:45:56,453 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=103890.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 06:46:00,956 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=103894.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 06:46:03,876 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.072e+02 4.634e+02 5.962e+02 7.073e+02 1.523e+03, threshold=1.192e+03, percent-clipped=2.0 2023-04-02 06:46:33,192 INFO [train.py:903] (1/4) Epoch 16, batch 1500, loss[loss=0.2748, simple_loss=0.342, pruned_loss=0.1038, over 19776.00 frames. ], tot_loss[loss=0.2208, simple_loss=0.2974, pruned_loss=0.07204, over 3823632.41 frames. ], batch size: 56, lr: 5.23e-03, grad_scale: 16.0 2023-04-02 06:47:20,908 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 2023-04-02 06:47:35,232 INFO [train.py:903] (1/4) Epoch 16, batch 1550, loss[loss=0.2275, simple_loss=0.3068, pruned_loss=0.07414, over 19522.00 frames. ], tot_loss[loss=0.2194, simple_loss=0.2964, pruned_loss=0.07122, over 3836461.14 frames. ], batch size: 54, lr: 5.22e-03, grad_scale: 8.0 2023-04-02 06:48:09,256 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.399e+02 5.500e+02 6.727e+02 8.636e+02 1.580e+03, threshold=1.345e+03, percent-clipped=8.0 2023-04-02 06:48:16,801 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 2023-04-02 06:48:24,314 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=104008.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 06:48:32,669 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=104015.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 06:48:39,357 INFO [train.py:903] (1/4) Epoch 16, batch 1600, loss[loss=0.1911, simple_loss=0.2684, pruned_loss=0.05691, over 19756.00 frames. ], tot_loss[loss=0.2177, simple_loss=0.295, pruned_loss=0.07018, over 3837580.84 frames. ], batch size: 51, lr: 5.22e-03, grad_scale: 8.0 2023-04-02 06:49:05,526 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9_rvb from training. Duration: 30.1555625 2023-04-02 06:49:10,531 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([0.9951, 1.1702, 1.4783, 0.5850, 2.0312, 2.4314, 2.1075, 2.5696], device='cuda:1'), covar=tensor([0.1544, 0.3720, 0.3248, 0.2615, 0.0561, 0.0255, 0.0343, 0.0338], device='cuda:1'), in_proj_covar=tensor([0.0265, 0.0307, 0.0335, 0.0257, 0.0228, 0.0172, 0.0209, 0.0229], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 06:49:38,437 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=104067.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 06:49:41,614 INFO [train.py:903] (1/4) Epoch 16, batch 1650, loss[loss=0.2013, simple_loss=0.2718, pruned_loss=0.06538, over 19759.00 frames. ], tot_loss[loss=0.2193, simple_loss=0.2966, pruned_loss=0.071, over 3838235.55 frames. ], batch size: 46, lr: 5.22e-03, grad_scale: 8.0 2023-04-02 06:50:14,808 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.025e+02 5.150e+02 6.179e+02 7.587e+02 1.568e+03, threshold=1.236e+03, percent-clipped=4.0 2023-04-02 06:50:18,560 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=104099.0, num_to_drop=1, layers_to_drop={0} 2023-04-02 06:50:21,715 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=104102.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 06:50:37,185 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.5677, 1.1360, 1.3875, 1.1838, 2.2059, 0.9742, 2.1028, 2.4186], device='cuda:1'), covar=tensor([0.0708, 0.2716, 0.2693, 0.1695, 0.0895, 0.2139, 0.0972, 0.0474], device='cuda:1'), in_proj_covar=tensor([0.0382, 0.0351, 0.0370, 0.0335, 0.0358, 0.0341, 0.0352, 0.0376], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 06:50:43,582 INFO [train.py:903] (1/4) Epoch 16, batch 1700, loss[loss=0.2527, simple_loss=0.3341, pruned_loss=0.08563, over 19492.00 frames. ], tot_loss[loss=0.2184, simple_loss=0.2957, pruned_loss=0.07052, over 3851268.59 frames. ], batch size: 64, lr: 5.22e-03, grad_scale: 8.0 2023-04-02 06:50:48,577 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=104124.0, num_to_drop=1, layers_to_drop={0} 2023-04-02 06:50:51,812 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=104127.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 06:51:25,773 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 2023-04-02 06:51:29,501 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9446, 1.8316, 1.6271, 2.1636, 1.9319, 1.8359, 1.7477, 2.0336], device='cuda:1'), covar=tensor([0.1000, 0.1607, 0.1377, 0.0919, 0.1261, 0.0507, 0.1248, 0.0665], device='cuda:1'), in_proj_covar=tensor([0.0263, 0.0356, 0.0300, 0.0249, 0.0303, 0.0249, 0.0295, 0.0248], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 06:51:44,976 INFO [train.py:903] (1/4) Epoch 16, batch 1750, loss[loss=0.1983, simple_loss=0.2826, pruned_loss=0.05693, over 19655.00 frames. ], tot_loss[loss=0.2194, simple_loss=0.2968, pruned_loss=0.07101, over 3848205.08 frames. ], batch size: 53, lr: 5.22e-03, grad_scale: 8.0 2023-04-02 06:52:19,031 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.135e+02 4.884e+02 5.867e+02 6.930e+02 2.034e+03, threshold=1.173e+03, percent-clipped=2.0 2023-04-02 06:52:44,495 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=104217.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 06:52:48,734 INFO [train.py:903] (1/4) Epoch 16, batch 1800, loss[loss=0.1966, simple_loss=0.2683, pruned_loss=0.06248, over 19740.00 frames. ], tot_loss[loss=0.2201, simple_loss=0.2975, pruned_loss=0.0714, over 3834691.82 frames. ], batch size: 47, lr: 5.22e-03, grad_scale: 8.0 2023-04-02 06:53:09,914 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=104238.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 06:53:11,571 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 2023-04-02 06:53:14,849 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=104242.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 06:53:46,191 WARNING [train.py:1073] (1/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9_rvb from training. Duration: 25.0944375 2023-04-02 06:53:49,827 INFO [train.py:903] (1/4) Epoch 16, batch 1850, loss[loss=0.1954, simple_loss=0.268, pruned_loss=0.06134, over 19383.00 frames. ], tot_loss[loss=0.2197, simple_loss=0.2966, pruned_loss=0.07138, over 3826133.71 frames. ], batch size: 47, lr: 5.22e-03, grad_scale: 8.0 2023-04-02 06:54:21,796 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9_rvb from training. Duration: 27.8166875 2023-04-02 06:54:22,914 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.793e+02 5.248e+02 6.749e+02 7.716e+02 1.558e+03, threshold=1.350e+03, percent-clipped=5.0 2023-04-02 06:54:26,627 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0 2023-04-02 06:54:46,616 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.1447, 1.2653, 1.6104, 1.3809, 2.7480, 1.1505, 2.1034, 2.9856], device='cuda:1'), covar=tensor([0.0534, 0.2612, 0.2434, 0.1681, 0.0722, 0.2136, 0.1113, 0.0336], device='cuda:1'), in_proj_covar=tensor([0.0381, 0.0349, 0.0367, 0.0334, 0.0357, 0.0339, 0.0351, 0.0374], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 06:54:51,925 INFO [train.py:903] (1/4) Epoch 16, batch 1900, loss[loss=0.2066, simple_loss=0.2983, pruned_loss=0.05743, over 19773.00 frames. ], tot_loss[loss=0.2196, simple_loss=0.2969, pruned_loss=0.07121, over 3828293.11 frames. ], batch size: 54, lr: 5.22e-03, grad_scale: 4.0 2023-04-02 06:55:09,094 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 2023-04-02 06:55:15,645 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9_rvb from training. Duration: 27.02225 2023-04-02 06:55:31,887 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=104352.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 06:55:34,127 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=104353.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 06:55:39,871 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 2023-04-02 06:55:41,199 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=104359.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 06:55:53,983 INFO [train.py:903] (1/4) Epoch 16, batch 1950, loss[loss=0.2352, simple_loss=0.3117, pruned_loss=0.07934, over 19600.00 frames. ], tot_loss[loss=0.2191, simple_loss=0.2964, pruned_loss=0.07095, over 3830148.45 frames. ], batch size: 57, lr: 5.21e-03, grad_scale: 4.0 2023-04-02 06:56:30,731 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.975e+02 4.688e+02 6.377e+02 8.120e+02 1.703e+03, threshold=1.275e+03, percent-clipped=4.0 2023-04-02 06:56:46,835 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=104411.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 06:56:58,267 INFO [train.py:903] (1/4) Epoch 16, batch 2000, loss[loss=0.2089, simple_loss=0.2957, pruned_loss=0.06109, over 19593.00 frames. ], tot_loss[loss=0.2184, simple_loss=0.2956, pruned_loss=0.07063, over 3813962.84 frames. ], batch size: 61, lr: 5.21e-03, grad_scale: 8.0 2023-04-02 06:57:57,350 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_rvb from training. Duration: 26.8349375 2023-04-02 06:57:57,641 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=104467.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 06:58:00,766 INFO [train.py:903] (1/4) Epoch 16, batch 2050, loss[loss=0.249, simple_loss=0.3136, pruned_loss=0.0922, over 13589.00 frames. ], tot_loss[loss=0.219, simple_loss=0.2963, pruned_loss=0.07089, over 3810883.42 frames. ], batch size: 136, lr: 5.21e-03, grad_scale: 8.0 2023-04-02 06:58:04,673 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=104473.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 06:58:05,641 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=104474.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 06:58:09,153 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.13 vs. limit=2.0 2023-04-02 06:58:14,090 WARNING [train.py:1073] (1/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 2023-04-02 06:58:15,354 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 2023-04-02 06:58:35,726 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.098e+02 4.885e+02 6.080e+02 8.555e+02 1.693e+03, threshold=1.216e+03, percent-clipped=6.0 2023-04-02 06:58:36,108 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=104498.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 06:58:36,139 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=104498.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 06:58:39,798 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.51 vs. limit=2.0 2023-04-02 06:58:40,119 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 2023-04-02 06:59:02,741 INFO [train.py:903] (1/4) Epoch 16, batch 2100, loss[loss=0.2647, simple_loss=0.3208, pruned_loss=0.1043, over 13239.00 frames. ], tot_loss[loss=0.2198, simple_loss=0.2968, pruned_loss=0.07141, over 3796068.47 frames. ], batch size: 135, lr: 5.21e-03, grad_scale: 8.0 2023-04-02 06:59:06,336 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=104523.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 06:59:09,922 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=104526.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 06:59:33,812 WARNING [train.py:1073] (1/4) Exclude cut with ID 453-131332-0000-131866_sp0.9_rvb from training. Duration: 25.3333125 2023-04-02 06:59:55,546 WARNING [train.py:1073] (1/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9_rvb from training. Duration: 26.6166875 2023-04-02 07:00:04,693 INFO [train.py:903] (1/4) Epoch 16, batch 2150, loss[loss=0.2644, simple_loss=0.3331, pruned_loss=0.09782, over 17320.00 frames. ], tot_loss[loss=0.2207, simple_loss=0.2978, pruned_loss=0.07179, over 3799515.79 frames. ], batch size: 101, lr: 5.21e-03, grad_scale: 8.0 2023-04-02 07:00:13,893 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4776, 1.3465, 1.4086, 1.8052, 1.4046, 1.7182, 1.7791, 1.5917], device='cuda:1'), covar=tensor([0.0913, 0.1033, 0.1074, 0.0701, 0.0822, 0.0771, 0.0832, 0.0717], device='cuda:1'), in_proj_covar=tensor([0.0209, 0.0222, 0.0223, 0.0243, 0.0225, 0.0210, 0.0190, 0.0202], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-02 07:00:27,137 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.9275, 4.4738, 2.6303, 3.8644, 0.7878, 4.4109, 4.3248, 4.4210], device='cuda:1'), covar=tensor([0.0575, 0.1041, 0.1993, 0.0742, 0.4076, 0.0601, 0.0756, 0.0890], device='cuda:1'), in_proj_covar=tensor([0.0470, 0.0383, 0.0462, 0.0331, 0.0391, 0.0398, 0.0393, 0.0426], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 07:00:27,628 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.82 vs. limit=2.0 2023-04-02 07:00:39,765 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.712e+02 5.547e+02 6.907e+02 8.298e+02 2.194e+03, threshold=1.381e+03, percent-clipped=3.0 2023-04-02 07:00:53,794 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=104609.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 07:01:08,301 INFO [train.py:903] (1/4) Epoch 16, batch 2200, loss[loss=0.2314, simple_loss=0.3085, pruned_loss=0.0772, over 19762.00 frames. ], tot_loss[loss=0.2212, simple_loss=0.2981, pruned_loss=0.07213, over 3808673.91 frames. ], batch size: 56, lr: 5.21e-03, grad_scale: 8.0 2023-04-02 07:01:11,124 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9916, 2.0792, 2.2363, 2.7431, 1.9791, 2.5119, 2.4241, 2.0702], device='cuda:1'), covar=tensor([0.3975, 0.3564, 0.1689, 0.2151, 0.3817, 0.1933, 0.3844, 0.2986], device='cuda:1'), in_proj_covar=tensor([0.0842, 0.0888, 0.0680, 0.0907, 0.0822, 0.0760, 0.0810, 0.0743], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 07:01:12,378 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.1530, 2.0172, 1.8632, 1.7262, 1.5128, 1.6798, 0.3954, 0.9695], device='cuda:1'), covar=tensor([0.0460, 0.0524, 0.0402, 0.0674, 0.1087, 0.0782, 0.1166, 0.0944], device='cuda:1'), in_proj_covar=tensor([0.0351, 0.0345, 0.0342, 0.0371, 0.0443, 0.0373, 0.0325, 0.0331], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 07:01:26,397 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=104634.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 07:02:12,264 INFO [train.py:903] (1/4) Epoch 16, batch 2250, loss[loss=0.2475, simple_loss=0.3161, pruned_loss=0.08939, over 18734.00 frames. ], tot_loss[loss=0.2207, simple_loss=0.298, pruned_loss=0.07173, over 3823448.40 frames. ], batch size: 74, lr: 5.21e-03, grad_scale: 8.0 2023-04-02 07:02:46,757 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.071e+02 4.921e+02 5.898e+02 6.952e+02 1.646e+03, threshold=1.180e+03, percent-clipped=1.0 2023-04-02 07:03:04,277 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2410, 1.1364, 1.2023, 1.3901, 1.1133, 1.2908, 1.3296, 1.2417], device='cuda:1'), covar=tensor([0.0874, 0.1029, 0.1067, 0.0640, 0.0810, 0.0879, 0.0826, 0.0815], device='cuda:1'), in_proj_covar=tensor([0.0207, 0.0221, 0.0222, 0.0241, 0.0225, 0.0209, 0.0190, 0.0202], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-02 07:03:15,202 INFO [train.py:903] (1/4) Epoch 16, batch 2300, loss[loss=0.2322, simple_loss=0.2996, pruned_loss=0.0824, over 19474.00 frames. ], tot_loss[loss=0.2215, simple_loss=0.2986, pruned_loss=0.07219, over 3824283.49 frames. ], batch size: 49, lr: 5.21e-03, grad_scale: 8.0 2023-04-02 07:03:19,228 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=104723.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 07:03:20,851 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 2023-04-02 07:03:27,440 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=104730.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 07:03:29,362 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_rvb from training. Duration: 26.205 2023-04-02 07:03:51,604 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=104748.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 07:03:59,864 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=104755.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 07:04:12,348 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.2904, 3.7378, 3.8516, 3.8669, 1.5483, 3.6038, 3.2079, 3.5786], device='cuda:1'), covar=tensor([0.1503, 0.1034, 0.0666, 0.0647, 0.5402, 0.0951, 0.0646, 0.1204], device='cuda:1'), in_proj_covar=tensor([0.0726, 0.0666, 0.0869, 0.0746, 0.0768, 0.0618, 0.0515, 0.0798], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-02 07:04:17,954 INFO [train.py:903] (1/4) Epoch 16, batch 2350, loss[loss=0.1652, simple_loss=0.243, pruned_loss=0.04369, over 19734.00 frames. ], tot_loss[loss=0.22, simple_loss=0.297, pruned_loss=0.07147, over 3820394.28 frames. ], batch size: 46, lr: 5.20e-03, grad_scale: 8.0 2023-04-02 07:04:34,915 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=104782.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 07:04:53,849 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.557e+02 5.048e+02 6.738e+02 8.844e+02 1.580e+03, threshold=1.348e+03, percent-clipped=5.0 2023-04-02 07:05:00,896 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_rvb from training. Duration: 25.775 2023-04-02 07:05:05,761 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=104807.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 07:05:18,350 WARNING [train.py:1073] (1/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9_rvb from training. Duration: 25.45 2023-04-02 07:05:22,503 INFO [train.py:903] (1/4) Epoch 16, batch 2400, loss[loss=0.2407, simple_loss=0.3081, pruned_loss=0.08668, over 19654.00 frames. ], tot_loss[loss=0.2199, simple_loss=0.297, pruned_loss=0.07145, over 3814611.76 frames. ], batch size: 53, lr: 5.20e-03, grad_scale: 8.0 2023-04-02 07:06:24,577 INFO [train.py:903] (1/4) Epoch 16, batch 2450, loss[loss=0.2199, simple_loss=0.3075, pruned_loss=0.06609, over 19667.00 frames. ], tot_loss[loss=0.2218, simple_loss=0.2988, pruned_loss=0.07238, over 3807296.50 frames. ], batch size: 55, lr: 5.20e-03, grad_scale: 8.0 2023-04-02 07:07:00,037 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.524e+02 5.877e+02 7.599e+02 9.302e+02 2.010e+03, threshold=1.520e+03, percent-clipped=8.0 2023-04-02 07:07:27,191 INFO [train.py:903] (1/4) Epoch 16, batch 2500, loss[loss=0.1821, simple_loss=0.2599, pruned_loss=0.05215, over 19738.00 frames. ], tot_loss[loss=0.2222, simple_loss=0.2991, pruned_loss=0.07263, over 3808593.90 frames. ], batch size: 45, lr: 5.20e-03, grad_scale: 8.0 2023-04-02 07:07:39,241 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.63 vs. limit=2.0 2023-04-02 07:08:26,454 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.3361, 3.0273, 2.2007, 2.7618, 0.7687, 2.8981, 2.8395, 2.9282], device='cuda:1'), covar=tensor([0.1053, 0.1273, 0.2098, 0.1015, 0.4069, 0.1014, 0.1176, 0.1303], device='cuda:1'), in_proj_covar=tensor([0.0469, 0.0384, 0.0459, 0.0329, 0.0390, 0.0396, 0.0391, 0.0424], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 07:08:29,688 INFO [train.py:903] (1/4) Epoch 16, batch 2550, loss[loss=0.2246, simple_loss=0.3108, pruned_loss=0.06923, over 19666.00 frames. ], tot_loss[loss=0.2204, simple_loss=0.2974, pruned_loss=0.07165, over 3819385.71 frames. ], batch size: 59, lr: 5.20e-03, grad_scale: 8.0 2023-04-02 07:09:05,175 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.330e+02 5.239e+02 6.384e+02 8.143e+02 1.987e+03, threshold=1.277e+03, percent-clipped=1.0 2023-04-02 07:09:24,251 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.45 vs. limit=2.0 2023-04-02 07:09:24,641 WARNING [train.py:1073] (1/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 2023-04-02 07:09:32,158 INFO [train.py:903] (1/4) Epoch 16, batch 2600, loss[loss=0.2375, simple_loss=0.3148, pruned_loss=0.08008, over 19696.00 frames. ], tot_loss[loss=0.2196, simple_loss=0.2968, pruned_loss=0.07123, over 3818462.06 frames. ], batch size: 59, lr: 5.20e-03, grad_scale: 8.0 2023-04-02 07:10:03,947 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.8648, 2.3911, 2.4823, 2.8939, 2.6095, 2.5482, 2.3056, 2.8984], device='cuda:1'), covar=tensor([0.0763, 0.1510, 0.1100, 0.0912, 0.1231, 0.0447, 0.1169, 0.0506], device='cuda:1'), in_proj_covar=tensor([0.0261, 0.0352, 0.0298, 0.0247, 0.0300, 0.0249, 0.0294, 0.0248], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 07:10:21,404 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2327, 2.0849, 1.6815, 1.3410, 1.9083, 1.3131, 1.1678, 1.7742], device='cuda:1'), covar=tensor([0.0876, 0.0744, 0.0942, 0.0825, 0.0458, 0.1177, 0.0693, 0.0388], device='cuda:1'), in_proj_covar=tensor([0.0297, 0.0313, 0.0329, 0.0258, 0.0243, 0.0333, 0.0294, 0.0268], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 07:10:35,320 INFO [train.py:903] (1/4) Epoch 16, batch 2650, loss[loss=0.2131, simple_loss=0.2977, pruned_loss=0.06428, over 19680.00 frames. ], tot_loss[loss=0.2193, simple_loss=0.2963, pruned_loss=0.07111, over 3830133.76 frames. ], batch size: 53, lr: 5.20e-03, grad_scale: 8.0 2023-04-02 07:10:54,817 WARNING [train.py:1073] (1/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9_rvb from training. Duration: 27.25 2023-04-02 07:11:09,898 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.780e+02 5.317e+02 6.316e+02 8.001e+02 1.365e+03, threshold=1.263e+03, percent-clipped=2.0 2023-04-02 07:11:36,915 INFO [train.py:903] (1/4) Epoch 16, batch 2700, loss[loss=0.246, simple_loss=0.3148, pruned_loss=0.08863, over 18291.00 frames. ], tot_loss[loss=0.2208, simple_loss=0.2971, pruned_loss=0.07226, over 3817780.88 frames. ], batch size: 83, lr: 5.20e-03, grad_scale: 8.0 2023-04-02 07:12:39,773 INFO [train.py:903] (1/4) Epoch 16, batch 2750, loss[loss=0.2196, simple_loss=0.282, pruned_loss=0.07855, over 19758.00 frames. ], tot_loss[loss=0.2211, simple_loss=0.2973, pruned_loss=0.07246, over 3826813.74 frames. ], batch size: 45, lr: 5.19e-03, grad_scale: 8.0 2023-04-02 07:13:15,057 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.123e+02 5.170e+02 6.445e+02 8.543e+02 2.677e+03, threshold=1.289e+03, percent-clipped=7.0 2023-04-02 07:13:41,800 INFO [train.py:903] (1/4) Epoch 16, batch 2800, loss[loss=0.2282, simple_loss=0.2993, pruned_loss=0.07859, over 19587.00 frames. ], tot_loss[loss=0.2197, simple_loss=0.2961, pruned_loss=0.07164, over 3832736.87 frames. ], batch size: 52, lr: 5.19e-03, grad_scale: 8.0 2023-04-02 07:14:44,336 INFO [train.py:903] (1/4) Epoch 16, batch 2850, loss[loss=0.2315, simple_loss=0.3066, pruned_loss=0.0782, over 18842.00 frames. ], tot_loss[loss=0.2189, simple_loss=0.2951, pruned_loss=0.0713, over 3827067.73 frames. ], batch size: 74, lr: 5.19e-03, grad_scale: 8.0 2023-04-02 07:15:18,939 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.908e+02 5.066e+02 6.361e+02 8.222e+02 2.548e+03, threshold=1.272e+03, percent-clipped=4.0 2023-04-02 07:15:46,241 INFO [train.py:903] (1/4) Epoch 16, batch 2900, loss[loss=0.2099, simple_loss=0.2965, pruned_loss=0.06162, over 19483.00 frames. ], tot_loss[loss=0.2201, simple_loss=0.2966, pruned_loss=0.07183, over 3819283.81 frames. ], batch size: 64, lr: 5.19e-03, grad_scale: 8.0 2023-04-02 07:15:46,268 WARNING [train.py:1073] (1/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 2023-04-02 07:16:48,796 INFO [train.py:903] (1/4) Epoch 16, batch 2950, loss[loss=0.2491, simple_loss=0.3284, pruned_loss=0.08488, over 19527.00 frames. ], tot_loss[loss=0.2193, simple_loss=0.2962, pruned_loss=0.07117, over 3818074.14 frames. ], batch size: 54, lr: 5.19e-03, grad_scale: 8.0 2023-04-02 07:17:23,862 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.147e+02 4.947e+02 6.195e+02 7.724e+02 2.015e+03, threshold=1.239e+03, percent-clipped=3.0 2023-04-02 07:17:24,309 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.1320, 1.8790, 1.7473, 2.1412, 1.9745, 1.8803, 1.6902, 2.0829], device='cuda:1'), covar=tensor([0.0982, 0.1644, 0.1399, 0.1005, 0.1312, 0.0512, 0.1383, 0.0644], device='cuda:1'), in_proj_covar=tensor([0.0262, 0.0353, 0.0299, 0.0247, 0.0299, 0.0250, 0.0295, 0.0248], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 07:17:33,573 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=105406.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 07:17:50,879 INFO [train.py:903] (1/4) Epoch 16, batch 3000, loss[loss=0.2751, simple_loss=0.3398, pruned_loss=0.1052, over 19571.00 frames. ], tot_loss[loss=0.22, simple_loss=0.2968, pruned_loss=0.07163, over 3807358.92 frames. ], batch size: 61, lr: 5.19e-03, grad_scale: 8.0 2023-04-02 07:17:50,879 INFO [train.py:928] (1/4) Computing validation loss 2023-04-02 07:18:04,140 INFO [train.py:937] (1/4) Epoch 16, validation: loss=0.1725, simple_loss=0.273, pruned_loss=0.03604, over 944034.00 frames. 2023-04-02 07:18:04,141 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18821MB 2023-04-02 07:18:07,769 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_rvb from training. Duration: 29.735 2023-04-02 07:19:07,121 INFO [train.py:903] (1/4) Epoch 16, batch 3050, loss[loss=0.1825, simple_loss=0.2512, pruned_loss=0.05696, over 18714.00 frames. ], tot_loss[loss=0.2202, simple_loss=0.2969, pruned_loss=0.07172, over 3794294.07 frames. ], batch size: 41, lr: 5.19e-03, grad_scale: 8.0 2023-04-02 07:19:41,628 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.577e+02 4.913e+02 6.190e+02 8.953e+02 2.496e+03, threshold=1.238e+03, percent-clipped=7.0 2023-04-02 07:19:49,971 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=105504.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 07:20:11,579 INFO [train.py:903] (1/4) Epoch 16, batch 3100, loss[loss=0.1893, simple_loss=0.2627, pruned_loss=0.05791, over 19770.00 frames. ], tot_loss[loss=0.2189, simple_loss=0.296, pruned_loss=0.0709, over 3799995.64 frames. ], batch size: 46, lr: 5.19e-03, grad_scale: 8.0 2023-04-02 07:20:11,899 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=105520.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 07:20:40,054 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.0971, 1.6611, 2.2141, 1.5846, 3.0593, 4.6439, 4.6570, 5.0501], device='cuda:1'), covar=tensor([0.1718, 0.3328, 0.2863, 0.2091, 0.0559, 0.0179, 0.0160, 0.0157], device='cuda:1'), in_proj_covar=tensor([0.0265, 0.0307, 0.0337, 0.0255, 0.0230, 0.0174, 0.0210, 0.0230], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 07:21:13,221 INFO [train.py:903] (1/4) Epoch 16, batch 3150, loss[loss=0.1849, simple_loss=0.2657, pruned_loss=0.05203, over 19463.00 frames. ], tot_loss[loss=0.2177, simple_loss=0.2949, pruned_loss=0.07023, over 3812831.70 frames. ], batch size: 49, lr: 5.18e-03, grad_scale: 8.0 2023-04-02 07:21:41,341 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 2023-04-02 07:21:46,677 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.213e+02 5.014e+02 6.069e+02 7.545e+02 2.509e+03, threshold=1.214e+03, percent-clipped=2.0 2023-04-02 07:22:12,922 INFO [train.py:903] (1/4) Epoch 16, batch 3200, loss[loss=0.2179, simple_loss=0.2861, pruned_loss=0.07485, over 19104.00 frames. ], tot_loss[loss=0.2194, simple_loss=0.296, pruned_loss=0.0714, over 3794327.14 frames. ], batch size: 42, lr: 5.18e-03, grad_scale: 8.0 2023-04-02 07:23:15,485 INFO [train.py:903] (1/4) Epoch 16, batch 3250, loss[loss=0.1834, simple_loss=0.2633, pruned_loss=0.05173, over 19391.00 frames. ], tot_loss[loss=0.2186, simple_loss=0.2956, pruned_loss=0.07075, over 3796008.79 frames. ], batch size: 48, lr: 5.18e-03, grad_scale: 8.0 2023-04-02 07:23:50,144 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.698e+02 5.141e+02 5.939e+02 7.859e+02 1.653e+03, threshold=1.188e+03, percent-clipped=2.0 2023-04-02 07:24:19,069 INFO [train.py:903] (1/4) Epoch 16, batch 3300, loss[loss=0.2602, simple_loss=0.3288, pruned_loss=0.09576, over 19059.00 frames. ], tot_loss[loss=0.2193, simple_loss=0.2962, pruned_loss=0.07124, over 3782308.83 frames. ], batch size: 69, lr: 5.18e-03, grad_scale: 8.0 2023-04-02 07:24:19,302 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.3679, 3.9794, 2.5421, 3.5669, 0.9845, 3.9194, 3.8212, 3.9207], device='cuda:1'), covar=tensor([0.0727, 0.0947, 0.2004, 0.0771, 0.3775, 0.0672, 0.0795, 0.1162], device='cuda:1'), in_proj_covar=tensor([0.0471, 0.0387, 0.0463, 0.0328, 0.0393, 0.0398, 0.0394, 0.0428], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 07:24:22,592 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 2023-04-02 07:24:48,424 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2926, 1.1662, 1.1576, 1.4837, 1.2521, 1.3852, 1.4166, 1.2838], device='cuda:1'), covar=tensor([0.0610, 0.0748, 0.0773, 0.0577, 0.0802, 0.0614, 0.0784, 0.0578], device='cuda:1'), in_proj_covar=tensor([0.0209, 0.0220, 0.0221, 0.0240, 0.0224, 0.0207, 0.0189, 0.0201], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-02 07:24:55,822 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=105750.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 07:25:21,450 INFO [train.py:903] (1/4) Epoch 16, batch 3350, loss[loss=0.2225, simple_loss=0.3027, pruned_loss=0.07114, over 19654.00 frames. ], tot_loss[loss=0.2186, simple_loss=0.2959, pruned_loss=0.07063, over 3795437.02 frames. ], batch size: 58, lr: 5.18e-03, grad_scale: 8.0 2023-04-02 07:25:31,928 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 2023-04-02 07:25:57,718 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.211e+02 5.066e+02 6.242e+02 8.095e+02 2.652e+03, threshold=1.248e+03, percent-clipped=5.0 2023-04-02 07:26:12,284 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=105810.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 07:26:20,504 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=105817.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 07:26:23,657 INFO [train.py:903] (1/4) Epoch 16, batch 3400, loss[loss=0.244, simple_loss=0.3237, pruned_loss=0.08211, over 18251.00 frames. ], tot_loss[loss=0.2198, simple_loss=0.2971, pruned_loss=0.07128, over 3786545.95 frames. ], batch size: 83, lr: 5.18e-03, grad_scale: 8.0 2023-04-02 07:26:29,233 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.74 vs. limit=2.0 2023-04-02 07:27:00,008 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=105848.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 07:27:18,403 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=105864.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 07:27:19,669 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=105865.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 07:27:26,112 INFO [train.py:903] (1/4) Epoch 16, batch 3450, loss[loss=0.1989, simple_loss=0.2794, pruned_loss=0.05923, over 19834.00 frames. ], tot_loss[loss=0.221, simple_loss=0.2982, pruned_loss=0.07191, over 3788802.84 frames. ], batch size: 52, lr: 5.18e-03, grad_scale: 8.0 2023-04-02 07:27:29,348 WARNING [train.py:1073] (1/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9_rvb from training. Duration: 25.2444375 2023-04-02 07:28:00,500 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.376e+02 5.355e+02 6.157e+02 7.690e+02 1.854e+03, threshold=1.231e+03, percent-clipped=4.0 2023-04-02 07:28:29,097 INFO [train.py:903] (1/4) Epoch 16, batch 3500, loss[loss=0.2034, simple_loss=0.2763, pruned_loss=0.0653, over 19777.00 frames. ], tot_loss[loss=0.2204, simple_loss=0.2977, pruned_loss=0.07155, over 3798944.42 frames. ], batch size: 48, lr: 5.18e-03, grad_scale: 8.0 2023-04-02 07:29:23,565 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=105963.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 07:29:31,025 INFO [train.py:903] (1/4) Epoch 16, batch 3550, loss[loss=0.1646, simple_loss=0.242, pruned_loss=0.04358, over 19697.00 frames. ], tot_loss[loss=0.2216, simple_loss=0.2987, pruned_loss=0.07226, over 3776598.39 frames. ], batch size: 45, lr: 5.17e-03, grad_scale: 8.0 2023-04-02 07:29:42,161 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=105979.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 07:30:05,948 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2945, 1.3819, 1.5540, 1.5128, 2.3326, 2.0433, 2.4473, 0.9436], device='cuda:1'), covar=tensor([0.2409, 0.4175, 0.2581, 0.1931, 0.1554, 0.2154, 0.1476, 0.4388], device='cuda:1'), in_proj_covar=tensor([0.0508, 0.0609, 0.0662, 0.0462, 0.0606, 0.0512, 0.0649, 0.0521], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 07:30:06,596 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.318e+02 5.018e+02 6.219e+02 7.272e+02 1.453e+03, threshold=1.244e+03, percent-clipped=3.0 2023-04-02 07:30:34,296 INFO [train.py:903] (1/4) Epoch 16, batch 3600, loss[loss=0.1793, simple_loss=0.2535, pruned_loss=0.05261, over 19770.00 frames. ], tot_loss[loss=0.2211, simple_loss=0.2985, pruned_loss=0.07188, over 3797660.95 frames. ], batch size: 45, lr: 5.17e-03, grad_scale: 8.0 2023-04-02 07:31:37,379 INFO [train.py:903] (1/4) Epoch 16, batch 3650, loss[loss=0.2176, simple_loss=0.304, pruned_loss=0.06562, over 19795.00 frames. ], tot_loss[loss=0.2201, simple_loss=0.2976, pruned_loss=0.07132, over 3803545.90 frames. ], batch size: 56, lr: 5.17e-03, grad_scale: 8.0 2023-04-02 07:32:09,753 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=106095.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 07:32:13,040 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.400e+02 5.371e+02 6.801e+02 8.277e+02 1.518e+03, threshold=1.360e+03, percent-clipped=5.0 2023-04-02 07:32:42,531 INFO [train.py:903] (1/4) Epoch 16, batch 3700, loss[loss=0.2206, simple_loss=0.2996, pruned_loss=0.07083, over 19760.00 frames. ], tot_loss[loss=0.2207, simple_loss=0.298, pruned_loss=0.07173, over 3802761.98 frames. ], batch size: 63, lr: 5.17e-03, grad_scale: 8.0 2023-04-02 07:32:44,093 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=106121.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 07:33:13,577 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=106146.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 07:33:23,621 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=106154.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 07:33:34,046 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=106161.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 07:33:44,490 INFO [train.py:903] (1/4) Epoch 16, batch 3750, loss[loss=0.241, simple_loss=0.3306, pruned_loss=0.07575, over 19756.00 frames. ], tot_loss[loss=0.2194, simple_loss=0.297, pruned_loss=0.07084, over 3808162.95 frames. ], batch size: 63, lr: 5.17e-03, grad_scale: 8.0 2023-04-02 07:34:19,089 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.576e+02 4.705e+02 5.517e+02 6.973e+02 1.532e+03, threshold=1.103e+03, percent-clipped=3.0 2023-04-02 07:34:24,801 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.09 vs. limit=2.0 2023-04-02 07:34:45,484 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=106219.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 07:34:46,193 INFO [train.py:903] (1/4) Epoch 16, batch 3800, loss[loss=0.2082, simple_loss=0.2824, pruned_loss=0.06707, over 19624.00 frames. ], tot_loss[loss=0.219, simple_loss=0.2966, pruned_loss=0.07066, over 3815065.65 frames. ], batch size: 50, lr: 5.17e-03, grad_scale: 8.0 2023-04-02 07:35:06,414 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=106235.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 07:35:18,148 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9_rvb from training. Duration: 29.1166875 2023-04-02 07:35:18,445 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=106244.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 07:35:36,721 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=106260.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 07:35:48,043 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=106269.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 07:35:48,820 INFO [train.py:903] (1/4) Epoch 16, batch 3850, loss[loss=0.1928, simple_loss=0.2772, pruned_loss=0.0542, over 19389.00 frames. ], tot_loss[loss=0.218, simple_loss=0.2959, pruned_loss=0.07006, over 3816986.69 frames. ], batch size: 48, lr: 5.17e-03, grad_scale: 8.0 2023-04-02 07:35:56,814 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=106276.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 07:36:09,579 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=106286.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 07:36:23,144 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.040e+02 5.267e+02 6.364e+02 7.734e+02 1.610e+03, threshold=1.273e+03, percent-clipped=5.0 2023-04-02 07:36:50,356 INFO [train.py:903] (1/4) Epoch 16, batch 3900, loss[loss=0.2484, simple_loss=0.3279, pruned_loss=0.08449, over 19601.00 frames. ], tot_loss[loss=0.2175, simple_loss=0.2959, pruned_loss=0.06961, over 3830057.63 frames. ], batch size: 57, lr: 5.17e-03, grad_scale: 16.0 2023-04-02 07:37:32,175 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=106354.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 07:37:53,351 INFO [train.py:903] (1/4) Epoch 16, batch 3950, loss[loss=0.1842, simple_loss=0.2612, pruned_loss=0.05359, over 19763.00 frames. ], tot_loss[loss=0.2172, simple_loss=0.2953, pruned_loss=0.06958, over 3821851.28 frames. ], batch size: 46, lr: 5.16e-03, grad_scale: 8.0 2023-04-02 07:37:58,036 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 2023-04-02 07:38:28,357 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.667e+02 4.931e+02 6.019e+02 7.839e+02 1.392e+03, threshold=1.204e+03, percent-clipped=3.0 2023-04-02 07:38:54,934 INFO [train.py:903] (1/4) Epoch 16, batch 4000, loss[loss=0.2286, simple_loss=0.3129, pruned_loss=0.07218, over 18020.00 frames. ], tot_loss[loss=0.2165, simple_loss=0.2944, pruned_loss=0.06924, over 3826190.36 frames. ], batch size: 83, lr: 5.16e-03, grad_scale: 8.0 2023-04-02 07:39:03,764 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 2023-04-02 07:39:19,323 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=106439.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 07:39:23,134 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.3865, 3.0994, 2.2673, 2.3767, 2.4090, 2.5857, 0.8220, 2.2109], device='cuda:1'), covar=tensor([0.0582, 0.0547, 0.0620, 0.0979, 0.0839, 0.0975, 0.1312, 0.0897], device='cuda:1'), in_proj_covar=tensor([0.0352, 0.0345, 0.0343, 0.0371, 0.0445, 0.0375, 0.0323, 0.0330], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 07:39:45,507 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1_rvb from training. Duration: 0.7545625 2023-04-02 07:39:56,761 INFO [train.py:903] (1/4) Epoch 16, batch 4050, loss[loss=0.1839, simple_loss=0.2667, pruned_loss=0.05052, over 19462.00 frames. ], tot_loss[loss=0.216, simple_loss=0.294, pruned_loss=0.06899, over 3824058.04 frames. ], batch size: 49, lr: 5.16e-03, grad_scale: 8.0 2023-04-02 07:39:59,431 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=106472.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 07:40:22,932 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.9086, 2.7011, 2.1197, 2.1458, 1.8796, 2.2783, 1.0390, 1.9837], device='cuda:1'), covar=tensor([0.0549, 0.0559, 0.0575, 0.0930, 0.1014, 0.1098, 0.1187, 0.0857], device='cuda:1'), in_proj_covar=tensor([0.0353, 0.0346, 0.0345, 0.0372, 0.0447, 0.0378, 0.0325, 0.0332], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 07:40:34,188 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.509e+02 4.912e+02 6.033e+02 8.007e+02 1.551e+03, threshold=1.207e+03, percent-clipped=4.0 2023-04-02 07:40:59,899 INFO [train.py:903] (1/4) Epoch 16, batch 4100, loss[loss=0.2009, simple_loss=0.288, pruned_loss=0.05692, over 19689.00 frames. ], tot_loss[loss=0.2162, simple_loss=0.2942, pruned_loss=0.0691, over 3822798.35 frames. ], batch size: 59, lr: 5.16e-03, grad_scale: 8.0 2023-04-02 07:41:07,079 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=106525.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 07:41:15,885 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=106532.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 07:41:34,839 WARNING [train.py:1073] (1/4) Exclude cut with ID 497-129325-0061-9566_sp1.1_rvb from training. Duration: 0.97725 2023-04-02 07:41:36,425 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=106550.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 07:41:41,945 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=106554.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 07:41:45,528 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=106557.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 07:42:02,445 INFO [train.py:903] (1/4) Epoch 16, batch 4150, loss[loss=0.2323, simple_loss=0.3106, pruned_loss=0.07694, over 19670.00 frames. ], tot_loss[loss=0.2158, simple_loss=0.2936, pruned_loss=0.06898, over 3835953.01 frames. ], batch size: 58, lr: 5.16e-03, grad_scale: 8.0 2023-04-02 07:42:20,407 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3804, 1.4338, 1.8033, 1.6420, 2.7371, 2.1747, 2.8627, 1.3682], device='cuda:1'), covar=tensor([0.2328, 0.4050, 0.2481, 0.1751, 0.1386, 0.2012, 0.1346, 0.3843], device='cuda:1'), in_proj_covar=tensor([0.0506, 0.0606, 0.0658, 0.0458, 0.0602, 0.0506, 0.0647, 0.0517], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 07:42:35,031 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0845, 1.9791, 1.7244, 1.5848, 1.4745, 1.6253, 0.4776, 1.0356], device='cuda:1'), covar=tensor([0.0556, 0.0589, 0.0436, 0.0757, 0.1095, 0.0942, 0.1117, 0.0932], device='cuda:1'), in_proj_covar=tensor([0.0352, 0.0346, 0.0343, 0.0371, 0.0446, 0.0375, 0.0324, 0.0331], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 07:42:36,771 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.179e+02 4.926e+02 6.073e+02 7.216e+02 1.422e+03, threshold=1.215e+03, percent-clipped=1.0 2023-04-02 07:43:03,954 INFO [train.py:903] (1/4) Epoch 16, batch 4200, loss[loss=0.1776, simple_loss=0.2693, pruned_loss=0.04292, over 19656.00 frames. ], tot_loss[loss=0.2158, simple_loss=0.2939, pruned_loss=0.06888, over 3835809.13 frames. ], batch size: 60, lr: 5.16e-03, grad_scale: 8.0 2023-04-02 07:43:11,039 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 2023-04-02 07:43:15,841 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=106630.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 07:44:05,134 INFO [train.py:903] (1/4) Epoch 16, batch 4250, loss[loss=0.1899, simple_loss=0.2698, pruned_loss=0.05501, over 19487.00 frames. ], tot_loss[loss=0.2176, simple_loss=0.2953, pruned_loss=0.06996, over 3816506.59 frames. ], batch size: 49, lr: 5.16e-03, grad_scale: 8.0 2023-04-02 07:44:20,457 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9_rvb from training. Duration: 29.816625 2023-04-02 07:44:32,738 WARNING [train.py:1073] (1/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 2023-04-02 07:44:41,264 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=106698.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 07:44:42,277 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.502e+02 4.922e+02 6.114e+02 7.451e+02 1.808e+03, threshold=1.223e+03, percent-clipped=2.0 2023-04-02 07:45:08,519 INFO [train.py:903] (1/4) Epoch 16, batch 4300, loss[loss=0.1911, simple_loss=0.264, pruned_loss=0.05907, over 19705.00 frames. ], tot_loss[loss=0.2173, simple_loss=0.295, pruned_loss=0.06976, over 3812884.95 frames. ], batch size: 46, lr: 5.16e-03, grad_scale: 8.0 2023-04-02 07:45:38,991 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=106745.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 07:45:58,078 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 2023-04-02 07:45:58,462 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9231, 3.4208, 2.0427, 2.1748, 3.0809, 1.6722, 1.3516, 2.0962], device='cuda:1'), covar=tensor([0.1347, 0.0561, 0.1012, 0.0737, 0.0515, 0.1180, 0.0974, 0.0673], device='cuda:1'), in_proj_covar=tensor([0.0297, 0.0311, 0.0329, 0.0254, 0.0244, 0.0330, 0.0291, 0.0266], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 07:46:11,653 INFO [train.py:903] (1/4) Epoch 16, batch 4350, loss[loss=0.2047, simple_loss=0.2812, pruned_loss=0.06413, over 19590.00 frames. ], tot_loss[loss=0.2173, simple_loss=0.2951, pruned_loss=0.06978, over 3811845.83 frames. ], batch size: 52, lr: 5.15e-03, grad_scale: 8.0 2023-04-02 07:46:46,926 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.450e+02 5.102e+02 6.153e+02 8.101e+02 2.041e+03, threshold=1.231e+03, percent-clipped=8.0 2023-04-02 07:47:03,114 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=106810.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 07:47:06,283 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=106813.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 07:47:09,625 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=106816.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 07:47:14,210 INFO [train.py:903] (1/4) Epoch 16, batch 4400, loss[loss=0.2188, simple_loss=0.3019, pruned_loss=0.06783, over 19475.00 frames. ], tot_loss[loss=0.2161, simple_loss=0.2936, pruned_loss=0.06931, over 3808844.69 frames. ], batch size: 64, lr: 5.15e-03, grad_scale: 8.0 2023-04-02 07:47:31,039 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1801, 1.0676, 1.1315, 1.4475, 1.0288, 1.2309, 1.3479, 1.2147], device='cuda:1'), covar=tensor([0.1051, 0.1277, 0.1276, 0.0685, 0.0945, 0.1045, 0.0925, 0.0909], device='cuda:1'), in_proj_covar=tensor([0.0211, 0.0223, 0.0224, 0.0246, 0.0227, 0.0209, 0.0192, 0.0203], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-02 07:47:33,133 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=106835.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 07:47:36,412 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_rvb from training. Duration: 25.285 2023-04-02 07:47:46,708 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 2023-04-02 07:47:52,783 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.6624, 2.2881, 2.3665, 2.8473, 2.5801, 2.3408, 2.1899, 2.7541], device='cuda:1'), covar=tensor([0.0906, 0.1630, 0.1300, 0.0900, 0.1195, 0.0466, 0.1232, 0.0520], device='cuda:1'), in_proj_covar=tensor([0.0261, 0.0351, 0.0298, 0.0242, 0.0296, 0.0245, 0.0290, 0.0246], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 07:48:17,291 INFO [train.py:903] (1/4) Epoch 16, batch 4450, loss[loss=0.2268, simple_loss=0.3021, pruned_loss=0.07579, over 19788.00 frames. ], tot_loss[loss=0.217, simple_loss=0.2946, pruned_loss=0.06971, over 3810179.29 frames. ], batch size: 56, lr: 5.15e-03, grad_scale: 8.0 2023-04-02 07:48:53,519 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.84 vs. limit=5.0 2023-04-02 07:48:53,922 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.305e+02 4.966e+02 6.259e+02 8.420e+02 1.632e+03, threshold=1.252e+03, percent-clipped=6.0 2023-04-02 07:49:18,988 INFO [train.py:903] (1/4) Epoch 16, batch 4500, loss[loss=0.2136, simple_loss=0.3001, pruned_loss=0.06355, over 18725.00 frames. ], tot_loss[loss=0.2159, simple_loss=0.2936, pruned_loss=0.0691, over 3818868.40 frames. ], batch size: 74, lr: 5.15e-03, grad_scale: 8.0 2023-04-02 07:49:34,394 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=106931.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 07:50:23,584 INFO [train.py:903] (1/4) Epoch 16, batch 4550, loss[loss=0.1809, simple_loss=0.2719, pruned_loss=0.04491, over 19675.00 frames. ], tot_loss[loss=0.2153, simple_loss=0.2927, pruned_loss=0.06891, over 3821370.71 frames. ], batch size: 53, lr: 5.15e-03, grad_scale: 8.0 2023-04-02 07:50:31,747 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9_rvb from training. Duration: 28.72225 2023-04-02 07:50:54,360 WARNING [train.py:1073] (1/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 2023-04-02 07:50:59,932 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.264e+02 4.883e+02 5.860e+02 7.136e+02 1.225e+03, threshold=1.172e+03, percent-clipped=0.0 2023-04-02 07:51:02,586 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=107001.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 07:51:27,842 INFO [train.py:903] (1/4) Epoch 16, batch 4600, loss[loss=0.2082, simple_loss=0.2926, pruned_loss=0.06185, over 19582.00 frames. ], tot_loss[loss=0.2164, simple_loss=0.2934, pruned_loss=0.0697, over 3809082.11 frames. ], batch size: 52, lr: 5.15e-03, grad_scale: 8.0 2023-04-02 07:51:34,896 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=107026.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 07:51:42,913 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.3930, 4.0159, 2.6311, 3.6244, 1.0768, 3.8583, 3.8381, 3.9171], device='cuda:1'), covar=tensor([0.0638, 0.1017, 0.1938, 0.0817, 0.3810, 0.0791, 0.0828, 0.0994], device='cuda:1'), in_proj_covar=tensor([0.0471, 0.0386, 0.0466, 0.0329, 0.0393, 0.0398, 0.0396, 0.0428], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 07:52:06,104 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3716, 1.4584, 1.7805, 1.6179, 2.6730, 2.2138, 2.7272, 1.0532], device='cuda:1'), covar=tensor([0.2307, 0.3955, 0.2486, 0.1814, 0.1337, 0.1942, 0.1305, 0.4054], device='cuda:1'), in_proj_covar=tensor([0.0509, 0.0607, 0.0660, 0.0459, 0.0605, 0.0509, 0.0648, 0.0520], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 07:52:17,745 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2567, 1.3564, 1.7874, 1.4430, 2.7235, 3.6931, 3.4296, 3.9130], device='cuda:1'), covar=tensor([0.1664, 0.3559, 0.3158, 0.2239, 0.0642, 0.0191, 0.0200, 0.0232], device='cuda:1'), in_proj_covar=tensor([0.0265, 0.0306, 0.0335, 0.0255, 0.0230, 0.0174, 0.0210, 0.0230], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 07:52:25,467 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3065, 1.3595, 1.6815, 1.5408, 2.1441, 1.8758, 2.1387, 1.0277], device='cuda:1'), covar=tensor([0.2366, 0.4004, 0.2339, 0.1884, 0.1533, 0.2293, 0.1508, 0.4157], device='cuda:1'), in_proj_covar=tensor([0.0506, 0.0604, 0.0658, 0.0457, 0.0602, 0.0507, 0.0646, 0.0518], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 07:52:29,058 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=107069.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 07:52:29,811 INFO [train.py:903] (1/4) Epoch 16, batch 4650, loss[loss=0.283, simple_loss=0.3386, pruned_loss=0.1137, over 13438.00 frames. ], tot_loss[loss=0.2169, simple_loss=0.2937, pruned_loss=0.07006, over 3820580.89 frames. ], batch size: 136, lr: 5.15e-03, grad_scale: 8.0 2023-04-02 07:52:47,216 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_rvb from training. Duration: 0.92 2023-04-02 07:52:59,803 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_rvb from training. Duration: 0.83 2023-04-02 07:53:01,388 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=107094.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 07:53:07,464 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.220e+02 5.459e+02 6.581e+02 8.910e+02 1.601e+03, threshold=1.316e+03, percent-clipped=6.0 2023-04-02 07:53:31,764 INFO [train.py:903] (1/4) Epoch 16, batch 4700, loss[loss=0.2325, simple_loss=0.3078, pruned_loss=0.07855, over 19854.00 frames. ], tot_loss[loss=0.2178, simple_loss=0.2947, pruned_loss=0.07047, over 3816317.58 frames. ], batch size: 52, lr: 5.15e-03, grad_scale: 8.0 2023-04-02 07:53:55,933 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9_rvb from training. Duration: 0.92225 2023-04-02 07:54:12,246 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0 2023-04-02 07:54:20,559 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=107158.0, num_to_drop=1, layers_to_drop={1} 2023-04-02 07:54:36,808 INFO [train.py:903] (1/4) Epoch 16, batch 4750, loss[loss=0.2027, simple_loss=0.2908, pruned_loss=0.05731, over 19646.00 frames. ], tot_loss[loss=0.2192, simple_loss=0.2962, pruned_loss=0.07112, over 3821176.81 frames. ], batch size: 55, lr: 5.15e-03, grad_scale: 8.0 2023-04-02 07:54:37,125 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=107170.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 07:54:58,973 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=107187.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 07:55:11,904 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.171e+02 5.545e+02 6.621e+02 8.650e+02 1.971e+03, threshold=1.324e+03, percent-clipped=6.0 2023-04-02 07:55:29,391 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=107212.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 07:55:40,217 INFO [train.py:903] (1/4) Epoch 16, batch 4800, loss[loss=0.2108, simple_loss=0.2973, pruned_loss=0.0621, over 19542.00 frames. ], tot_loss[loss=0.2185, simple_loss=0.2958, pruned_loss=0.0706, over 3833966.97 frames. ], batch size: 56, lr: 5.14e-03, grad_scale: 8.0 2023-04-02 07:55:50,298 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0 2023-04-02 07:56:41,879 INFO [train.py:903] (1/4) Epoch 16, batch 4850, loss[loss=0.2657, simple_loss=0.3334, pruned_loss=0.09897, over 13040.00 frames. ], tot_loss[loss=0.2196, simple_loss=0.2969, pruned_loss=0.0711, over 3823595.05 frames. ], batch size: 136, lr: 5.14e-03, grad_scale: 4.0 2023-04-02 07:57:07,068 WARNING [train.py:1073] (1/4) Exclude cut with ID 774-127930-0014-48411_sp1.1_rvb from training. Duration: 0.95 2023-04-02 07:57:19,837 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.329e+02 4.930e+02 6.295e+02 8.478e+02 1.665e+03, threshold=1.259e+03, percent-clipped=1.0 2023-04-02 07:57:27,485 WARNING [train.py:1073] (1/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1_rvb from training. Duration: 0.9409375 2023-04-02 07:57:32,774 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 2023-04-02 07:57:32,801 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 2023-04-02 07:57:43,099 INFO [train.py:903] (1/4) Epoch 16, batch 4900, loss[loss=0.2052, simple_loss=0.2957, pruned_loss=0.05732, over 19698.00 frames. ], tot_loss[loss=0.2194, simple_loss=0.2965, pruned_loss=0.07112, over 3834193.82 frames. ], batch size: 59, lr: 5.14e-03, grad_scale: 4.0 2023-04-02 07:57:43,112 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1_rvb from training. Duration: 25.3818125 2023-04-02 07:57:43,918 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=4.11 vs. limit=5.0 2023-04-02 07:58:04,195 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 2023-04-02 07:58:46,426 INFO [train.py:903] (1/4) Epoch 16, batch 4950, loss[loss=0.2458, simple_loss=0.3223, pruned_loss=0.08469, over 19326.00 frames. ], tot_loss[loss=0.2195, simple_loss=0.2965, pruned_loss=0.07122, over 3831373.16 frames. ], batch size: 66, lr: 5.14e-03, grad_scale: 4.0 2023-04-02 07:59:04,257 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 2023-04-02 07:59:22,735 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.794e+02 5.680e+02 6.678e+02 8.404e+02 2.020e+03, threshold=1.336e+03, percent-clipped=4.0 2023-04-02 07:59:27,613 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 2023-04-02 07:59:48,958 INFO [train.py:903] (1/4) Epoch 16, batch 5000, loss[loss=0.2319, simple_loss=0.3084, pruned_loss=0.07774, over 19085.00 frames. ], tot_loss[loss=0.2194, simple_loss=0.2965, pruned_loss=0.07121, over 3835928.00 frames. ], batch size: 69, lr: 5.14e-03, grad_scale: 4.0 2023-04-02 07:59:58,933 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_rvb from training. Duration: 27.14 2023-04-02 08:00:08,997 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 2023-04-02 08:00:50,459 INFO [train.py:903] (1/4) Epoch 16, batch 5050, loss[loss=0.2255, simple_loss=0.3008, pruned_loss=0.0751, over 18097.00 frames. ], tot_loss[loss=0.2187, simple_loss=0.2957, pruned_loss=0.07086, over 3845258.33 frames. ], batch size: 83, lr: 5.14e-03, grad_scale: 4.0 2023-04-02 08:01:27,875 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.607e+02 5.470e+02 6.454e+02 8.047e+02 2.188e+03, threshold=1.291e+03, percent-clipped=2.0 2023-04-02 08:01:27,918 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9_rvb from training. Duration: 26.62775 2023-04-02 08:01:30,416 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=107502.0, num_to_drop=1, layers_to_drop={1} 2023-04-02 08:01:44,956 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=107514.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 08:01:51,918 INFO [train.py:903] (1/4) Epoch 16, batch 5100, loss[loss=0.2685, simple_loss=0.331, pruned_loss=0.103, over 13597.00 frames. ], tot_loss[loss=0.2191, simple_loss=0.296, pruned_loss=0.0711, over 3819276.66 frames. ], batch size: 136, lr: 5.14e-03, grad_scale: 4.0 2023-04-02 08:02:02,170 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=107528.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 08:02:04,295 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 2023-04-02 08:02:08,825 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_rvb from training. Duration: 25.035 2023-04-02 08:02:13,298 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_rvb from training. Duration: 27.92 2023-04-02 08:02:54,731 INFO [train.py:903] (1/4) Epoch 16, batch 5150, loss[loss=0.2003, simple_loss=0.2904, pruned_loss=0.05514, over 19664.00 frames. ], tot_loss[loss=0.2181, simple_loss=0.2953, pruned_loss=0.07046, over 3820138.57 frames. ], batch size: 55, lr: 5.14e-03, grad_scale: 4.0 2023-04-02 08:03:08,979 WARNING [train.py:1073] (1/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 2023-04-02 08:03:31,791 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.461e+02 4.950e+02 6.087e+02 7.766e+02 1.818e+03, threshold=1.217e+03, percent-clipped=6.0 2023-04-02 08:03:43,252 WARNING [train.py:1073] (1/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1_rvb from training. Duration: 0.9681875 2023-04-02 08:03:54,404 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=107617.0, num_to_drop=1, layers_to_drop={0} 2023-04-02 08:03:57,186 INFO [train.py:903] (1/4) Epoch 16, batch 5200, loss[loss=0.2071, simple_loss=0.2771, pruned_loss=0.06852, over 19726.00 frames. ], tot_loss[loss=0.2187, simple_loss=0.2956, pruned_loss=0.07089, over 3802192.45 frames. ], batch size: 46, lr: 5.13e-03, grad_scale: 8.0 2023-04-02 08:04:08,703 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=107629.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 08:04:09,526 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 2023-04-02 08:04:55,315 WARNING [train.py:1073] (1/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 2023-04-02 08:04:59,581 INFO [train.py:903] (1/4) Epoch 16, batch 5250, loss[loss=0.1953, simple_loss=0.269, pruned_loss=0.06076, over 19769.00 frames. ], tot_loss[loss=0.2184, simple_loss=0.2951, pruned_loss=0.07087, over 3799301.54 frames. ], batch size: 47, lr: 5.13e-03, grad_scale: 8.0 2023-04-02 08:05:07,708 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=107677.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 08:05:36,450 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.729e+02 5.464e+02 6.488e+02 8.647e+02 1.622e+03, threshold=1.298e+03, percent-clipped=8.0 2023-04-02 08:05:42,454 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6516, 1.5487, 1.4868, 2.2065, 1.7111, 1.9676, 2.0791, 1.8097], device='cuda:1'), covar=tensor([0.0812, 0.0914, 0.1025, 0.0712, 0.0790, 0.0707, 0.0753, 0.0658], device='cuda:1'), in_proj_covar=tensor([0.0209, 0.0221, 0.0222, 0.0245, 0.0225, 0.0207, 0.0189, 0.0201], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-02 08:06:00,480 INFO [train.py:903] (1/4) Epoch 16, batch 5300, loss[loss=0.1901, simple_loss=0.2632, pruned_loss=0.05847, over 17293.00 frames. ], tot_loss[loss=0.2187, simple_loss=0.2952, pruned_loss=0.0711, over 3803253.43 frames. ], batch size: 38, lr: 5.13e-03, grad_scale: 8.0 2023-04-02 08:06:19,332 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 2023-04-02 08:06:28,034 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.3498, 3.0591, 2.2620, 2.8184, 0.7706, 2.9475, 2.8807, 2.9668], device='cuda:1'), covar=tensor([0.1096, 0.1266, 0.2062, 0.0986, 0.3957, 0.1017, 0.1085, 0.1293], device='cuda:1'), in_proj_covar=tensor([0.0470, 0.0385, 0.0463, 0.0328, 0.0391, 0.0399, 0.0398, 0.0427], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 08:06:28,315 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3488, 1.3522, 1.3840, 1.4622, 1.7912, 1.8728, 1.6629, 0.5985], device='cuda:1'), covar=tensor([0.2285, 0.4208, 0.2614, 0.1869, 0.1550, 0.2107, 0.1457, 0.4271], device='cuda:1'), in_proj_covar=tensor([0.0507, 0.0609, 0.0659, 0.0461, 0.0607, 0.0508, 0.0649, 0.0520], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 08:06:45,139 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=107755.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 08:07:03,282 INFO [train.py:903] (1/4) Epoch 16, batch 5350, loss[loss=0.247, simple_loss=0.3267, pruned_loss=0.08366, over 19670.00 frames. ], tot_loss[loss=0.2192, simple_loss=0.2957, pruned_loss=0.07132, over 3811656.40 frames. ], batch size: 60, lr: 5.13e-03, grad_scale: 8.0 2023-04-02 08:07:37,241 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9_rvb from training. Duration: 26.438875 2023-04-02 08:07:39,720 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.3603, 3.9301, 2.7640, 3.6303, 0.7419, 3.8805, 3.7966, 3.9236], device='cuda:1'), covar=tensor([0.0667, 0.1029, 0.1714, 0.0674, 0.4168, 0.0682, 0.0830, 0.0930], device='cuda:1'), in_proj_covar=tensor([0.0470, 0.0384, 0.0463, 0.0328, 0.0391, 0.0399, 0.0399, 0.0428], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 08:07:40,607 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.374e+02 4.506e+02 5.884e+02 6.805e+02 1.610e+03, threshold=1.177e+03, percent-clipped=2.0 2023-04-02 08:08:06,503 INFO [train.py:903] (1/4) Epoch 16, batch 5400, loss[loss=0.2401, simple_loss=0.3107, pruned_loss=0.08472, over 19782.00 frames. ], tot_loss[loss=0.2195, simple_loss=0.2963, pruned_loss=0.07138, over 3823117.61 frames. ], batch size: 54, lr: 5.13e-03, grad_scale: 8.0 2023-04-02 08:08:56,404 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 2023-04-02 08:09:02,902 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4321, 1.6292, 1.8804, 1.6757, 3.2087, 2.6815, 3.4841, 1.6683], device='cuda:1'), covar=tensor([0.2446, 0.4067, 0.2708, 0.1953, 0.1576, 0.1919, 0.1570, 0.3824], device='cuda:1'), in_proj_covar=tensor([0.0509, 0.0609, 0.0661, 0.0462, 0.0608, 0.0511, 0.0649, 0.0519], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 08:09:08,381 INFO [train.py:903] (1/4) Epoch 16, batch 5450, loss[loss=0.2256, simple_loss=0.3031, pruned_loss=0.07402, over 19681.00 frames. ], tot_loss[loss=0.22, simple_loss=0.2971, pruned_loss=0.07142, over 3817135.01 frames. ], batch size: 53, lr: 5.13e-03, grad_scale: 8.0 2023-04-02 08:09:10,571 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=107872.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 08:09:11,938 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=107873.0, num_to_drop=1, layers_to_drop={0} 2023-04-02 08:09:26,833 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=107885.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 08:09:44,570 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=107898.0, num_to_drop=1, layers_to_drop={0} 2023-04-02 08:09:46,448 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.026e+02 5.228e+02 6.592e+02 8.752e+02 1.860e+03, threshold=1.318e+03, percent-clipped=11.0 2023-04-02 08:09:58,386 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=107910.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 08:10:10,022 INFO [train.py:903] (1/4) Epoch 16, batch 5500, loss[loss=0.2481, simple_loss=0.3239, pruned_loss=0.0861, over 19299.00 frames. ], tot_loss[loss=0.2199, simple_loss=0.2971, pruned_loss=0.07133, over 3826448.90 frames. ], batch size: 66, lr: 5.13e-03, grad_scale: 8.0 2023-04-02 08:10:12,820 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0049, 2.0542, 2.2282, 2.6796, 2.0956, 2.5751, 2.3376, 2.0995], device='cuda:1'), covar=tensor([0.3427, 0.2918, 0.1364, 0.1700, 0.3018, 0.1434, 0.3227, 0.2390], device='cuda:1'), in_proj_covar=tensor([0.0846, 0.0892, 0.0682, 0.0905, 0.0826, 0.0763, 0.0812, 0.0746], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 08:10:28,510 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.9086, 4.3198, 4.6185, 4.5856, 1.6286, 4.3141, 3.7660, 4.3039], device='cuda:1'), covar=tensor([0.1422, 0.0861, 0.0589, 0.0602, 0.5516, 0.0711, 0.0647, 0.1086], device='cuda:1'), in_proj_covar=tensor([0.0736, 0.0678, 0.0879, 0.0761, 0.0784, 0.0630, 0.0524, 0.0817], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-02 08:10:34,926 WARNING [train.py:1073] (1/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9_rvb from training. Duration: 27.47775 2023-04-02 08:10:43,671 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.2714, 1.7932, 1.8907, 2.1180, 1.8863, 1.8733, 1.7051, 2.1160], device='cuda:1'), covar=tensor([0.0899, 0.1623, 0.1338, 0.1008, 0.1355, 0.0529, 0.1349, 0.0658], device='cuda:1'), in_proj_covar=tensor([0.0264, 0.0354, 0.0302, 0.0244, 0.0299, 0.0248, 0.0292, 0.0247], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 08:11:00,088 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 2023-04-02 08:11:13,100 INFO [train.py:903] (1/4) Epoch 16, batch 5550, loss[loss=0.2159, simple_loss=0.3004, pruned_loss=0.06573, over 19527.00 frames. ], tot_loss[loss=0.2196, simple_loss=0.2965, pruned_loss=0.07132, over 3802761.70 frames. ], batch size: 54, lr: 5.13e-03, grad_scale: 8.0 2023-04-02 08:11:18,615 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 2023-04-02 08:11:34,450 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=107987.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 08:11:50,032 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.970e+02 4.933e+02 6.287e+02 7.608e+02 2.106e+03, threshold=1.257e+03, percent-clipped=3.0 2023-04-02 08:12:10,704 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 2023-04-02 08:12:17,773 INFO [train.py:903] (1/4) Epoch 16, batch 5600, loss[loss=0.1633, simple_loss=0.2429, pruned_loss=0.04192, over 19740.00 frames. ], tot_loss[loss=0.22, simple_loss=0.2967, pruned_loss=0.07169, over 3793915.20 frames. ], batch size: 45, lr: 5.13e-03, grad_scale: 8.0 2023-04-02 08:12:19,204 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=108021.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 08:12:44,606 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.7406, 1.5479, 1.6027, 2.3782, 1.6635, 1.9645, 2.1608, 1.8471], device='cuda:1'), covar=tensor([0.0777, 0.0959, 0.0998, 0.0755, 0.0884, 0.0785, 0.0774, 0.0663], device='cuda:1'), in_proj_covar=tensor([0.0209, 0.0221, 0.0223, 0.0245, 0.0227, 0.0207, 0.0188, 0.0201], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-02 08:13:17,754 INFO [train.py:903] (1/4) Epoch 16, batch 5650, loss[loss=0.2529, simple_loss=0.3198, pruned_loss=0.09306, over 19539.00 frames. ], tot_loss[loss=0.2197, simple_loss=0.2963, pruned_loss=0.07154, over 3814480.61 frames. ], batch size: 54, lr: 5.12e-03, grad_scale: 8.0 2023-04-02 08:13:55,641 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=108099.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 08:13:56,646 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.333e+02 5.688e+02 6.420e+02 8.034e+02 1.662e+03, threshold=1.284e+03, percent-clipped=4.0 2023-04-02 08:14:03,618 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9_rvb from training. Duration: 28.0944375 2023-04-02 08:14:04,428 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=4.01 vs. limit=5.0 2023-04-02 08:14:20,999 INFO [train.py:903] (1/4) Epoch 16, batch 5700, loss[loss=0.2554, simple_loss=0.3286, pruned_loss=0.09111, over 18070.00 frames. ], tot_loss[loss=0.2184, simple_loss=0.2951, pruned_loss=0.07079, over 3809486.34 frames. ], batch size: 83, lr: 5.12e-03, grad_scale: 8.0 2023-04-02 08:14:41,758 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=108136.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 08:15:22,678 INFO [train.py:903] (1/4) Epoch 16, batch 5750, loss[loss=0.1876, simple_loss=0.267, pruned_loss=0.05413, over 19412.00 frames. ], tot_loss[loss=0.2182, simple_loss=0.2951, pruned_loss=0.0707, over 3808986.09 frames. ], batch size: 48, lr: 5.12e-03, grad_scale: 8.0 2023-04-02 08:15:22,702 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9_rvb from training. Duration: 33.038875 2023-04-02 08:15:33,048 WARNING [train.py:1073] (1/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 2023-04-02 08:15:36,703 WARNING [train.py:1073] (1/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 2023-04-02 08:16:00,583 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.754e+02 4.860e+02 6.382e+02 7.922e+02 1.732e+03, threshold=1.276e+03, percent-clipped=4.0 2023-04-02 08:16:19,198 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=108214.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 08:16:26,739 INFO [train.py:903] (1/4) Epoch 16, batch 5800, loss[loss=0.2499, simple_loss=0.3206, pruned_loss=0.08958, over 19291.00 frames. ], tot_loss[loss=0.2186, simple_loss=0.2955, pruned_loss=0.07089, over 3813051.80 frames. ], batch size: 66, lr: 5.12e-03, grad_scale: 8.0 2023-04-02 08:16:54,050 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=108243.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 08:17:25,795 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=108268.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 08:17:27,717 INFO [train.py:903] (1/4) Epoch 16, batch 5850, loss[loss=0.2365, simple_loss=0.3144, pruned_loss=0.07933, over 19778.00 frames. ], tot_loss[loss=0.218, simple_loss=0.2949, pruned_loss=0.07058, over 3818882.62 frames. ], batch size: 56, lr: 5.12e-03, grad_scale: 8.0 2023-04-02 08:17:53,820 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 2023-04-02 08:18:05,296 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.738e+02 5.463e+02 6.888e+02 8.802e+02 1.964e+03, threshold=1.378e+03, percent-clipped=5.0 2023-04-02 08:18:12,668 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.5325, 2.9579, 3.0049, 3.0025, 1.2836, 2.8919, 2.5245, 2.8035], device='cuda:1'), covar=tensor([0.1464, 0.0775, 0.0728, 0.0818, 0.4646, 0.0770, 0.0704, 0.1213], device='cuda:1'), in_proj_covar=tensor([0.0730, 0.0673, 0.0877, 0.0758, 0.0777, 0.0630, 0.0523, 0.0812], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-02 08:18:24,146 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.14 vs. limit=2.0 2023-04-02 08:18:28,344 WARNING [train.py:1073] (1/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 2023-04-02 08:18:29,521 INFO [train.py:903] (1/4) Epoch 16, batch 5900, loss[loss=0.2017, simple_loss=0.2753, pruned_loss=0.06409, over 19422.00 frames. ], tot_loss[loss=0.2174, simple_loss=0.2943, pruned_loss=0.07029, over 3809400.63 frames. ], batch size: 48, lr: 5.12e-03, grad_scale: 8.0 2023-04-02 08:18:52,005 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9_rvb from training. Duration: 27.511125 2023-04-02 08:18:59,402 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=108343.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 08:19:00,959 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.52 vs. limit=5.0 2023-04-02 08:19:30,990 INFO [train.py:903] (1/4) Epoch 16, batch 5950, loss[loss=0.2422, simple_loss=0.317, pruned_loss=0.08372, over 13577.00 frames. ], tot_loss[loss=0.2171, simple_loss=0.2944, pruned_loss=0.06988, over 3811594.83 frames. ], batch size: 135, lr: 5.12e-03, grad_scale: 8.0 2023-04-02 08:19:59,911 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=108392.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 08:20:09,619 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.801e+02 5.248e+02 6.439e+02 7.965e+02 2.252e+03, threshold=1.288e+03, percent-clipped=3.0 2023-04-02 08:20:30,453 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=108417.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 08:20:35,482 INFO [train.py:903] (1/4) Epoch 16, batch 6000, loss[loss=0.204, simple_loss=0.2755, pruned_loss=0.06621, over 19759.00 frames. ], tot_loss[loss=0.2175, simple_loss=0.2946, pruned_loss=0.07024, over 3818487.01 frames. ], batch size: 45, lr: 5.12e-03, grad_scale: 8.0 2023-04-02 08:20:35,482 INFO [train.py:928] (1/4) Computing validation loss 2023-04-02 08:20:47,901 INFO [train.py:937] (1/4) Epoch 16, validation: loss=0.1716, simple_loss=0.2723, pruned_loss=0.03545, over 944034.00 frames. 2023-04-02 08:20:47,902 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18821MB 2023-04-02 08:21:18,664 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([6.0433, 5.4330, 2.8389, 4.7360, 0.9379, 5.4484, 5.3591, 5.6381], device='cuda:1'), covar=tensor([0.0390, 0.0770, 0.1915, 0.0622, 0.4034, 0.0559, 0.0762, 0.0872], device='cuda:1'), in_proj_covar=tensor([0.0469, 0.0385, 0.0463, 0.0330, 0.0391, 0.0402, 0.0399, 0.0426], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 08:21:25,597 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=108450.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 08:21:51,718 INFO [train.py:903] (1/4) Epoch 16, batch 6050, loss[loss=0.2094, simple_loss=0.2909, pruned_loss=0.06391, over 19770.00 frames. ], tot_loss[loss=0.2178, simple_loss=0.2951, pruned_loss=0.07023, over 3819232.12 frames. ], batch size: 54, lr: 5.11e-03, grad_scale: 4.0 2023-04-02 08:21:52,082 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=108470.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 08:22:22,230 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=108495.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 08:22:28,594 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.284e+02 5.272e+02 6.402e+02 8.353e+02 1.883e+03, threshold=1.280e+03, percent-clipped=4.0 2023-04-02 08:22:53,805 INFO [train.py:903] (1/4) Epoch 16, batch 6100, loss[loss=0.2419, simple_loss=0.3148, pruned_loss=0.08453, over 19768.00 frames. ], tot_loss[loss=0.2168, simple_loss=0.2943, pruned_loss=0.06964, over 3820823.38 frames. ], batch size: 56, lr: 5.11e-03, grad_scale: 4.0 2023-04-02 08:23:41,951 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([0.9534, 1.2147, 1.5497, 0.6577, 2.2073, 2.4712, 2.2128, 2.6412], device='cuda:1'), covar=tensor([0.1554, 0.3611, 0.3194, 0.2566, 0.0540, 0.0262, 0.0332, 0.0324], device='cuda:1'), in_proj_covar=tensor([0.0264, 0.0305, 0.0334, 0.0254, 0.0230, 0.0173, 0.0206, 0.0231], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 08:23:52,154 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.7538, 1.8219, 2.0632, 2.3390, 1.6401, 2.1885, 2.1563, 1.9258], device='cuda:1'), covar=tensor([0.3948, 0.3618, 0.1797, 0.2092, 0.3620, 0.1886, 0.4384, 0.3173], device='cuda:1'), in_proj_covar=tensor([0.0850, 0.0894, 0.0682, 0.0906, 0.0828, 0.0768, 0.0814, 0.0748], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 08:23:56,079 INFO [train.py:903] (1/4) Epoch 16, batch 6150, loss[loss=0.2315, simple_loss=0.3123, pruned_loss=0.0753, over 19535.00 frames. ], tot_loss[loss=0.2173, simple_loss=0.2945, pruned_loss=0.07002, over 3820060.60 frames. ], batch size: 56, lr: 5.11e-03, grad_scale: 4.0 2023-04-02 08:24:26,221 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9_rvb from training. Duration: 31.02225 2023-04-02 08:24:28,857 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=108595.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 08:24:35,758 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.970e+02 4.845e+02 6.012e+02 7.583e+02 1.796e+03, threshold=1.202e+03, percent-clipped=3.0 2023-04-02 08:24:58,796 INFO [train.py:903] (1/4) Epoch 16, batch 6200, loss[loss=0.2492, simple_loss=0.3236, pruned_loss=0.08737, over 18090.00 frames. ], tot_loss[loss=0.2173, simple_loss=0.2947, pruned_loss=0.06993, over 3829612.68 frames. ], batch size: 83, lr: 5.11e-03, grad_scale: 4.0 2023-04-02 08:26:02,209 INFO [train.py:903] (1/4) Epoch 16, batch 6250, loss[loss=0.2222, simple_loss=0.3023, pruned_loss=0.07109, over 19691.00 frames. ], tot_loss[loss=0.2173, simple_loss=0.2951, pruned_loss=0.0697, over 3834041.82 frames. ], batch size: 60, lr: 5.11e-03, grad_scale: 4.0 2023-04-02 08:26:22,928 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=108687.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 08:26:27,135 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 2023-04-02 08:26:29,318 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.3289, 2.3260, 2.6297, 3.2631, 2.2592, 3.0304, 2.8290, 2.4609], device='cuda:1'), covar=tensor([0.4007, 0.3808, 0.1684, 0.2380, 0.4270, 0.1856, 0.3667, 0.2888], device='cuda:1'), in_proj_covar=tensor([0.0847, 0.0891, 0.0679, 0.0906, 0.0826, 0.0764, 0.0813, 0.0745], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 08:26:34,487 WARNING [train.py:1073] (1/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9_rvb from training. Duration: 25.988875 2023-04-02 08:26:40,079 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.877e+02 4.966e+02 6.026e+02 7.805e+02 1.726e+03, threshold=1.205e+03, percent-clipped=3.0 2023-04-02 08:27:04,698 INFO [train.py:903] (1/4) Epoch 16, batch 6300, loss[loss=0.1929, simple_loss=0.2703, pruned_loss=0.05776, over 19619.00 frames. ], tot_loss[loss=0.2176, simple_loss=0.295, pruned_loss=0.07008, over 3816016.29 frames. ], batch size: 50, lr: 5.11e-03, grad_scale: 4.0 2023-04-02 08:27:07,181 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.9065, 4.3391, 4.5865, 4.5865, 1.6222, 4.3422, 3.6593, 4.2572], device='cuda:1'), covar=tensor([0.1483, 0.0913, 0.0629, 0.0615, 0.5936, 0.0825, 0.0666, 0.1175], device='cuda:1'), in_proj_covar=tensor([0.0741, 0.0684, 0.0886, 0.0768, 0.0789, 0.0638, 0.0532, 0.0820], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-02 08:27:10,265 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.5369, 4.1297, 2.5919, 3.6817, 0.7403, 3.9161, 3.8828, 3.9747], device='cuda:1'), covar=tensor([0.0665, 0.0934, 0.2037, 0.0799, 0.4433, 0.0777, 0.0937, 0.1088], device='cuda:1'), in_proj_covar=tensor([0.0466, 0.0381, 0.0461, 0.0329, 0.0388, 0.0398, 0.0397, 0.0424], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 08:28:06,342 INFO [train.py:903] (1/4) Epoch 16, batch 6350, loss[loss=0.2199, simple_loss=0.2991, pruned_loss=0.07032, over 18839.00 frames. ], tot_loss[loss=0.218, simple_loss=0.2954, pruned_loss=0.07025, over 3801672.79 frames. ], batch size: 74, lr: 5.11e-03, grad_scale: 4.0 2023-04-02 08:28:38,938 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=108794.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 08:28:47,226 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.196e+02 5.125e+02 6.267e+02 8.166e+02 1.468e+03, threshold=1.253e+03, percent-clipped=8.0 2023-04-02 08:28:48,878 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=108802.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 08:28:54,168 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.61 vs. limit=2.0 2023-04-02 08:28:56,965 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=108809.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 08:29:01,782 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([0.9808, 1.3445, 1.0745, 1.0311, 1.1739, 1.0411, 0.9575, 1.2243], device='cuda:1'), covar=tensor([0.0537, 0.0774, 0.1018, 0.0674, 0.0516, 0.1240, 0.0572, 0.0468], device='cuda:1'), in_proj_covar=tensor([0.0295, 0.0306, 0.0330, 0.0254, 0.0242, 0.0329, 0.0289, 0.0265], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 08:29:09,799 INFO [train.py:903] (1/4) Epoch 16, batch 6400, loss[loss=0.2002, simple_loss=0.2689, pruned_loss=0.06579, over 19785.00 frames. ], tot_loss[loss=0.2177, simple_loss=0.2954, pruned_loss=0.07001, over 3808162.71 frames. ], batch size: 48, lr: 5.11e-03, grad_scale: 8.0 2023-04-02 08:29:23,228 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.63 vs. limit=2.0 2023-04-02 08:30:12,311 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.5707, 1.1546, 1.3412, 1.2912, 2.2077, 0.9804, 2.0247, 2.4454], device='cuda:1'), covar=tensor([0.0643, 0.2691, 0.2794, 0.1511, 0.0858, 0.1997, 0.1005, 0.0442], device='cuda:1'), in_proj_covar=tensor([0.0384, 0.0352, 0.0373, 0.0332, 0.0358, 0.0339, 0.0358, 0.0375], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 08:30:14,212 INFO [train.py:903] (1/4) Epoch 16, batch 6450, loss[loss=0.2556, simple_loss=0.3302, pruned_loss=0.0905, over 19715.00 frames. ], tot_loss[loss=0.2191, simple_loss=0.2966, pruned_loss=0.07082, over 3807653.68 frames. ], batch size: 63, lr: 5.11e-03, grad_scale: 8.0 2023-04-02 08:30:52,179 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.276e+02 5.109e+02 6.250e+02 7.655e+02 1.750e+03, threshold=1.250e+03, percent-clipped=4.0 2023-04-02 08:31:00,316 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9_rvb from training. Duration: 28.638875 2023-04-02 08:31:04,028 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=108909.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 08:31:18,890 INFO [train.py:903] (1/4) Epoch 16, batch 6500, loss[loss=0.1814, simple_loss=0.2671, pruned_loss=0.04787, over 19410.00 frames. ], tot_loss[loss=0.2179, simple_loss=0.2957, pruned_loss=0.07006, over 3819547.15 frames. ], batch size: 48, lr: 5.10e-03, grad_scale: 8.0 2023-04-02 08:31:24,486 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1_rvb from training. Duration: 0.836375 2023-04-02 08:31:40,938 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=108939.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 08:32:20,051 INFO [train.py:903] (1/4) Epoch 16, batch 6550, loss[loss=0.2018, simple_loss=0.2896, pruned_loss=0.05696, over 19391.00 frames. ], tot_loss[loss=0.2175, simple_loss=0.296, pruned_loss=0.06955, over 3821299.88 frames. ], batch size: 70, lr: 5.10e-03, grad_scale: 8.0 2023-04-02 08:32:53,564 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9419, 2.0399, 2.2299, 2.7586, 1.9292, 2.5491, 2.4036, 2.0687], device='cuda:1'), covar=tensor([0.3944, 0.3666, 0.1731, 0.2119, 0.3838, 0.1834, 0.3918, 0.3025], device='cuda:1'), in_proj_covar=tensor([0.0849, 0.0892, 0.0682, 0.0905, 0.0829, 0.0767, 0.0815, 0.0746], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 08:32:58,931 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.314e+02 4.624e+02 5.967e+02 6.821e+02 1.232e+03, threshold=1.193e+03, percent-clipped=0.0 2023-04-02 08:33:21,266 INFO [train.py:903] (1/4) Epoch 16, batch 6600, loss[loss=0.2165, simple_loss=0.2957, pruned_loss=0.0687, over 19676.00 frames. ], tot_loss[loss=0.2171, simple_loss=0.2952, pruned_loss=0.06951, over 3830861.79 frames. ], batch size: 53, lr: 5.10e-03, grad_scale: 8.0 2023-04-02 08:33:35,339 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.85 vs. limit=2.0 2023-04-02 08:33:52,875 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.8356, 1.9432, 2.1921, 2.5132, 1.7532, 2.4024, 2.3419, 2.0484], device='cuda:1'), covar=tensor([0.4042, 0.3688, 0.1687, 0.2057, 0.3691, 0.1855, 0.4145, 0.3094], device='cuda:1'), in_proj_covar=tensor([0.0847, 0.0892, 0.0682, 0.0905, 0.0828, 0.0766, 0.0815, 0.0747], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 08:34:03,940 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=109054.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 08:34:05,035 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=109055.0, num_to_drop=1, layers_to_drop={0} 2023-04-02 08:34:08,552 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=109058.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 08:34:22,843 INFO [train.py:903] (1/4) Epoch 16, batch 6650, loss[loss=0.209, simple_loss=0.2924, pruned_loss=0.06277, over 19701.00 frames. ], tot_loss[loss=0.2179, simple_loss=0.2954, pruned_loss=0.07019, over 3799559.73 frames. ], batch size: 63, lr: 5.10e-03, grad_scale: 8.0 2023-04-02 08:34:41,480 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=109083.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 08:35:01,848 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.091e+02 5.420e+02 6.464e+02 8.289e+02 2.034e+03, threshold=1.293e+03, percent-clipped=5.0 2023-04-02 08:35:27,580 INFO [train.py:903] (1/4) Epoch 16, batch 6700, loss[loss=0.2568, simple_loss=0.3312, pruned_loss=0.09123, over 19530.00 frames. ], tot_loss[loss=0.2181, simple_loss=0.2958, pruned_loss=0.07027, over 3801508.99 frames. ], batch size: 54, lr: 5.10e-03, grad_scale: 8.0 2023-04-02 08:35:32,610 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.8600, 1.7226, 1.7145, 2.2927, 1.7540, 2.1751, 2.1857, 1.9741], device='cuda:1'), covar=tensor([0.0720, 0.0857, 0.0936, 0.0698, 0.0818, 0.0673, 0.0801, 0.0593], device='cuda:1'), in_proj_covar=tensor([0.0208, 0.0220, 0.0224, 0.0243, 0.0225, 0.0206, 0.0188, 0.0200], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-02 08:36:06,060 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=109153.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 08:36:19,840 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=109165.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 08:36:25,886 INFO [train.py:903] (1/4) Epoch 16, batch 6750, loss[loss=0.241, simple_loss=0.3153, pruned_loss=0.08335, over 17442.00 frames. ], tot_loss[loss=0.2176, simple_loss=0.2955, pruned_loss=0.06989, over 3800987.66 frames. ], batch size: 101, lr: 5.10e-03, grad_scale: 8.0 2023-04-02 08:36:50,456 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=109190.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 08:37:03,549 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.450e+02 4.897e+02 5.778e+02 7.062e+02 1.289e+03, threshold=1.156e+03, percent-clipped=0.0 2023-04-02 08:37:15,740 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=109212.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 08:37:24,175 INFO [train.py:903] (1/4) Epoch 16, batch 6800, loss[loss=0.2436, simple_loss=0.3141, pruned_loss=0.08651, over 19651.00 frames. ], tot_loss[loss=0.2176, simple_loss=0.2953, pruned_loss=0.06997, over 3807624.17 frames. ], batch size: 58, lr: 5.10e-03, grad_scale: 8.0 2023-04-02 08:37:45,699 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=2.04 vs. limit=2.0 2023-04-02 08:38:09,402 WARNING [train.py:1073] (1/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 2023-04-02 08:38:09,837 WARNING [train.py:1073] (1/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 2023-04-02 08:38:13,472 INFO [train.py:903] (1/4) Epoch 17, batch 0, loss[loss=0.2199, simple_loss=0.3108, pruned_loss=0.06449, over 19660.00 frames. ], tot_loss[loss=0.2199, simple_loss=0.3108, pruned_loss=0.06449, over 19660.00 frames. ], batch size: 55, lr: 4.94e-03, grad_scale: 8.0 2023-04-02 08:38:13,473 INFO [train.py:928] (1/4) Computing validation loss 2023-04-02 08:38:24,297 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3524, 1.4079, 1.6580, 1.5492, 2.2272, 2.0449, 2.3042, 0.8950], device='cuda:1'), covar=tensor([0.2421, 0.4305, 0.2734, 0.1850, 0.1534, 0.2221, 0.1382, 0.4378], device='cuda:1'), in_proj_covar=tensor([0.0509, 0.0612, 0.0664, 0.0460, 0.0606, 0.0512, 0.0650, 0.0520], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 08:38:26,005 INFO [train.py:937] (1/4) Epoch 17, validation: loss=0.1721, simple_loss=0.2728, pruned_loss=0.03571, over 944034.00 frames. 2023-04-02 08:38:26,006 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18821MB 2023-04-02 08:38:39,466 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 2023-04-02 08:38:51,285 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=109268.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 08:39:29,171 INFO [train.py:903] (1/4) Epoch 17, batch 50, loss[loss=0.199, simple_loss=0.2721, pruned_loss=0.06296, over 19062.00 frames. ], tot_loss[loss=0.2205, simple_loss=0.2975, pruned_loss=0.07174, over 866563.93 frames. ], batch size: 42, lr: 4.94e-03, grad_scale: 8.0 2023-04-02 08:39:32,720 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.530e+02 5.319e+02 6.338e+02 7.961e+02 1.981e+03, threshold=1.268e+03, percent-clipped=4.0 2023-04-02 08:39:40,271 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4010, 1.5523, 2.0832, 1.6790, 3.2698, 2.5348, 3.5436, 1.7285], device='cuda:1'), covar=tensor([0.2425, 0.4250, 0.2578, 0.1858, 0.1426, 0.2073, 0.1561, 0.3850], device='cuda:1'), in_proj_covar=tensor([0.0508, 0.0611, 0.0663, 0.0461, 0.0604, 0.0511, 0.0651, 0.0521], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 08:39:43,519 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=109310.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 08:39:45,711 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=109312.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 08:39:59,852 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1_rvb from training. Duration: 27.0318125 2023-04-02 08:40:13,459 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=109335.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 08:40:14,426 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=109336.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 08:40:28,821 INFO [train.py:903] (1/4) Epoch 17, batch 100, loss[loss=0.2218, simple_loss=0.3088, pruned_loss=0.06745, over 19487.00 frames. ], tot_loss[loss=0.2187, simple_loss=0.2956, pruned_loss=0.07094, over 1514802.86 frames. ], batch size: 64, lr: 4.94e-03, grad_scale: 8.0 2023-04-02 08:40:36,784 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 2023-04-02 08:41:16,025 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5844, 1.5608, 1.4937, 2.1704, 1.7965, 1.9440, 2.0441, 1.8139], device='cuda:1'), covar=tensor([0.0825, 0.0917, 0.1091, 0.0771, 0.0803, 0.0744, 0.0857, 0.0668], device='cuda:1'), in_proj_covar=tensor([0.0209, 0.0222, 0.0225, 0.0245, 0.0226, 0.0207, 0.0190, 0.0201], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-02 08:41:29,301 INFO [train.py:903] (1/4) Epoch 17, batch 150, loss[loss=0.2106, simple_loss=0.2811, pruned_loss=0.07008, over 19396.00 frames. ], tot_loss[loss=0.2171, simple_loss=0.2941, pruned_loss=0.07003, over 2031858.37 frames. ], batch size: 47, lr: 4.94e-03, grad_scale: 8.0 2023-04-02 08:41:30,556 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=109399.0, num_to_drop=1, layers_to_drop={0} 2023-04-02 08:41:32,668 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.150e+02 5.062e+02 6.078e+02 8.331e+02 1.364e+03, threshold=1.216e+03, percent-clipped=3.0 2023-04-02 08:41:46,344 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.6413, 1.3903, 1.5808, 1.4366, 3.2170, 0.9986, 2.2217, 3.5906], device='cuda:1'), covar=tensor([0.0454, 0.2609, 0.2702, 0.1822, 0.0651, 0.2585, 0.1388, 0.0262], device='cuda:1'), in_proj_covar=tensor([0.0384, 0.0350, 0.0373, 0.0333, 0.0359, 0.0339, 0.0359, 0.0377], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 08:42:22,303 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 2023-04-02 08:42:29,334 INFO [train.py:903] (1/4) Epoch 17, batch 200, loss[loss=0.1872, simple_loss=0.2534, pruned_loss=0.06045, over 19745.00 frames. ], tot_loss[loss=0.2189, simple_loss=0.2956, pruned_loss=0.07112, over 2427074.41 frames. ], batch size: 45, lr: 4.94e-03, grad_scale: 8.0 2023-04-02 08:43:27,601 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9775, 1.8507, 1.7925, 1.6145, 1.4905, 1.5897, 0.3442, 0.8310], device='cuda:1'), covar=tensor([0.0510, 0.0545, 0.0348, 0.0553, 0.0962, 0.0684, 0.1078, 0.0897], device='cuda:1'), in_proj_covar=tensor([0.0342, 0.0343, 0.0340, 0.0365, 0.0439, 0.0371, 0.0320, 0.0328], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 08:43:32,097 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3649, 1.3474, 1.5693, 1.5145, 2.4061, 2.1144, 2.5518, 0.9045], device='cuda:1'), covar=tensor([0.2227, 0.4032, 0.2488, 0.1884, 0.1500, 0.2009, 0.1409, 0.4222], device='cuda:1'), in_proj_covar=tensor([0.0503, 0.0606, 0.0658, 0.0457, 0.0601, 0.0507, 0.0645, 0.0515], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 08:43:32,829 INFO [train.py:903] (1/4) Epoch 17, batch 250, loss[loss=0.2267, simple_loss=0.3056, pruned_loss=0.0739, over 19624.00 frames. ], tot_loss[loss=0.2171, simple_loss=0.2943, pruned_loss=0.07002, over 2750112.49 frames. ], batch size: 57, lr: 4.94e-03, grad_scale: 8.0 2023-04-02 08:43:36,247 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.403e+02 5.064e+02 6.047e+02 7.271e+02 1.663e+03, threshold=1.209e+03, percent-clipped=2.0 2023-04-02 08:43:52,395 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=109514.0, num_to_drop=1, layers_to_drop={1} 2023-04-02 08:44:04,273 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=109524.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 08:44:35,510 INFO [train.py:903] (1/4) Epoch 17, batch 300, loss[loss=0.2736, simple_loss=0.3363, pruned_loss=0.1054, over 13654.00 frames. ], tot_loss[loss=0.2169, simple_loss=0.2939, pruned_loss=0.06998, over 2959453.54 frames. ], batch size: 136, lr: 4.94e-03, grad_scale: 8.0 2023-04-02 08:44:37,124 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=109549.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 08:44:45,070 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=109556.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 08:44:56,641 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.3245, 2.7370, 2.7612, 3.1852, 2.9355, 2.7261, 2.4472, 3.1534], device='cuda:1'), covar=tensor([0.0612, 0.1302, 0.1069, 0.0719, 0.1027, 0.0416, 0.1135, 0.0465], device='cuda:1'), in_proj_covar=tensor([0.0264, 0.0353, 0.0300, 0.0244, 0.0296, 0.0249, 0.0294, 0.0247], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 08:45:36,099 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=109597.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 08:45:36,923 INFO [train.py:903] (1/4) Epoch 17, batch 350, loss[loss=0.1918, simple_loss=0.2778, pruned_loss=0.05286, over 19862.00 frames. ], tot_loss[loss=0.2151, simple_loss=0.2925, pruned_loss=0.06887, over 3160822.74 frames. ], batch size: 52, lr: 4.93e-03, grad_scale: 8.0 2023-04-02 08:45:38,106 WARNING [train.py:1073] (1/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1_rvb from training. Duration: 0.9681875 2023-04-02 08:45:40,567 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.272e+02 4.783e+02 5.858e+02 7.548e+02 1.929e+03, threshold=1.172e+03, percent-clipped=3.0 2023-04-02 08:45:49,724 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=2.05 vs. limit=2.0 2023-04-02 08:46:38,845 INFO [train.py:903] (1/4) Epoch 17, batch 400, loss[loss=0.1811, simple_loss=0.2633, pruned_loss=0.04944, over 19737.00 frames. ], tot_loss[loss=0.2151, simple_loss=0.2924, pruned_loss=0.06889, over 3309559.72 frames. ], batch size: 51, lr: 4.93e-03, grad_scale: 8.0 2023-04-02 08:46:49,103 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=109656.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 08:47:09,952 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=109671.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 08:47:16,747 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=109677.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 08:47:20,118 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=109680.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 08:47:40,321 INFO [train.py:903] (1/4) Epoch 17, batch 450, loss[loss=0.2196, simple_loss=0.3045, pruned_loss=0.06731, over 19651.00 frames. ], tot_loss[loss=0.2155, simple_loss=0.2932, pruned_loss=0.06894, over 3427267.59 frames. ], batch size: 53, lr: 4.93e-03, grad_scale: 8.0 2023-04-02 08:47:44,686 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.768e+02 5.202e+02 6.539e+02 8.058e+02 1.631e+03, threshold=1.308e+03, percent-clipped=6.0 2023-04-02 08:48:12,289 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 2023-04-02 08:48:12,747 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.8941, 2.0028, 2.2279, 2.6877, 1.9358, 2.4971, 2.3401, 2.0562], device='cuda:1'), covar=tensor([0.4081, 0.3676, 0.1688, 0.2256, 0.4031, 0.2015, 0.4109, 0.3050], device='cuda:1'), in_proj_covar=tensor([0.0851, 0.0896, 0.0681, 0.0905, 0.0832, 0.0768, 0.0815, 0.0748], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 08:48:13,422 WARNING [train.py:1073] (1/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 2023-04-02 08:48:44,997 INFO [train.py:903] (1/4) Epoch 17, batch 500, loss[loss=0.2004, simple_loss=0.287, pruned_loss=0.05687, over 19774.00 frames. ], tot_loss[loss=0.2173, simple_loss=0.2943, pruned_loss=0.07013, over 3525074.49 frames. ], batch size: 56, lr: 4.93e-03, grad_scale: 8.0 2023-04-02 08:48:48,151 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 2023-04-02 08:48:51,245 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2177, 1.2458, 1.7171, 1.1389, 2.5009, 3.3199, 3.0466, 3.5114], device='cuda:1'), covar=tensor([0.1590, 0.3664, 0.3147, 0.2527, 0.0588, 0.0228, 0.0216, 0.0262], device='cuda:1'), in_proj_covar=tensor([0.0266, 0.0309, 0.0337, 0.0257, 0.0231, 0.0175, 0.0208, 0.0233], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 08:49:01,142 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.69 vs. limit=2.0 2023-04-02 08:49:11,991 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=109770.0, num_to_drop=1, layers_to_drop={1} 2023-04-02 08:49:12,970 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=109771.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 08:49:33,449 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.12 vs. limit=2.0 2023-04-02 08:49:44,292 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=109795.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 08:49:44,336 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=109795.0, num_to_drop=1, layers_to_drop={0} 2023-04-02 08:49:47,259 INFO [train.py:903] (1/4) Epoch 17, batch 550, loss[loss=0.2308, simple_loss=0.2965, pruned_loss=0.08249, over 19370.00 frames. ], tot_loss[loss=0.2172, simple_loss=0.2942, pruned_loss=0.07014, over 3593774.42 frames. ], batch size: 47, lr: 4.93e-03, grad_scale: 8.0 2023-04-02 08:49:50,690 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.960e+02 5.345e+02 7.483e+02 9.617e+02 2.288e+03, threshold=1.497e+03, percent-clipped=10.0 2023-04-02 08:50:48,347 INFO [train.py:903] (1/4) Epoch 17, batch 600, loss[loss=0.1754, simple_loss=0.2547, pruned_loss=0.04804, over 19745.00 frames. ], tot_loss[loss=0.2172, simple_loss=0.2941, pruned_loss=0.07015, over 3641712.25 frames. ], batch size: 47, lr: 4.93e-03, grad_scale: 8.0 2023-04-02 08:51:27,268 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 2023-04-02 08:51:49,465 INFO [train.py:903] (1/4) Epoch 17, batch 650, loss[loss=0.2644, simple_loss=0.3224, pruned_loss=0.1032, over 13574.00 frames. ], tot_loss[loss=0.2182, simple_loss=0.295, pruned_loss=0.07074, over 3688105.87 frames. ], batch size: 135, lr: 4.93e-03, grad_scale: 8.0 2023-04-02 08:51:53,021 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.996e+02 5.642e+02 6.698e+02 8.791e+02 1.815e+03, threshold=1.340e+03, percent-clipped=2.0 2023-04-02 08:52:27,860 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=109927.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 08:52:44,024 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=109941.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 08:52:51,841 INFO [train.py:903] (1/4) Epoch 17, batch 700, loss[loss=0.2228, simple_loss=0.2864, pruned_loss=0.07956, over 19755.00 frames. ], tot_loss[loss=0.2181, simple_loss=0.2948, pruned_loss=0.07074, over 3711315.13 frames. ], batch size: 47, lr: 4.93e-03, grad_scale: 8.0 2023-04-02 08:52:58,959 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=109952.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 08:53:13,587 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([5.2778, 5.2080, 6.0784, 6.0226, 2.2641, 5.7206, 4.8592, 5.6763], device='cuda:1'), covar=tensor([0.1392, 0.0631, 0.0467, 0.0483, 0.5710, 0.0580, 0.0532, 0.1062], device='cuda:1'), in_proj_covar=tensor([0.0738, 0.0678, 0.0880, 0.0764, 0.0788, 0.0629, 0.0527, 0.0810], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-02 08:53:57,850 INFO [train.py:903] (1/4) Epoch 17, batch 750, loss[loss=0.1985, simple_loss=0.2895, pruned_loss=0.0538, over 19680.00 frames. ], tot_loss[loss=0.2182, simple_loss=0.2953, pruned_loss=0.0706, over 3742252.53 frames. ], batch size: 59, lr: 4.93e-03, grad_scale: 8.0 2023-04-02 08:54:02,550 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.920e+02 4.697e+02 5.712e+02 7.217e+02 1.165e+03, threshold=1.142e+03, percent-clipped=0.0 2023-04-02 08:54:25,901 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=110021.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 08:54:33,276 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=110027.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 08:54:44,193 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=110035.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 08:54:44,330 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.0947, 2.8151, 2.0567, 2.0502, 1.8891, 2.3542, 0.8762, 1.9412], device='cuda:1'), covar=tensor([0.0621, 0.0578, 0.0744, 0.1104, 0.1068, 0.1045, 0.1228, 0.1007], device='cuda:1'), in_proj_covar=tensor([0.0351, 0.0349, 0.0347, 0.0371, 0.0448, 0.0378, 0.0326, 0.0333], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 08:55:00,581 INFO [train.py:903] (1/4) Epoch 17, batch 800, loss[loss=0.1767, simple_loss=0.2548, pruned_loss=0.0493, over 19392.00 frames. ], tot_loss[loss=0.2179, simple_loss=0.2949, pruned_loss=0.07048, over 3769981.01 frames. ], batch size: 48, lr: 4.92e-03, grad_scale: 8.0 2023-04-02 08:55:04,546 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=110051.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 08:55:05,665 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=110052.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 08:55:10,573 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=110056.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 08:55:14,906 WARNING [train.py:1073] (1/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 2023-04-02 08:55:17,656 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=110062.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 08:55:35,628 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=110076.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 08:55:54,368 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9604, 2.0279, 2.2254, 2.7204, 1.9548, 2.5419, 2.3276, 1.9666], device='cuda:1'), covar=tensor([0.4032, 0.3712, 0.1730, 0.2220, 0.4061, 0.1931, 0.4287, 0.3190], device='cuda:1'), in_proj_covar=tensor([0.0848, 0.0897, 0.0680, 0.0903, 0.0830, 0.0764, 0.0811, 0.0746], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 08:56:03,225 INFO [train.py:903] (1/4) Epoch 17, batch 850, loss[loss=0.1692, simple_loss=0.2407, pruned_loss=0.04887, over 19291.00 frames. ], tot_loss[loss=0.2182, simple_loss=0.2953, pruned_loss=0.07051, over 3785811.30 frames. ], batch size: 44, lr: 4.92e-03, grad_scale: 8.0 2023-04-02 08:56:06,192 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.347e+02 5.213e+02 6.563e+02 8.363e+02 2.159e+03, threshold=1.313e+03, percent-clipped=10.0 2023-04-02 08:56:42,128 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=110128.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 08:56:47,702 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=110133.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 08:56:51,195 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=110136.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 08:56:56,474 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9_rvb from training. Duration: 25.061125 2023-04-02 08:57:04,359 INFO [train.py:903] (1/4) Epoch 17, batch 900, loss[loss=0.1803, simple_loss=0.2543, pruned_loss=0.05321, over 19721.00 frames. ], tot_loss[loss=0.2181, simple_loss=0.2949, pruned_loss=0.07064, over 3793872.48 frames. ], batch size: 46, lr: 4.92e-03, grad_scale: 8.0 2023-04-02 08:57:44,481 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.1353, 2.1221, 1.6816, 2.1737, 2.1473, 1.7136, 1.7010, 2.0977], device='cuda:1'), covar=tensor([0.1095, 0.1620, 0.1738, 0.1118, 0.1372, 0.0950, 0.1689, 0.0913], device='cuda:1'), in_proj_covar=tensor([0.0266, 0.0356, 0.0304, 0.0245, 0.0299, 0.0250, 0.0297, 0.0248], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 08:58:05,928 INFO [train.py:903] (1/4) Epoch 17, batch 950, loss[loss=0.2309, simple_loss=0.3101, pruned_loss=0.07586, over 19345.00 frames. ], tot_loss[loss=0.2205, simple_loss=0.2972, pruned_loss=0.07185, over 3792219.71 frames. ], batch size: 70, lr: 4.92e-03, grad_scale: 8.0 2023-04-02 08:58:08,348 WARNING [train.py:1073] (1/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9_rvb from training. Duration: 26.32775 2023-04-02 08:58:09,604 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.667e+02 4.867e+02 6.255e+02 8.229e+02 2.250e+03, threshold=1.251e+03, percent-clipped=5.0 2023-04-02 08:58:15,915 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5921, 1.4364, 1.3518, 2.0851, 1.5934, 1.9065, 1.9253, 1.6573], device='cuda:1'), covar=tensor([0.0815, 0.0917, 0.1073, 0.0730, 0.0873, 0.0715, 0.0772, 0.0698], device='cuda:1'), in_proj_covar=tensor([0.0207, 0.0219, 0.0222, 0.0244, 0.0225, 0.0206, 0.0186, 0.0200], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-02 08:59:08,997 INFO [train.py:903] (1/4) Epoch 17, batch 1000, loss[loss=0.2064, simple_loss=0.2918, pruned_loss=0.06055, over 19693.00 frames. ], tot_loss[loss=0.2183, simple_loss=0.2953, pruned_loss=0.07059, over 3814593.99 frames. ], batch size: 59, lr: 4.92e-03, grad_scale: 4.0 2023-04-02 08:59:44,523 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.4112, 1.9283, 2.0171, 2.7944, 2.1236, 2.4545, 2.4877, 2.4495], device='cuda:1'), covar=tensor([0.0660, 0.0835, 0.0890, 0.0774, 0.0811, 0.0691, 0.0803, 0.0561], device='cuda:1'), in_proj_covar=tensor([0.0208, 0.0220, 0.0223, 0.0244, 0.0226, 0.0207, 0.0187, 0.0200], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-02 09:00:05,667 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 2023-04-02 09:00:11,647 INFO [train.py:903] (1/4) Epoch 17, batch 1050, loss[loss=0.2138, simple_loss=0.2966, pruned_loss=0.06552, over 19369.00 frames. ], tot_loss[loss=0.2186, simple_loss=0.2958, pruned_loss=0.07073, over 3816321.33 frames. ], batch size: 70, lr: 4.92e-03, grad_scale: 4.0 2023-04-02 09:00:15,438 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=110301.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:00:16,232 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.124e+02 4.918e+02 6.228e+02 8.050e+02 1.872e+03, threshold=1.246e+03, percent-clipped=2.0 2023-04-02 09:00:28,170 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=110312.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:00:40,483 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=110321.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:00:44,971 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 2023-04-02 09:01:00,216 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=110336.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:01:01,475 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=110337.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:01:13,838 INFO [train.py:903] (1/4) Epoch 17, batch 1100, loss[loss=0.2009, simple_loss=0.2903, pruned_loss=0.05572, over 19705.00 frames. ], tot_loss[loss=0.2194, simple_loss=0.2964, pruned_loss=0.07121, over 3810931.86 frames. ], batch size: 59, lr: 4.92e-03, grad_scale: 4.0 2023-04-02 09:01:36,074 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.7496, 1.2187, 1.4982, 1.3458, 3.2857, 1.0530, 2.3692, 3.6914], device='cuda:1'), covar=tensor([0.0460, 0.2878, 0.2891, 0.1966, 0.0726, 0.2542, 0.1257, 0.0258], device='cuda:1'), in_proj_covar=tensor([0.0389, 0.0355, 0.0375, 0.0335, 0.0360, 0.0342, 0.0357, 0.0381], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 09:01:54,277 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=110379.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:02:09,872 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=110392.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:02:16,596 INFO [train.py:903] (1/4) Epoch 17, batch 1150, loss[loss=0.2202, simple_loss=0.3061, pruned_loss=0.06716, over 19651.00 frames. ], tot_loss[loss=0.2165, simple_loss=0.2943, pruned_loss=0.06936, over 3825382.46 frames. ], batch size: 55, lr: 4.92e-03, grad_scale: 4.0 2023-04-02 09:02:21,350 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.021e+02 4.729e+02 5.705e+02 7.310e+02 1.426e+03, threshold=1.141e+03, percent-clipped=3.0 2023-04-02 09:02:28,350 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=110406.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:02:43,159 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=110417.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:03:20,433 INFO [train.py:903] (1/4) Epoch 17, batch 1200, loss[loss=0.223, simple_loss=0.2983, pruned_loss=0.07385, over 19305.00 frames. ], tot_loss[loss=0.2154, simple_loss=0.2934, pruned_loss=0.06871, over 3822480.95 frames. ], batch size: 66, lr: 4.92e-03, grad_scale: 8.0 2023-04-02 09:03:22,087 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.05 vs. limit=5.0 2023-04-02 09:03:49,892 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=110472.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:03:53,250 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 2023-04-02 09:03:55,645 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=110477.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:04:19,573 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=110494.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:04:23,819 INFO [train.py:903] (1/4) Epoch 17, batch 1250, loss[loss=0.2214, simple_loss=0.3047, pruned_loss=0.06902, over 19321.00 frames. ], tot_loss[loss=0.2168, simple_loss=0.2947, pruned_loss=0.06942, over 3811189.08 frames. ], batch size: 66, lr: 4.91e-03, grad_scale: 8.0 2023-04-02 09:04:28,265 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.549e+02 5.334e+02 6.614e+02 7.905e+02 1.343e+03, threshold=1.323e+03, percent-clipped=1.0 2023-04-02 09:04:48,487 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.8276, 4.3731, 2.6547, 3.8444, 1.1225, 4.2368, 4.1836, 4.2746], device='cuda:1'), covar=tensor([0.0578, 0.1037, 0.2053, 0.0856, 0.3879, 0.0637, 0.0795, 0.1065], device='cuda:1'), in_proj_covar=tensor([0.0471, 0.0385, 0.0465, 0.0331, 0.0389, 0.0402, 0.0400, 0.0429], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 09:04:50,965 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=110521.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:05:25,879 INFO [train.py:903] (1/4) Epoch 17, batch 1300, loss[loss=0.2168, simple_loss=0.2974, pruned_loss=0.06809, over 17369.00 frames. ], tot_loss[loss=0.2169, simple_loss=0.2947, pruned_loss=0.06949, over 3816374.21 frames. ], batch size: 101, lr: 4.91e-03, grad_scale: 8.0 2023-04-02 09:06:02,876 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.60 vs. limit=5.0 2023-04-02 09:06:14,229 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=110587.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:06:20,236 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=110592.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:06:26,784 INFO [train.py:903] (1/4) Epoch 17, batch 1350, loss[loss=0.1963, simple_loss=0.272, pruned_loss=0.06035, over 19720.00 frames. ], tot_loss[loss=0.2168, simple_loss=0.2946, pruned_loss=0.06951, over 3814627.65 frames. ], batch size: 51, lr: 4.91e-03, grad_scale: 8.0 2023-04-02 09:06:31,255 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.362e+02 4.458e+02 6.078e+02 8.226e+02 1.667e+03, threshold=1.216e+03, percent-clipped=3.0 2023-04-02 09:06:34,198 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3808, 1.4204, 1.8240, 1.6496, 2.6602, 2.1746, 2.7222, 1.3431], device='cuda:1'), covar=tensor([0.2669, 0.4659, 0.2919, 0.2078, 0.1752, 0.2364, 0.1803, 0.4370], device='cuda:1'), in_proj_covar=tensor([0.0515, 0.0621, 0.0675, 0.0467, 0.0612, 0.0518, 0.0658, 0.0525], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 09:07:07,100 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=110629.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:07:26,707 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=110645.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:07:28,059 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3598, 1.3438, 1.8029, 1.3353, 2.7184, 3.6919, 3.4685, 3.8859], device='cuda:1'), covar=tensor([0.1541, 0.3633, 0.3074, 0.2290, 0.0557, 0.0185, 0.0195, 0.0242], device='cuda:1'), in_proj_covar=tensor([0.0266, 0.0308, 0.0337, 0.0258, 0.0232, 0.0175, 0.0209, 0.0234], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 09:07:29,984 INFO [train.py:903] (1/4) Epoch 17, batch 1400, loss[loss=0.1992, simple_loss=0.2732, pruned_loss=0.06258, over 19841.00 frames. ], tot_loss[loss=0.2159, simple_loss=0.2935, pruned_loss=0.06912, over 3818417.72 frames. ], batch size: 52, lr: 4.91e-03, grad_scale: 4.0 2023-04-02 09:07:51,398 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=110665.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:08:08,774 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=110680.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:08:32,849 INFO [train.py:903] (1/4) Epoch 17, batch 1450, loss[loss=0.2054, simple_loss=0.2692, pruned_loss=0.07082, over 19772.00 frames. ], tot_loss[loss=0.2159, simple_loss=0.2932, pruned_loss=0.06933, over 3828179.74 frames. ], batch size: 46, lr: 4.91e-03, grad_scale: 4.0 2023-04-02 09:08:32,898 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_rvb from training. Duration: 25.85 2023-04-02 09:08:38,609 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.085e+02 5.047e+02 5.721e+02 7.117e+02 1.861e+03, threshold=1.144e+03, percent-clipped=3.0 2023-04-02 09:09:35,379 INFO [train.py:903] (1/4) Epoch 17, batch 1500, loss[loss=0.1962, simple_loss=0.2732, pruned_loss=0.05963, over 19784.00 frames. ], tot_loss[loss=0.2148, simple_loss=0.2923, pruned_loss=0.06866, over 3833821.49 frames. ], batch size: 48, lr: 4.91e-03, grad_scale: 4.0 2023-04-02 09:09:38,234 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=110750.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:09:42,863 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=110754.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:09:50,121 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=110760.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:10:11,746 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=110775.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:10:14,292 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=110777.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:10:16,477 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.2567, 3.7606, 3.8760, 3.8808, 1.5461, 3.6943, 3.2202, 3.6321], device='cuda:1'), covar=tensor([0.1523, 0.0903, 0.0710, 0.0749, 0.5597, 0.0920, 0.0660, 0.1191], device='cuda:1'), in_proj_covar=tensor([0.0742, 0.0686, 0.0885, 0.0771, 0.0792, 0.0638, 0.0534, 0.0818], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-02 09:10:17,711 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=110780.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:10:27,703 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 2023-04-02 09:10:35,429 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=110795.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:10:38,194 INFO [train.py:903] (1/4) Epoch 17, batch 1550, loss[loss=0.2367, simple_loss=0.3151, pruned_loss=0.07912, over 19740.00 frames. ], tot_loss[loss=0.2155, simple_loss=0.2927, pruned_loss=0.0691, over 3825323.25 frames. ], batch size: 63, lr: 4.91e-03, grad_scale: 4.0 2023-04-02 09:10:43,467 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=110802.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:10:44,892 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.880e+02 4.421e+02 5.256e+02 6.667e+02 1.625e+03, threshold=1.051e+03, percent-clipped=2.0 2023-04-02 09:11:34,852 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=110843.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:11:42,180 INFO [train.py:903] (1/4) Epoch 17, batch 1600, loss[loss=0.2287, simple_loss=0.3047, pruned_loss=0.07637, over 18807.00 frames. ], tot_loss[loss=0.2171, simple_loss=0.2945, pruned_loss=0.06979, over 3803465.98 frames. ], batch size: 74, lr: 4.91e-03, grad_scale: 8.0 2023-04-02 09:11:42,652 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=110848.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:12:05,497 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9_rvb from training. Duration: 30.1555625 2023-04-02 09:12:05,884 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5432, 1.3947, 1.3873, 1.7967, 1.4103, 1.6619, 1.6690, 1.5756], device='cuda:1'), covar=tensor([0.0808, 0.0988, 0.1077, 0.0703, 0.0791, 0.0787, 0.0811, 0.0735], device='cuda:1'), in_proj_covar=tensor([0.0208, 0.0220, 0.0222, 0.0244, 0.0226, 0.0208, 0.0188, 0.0201], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-02 09:12:07,077 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=110868.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:12:12,588 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=110873.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:12:29,566 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=110886.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:12:45,459 INFO [train.py:903] (1/4) Epoch 17, batch 1650, loss[loss=0.2325, simple_loss=0.3198, pruned_loss=0.07264, over 19678.00 frames. ], tot_loss[loss=0.2178, simple_loss=0.295, pruned_loss=0.07026, over 3788473.74 frames. ], batch size: 59, lr: 4.91e-03, grad_scale: 8.0 2023-04-02 09:12:51,287 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.530e+02 5.600e+02 6.791e+02 9.379e+02 3.114e+03, threshold=1.358e+03, percent-clipped=15.0 2023-04-02 09:13:30,788 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0461, 1.6074, 1.6306, 2.5360, 2.0342, 2.2592, 2.3984, 2.1206], device='cuda:1'), covar=tensor([0.0805, 0.1028, 0.1078, 0.0911, 0.0853, 0.0783, 0.0881, 0.0680], device='cuda:1'), in_proj_covar=tensor([0.0209, 0.0221, 0.0224, 0.0245, 0.0227, 0.0209, 0.0189, 0.0202], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-02 09:13:47,398 INFO [train.py:903] (1/4) Epoch 17, batch 1700, loss[loss=0.2199, simple_loss=0.2963, pruned_loss=0.07176, over 19675.00 frames. ], tot_loss[loss=0.2178, simple_loss=0.2956, pruned_loss=0.07, over 3805663.25 frames. ], batch size: 53, lr: 4.90e-03, grad_scale: 8.0 2023-04-02 09:14:19,103 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=110973.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:14:24,132 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=110976.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:14:29,659 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 2023-04-02 09:14:49,217 INFO [train.py:903] (1/4) Epoch 17, batch 1750, loss[loss=0.2414, simple_loss=0.3151, pruned_loss=0.08381, over 19377.00 frames. ], tot_loss[loss=0.2187, simple_loss=0.2963, pruned_loss=0.07055, over 3806614.39 frames. ], batch size: 66, lr: 4.90e-03, grad_scale: 8.0 2023-04-02 09:14:55,347 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.686e+02 4.870e+02 5.792e+02 7.151e+02 1.845e+03, threshold=1.158e+03, percent-clipped=1.0 2023-04-02 09:15:03,695 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.7292, 1.5510, 1.5385, 2.0118, 1.5597, 1.9781, 2.0069, 1.7867], device='cuda:1'), covar=tensor([0.0759, 0.0912, 0.0982, 0.0722, 0.0798, 0.0696, 0.0713, 0.0664], device='cuda:1'), in_proj_covar=tensor([0.0209, 0.0221, 0.0223, 0.0245, 0.0227, 0.0208, 0.0188, 0.0202], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-02 09:15:14,951 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=111016.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:15:37,648 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=111036.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:15:39,734 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.9950, 5.0475, 5.8335, 5.7967, 2.1092, 5.4971, 4.6345, 5.4047], device='cuda:1'), covar=tensor([0.1529, 0.0814, 0.0577, 0.0569, 0.5746, 0.0671, 0.0626, 0.1207], device='cuda:1'), in_proj_covar=tensor([0.0738, 0.0679, 0.0883, 0.0766, 0.0788, 0.0638, 0.0530, 0.0809], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-02 09:15:43,367 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=111041.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:15:51,900 INFO [train.py:903] (1/4) Epoch 17, batch 1800, loss[loss=0.1968, simple_loss=0.2778, pruned_loss=0.05791, over 19726.00 frames. ], tot_loss[loss=0.2173, simple_loss=0.2949, pruned_loss=0.06986, over 3817075.01 frames. ], batch size: 46, lr: 4.90e-03, grad_scale: 8.0 2023-04-02 09:15:57,923 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=111051.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:16:10,639 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=111061.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:16:28,053 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=111076.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:16:44,395 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=111088.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:16:51,939 WARNING [train.py:1073] (1/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9_rvb from training. Duration: 25.0944375 2023-04-02 09:16:56,868 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0273, 2.0990, 2.2784, 2.7957, 2.0757, 2.6802, 2.3865, 2.1114], device='cuda:1'), covar=tensor([0.3778, 0.3509, 0.1681, 0.2123, 0.3719, 0.1812, 0.3990, 0.2998], device='cuda:1'), in_proj_covar=tensor([0.0853, 0.0902, 0.0682, 0.0909, 0.0833, 0.0770, 0.0812, 0.0752], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 09:16:57,618 INFO [train.py:903] (1/4) Epoch 17, batch 1850, loss[loss=0.233, simple_loss=0.3079, pruned_loss=0.07903, over 19355.00 frames. ], tot_loss[loss=0.2174, simple_loss=0.2949, pruned_loss=0.06992, over 3786589.02 frames. ], batch size: 70, lr: 4.90e-03, grad_scale: 8.0 2023-04-02 09:16:57,849 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=111098.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:17:03,773 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.487e+02 5.241e+02 6.549e+02 7.790e+02 1.838e+03, threshold=1.310e+03, percent-clipped=3.0 2023-04-02 09:17:29,421 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9_rvb from training. Duration: 27.8166875 2023-04-02 09:18:01,005 INFO [train.py:903] (1/4) Epoch 17, batch 1900, loss[loss=0.1788, simple_loss=0.2586, pruned_loss=0.04953, over 19766.00 frames. ], tot_loss[loss=0.2168, simple_loss=0.2941, pruned_loss=0.06979, over 3773738.94 frames. ], batch size: 47, lr: 4.90e-03, grad_scale: 8.0 2023-04-02 09:18:17,330 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 2023-04-02 09:18:24,352 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9_rvb from training. Duration: 27.02225 2023-04-02 09:18:34,785 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2430, 1.3586, 1.7891, 1.5289, 2.7281, 2.1423, 2.8361, 1.2003], device='cuda:1'), covar=tensor([0.2493, 0.4220, 0.2538, 0.1926, 0.1417, 0.2136, 0.1371, 0.4121], device='cuda:1'), in_proj_covar=tensor([0.0506, 0.0613, 0.0670, 0.0462, 0.0608, 0.0513, 0.0651, 0.0521], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 09:18:49,754 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 2023-04-02 09:19:02,971 INFO [train.py:903] (1/4) Epoch 17, batch 1950, loss[loss=0.2045, simple_loss=0.2889, pruned_loss=0.06004, over 19790.00 frames. ], tot_loss[loss=0.2176, simple_loss=0.2952, pruned_loss=0.07001, over 3783745.88 frames. ], batch size: 56, lr: 4.90e-03, grad_scale: 8.0 2023-04-02 09:19:08,707 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.803e+02 5.118e+02 6.155e+02 7.161e+02 1.490e+03, threshold=1.231e+03, percent-clipped=2.0 2023-04-02 09:19:10,543 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.58 vs. limit=2.0 2023-04-02 09:19:23,889 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=111213.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:19:24,912 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3440, 1.4606, 1.8402, 1.8207, 2.9412, 3.9174, 3.8064, 4.4088], device='cuda:1'), covar=tensor([0.1892, 0.4796, 0.4101, 0.2272, 0.0796, 0.0346, 0.0282, 0.0260], device='cuda:1'), in_proj_covar=tensor([0.0266, 0.0309, 0.0339, 0.0259, 0.0233, 0.0176, 0.0210, 0.0234], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-02 09:19:44,079 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=111230.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:19:59,226 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=111243.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:20:05,434 INFO [train.py:903] (1/4) Epoch 17, batch 2000, loss[loss=0.2403, simple_loss=0.3106, pruned_loss=0.08503, over 13436.00 frames. ], tot_loss[loss=0.2182, simple_loss=0.2958, pruned_loss=0.07025, over 3784233.00 frames. ], batch size: 135, lr: 4.90e-03, grad_scale: 8.0 2023-04-02 09:21:04,238 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_rvb from training. Duration: 26.8349375 2023-04-02 09:21:07,755 INFO [train.py:903] (1/4) Epoch 17, batch 2050, loss[loss=0.2306, simple_loss=0.3062, pruned_loss=0.07749, over 13802.00 frames. ], tot_loss[loss=0.2186, simple_loss=0.2963, pruned_loss=0.07044, over 3791054.25 frames. ], batch size: 136, lr: 4.90e-03, grad_scale: 8.0 2023-04-02 09:21:14,881 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.549e+02 5.413e+02 6.604e+02 7.706e+02 1.674e+03, threshold=1.321e+03, percent-clipped=5.0 2023-04-02 09:21:22,801 WARNING [train.py:1073] (1/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 2023-04-02 09:21:24,048 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 2023-04-02 09:21:34,576 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=111320.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:21:43,893 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 2023-04-02 09:22:05,020 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=111344.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:22:06,113 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=111345.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:22:09,339 INFO [train.py:903] (1/4) Epoch 17, batch 2100, loss[loss=0.1801, simple_loss=0.2625, pruned_loss=0.04882, over 19583.00 frames. ], tot_loss[loss=0.2184, simple_loss=0.2961, pruned_loss=0.07038, over 3797922.66 frames. ], batch size: 52, lr: 4.90e-03, grad_scale: 8.0 2023-04-02 09:22:34,065 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=111369.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:22:38,047 WARNING [train.py:1073] (1/4) Exclude cut with ID 453-131332-0000-131866_sp0.9_rvb from training. Duration: 25.3333125 2023-04-02 09:22:55,859 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.3319, 2.1827, 1.9994, 1.8332, 1.5677, 1.8157, 0.4753, 1.2428], device='cuda:1'), covar=tensor([0.0484, 0.0568, 0.0441, 0.0770, 0.1194, 0.0840, 0.1315, 0.0940], device='cuda:1'), in_proj_covar=tensor([0.0351, 0.0350, 0.0348, 0.0374, 0.0450, 0.0382, 0.0326, 0.0335], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 09:23:01,368 WARNING [train.py:1073] (1/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9_rvb from training. Duration: 26.6166875 2023-04-02 09:23:10,553 INFO [train.py:903] (1/4) Epoch 17, batch 2150, loss[loss=0.2159, simple_loss=0.3077, pruned_loss=0.06202, over 19673.00 frames. ], tot_loss[loss=0.2165, simple_loss=0.2945, pruned_loss=0.06921, over 3821679.79 frames. ], batch size: 59, lr: 4.89e-03, grad_scale: 8.0 2023-04-02 09:23:15,799 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2933, 1.3499, 1.5771, 1.4848, 2.2537, 1.9907, 2.3327, 0.9266], device='cuda:1'), covar=tensor([0.2372, 0.4253, 0.2589, 0.1892, 0.1476, 0.2149, 0.1352, 0.4258], device='cuda:1'), in_proj_covar=tensor([0.0508, 0.0615, 0.0671, 0.0463, 0.0611, 0.0514, 0.0653, 0.0522], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 09:23:16,490 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.230e+02 4.943e+02 5.975e+02 7.359e+02 1.553e+03, threshold=1.195e+03, percent-clipped=3.0 2023-04-02 09:23:56,877 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=111435.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:24:11,375 INFO [train.py:903] (1/4) Epoch 17, batch 2200, loss[loss=0.2265, simple_loss=0.2997, pruned_loss=0.07661, over 17623.00 frames. ], tot_loss[loss=0.2177, simple_loss=0.2956, pruned_loss=0.06997, over 3813876.21 frames. ], batch size: 101, lr: 4.89e-03, grad_scale: 8.0 2023-04-02 09:24:38,950 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=111469.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:24:58,506 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=111486.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:25:09,823 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=111494.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:25:15,240 INFO [train.py:903] (1/4) Epoch 17, batch 2250, loss[loss=0.1957, simple_loss=0.2848, pruned_loss=0.05328, over 18178.00 frames. ], tot_loss[loss=0.2165, simple_loss=0.2944, pruned_loss=0.06934, over 3805276.96 frames. ], batch size: 84, lr: 4.89e-03, grad_scale: 8.0 2023-04-02 09:25:22,030 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.234e+02 5.257e+02 6.827e+02 8.687e+02 2.303e+03, threshold=1.365e+03, percent-clipped=8.0 2023-04-02 09:26:16,852 INFO [train.py:903] (1/4) Epoch 17, batch 2300, loss[loss=0.1735, simple_loss=0.2513, pruned_loss=0.0478, over 19372.00 frames. ], tot_loss[loss=0.2164, simple_loss=0.2946, pruned_loss=0.06911, over 3812078.13 frames. ], batch size: 47, lr: 4.89e-03, grad_scale: 8.0 2023-04-02 09:26:27,082 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_rvb from training. Duration: 26.205 2023-04-02 09:26:38,553 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 2023-04-02 09:27:05,561 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=111587.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:27:18,439 INFO [train.py:903] (1/4) Epoch 17, batch 2350, loss[loss=0.23, simple_loss=0.3126, pruned_loss=0.07369, over 19276.00 frames. ], tot_loss[loss=0.2176, simple_loss=0.2956, pruned_loss=0.06981, over 3805705.53 frames. ], batch size: 66, lr: 4.89e-03, grad_scale: 8.0 2023-04-02 09:27:22,391 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=111601.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:27:24,245 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.974e+02 5.149e+02 5.946e+02 7.803e+02 1.982e+03, threshold=1.189e+03, percent-clipped=4.0 2023-04-02 09:27:54,520 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=111626.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:27:58,768 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_rvb from training. Duration: 25.775 2023-04-02 09:28:06,321 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.35 vs. limit=5.0 2023-04-02 09:28:14,864 WARNING [train.py:1073] (1/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9_rvb from training. Duration: 25.45 2023-04-02 09:28:19,175 INFO [train.py:903] (1/4) Epoch 17, batch 2400, loss[loss=0.2162, simple_loss=0.2857, pruned_loss=0.07332, over 19737.00 frames. ], tot_loss[loss=0.2175, simple_loss=0.2954, pruned_loss=0.06981, over 3815240.23 frames. ], batch size: 51, lr: 4.89e-03, grad_scale: 8.0 2023-04-02 09:28:24,021 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0077, 2.1236, 2.3328, 2.7576, 1.9581, 2.6007, 2.4034, 2.1952], device='cuda:1'), covar=tensor([0.4174, 0.3806, 0.1794, 0.2231, 0.4073, 0.1928, 0.4509, 0.3176], device='cuda:1'), in_proj_covar=tensor([0.0854, 0.0902, 0.0683, 0.0909, 0.0833, 0.0768, 0.0816, 0.0752], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 09:29:15,392 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=111691.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:29:24,915 INFO [train.py:903] (1/4) Epoch 17, batch 2450, loss[loss=0.2241, simple_loss=0.3083, pruned_loss=0.06994, over 19664.00 frames. ], tot_loss[loss=0.2172, simple_loss=0.2955, pruned_loss=0.06946, over 3817966.84 frames. ], batch size: 58, lr: 4.89e-03, grad_scale: 4.0 2023-04-02 09:29:29,901 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=111702.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:29:30,394 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0 2023-04-02 09:29:32,591 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.040e+02 4.976e+02 6.292e+02 8.323e+02 1.636e+03, threshold=1.258e+03, percent-clipped=0.0 2023-04-02 09:29:33,020 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0183, 1.7000, 1.5990, 1.9752, 1.7045, 1.7066, 1.5377, 1.8553], device='cuda:1'), covar=tensor([0.0982, 0.1451, 0.1485, 0.1010, 0.1291, 0.0539, 0.1389, 0.0687], device='cuda:1'), in_proj_covar=tensor([0.0264, 0.0354, 0.0302, 0.0245, 0.0299, 0.0248, 0.0297, 0.0247], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 09:29:47,077 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=111716.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:30:18,841 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9582, 1.9997, 2.1993, 2.6553, 1.9418, 2.5188, 2.2847, 2.0710], device='cuda:1'), covar=tensor([0.3881, 0.3624, 0.1669, 0.2176, 0.3684, 0.1884, 0.4380, 0.2917], device='cuda:1'), in_proj_covar=tensor([0.0857, 0.0907, 0.0687, 0.0914, 0.0836, 0.0773, 0.0818, 0.0753], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 09:30:27,993 INFO [train.py:903] (1/4) Epoch 17, batch 2500, loss[loss=0.2426, simple_loss=0.3094, pruned_loss=0.0879, over 13435.00 frames. ], tot_loss[loss=0.2175, simple_loss=0.2955, pruned_loss=0.06976, over 3819462.35 frames. ], batch size: 136, lr: 4.89e-03, grad_scale: 4.0 2023-04-02 09:30:51,928 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.31 vs. limit=2.0 2023-04-02 09:31:31,116 INFO [train.py:903] (1/4) Epoch 17, batch 2550, loss[loss=0.2497, simple_loss=0.3258, pruned_loss=0.08677, over 19508.00 frames. ], tot_loss[loss=0.2174, simple_loss=0.2955, pruned_loss=0.06966, over 3823663.59 frames. ], batch size: 64, lr: 4.89e-03, grad_scale: 4.0 2023-04-02 09:31:36,577 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9519, 1.6889, 1.5177, 1.9858, 1.6210, 1.6684, 1.5362, 1.8178], device='cuda:1'), covar=tensor([0.0951, 0.1359, 0.1435, 0.0924, 0.1233, 0.0527, 0.1289, 0.0714], device='cuda:1'), in_proj_covar=tensor([0.0266, 0.0355, 0.0304, 0.0246, 0.0301, 0.0250, 0.0299, 0.0249], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 09:31:38,617 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.039e+02 5.268e+02 6.331e+02 7.722e+02 1.710e+03, threshold=1.266e+03, percent-clipped=3.0 2023-04-02 09:31:44,790 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=111809.0, num_to_drop=1, layers_to_drop={1} 2023-04-02 09:32:13,241 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=111830.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:32:28,429 WARNING [train.py:1073] (1/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 2023-04-02 09:32:34,316 INFO [train.py:903] (1/4) Epoch 17, batch 2600, loss[loss=0.2227, simple_loss=0.3019, pruned_loss=0.07176, over 19663.00 frames. ], tot_loss[loss=0.217, simple_loss=0.295, pruned_loss=0.06955, over 3824891.83 frames. ], batch size: 55, lr: 4.88e-03, grad_scale: 4.0 2023-04-02 09:32:50,533 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([6.1198, 5.5866, 2.9276, 4.7578, 0.9735, 5.5716, 5.5585, 5.7594], device='cuda:1'), covar=tensor([0.0390, 0.0856, 0.1863, 0.0700, 0.3946, 0.0486, 0.0683, 0.0814], device='cuda:1'), in_proj_covar=tensor([0.0481, 0.0392, 0.0475, 0.0337, 0.0395, 0.0410, 0.0410, 0.0438], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 09:33:07,602 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([5.1143, 5.1061, 5.9560, 5.9452, 1.9468, 5.6105, 4.8278, 5.5639], device='cuda:1'), covar=tensor([0.1489, 0.0710, 0.0511, 0.0544, 0.5810, 0.0659, 0.0525, 0.1065], device='cuda:1'), in_proj_covar=tensor([0.0748, 0.0684, 0.0891, 0.0775, 0.0796, 0.0641, 0.0534, 0.0820], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-02 09:33:38,517 INFO [train.py:903] (1/4) Epoch 17, batch 2650, loss[loss=0.1949, simple_loss=0.2571, pruned_loss=0.06638, over 19757.00 frames. ], tot_loss[loss=0.2164, simple_loss=0.2941, pruned_loss=0.06942, over 3825672.10 frames. ], batch size: 46, lr: 4.88e-03, grad_scale: 4.0 2023-04-02 09:33:46,334 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.387e+02 5.132e+02 6.430e+02 7.842e+02 1.964e+03, threshold=1.286e+03, percent-clipped=4.0 2023-04-02 09:33:58,643 WARNING [train.py:1073] (1/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9_rvb from training. Duration: 27.25 2023-04-02 09:34:09,585 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.4474, 1.8482, 2.0253, 1.9568, 3.1809, 1.5909, 2.7068, 3.3543], device='cuda:1'), covar=tensor([0.0446, 0.2137, 0.2122, 0.1468, 0.0549, 0.2075, 0.1393, 0.0317], device='cuda:1'), in_proj_covar=tensor([0.0386, 0.0351, 0.0372, 0.0334, 0.0360, 0.0341, 0.0358, 0.0378], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 09:34:38,390 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=111945.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:34:41,295 INFO [train.py:903] (1/4) Epoch 17, batch 2700, loss[loss=0.1875, simple_loss=0.2601, pruned_loss=0.05744, over 19729.00 frames. ], tot_loss[loss=0.2177, simple_loss=0.2952, pruned_loss=0.07008, over 3828967.33 frames. ], batch size: 46, lr: 4.88e-03, grad_scale: 4.0 2023-04-02 09:34:54,264 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=111958.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:34:58,877 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.7802, 1.5950, 1.3961, 1.7693, 1.5008, 1.5406, 1.3927, 1.6581], device='cuda:1'), covar=tensor([0.1109, 0.1301, 0.1544, 0.1030, 0.1312, 0.0576, 0.1392, 0.0773], device='cuda:1'), in_proj_covar=tensor([0.0265, 0.0355, 0.0303, 0.0246, 0.0300, 0.0249, 0.0298, 0.0249], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 09:35:26,054 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=111983.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:35:43,883 INFO [train.py:903] (1/4) Epoch 17, batch 2750, loss[loss=0.1857, simple_loss=0.2624, pruned_loss=0.05453, over 19737.00 frames. ], tot_loss[loss=0.2191, simple_loss=0.2962, pruned_loss=0.07096, over 3815557.03 frames. ], batch size: 51, lr: 4.88e-03, grad_scale: 4.0 2023-04-02 09:35:52,117 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.760e+02 5.468e+02 6.814e+02 8.726e+02 1.544e+03, threshold=1.363e+03, percent-clipped=3.0 2023-04-02 09:36:45,022 INFO [train.py:903] (1/4) Epoch 17, batch 2800, loss[loss=0.1838, simple_loss=0.2571, pruned_loss=0.05532, over 19297.00 frames. ], tot_loss[loss=0.2199, simple_loss=0.2968, pruned_loss=0.07149, over 3812285.38 frames. ], batch size: 44, lr: 4.88e-03, grad_scale: 8.0 2023-04-02 09:37:26,092 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.0753, 1.2387, 1.6673, 1.2019, 2.4622, 3.2826, 3.0355, 3.4936], device='cuda:1'), covar=tensor([0.1734, 0.3720, 0.3134, 0.2398, 0.0614, 0.0227, 0.0228, 0.0278], device='cuda:1'), in_proj_covar=tensor([0.0263, 0.0307, 0.0337, 0.0256, 0.0229, 0.0174, 0.0208, 0.0233], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 09:37:29,672 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=112084.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:37:48,059 INFO [train.py:903] (1/4) Epoch 17, batch 2850, loss[loss=0.2064, simple_loss=0.2822, pruned_loss=0.06528, over 19393.00 frames. ], tot_loss[loss=0.2202, simple_loss=0.2973, pruned_loss=0.07154, over 3810619.04 frames. ], batch size: 48, lr: 4.88e-03, grad_scale: 8.0 2023-04-02 09:37:54,817 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.003e+02 5.207e+02 6.661e+02 8.674e+02 1.797e+03, threshold=1.332e+03, percent-clipped=6.0 2023-04-02 09:37:57,423 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.6878, 1.5011, 1.6386, 1.5986, 3.2860, 1.1852, 2.4749, 3.6003], device='cuda:1'), covar=tensor([0.0448, 0.2598, 0.2756, 0.1865, 0.0635, 0.2523, 0.1354, 0.0278], device='cuda:1'), in_proj_covar=tensor([0.0386, 0.0349, 0.0371, 0.0332, 0.0358, 0.0340, 0.0357, 0.0378], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 09:38:46,051 WARNING [train.py:1073] (1/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 2023-04-02 09:38:49,623 INFO [train.py:903] (1/4) Epoch 17, batch 2900, loss[loss=0.1762, simple_loss=0.2571, pruned_loss=0.04769, over 19603.00 frames. ], tot_loss[loss=0.2187, simple_loss=0.2959, pruned_loss=0.07077, over 3816036.82 frames. ], batch size: 50, lr: 4.88e-03, grad_scale: 8.0 2023-04-02 09:38:56,302 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=112153.0, num_to_drop=1, layers_to_drop={1} 2023-04-02 09:39:51,804 INFO [train.py:903] (1/4) Epoch 17, batch 2950, loss[loss=0.2407, simple_loss=0.3103, pruned_loss=0.08558, over 19415.00 frames. ], tot_loss[loss=0.2192, simple_loss=0.2962, pruned_loss=0.07113, over 3820264.87 frames. ], batch size: 70, lr: 4.88e-03, grad_scale: 8.0 2023-04-02 09:39:55,987 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=112201.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:39:58,770 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.039e+02 4.879e+02 6.137e+02 7.850e+02 1.399e+03, threshold=1.227e+03, percent-clipped=1.0 2023-04-02 09:40:27,813 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=112226.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:40:36,012 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4019, 1.5034, 1.7438, 1.6115, 2.6127, 2.2906, 2.8224, 1.2128], device='cuda:1'), covar=tensor([0.2299, 0.4004, 0.2456, 0.1792, 0.1457, 0.1944, 0.1316, 0.3974], device='cuda:1'), in_proj_covar=tensor([0.0507, 0.0610, 0.0664, 0.0461, 0.0603, 0.0513, 0.0649, 0.0522], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 09:40:54,421 INFO [train.py:903] (1/4) Epoch 17, batch 3000, loss[loss=0.1769, simple_loss=0.2648, pruned_loss=0.04446, over 19749.00 frames. ], tot_loss[loss=0.2182, simple_loss=0.2952, pruned_loss=0.0706, over 3830579.51 frames. ], batch size: 51, lr: 4.88e-03, grad_scale: 8.0 2023-04-02 09:40:54,422 INFO [train.py:928] (1/4) Computing validation loss 2023-04-02 09:41:09,008 INFO [train.py:937] (1/4) Epoch 17, validation: loss=0.1717, simple_loss=0.272, pruned_loss=0.03576, over 944034.00 frames. 2023-04-02 09:41:09,010 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18821MB 2023-04-02 09:41:13,758 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_rvb from training. Duration: 29.735 2023-04-02 09:41:33,188 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=112268.0, num_to_drop=1, layers_to_drop={0} 2023-04-02 09:42:07,701 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([6.2220, 5.5281, 3.2692, 4.7369, 0.8381, 5.7467, 5.5298, 5.8069], device='cuda:1'), covar=tensor([0.0329, 0.0824, 0.1660, 0.0719, 0.4328, 0.0440, 0.0654, 0.0744], device='cuda:1'), in_proj_covar=tensor([0.0475, 0.0388, 0.0472, 0.0333, 0.0392, 0.0406, 0.0405, 0.0434], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 09:42:09,759 INFO [train.py:903] (1/4) Epoch 17, batch 3050, loss[loss=0.252, simple_loss=0.3292, pruned_loss=0.08736, over 19724.00 frames. ], tot_loss[loss=0.2187, simple_loss=0.2956, pruned_loss=0.07088, over 3817397.95 frames. ], batch size: 63, lr: 4.87e-03, grad_scale: 8.0 2023-04-02 09:42:16,485 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.397e+02 5.197e+02 6.217e+02 9.038e+02 1.667e+03, threshold=1.243e+03, percent-clipped=8.0 2023-04-02 09:43:10,132 INFO [train.py:903] (1/4) Epoch 17, batch 3100, loss[loss=0.1818, simple_loss=0.2604, pruned_loss=0.05158, over 16476.00 frames. ], tot_loss[loss=0.2189, simple_loss=0.2956, pruned_loss=0.0711, over 3822888.52 frames. ], batch size: 36, lr: 4.87e-03, grad_scale: 8.0 2023-04-02 09:44:14,541 INFO [train.py:903] (1/4) Epoch 17, batch 3150, loss[loss=0.2633, simple_loss=0.3371, pruned_loss=0.09477, over 18780.00 frames. ], tot_loss[loss=0.2196, simple_loss=0.2962, pruned_loss=0.0715, over 3821384.81 frames. ], batch size: 74, lr: 4.87e-03, grad_scale: 8.0 2023-04-02 09:44:21,822 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.470e+02 5.016e+02 6.190e+02 7.660e+02 1.883e+03, threshold=1.238e+03, percent-clipped=9.0 2023-04-02 09:44:25,517 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=112407.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:44:42,669 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 2023-04-02 09:44:51,283 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=112428.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:45:17,098 INFO [train.py:903] (1/4) Epoch 17, batch 3200, loss[loss=0.2696, simple_loss=0.338, pruned_loss=0.1006, over 19667.00 frames. ], tot_loss[loss=0.2186, simple_loss=0.2954, pruned_loss=0.07085, over 3808411.38 frames. ], batch size: 55, lr: 4.87e-03, grad_scale: 8.0 2023-04-02 09:46:19,039 INFO [train.py:903] (1/4) Epoch 17, batch 3250, loss[loss=0.2063, simple_loss=0.2813, pruned_loss=0.06568, over 19378.00 frames. ], tot_loss[loss=0.2188, simple_loss=0.2959, pruned_loss=0.07082, over 3814198.69 frames. ], batch size: 48, lr: 4.87e-03, grad_scale: 8.0 2023-04-02 09:46:26,171 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.294e+02 4.958e+02 6.274e+02 7.840e+02 2.025e+03, threshold=1.255e+03, percent-clipped=2.0 2023-04-02 09:46:42,256 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4129, 1.3347, 1.3273, 1.7435, 1.3748, 1.6582, 1.6920, 1.5585], device='cuda:1'), covar=tensor([0.0786, 0.0881, 0.1024, 0.0689, 0.0818, 0.0702, 0.0755, 0.0666], device='cuda:1'), in_proj_covar=tensor([0.0211, 0.0222, 0.0223, 0.0243, 0.0227, 0.0209, 0.0188, 0.0204], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-02 09:46:51,208 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=112524.0, num_to_drop=1, layers_to_drop={1} 2023-04-02 09:47:13,554 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=112543.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:47:18,725 INFO [train.py:903] (1/4) Epoch 17, batch 3300, loss[loss=0.2382, simple_loss=0.3149, pruned_loss=0.0807, over 17539.00 frames. ], tot_loss[loss=0.2193, simple_loss=0.296, pruned_loss=0.07134, over 3796720.51 frames. ], batch size: 101, lr: 4.87e-03, grad_scale: 8.0 2023-04-02 09:47:20,384 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=112549.0, num_to_drop=1, layers_to_drop={0} 2023-04-02 09:47:25,359 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 2023-04-02 09:48:00,388 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=112580.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:48:20,961 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=112595.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:48:24,058 INFO [train.py:903] (1/4) Epoch 17, batch 3350, loss[loss=0.2462, simple_loss=0.3185, pruned_loss=0.08695, over 19592.00 frames. ], tot_loss[loss=0.2176, simple_loss=0.2943, pruned_loss=0.07039, over 3802877.43 frames. ], batch size: 61, lr: 4.87e-03, grad_scale: 8.0 2023-04-02 09:48:31,318 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.087e+02 5.436e+02 6.846e+02 8.612e+02 1.565e+03, threshold=1.369e+03, percent-clipped=5.0 2023-04-02 09:49:09,924 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=112635.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:49:24,116 INFO [train.py:903] (1/4) Epoch 17, batch 3400, loss[loss=0.2556, simple_loss=0.3286, pruned_loss=0.0913, over 19758.00 frames. ], tot_loss[loss=0.2178, simple_loss=0.2946, pruned_loss=0.07052, over 3792201.00 frames. ], batch size: 63, lr: 4.87e-03, grad_scale: 8.0 2023-04-02 09:50:25,917 INFO [train.py:903] (1/4) Epoch 17, batch 3450, loss[loss=0.2222, simple_loss=0.2921, pruned_loss=0.07613, over 19830.00 frames. ], tot_loss[loss=0.2168, simple_loss=0.2941, pruned_loss=0.06975, over 3813079.46 frames. ], batch size: 52, lr: 4.87e-03, grad_scale: 8.0 2023-04-02 09:50:28,239 WARNING [train.py:1073] (1/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9_rvb from training. Duration: 25.2444375 2023-04-02 09:50:33,005 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.425e+02 4.932e+02 6.092e+02 9.481e+02 2.200e+03, threshold=1.218e+03, percent-clipped=6.0 2023-04-02 09:50:45,720 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.7733, 1.2949, 1.6246, 1.5132, 3.3284, 1.0368, 2.3911, 3.7797], device='cuda:1'), covar=tensor([0.0486, 0.2840, 0.2748, 0.1880, 0.0724, 0.2669, 0.1291, 0.0222], device='cuda:1'), in_proj_covar=tensor([0.0388, 0.0351, 0.0371, 0.0331, 0.0359, 0.0340, 0.0356, 0.0378], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 09:51:27,355 INFO [train.py:903] (1/4) Epoch 17, batch 3500, loss[loss=0.2208, simple_loss=0.3018, pruned_loss=0.06991, over 18704.00 frames. ], tot_loss[loss=0.2182, simple_loss=0.2953, pruned_loss=0.07058, over 3810503.70 frames. ], batch size: 74, lr: 4.87e-03, grad_scale: 8.0 2023-04-02 09:51:32,019 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=112751.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:52:31,117 INFO [train.py:903] (1/4) Epoch 17, batch 3550, loss[loss=0.2208, simple_loss=0.2972, pruned_loss=0.07226, over 19674.00 frames. ], tot_loss[loss=0.2177, simple_loss=0.2949, pruned_loss=0.0703, over 3793791.10 frames. ], batch size: 58, lr: 4.86e-03, grad_scale: 8.0 2023-04-02 09:52:32,892 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=112799.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:52:38,380 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.345e+02 4.759e+02 5.980e+02 7.566e+02 1.638e+03, threshold=1.196e+03, percent-clipped=2.0 2023-04-02 09:53:03,320 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=112824.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:53:10,140 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([0.8854, 1.3465, 1.0630, 0.9420, 1.1709, 1.0011, 0.8964, 1.2639], device='cuda:1'), covar=tensor([0.0627, 0.0769, 0.1148, 0.0695, 0.0523, 0.1235, 0.0621, 0.0440], device='cuda:1'), in_proj_covar=tensor([0.0299, 0.0311, 0.0333, 0.0258, 0.0245, 0.0333, 0.0296, 0.0268], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 09:53:33,312 INFO [train.py:903] (1/4) Epoch 17, batch 3600, loss[loss=0.2122, simple_loss=0.2992, pruned_loss=0.06259, over 19674.00 frames. ], tot_loss[loss=0.2167, simple_loss=0.2941, pruned_loss=0.06966, over 3817591.09 frames. ], batch size: 59, lr: 4.86e-03, grad_scale: 8.0 2023-04-02 09:53:55,561 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=112866.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:54:35,715 INFO [train.py:903] (1/4) Epoch 17, batch 3650, loss[loss=0.2028, simple_loss=0.2793, pruned_loss=0.06319, over 19604.00 frames. ], tot_loss[loss=0.2168, simple_loss=0.2946, pruned_loss=0.06951, over 3818795.46 frames. ], batch size: 50, lr: 4.86e-03, grad_scale: 8.0 2023-04-02 09:54:43,571 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.951e+02 4.960e+02 5.826e+02 7.647e+02 1.614e+03, threshold=1.165e+03, percent-clipped=2.0 2023-04-02 09:55:09,846 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=112924.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:55:28,601 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=112939.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:55:28,734 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=112939.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:55:38,877 INFO [train.py:903] (1/4) Epoch 17, batch 3700, loss[loss=0.1806, simple_loss=0.2625, pruned_loss=0.04933, over 19487.00 frames. ], tot_loss[loss=0.2173, simple_loss=0.2954, pruned_loss=0.06959, over 3825492.71 frames. ], batch size: 49, lr: 4.86e-03, grad_scale: 8.0 2023-04-02 09:55:55,798 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1076, 1.3442, 1.7366, 1.4559, 2.7579, 3.7048, 3.4955, 3.9388], device='cuda:1'), covar=tensor([0.1725, 0.3550, 0.3231, 0.2283, 0.0614, 0.0233, 0.0183, 0.0216], device='cuda:1'), in_proj_covar=tensor([0.0266, 0.0311, 0.0342, 0.0259, 0.0232, 0.0179, 0.0211, 0.0236], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-02 09:56:17,480 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=112979.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:56:28,228 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 2023-04-02 09:56:42,247 INFO [train.py:903] (1/4) Epoch 17, batch 3750, loss[loss=0.1706, simple_loss=0.2475, pruned_loss=0.04681, over 17314.00 frames. ], tot_loss[loss=0.2167, simple_loss=0.2948, pruned_loss=0.06937, over 3826874.35 frames. ], batch size: 38, lr: 4.86e-03, grad_scale: 8.0 2023-04-02 09:56:49,255 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.286e+02 4.723e+02 6.001e+02 7.947e+02 1.345e+03, threshold=1.200e+03, percent-clipped=4.0 2023-04-02 09:57:32,631 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=113039.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:57:42,418 INFO [train.py:903] (1/4) Epoch 17, batch 3800, loss[loss=0.2342, simple_loss=0.3019, pruned_loss=0.0833, over 19745.00 frames. ], tot_loss[loss=0.2178, simple_loss=0.2955, pruned_loss=0.07002, over 3819459.70 frames. ], batch size: 51, lr: 4.86e-03, grad_scale: 8.0 2023-04-02 09:57:49,549 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=113054.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:58:14,097 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9_rvb from training. Duration: 29.1166875 2023-04-02 09:58:39,370 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=113094.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:58:43,325 INFO [train.py:903] (1/4) Epoch 17, batch 3850, loss[loss=0.1927, simple_loss=0.2716, pruned_loss=0.05694, over 19487.00 frames. ], tot_loss[loss=0.2188, simple_loss=0.296, pruned_loss=0.0708, over 3805852.84 frames. ], batch size: 49, lr: 4.86e-03, grad_scale: 8.0 2023-04-02 09:58:51,557 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.627e+02 5.316e+02 6.326e+02 9.097e+02 1.552e+03, threshold=1.265e+03, percent-clipped=8.0 2023-04-02 09:59:14,526 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=113122.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:59:29,828 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2740, 1.2608, 1.8587, 1.2519, 2.6540, 3.8015, 3.4808, 3.9089], device='cuda:1'), covar=tensor([0.1521, 0.3666, 0.2955, 0.2373, 0.0591, 0.0163, 0.0184, 0.0210], device='cuda:1'), in_proj_covar=tensor([0.0263, 0.0309, 0.0338, 0.0257, 0.0230, 0.0176, 0.0209, 0.0234], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-02 09:59:44,563 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=113147.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 09:59:45,346 INFO [train.py:903] (1/4) Epoch 17, batch 3900, loss[loss=0.1833, simple_loss=0.2634, pruned_loss=0.05155, over 19567.00 frames. ], tot_loss[loss=0.2173, simple_loss=0.2949, pruned_loss=0.06988, over 3808669.32 frames. ], batch size: 52, lr: 4.86e-03, grad_scale: 8.0 2023-04-02 09:59:56,874 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1977, 1.1492, 1.5693, 1.0772, 2.4415, 3.5213, 3.1918, 3.6463], device='cuda:1'), covar=tensor([0.1677, 0.4033, 0.3590, 0.2630, 0.0642, 0.0184, 0.0209, 0.0241], device='cuda:1'), in_proj_covar=tensor([0.0265, 0.0310, 0.0339, 0.0258, 0.0230, 0.0177, 0.0210, 0.0235], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-02 10:00:48,760 INFO [train.py:903] (1/4) Epoch 17, batch 3950, loss[loss=0.2404, simple_loss=0.3143, pruned_loss=0.08332, over 19429.00 frames. ], tot_loss[loss=0.2163, simple_loss=0.2943, pruned_loss=0.06921, over 3802040.45 frames. ], batch size: 70, lr: 4.86e-03, grad_scale: 8.0 2023-04-02 10:00:56,124 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 2023-04-02 10:00:57,247 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.297e+02 4.545e+02 5.288e+02 6.585e+02 1.560e+03, threshold=1.058e+03, percent-clipped=1.0 2023-04-02 10:01:51,456 INFO [train.py:903] (1/4) Epoch 17, batch 4000, loss[loss=0.2137, simple_loss=0.2987, pruned_loss=0.06433, over 19526.00 frames. ], tot_loss[loss=0.2181, simple_loss=0.2956, pruned_loss=0.07023, over 3794558.80 frames. ], batch size: 54, lr: 4.85e-03, grad_scale: 8.0 2023-04-02 10:01:56,567 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=113252.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 10:02:11,594 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 2023-04-02 10:02:35,244 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=113283.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 10:02:39,812 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1_rvb from training. Duration: 0.7545625 2023-04-02 10:02:44,786 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2056, 1.1020, 1.0945, 1.4448, 1.2548, 1.3789, 1.3809, 1.2195], device='cuda:1'), covar=tensor([0.0653, 0.0755, 0.0807, 0.0591, 0.0773, 0.0626, 0.0747, 0.0600], device='cuda:1'), in_proj_covar=tensor([0.0210, 0.0221, 0.0222, 0.0241, 0.0226, 0.0207, 0.0187, 0.0202], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-02 10:02:49,658 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=113295.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 10:02:52,759 INFO [train.py:903] (1/4) Epoch 17, batch 4050, loss[loss=0.193, simple_loss=0.2669, pruned_loss=0.05957, over 15083.00 frames. ], tot_loss[loss=0.2168, simple_loss=0.2943, pruned_loss=0.06966, over 3797576.25 frames. ], batch size: 33, lr: 4.85e-03, grad_scale: 8.0 2023-04-02 10:03:00,896 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.601e+02 4.703e+02 5.716e+02 7.594e+02 1.568e+03, threshold=1.143e+03, percent-clipped=5.0 2023-04-02 10:03:08,237 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=113310.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 10:03:21,618 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=113320.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 10:03:38,671 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=113335.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 10:03:54,677 INFO [train.py:903] (1/4) Epoch 17, batch 4100, loss[loss=0.2155, simple_loss=0.3006, pruned_loss=0.06523, over 19671.00 frames. ], tot_loss[loss=0.2175, simple_loss=0.2949, pruned_loss=0.0701, over 3796029.40 frames. ], batch size: 53, lr: 4.85e-03, grad_scale: 8.0 2023-04-02 10:03:57,604 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=113350.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 10:04:27,858 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.12 vs. limit=2.0 2023-04-02 10:04:28,699 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=113375.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 10:04:31,497 WARNING [train.py:1073] (1/4) Exclude cut with ID 497-129325-0061-9566_sp1.1_rvb from training. Duration: 0.97725 2023-04-02 10:04:56,296 INFO [train.py:903] (1/4) Epoch 17, batch 4150, loss[loss=0.2258, simple_loss=0.3089, pruned_loss=0.07134, over 19035.00 frames. ], tot_loss[loss=0.2167, simple_loss=0.2941, pruned_loss=0.06968, over 3814358.69 frames. ], batch size: 75, lr: 4.85e-03, grad_scale: 8.0 2023-04-02 10:04:56,639 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=113398.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 10:05:03,828 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.156e+02 5.375e+02 6.520e+02 8.152e+02 2.133e+03, threshold=1.304e+03, percent-clipped=6.0 2023-04-02 10:05:39,332 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.1018, 1.2674, 1.6302, 1.3908, 2.7202, 1.1378, 2.2669, 3.0300], device='cuda:1'), covar=tensor([0.0591, 0.2699, 0.2552, 0.1770, 0.0753, 0.2284, 0.1047, 0.0336], device='cuda:1'), in_proj_covar=tensor([0.0389, 0.0353, 0.0374, 0.0334, 0.0361, 0.0342, 0.0359, 0.0379], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 10:05:57,673 INFO [train.py:903] (1/4) Epoch 17, batch 4200, loss[loss=0.2258, simple_loss=0.3107, pruned_loss=0.07047, over 18377.00 frames. ], tot_loss[loss=0.2172, simple_loss=0.2944, pruned_loss=0.07003, over 3802454.79 frames. ], batch size: 84, lr: 4.85e-03, grad_scale: 8.0 2023-04-02 10:06:02,341 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 2023-04-02 10:06:54,131 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([6.2068, 5.6280, 2.8684, 4.8461, 1.2641, 5.7591, 5.6692, 5.7741], device='cuda:1'), covar=tensor([0.0415, 0.0784, 0.1916, 0.0675, 0.3586, 0.0540, 0.0681, 0.0796], device='cuda:1'), in_proj_covar=tensor([0.0473, 0.0387, 0.0468, 0.0330, 0.0391, 0.0406, 0.0403, 0.0428], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 10:06:59,520 INFO [train.py:903] (1/4) Epoch 17, batch 4250, loss[loss=0.2058, simple_loss=0.2721, pruned_loss=0.06978, over 19087.00 frames. ], tot_loss[loss=0.2169, simple_loss=0.294, pruned_loss=0.06987, over 3802634.12 frames. ], batch size: 42, lr: 4.85e-03, grad_scale: 8.0 2023-04-02 10:07:06,465 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.137e+02 4.808e+02 5.898e+02 7.585e+02 1.571e+03, threshold=1.180e+03, percent-clipped=5.0 2023-04-02 10:07:13,472 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9_rvb from training. Duration: 29.816625 2023-04-02 10:07:25,853 WARNING [train.py:1073] (1/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 2023-04-02 10:08:02,056 INFO [train.py:903] (1/4) Epoch 17, batch 4300, loss[loss=0.2216, simple_loss=0.3029, pruned_loss=0.07014, over 18380.00 frames. ], tot_loss[loss=0.2166, simple_loss=0.2939, pruned_loss=0.06965, over 3811476.29 frames. ], batch size: 84, lr: 4.85e-03, grad_scale: 8.0 2023-04-02 10:08:55,439 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 2023-04-02 10:09:02,221 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=113596.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 10:09:04,353 INFO [train.py:903] (1/4) Epoch 17, batch 4350, loss[loss=0.2369, simple_loss=0.3108, pruned_loss=0.08153, over 19623.00 frames. ], tot_loss[loss=0.2157, simple_loss=0.2936, pruned_loss=0.06897, over 3824252.84 frames. ], batch size: 57, lr: 4.85e-03, grad_scale: 8.0 2023-04-02 10:09:12,445 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.419e+02 4.847e+02 6.118e+02 7.738e+02 1.753e+03, threshold=1.224e+03, percent-clipped=4.0 2023-04-02 10:09:31,777 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.04 vs. limit=5.0 2023-04-02 10:09:49,719 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=113635.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 10:10:07,376 INFO [train.py:903] (1/4) Epoch 17, batch 4400, loss[loss=0.1927, simple_loss=0.2736, pruned_loss=0.05586, over 19763.00 frames. ], tot_loss[loss=0.2163, simple_loss=0.2937, pruned_loss=0.06939, over 3815882.29 frames. ], batch size: 54, lr: 4.85e-03, grad_scale: 8.0 2023-04-02 10:10:14,904 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=113654.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 10:10:26,357 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=113664.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 10:10:33,143 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_rvb from training. Duration: 25.285 2023-04-02 10:10:43,125 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 2023-04-02 10:10:45,258 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=113679.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 10:11:07,786 INFO [train.py:903] (1/4) Epoch 17, batch 4450, loss[loss=0.248, simple_loss=0.323, pruned_loss=0.08652, over 19480.00 frames. ], tot_loss[loss=0.2165, simple_loss=0.2941, pruned_loss=0.06948, over 3824407.00 frames. ], batch size: 64, lr: 4.84e-03, grad_scale: 16.0 2023-04-02 10:11:14,459 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.974e+02 5.100e+02 6.811e+02 8.906e+02 1.680e+03, threshold=1.362e+03, percent-clipped=7.0 2023-04-02 10:11:23,001 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=113711.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 10:12:07,715 INFO [train.py:903] (1/4) Epoch 17, batch 4500, loss[loss=0.2425, simple_loss=0.3185, pruned_loss=0.0832, over 19525.00 frames. ], tot_loss[loss=0.2179, simple_loss=0.2949, pruned_loss=0.07045, over 3810318.91 frames. ], batch size: 56, lr: 4.84e-03, grad_scale: 8.0 2023-04-02 10:12:12,819 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.1138, 1.9985, 1.8863, 1.7045, 1.4245, 1.7278, 0.6623, 1.0938], device='cuda:1'), covar=tensor([0.0608, 0.0566, 0.0383, 0.0698, 0.1124, 0.0745, 0.1078, 0.0921], device='cuda:1'), in_proj_covar=tensor([0.0347, 0.0345, 0.0345, 0.0373, 0.0445, 0.0377, 0.0322, 0.0330], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 10:12:15,933 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.7001, 4.1133, 4.3418, 4.3438, 2.0254, 4.0694, 3.6556, 4.0719], device='cuda:1'), covar=tensor([0.1446, 0.1386, 0.0571, 0.0648, 0.5033, 0.0900, 0.0585, 0.0995], device='cuda:1'), in_proj_covar=tensor([0.0745, 0.0691, 0.0896, 0.0778, 0.0799, 0.0648, 0.0539, 0.0822], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-02 10:13:09,293 INFO [train.py:903] (1/4) Epoch 17, batch 4550, loss[loss=0.242, simple_loss=0.3169, pruned_loss=0.08356, over 19312.00 frames. ], tot_loss[loss=0.218, simple_loss=0.2951, pruned_loss=0.07049, over 3810285.99 frames. ], batch size: 66, lr: 4.84e-03, grad_scale: 4.0 2023-04-02 10:13:19,358 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9_rvb from training. Duration: 28.72225 2023-04-02 10:13:20,452 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.619e+02 5.090e+02 6.214e+02 7.749e+02 1.433e+03, threshold=1.243e+03, percent-clipped=2.0 2023-04-02 10:13:42,463 WARNING [train.py:1073] (1/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 2023-04-02 10:14:12,730 INFO [train.py:903] (1/4) Epoch 17, batch 4600, loss[loss=0.2323, simple_loss=0.3102, pruned_loss=0.07719, over 19692.00 frames. ], tot_loss[loss=0.2187, simple_loss=0.2957, pruned_loss=0.07083, over 3800298.42 frames. ], batch size: 59, lr: 4.84e-03, grad_scale: 4.0 2023-04-02 10:15:14,385 INFO [train.py:903] (1/4) Epoch 17, batch 4650, loss[loss=0.2446, simple_loss=0.3261, pruned_loss=0.08158, over 19535.00 frames. ], tot_loss[loss=0.219, simple_loss=0.296, pruned_loss=0.07097, over 3805122.97 frames. ], batch size: 56, lr: 4.84e-03, grad_scale: 4.0 2023-04-02 10:15:23,610 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.367e+02 5.276e+02 6.482e+02 7.907e+02 1.823e+03, threshold=1.296e+03, percent-clipped=2.0 2023-04-02 10:15:31,174 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1324, 1.4414, 1.9926, 1.5772, 3.0691, 4.6419, 4.4975, 5.0032], device='cuda:1'), covar=tensor([0.1700, 0.3596, 0.3034, 0.2150, 0.0575, 0.0159, 0.0153, 0.0154], device='cuda:1'), in_proj_covar=tensor([0.0266, 0.0313, 0.0342, 0.0260, 0.0233, 0.0178, 0.0212, 0.0236], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-02 10:15:32,011 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_rvb from training. Duration: 0.92 2023-04-02 10:15:44,652 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_rvb from training. Duration: 0.83 2023-04-02 10:15:55,314 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4479, 1.5125, 1.9142, 1.5964, 2.9313, 2.5879, 3.3291, 1.4680], device='cuda:1'), covar=tensor([0.2303, 0.4094, 0.2636, 0.1854, 0.1590, 0.1951, 0.1506, 0.4018], device='cuda:1'), in_proj_covar=tensor([0.0516, 0.0616, 0.0670, 0.0463, 0.0607, 0.0515, 0.0650, 0.0525], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 10:16:16,809 INFO [train.py:903] (1/4) Epoch 17, batch 4700, loss[loss=0.2295, simple_loss=0.3073, pruned_loss=0.07586, over 19720.00 frames. ], tot_loss[loss=0.2177, simple_loss=0.2948, pruned_loss=0.07027, over 3808497.25 frames. ], batch size: 63, lr: 4.84e-03, grad_scale: 4.0 2023-04-02 10:16:42,948 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=113967.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 10:16:43,741 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9_rvb from training. Duration: 0.92225 2023-04-02 10:16:56,546 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=113979.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 10:17:12,201 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=113992.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 10:17:19,851 INFO [train.py:903] (1/4) Epoch 17, batch 4750, loss[loss=0.2372, simple_loss=0.316, pruned_loss=0.07915, over 19644.00 frames. ], tot_loss[loss=0.2167, simple_loss=0.2942, pruned_loss=0.0696, over 3818661.92 frames. ], batch size: 60, lr: 4.84e-03, grad_scale: 4.0 2023-04-02 10:17:32,731 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.044e+02 4.836e+02 6.122e+02 7.624e+02 1.576e+03, threshold=1.224e+03, percent-clipped=1.0 2023-04-02 10:17:35,075 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=114008.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 10:17:53,581 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6192, 1.4792, 1.4466, 1.9054, 1.6161, 1.8166, 1.8540, 1.6657], device='cuda:1'), covar=tensor([0.0768, 0.0906, 0.0995, 0.0761, 0.0826, 0.0755, 0.0860, 0.0717], device='cuda:1'), in_proj_covar=tensor([0.0211, 0.0223, 0.0225, 0.0244, 0.0228, 0.0209, 0.0190, 0.0205], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-02 10:18:02,761 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.2931, 2.1106, 1.9250, 1.8760, 1.5626, 1.7210, 0.6685, 1.2765], device='cuda:1'), covar=tensor([0.0521, 0.0546, 0.0436, 0.0737, 0.1150, 0.0862, 0.1130, 0.0951], device='cuda:1'), in_proj_covar=tensor([0.0347, 0.0346, 0.0345, 0.0371, 0.0444, 0.0377, 0.0323, 0.0331], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 10:18:24,446 INFO [train.py:903] (1/4) Epoch 17, batch 4800, loss[loss=0.2274, simple_loss=0.3005, pruned_loss=0.07715, over 18311.00 frames. ], tot_loss[loss=0.2178, simple_loss=0.2949, pruned_loss=0.07033, over 3812974.64 frames. ], batch size: 83, lr: 4.84e-03, grad_scale: 8.0 2023-04-02 10:19:22,128 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=114094.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 10:19:25,654 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.2936, 3.7729, 3.8814, 3.8991, 1.5370, 3.7064, 3.2398, 3.6115], device='cuda:1'), covar=tensor([0.1607, 0.1053, 0.0697, 0.0739, 0.5411, 0.0925, 0.0706, 0.1219], device='cuda:1'), in_proj_covar=tensor([0.0739, 0.0691, 0.0887, 0.0772, 0.0791, 0.0643, 0.0536, 0.0814], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-02 10:19:26,568 INFO [train.py:903] (1/4) Epoch 17, batch 4850, loss[loss=0.2459, simple_loss=0.3188, pruned_loss=0.08655, over 19783.00 frames. ], tot_loss[loss=0.2172, simple_loss=0.2942, pruned_loss=0.07005, over 3815366.79 frames. ], batch size: 56, lr: 4.84e-03, grad_scale: 8.0 2023-04-02 10:19:35,426 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.185e+02 5.130e+02 6.675e+02 8.728e+02 1.864e+03, threshold=1.335e+03, percent-clipped=11.0 2023-04-02 10:19:52,822 WARNING [train.py:1073] (1/4) Exclude cut with ID 774-127930-0014-48411_sp1.1_rvb from training. Duration: 0.95 2023-04-02 10:19:57,405 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=114123.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 10:20:08,928 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.8208, 3.3035, 3.3517, 3.3473, 1.3386, 3.2049, 2.8220, 3.1209], device='cuda:1'), covar=tensor([0.1648, 0.0921, 0.0781, 0.0846, 0.5325, 0.0871, 0.0781, 0.1241], device='cuda:1'), in_proj_covar=tensor([0.0742, 0.0694, 0.0892, 0.0777, 0.0798, 0.0646, 0.0539, 0.0819], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-02 10:20:14,647 WARNING [train.py:1073] (1/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1_rvb from training. Duration: 0.9409375 2023-04-02 10:20:19,129 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 2023-04-02 10:20:20,353 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 2023-04-02 10:20:25,163 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=114145.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 10:20:28,431 INFO [train.py:903] (1/4) Epoch 17, batch 4900, loss[loss=0.2534, simple_loss=0.3304, pruned_loss=0.08815, over 17238.00 frames. ], tot_loss[loss=0.2184, simple_loss=0.2955, pruned_loss=0.07062, over 3803424.46 frames. ], batch size: 101, lr: 4.84e-03, grad_scale: 8.0 2023-04-02 10:20:28,494 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1_rvb from training. Duration: 25.3818125 2023-04-02 10:20:45,337 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0215, 2.0594, 2.2436, 2.6472, 1.8885, 2.4676, 2.3252, 2.0997], device='cuda:1'), covar=tensor([0.3888, 0.3409, 0.1712, 0.2183, 0.3814, 0.1879, 0.4158, 0.3037], device='cuda:1'), in_proj_covar=tensor([0.0856, 0.0905, 0.0687, 0.0913, 0.0836, 0.0773, 0.0817, 0.0755], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 10:20:48,124 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 2023-04-02 10:21:29,546 INFO [train.py:903] (1/4) Epoch 17, batch 4950, loss[loss=0.2312, simple_loss=0.3129, pruned_loss=0.07481, over 13506.00 frames. ], tot_loss[loss=0.2179, simple_loss=0.2955, pruned_loss=0.07018, over 3802622.77 frames. ], batch size: 136, lr: 4.83e-03, grad_scale: 8.0 2023-04-02 10:21:41,951 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.237e+02 5.073e+02 6.090e+02 7.599e+02 1.461e+03, threshold=1.218e+03, percent-clipped=1.0 2023-04-02 10:21:48,500 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 2023-04-02 10:22:09,556 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 2023-04-02 10:22:31,813 INFO [train.py:903] (1/4) Epoch 17, batch 5000, loss[loss=0.2303, simple_loss=0.3032, pruned_loss=0.07877, over 19663.00 frames. ], tot_loss[loss=0.2191, simple_loss=0.2965, pruned_loss=0.0709, over 3801590.79 frames. ], batch size: 55, lr: 4.83e-03, grad_scale: 8.0 2023-04-02 10:22:39,601 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_rvb from training. Duration: 27.14 2023-04-02 10:22:50,090 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 2023-04-02 10:23:32,992 INFO [train.py:903] (1/4) Epoch 17, batch 5050, loss[loss=0.1773, simple_loss=0.2541, pruned_loss=0.05025, over 19777.00 frames. ], tot_loss[loss=0.2176, simple_loss=0.295, pruned_loss=0.07011, over 3811149.52 frames. ], batch size: 47, lr: 4.83e-03, grad_scale: 8.0 2023-04-02 10:23:42,367 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.900e+02 5.068e+02 6.244e+02 7.899e+02 1.430e+03, threshold=1.249e+03, percent-clipped=5.0 2023-04-02 10:24:10,958 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9_rvb from training. Duration: 26.62775 2023-04-02 10:24:34,969 INFO [train.py:903] (1/4) Epoch 17, batch 5100, loss[loss=0.1933, simple_loss=0.2795, pruned_loss=0.05358, over 19544.00 frames. ], tot_loss[loss=0.2173, simple_loss=0.2949, pruned_loss=0.0699, over 3806084.20 frames. ], batch size: 56, lr: 4.83e-03, grad_scale: 8.0 2023-04-02 10:24:37,695 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=114350.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 10:24:44,352 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 2023-04-02 10:24:46,809 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_rvb from training. Duration: 25.035 2023-04-02 10:24:52,452 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_rvb from training. Duration: 27.92 2023-04-02 10:25:10,346 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=114375.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 10:25:14,824 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=114379.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 10:25:19,674 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5341, 2.1751, 1.6526, 1.4831, 2.1101, 1.3660, 1.4593, 1.8548], device='cuda:1'), covar=tensor([0.0997, 0.0757, 0.0899, 0.0815, 0.0466, 0.1140, 0.0683, 0.0477], device='cuda:1'), in_proj_covar=tensor([0.0298, 0.0309, 0.0329, 0.0256, 0.0245, 0.0328, 0.0292, 0.0268], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 10:25:36,284 INFO [train.py:903] (1/4) Epoch 17, batch 5150, loss[loss=0.2285, simple_loss=0.3039, pruned_loss=0.07655, over 19569.00 frames. ], tot_loss[loss=0.2186, simple_loss=0.2961, pruned_loss=0.07052, over 3782537.02 frames. ], batch size: 52, lr: 4.83e-03, grad_scale: 8.0 2023-04-02 10:25:46,456 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=114404.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 10:25:49,640 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.419e+02 5.288e+02 6.652e+02 7.839e+02 1.735e+03, threshold=1.330e+03, percent-clipped=3.0 2023-04-02 10:25:50,836 WARNING [train.py:1073] (1/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 2023-04-02 10:26:24,427 WARNING [train.py:1073] (1/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1_rvb from training. Duration: 0.9681875 2023-04-02 10:26:41,819 INFO [train.py:903] (1/4) Epoch 17, batch 5200, loss[loss=0.2462, simple_loss=0.3271, pruned_loss=0.08267, over 19549.00 frames. ], tot_loss[loss=0.2185, simple_loss=0.2958, pruned_loss=0.07063, over 3794853.51 frames. ], batch size: 56, lr: 4.83e-03, grad_scale: 8.0 2023-04-02 10:26:47,159 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.13 vs. limit=2.0 2023-04-02 10:26:54,845 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 2023-04-02 10:27:33,116 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=114489.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 10:27:35,863 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2398, 1.2975, 1.3001, 1.0611, 1.1134, 1.1489, 0.0642, 0.3979], device='cuda:1'), covar=tensor([0.0643, 0.0584, 0.0381, 0.0503, 0.1224, 0.0544, 0.1096, 0.0971], device='cuda:1'), in_proj_covar=tensor([0.0349, 0.0346, 0.0345, 0.0374, 0.0448, 0.0379, 0.0325, 0.0333], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 10:27:37,842 WARNING [train.py:1073] (1/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 2023-04-02 10:27:43,618 INFO [train.py:903] (1/4) Epoch 17, batch 5250, loss[loss=0.1979, simple_loss=0.2813, pruned_loss=0.05723, over 19531.00 frames. ], tot_loss[loss=0.2187, simple_loss=0.2959, pruned_loss=0.07073, over 3783956.13 frames. ], batch size: 54, lr: 4.83e-03, grad_scale: 8.0 2023-04-02 10:27:53,070 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.560e+02 4.802e+02 5.852e+02 7.465e+02 1.395e+03, threshold=1.170e+03, percent-clipped=1.0 2023-04-02 10:28:03,714 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.1212, 1.9562, 1.6828, 2.0794, 1.9128, 1.8036, 1.6032, 1.9948], device='cuda:1'), covar=tensor([0.0955, 0.1325, 0.1397, 0.1012, 0.1284, 0.0518, 0.1344, 0.0638], device='cuda:1'), in_proj_covar=tensor([0.0262, 0.0352, 0.0299, 0.0243, 0.0297, 0.0246, 0.0293, 0.0245], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 10:28:26,787 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.80 vs. limit=2.0 2023-04-02 10:28:44,596 INFO [train.py:903] (1/4) Epoch 17, batch 5300, loss[loss=0.2167, simple_loss=0.3016, pruned_loss=0.06587, over 19740.00 frames. ], tot_loss[loss=0.2185, simple_loss=0.2956, pruned_loss=0.07073, over 3781400.01 frames. ], batch size: 63, lr: 4.83e-03, grad_scale: 8.0 2023-04-02 10:28:59,050 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 2023-04-02 10:29:44,123 INFO [train.py:903] (1/4) Epoch 17, batch 5350, loss[loss=0.2129, simple_loss=0.2976, pruned_loss=0.06411, over 19271.00 frames. ], tot_loss[loss=0.2164, simple_loss=0.2937, pruned_loss=0.06957, over 3788057.59 frames. ], batch size: 66, lr: 4.83e-03, grad_scale: 8.0 2023-04-02 10:29:51,244 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=114604.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 10:29:54,895 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.137e+02 5.171e+02 6.688e+02 9.089e+02 2.274e+03, threshold=1.338e+03, percent-clipped=9.0 2023-04-02 10:30:19,031 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9_rvb from training. Duration: 26.438875 2023-04-02 10:30:46,234 INFO [train.py:903] (1/4) Epoch 17, batch 5400, loss[loss=0.1696, simple_loss=0.2502, pruned_loss=0.0445, over 19729.00 frames. ], tot_loss[loss=0.2156, simple_loss=0.2932, pruned_loss=0.069, over 3790950.87 frames. ], batch size: 51, lr: 4.82e-03, grad_scale: 8.0 2023-04-02 10:31:01,004 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.9177, 4.4772, 2.8918, 3.8325, 0.8072, 4.4329, 4.3047, 4.4083], device='cuda:1'), covar=tensor([0.0522, 0.0878, 0.1769, 0.0821, 0.4169, 0.0586, 0.0798, 0.1040], device='cuda:1'), in_proj_covar=tensor([0.0474, 0.0390, 0.0474, 0.0336, 0.0394, 0.0410, 0.0408, 0.0435], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 10:31:47,161 INFO [train.py:903] (1/4) Epoch 17, batch 5450, loss[loss=0.2163, simple_loss=0.3005, pruned_loss=0.06606, over 19674.00 frames. ], tot_loss[loss=0.216, simple_loss=0.294, pruned_loss=0.06903, over 3788638.79 frames. ], batch size: 55, lr: 4.82e-03, grad_scale: 8.0 2023-04-02 10:31:56,196 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.018e+02 4.503e+02 5.761e+02 7.243e+02 1.420e+03, threshold=1.152e+03, percent-clipped=1.0 2023-04-02 10:32:31,011 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=114734.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 10:32:47,136 INFO [train.py:903] (1/4) Epoch 17, batch 5500, loss[loss=0.1966, simple_loss=0.2792, pruned_loss=0.05701, over 19582.00 frames. ], tot_loss[loss=0.2146, simple_loss=0.2928, pruned_loss=0.06817, over 3802307.48 frames. ], batch size: 52, lr: 4.82e-03, grad_scale: 8.0 2023-04-02 10:33:10,830 WARNING [train.py:1073] (1/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9_rvb from training. Duration: 27.47775 2023-04-02 10:33:45,881 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.7707, 4.2056, 4.4406, 4.4601, 1.7068, 4.1774, 3.5775, 4.1392], device='cuda:1'), covar=tensor([0.1672, 0.0967, 0.0658, 0.0684, 0.6033, 0.0862, 0.0756, 0.1188], device='cuda:1'), in_proj_covar=tensor([0.0748, 0.0696, 0.0898, 0.0783, 0.0802, 0.0648, 0.0543, 0.0827], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-02 10:33:46,749 INFO [train.py:903] (1/4) Epoch 17, batch 5550, loss[loss=0.2191, simple_loss=0.299, pruned_loss=0.06955, over 19582.00 frames. ], tot_loss[loss=0.2154, simple_loss=0.293, pruned_loss=0.06887, over 3813966.50 frames. ], batch size: 61, lr: 4.82e-03, grad_scale: 8.0 2023-04-02 10:33:54,781 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 2023-04-02 10:33:55,925 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.429e+02 4.950e+02 6.230e+02 7.289e+02 1.704e+03, threshold=1.246e+03, percent-clipped=5.0 2023-04-02 10:34:41,616 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 2023-04-02 10:34:48,925 INFO [train.py:903] (1/4) Epoch 17, batch 5600, loss[loss=0.2207, simple_loss=0.3056, pruned_loss=0.06791, over 17384.00 frames. ], tot_loss[loss=0.2148, simple_loss=0.2926, pruned_loss=0.06854, over 3813837.80 frames. ], batch size: 101, lr: 4.82e-03, grad_scale: 8.0 2023-04-02 10:34:54,822 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6578, 1.4376, 1.4133, 2.2234, 1.7063, 1.9265, 2.0206, 1.6754], device='cuda:1'), covar=tensor([0.0846, 0.1012, 0.1109, 0.0850, 0.0876, 0.0818, 0.0893, 0.0798], device='cuda:1'), in_proj_covar=tensor([0.0209, 0.0222, 0.0224, 0.0243, 0.0227, 0.0210, 0.0188, 0.0204], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-02 10:35:03,648 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=114860.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 10:35:33,292 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=114885.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 10:35:39,369 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=114889.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 10:35:50,182 INFO [train.py:903] (1/4) Epoch 17, batch 5650, loss[loss=0.2251, simple_loss=0.2991, pruned_loss=0.07556, over 19327.00 frames. ], tot_loss[loss=0.2152, simple_loss=0.2928, pruned_loss=0.06879, over 3823574.42 frames. ], batch size: 66, lr: 4.82e-03, grad_scale: 8.0 2023-04-02 10:35:59,363 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.231e+02 5.160e+02 6.498e+02 8.575e+02 1.504e+03, threshold=1.300e+03, percent-clipped=5.0 2023-04-02 10:36:36,126 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9_rvb from training. Duration: 28.0944375 2023-04-02 10:36:51,138 INFO [train.py:903] (1/4) Epoch 17, batch 5700, loss[loss=0.2399, simple_loss=0.3133, pruned_loss=0.08319, over 19441.00 frames. ], tot_loss[loss=0.2155, simple_loss=0.2933, pruned_loss=0.06888, over 3830994.11 frames. ], batch size: 70, lr: 4.82e-03, grad_scale: 8.0 2023-04-02 10:37:50,261 INFO [train.py:903] (1/4) Epoch 17, batch 5750, loss[loss=0.1767, simple_loss=0.2485, pruned_loss=0.05251, over 19732.00 frames. ], tot_loss[loss=0.2149, simple_loss=0.2931, pruned_loss=0.06839, over 3839686.45 frames. ], batch size: 45, lr: 4.82e-03, grad_scale: 8.0 2023-04-02 10:37:50,274 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9_rvb from training. Duration: 33.038875 2023-04-02 10:37:57,221 WARNING [train.py:1073] (1/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 2023-04-02 10:37:59,478 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.823e+02 5.221e+02 6.429e+02 7.572e+02 1.818e+03, threshold=1.286e+03, percent-clipped=4.0 2023-04-02 10:38:04,571 WARNING [train.py:1073] (1/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 2023-04-02 10:38:50,579 INFO [train.py:903] (1/4) Epoch 17, batch 5800, loss[loss=0.1889, simple_loss=0.2677, pruned_loss=0.05507, over 19498.00 frames. ], tot_loss[loss=0.2146, simple_loss=0.2929, pruned_loss=0.06813, over 3841647.08 frames. ], batch size: 49, lr: 4.82e-03, grad_scale: 8.0 2023-04-02 10:38:52,004 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.2152, 3.7319, 3.8479, 3.8873, 1.6047, 3.6476, 3.1777, 3.5783], device='cuda:1'), covar=tensor([0.1707, 0.1062, 0.0709, 0.0762, 0.5358, 0.1062, 0.0773, 0.1179], device='cuda:1'), in_proj_covar=tensor([0.0741, 0.0691, 0.0891, 0.0775, 0.0794, 0.0645, 0.0536, 0.0819], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-02 10:39:27,226 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=115078.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 10:39:28,586 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5800, 1.2035, 1.2031, 1.4646, 1.1496, 1.3698, 1.1941, 1.4313], device='cuda:1'), covar=tensor([0.1051, 0.1163, 0.1561, 0.0980, 0.1269, 0.0609, 0.1430, 0.0782], device='cuda:1'), in_proj_covar=tensor([0.0264, 0.0352, 0.0302, 0.0245, 0.0298, 0.0247, 0.0294, 0.0249], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 10:39:35,785 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.09 vs. limit=2.0 2023-04-02 10:39:52,203 INFO [train.py:903] (1/4) Epoch 17, batch 5850, loss[loss=0.2343, simple_loss=0.3192, pruned_loss=0.07471, over 19589.00 frames. ], tot_loss[loss=0.2134, simple_loss=0.2919, pruned_loss=0.06746, over 3847981.98 frames. ], batch size: 61, lr: 4.82e-03, grad_scale: 8.0 2023-04-02 10:39:59,380 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=115104.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 10:40:01,418 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.460e+02 4.663e+02 6.050e+02 7.882e+02 1.454e+03, threshold=1.210e+03, percent-clipped=2.0 2023-04-02 10:40:51,600 INFO [train.py:903] (1/4) Epoch 17, batch 5900, loss[loss=0.1935, simple_loss=0.2664, pruned_loss=0.06033, over 15655.00 frames. ], tot_loss[loss=0.2137, simple_loss=0.2924, pruned_loss=0.06752, over 3848835.24 frames. ], batch size: 34, lr: 4.81e-03, grad_scale: 8.0 2023-04-02 10:40:55,184 WARNING [train.py:1073] (1/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 2023-04-02 10:40:55,561 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9702, 3.3092, 1.9334, 2.0048, 2.9243, 1.6517, 1.4165, 2.0881], device='cuda:1'), covar=tensor([0.1419, 0.0526, 0.0974, 0.0783, 0.0570, 0.1164, 0.0941, 0.0644], device='cuda:1'), in_proj_covar=tensor([0.0297, 0.0309, 0.0328, 0.0257, 0.0245, 0.0327, 0.0290, 0.0268], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 10:41:13,977 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9_rvb from training. Duration: 27.511125 2023-04-02 10:41:45,878 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=115193.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 10:41:51,149 INFO [train.py:903] (1/4) Epoch 17, batch 5950, loss[loss=0.2798, simple_loss=0.3469, pruned_loss=0.1064, over 18397.00 frames. ], tot_loss[loss=0.2139, simple_loss=0.2925, pruned_loss=0.06762, over 3853119.08 frames. ], batch size: 84, lr: 4.81e-03, grad_scale: 8.0 2023-04-02 10:42:00,467 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.984e+02 4.938e+02 6.318e+02 8.201e+02 2.090e+03, threshold=1.264e+03, percent-clipped=7.0 2023-04-02 10:42:20,332 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3247, 1.3426, 1.6336, 1.4896, 2.2210, 2.0488, 2.3176, 0.9426], device='cuda:1'), covar=tensor([0.2389, 0.4156, 0.2519, 0.1918, 0.1475, 0.2143, 0.1348, 0.4168], device='cuda:1'), in_proj_covar=tensor([0.0514, 0.0615, 0.0674, 0.0464, 0.0611, 0.0518, 0.0652, 0.0527], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 10:42:35,071 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=115233.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 10:42:43,156 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([6.0106, 5.4077, 3.0874, 4.7719, 1.3137, 5.4746, 5.3885, 5.5434], device='cuda:1'), covar=tensor([0.0386, 0.0817, 0.1673, 0.0580, 0.3526, 0.0520, 0.0716, 0.0914], device='cuda:1'), in_proj_covar=tensor([0.0471, 0.0388, 0.0469, 0.0336, 0.0391, 0.0406, 0.0404, 0.0433], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 10:42:51,742 INFO [train.py:903] (1/4) Epoch 17, batch 6000, loss[loss=0.1948, simple_loss=0.2706, pruned_loss=0.05949, over 19769.00 frames. ], tot_loss[loss=0.2148, simple_loss=0.2934, pruned_loss=0.06806, over 3848958.54 frames. ], batch size: 47, lr: 4.81e-03, grad_scale: 8.0 2023-04-02 10:42:51,742 INFO [train.py:928] (1/4) Computing validation loss 2023-04-02 10:43:04,253 INFO [train.py:937] (1/4) Epoch 17, validation: loss=0.1707, simple_loss=0.2712, pruned_loss=0.03505, over 944034.00 frames. 2023-04-02 10:43:04,254 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18821MB 2023-04-02 10:44:04,145 INFO [train.py:903] (1/4) Epoch 17, batch 6050, loss[loss=0.1771, simple_loss=0.2619, pruned_loss=0.04612, over 19845.00 frames. ], tot_loss[loss=0.2133, simple_loss=0.2922, pruned_loss=0.06724, over 3847207.60 frames. ], batch size: 52, lr: 4.81e-03, grad_scale: 8.0 2023-04-02 10:44:15,951 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.299e+02 5.156e+02 6.136e+02 7.598e+02 1.906e+03, threshold=1.227e+03, percent-clipped=4.0 2023-04-02 10:45:06,510 INFO [train.py:903] (1/4) Epoch 17, batch 6100, loss[loss=0.2105, simple_loss=0.2993, pruned_loss=0.06088, over 19667.00 frames. ], tot_loss[loss=0.2133, simple_loss=0.2921, pruned_loss=0.06726, over 3851269.35 frames. ], batch size: 55, lr: 4.81e-03, grad_scale: 8.0 2023-04-02 10:45:06,885 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=115348.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 10:45:32,783 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9782, 2.5651, 1.8296, 1.8872, 2.3844, 1.7394, 1.6447, 2.0857], device='cuda:1'), covar=tensor([0.0965, 0.0678, 0.0794, 0.0634, 0.0476, 0.0950, 0.0642, 0.0501], device='cuda:1'), in_proj_covar=tensor([0.0294, 0.0308, 0.0328, 0.0257, 0.0243, 0.0325, 0.0288, 0.0267], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 10:45:54,962 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=115388.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 10:45:56,148 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=115389.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 10:46:06,535 INFO [train.py:903] (1/4) Epoch 17, batch 6150, loss[loss=0.2026, simple_loss=0.2779, pruned_loss=0.06368, over 19621.00 frames. ], tot_loss[loss=0.2134, simple_loss=0.292, pruned_loss=0.06745, over 3854343.30 frames. ], batch size: 50, lr: 4.81e-03, grad_scale: 8.0 2023-04-02 10:46:15,589 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.636e+02 5.209e+02 6.440e+02 8.380e+02 1.538e+03, threshold=1.288e+03, percent-clipped=5.0 2023-04-02 10:46:33,785 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9_rvb from training. Duration: 31.02225 2023-04-02 10:46:55,033 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6821, 1.3618, 1.2811, 1.5345, 1.2614, 1.4563, 1.3081, 1.5114], device='cuda:1'), covar=tensor([0.1005, 0.0954, 0.1491, 0.0946, 0.1144, 0.0568, 0.1329, 0.0722], device='cuda:1'), in_proj_covar=tensor([0.0265, 0.0353, 0.0303, 0.0245, 0.0297, 0.0247, 0.0294, 0.0249], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 10:47:07,332 INFO [train.py:903] (1/4) Epoch 17, batch 6200, loss[loss=0.1956, simple_loss=0.2627, pruned_loss=0.06428, over 19745.00 frames. ], tot_loss[loss=0.2137, simple_loss=0.2921, pruned_loss=0.0677, over 3856513.16 frames. ], batch size: 45, lr: 4.81e-03, grad_scale: 8.0 2023-04-02 10:47:07,478 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=115448.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 10:47:08,911 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=115449.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 10:47:13,991 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.50 vs. limit=5.0 2023-04-02 10:47:39,845 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=115474.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 10:48:07,512 INFO [train.py:903] (1/4) Epoch 17, batch 6250, loss[loss=0.2123, simple_loss=0.2962, pruned_loss=0.06422, over 19674.00 frames. ], tot_loss[loss=0.2142, simple_loss=0.2928, pruned_loss=0.06777, over 3848233.12 frames. ], batch size: 58, lr: 4.81e-03, grad_scale: 8.0 2023-04-02 10:48:16,579 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.081e+02 4.690e+02 5.769e+02 7.890e+02 2.007e+03, threshold=1.154e+03, percent-clipped=3.0 2023-04-02 10:48:37,593 WARNING [train.py:1073] (1/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9_rvb from training. Duration: 25.988875 2023-04-02 10:48:38,468 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0 2023-04-02 10:48:52,594 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.5382, 4.7362, 5.2989, 5.3011, 2.3639, 4.9871, 4.3697, 4.9599], device='cuda:1'), covar=tensor([0.1486, 0.1398, 0.0507, 0.0573, 0.5079, 0.0732, 0.0596, 0.1009], device='cuda:1'), in_proj_covar=tensor([0.0738, 0.0689, 0.0891, 0.0772, 0.0792, 0.0644, 0.0534, 0.0819], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-02 10:49:00,686 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.2582, 2.0952, 1.8939, 1.7511, 1.6363, 1.8092, 0.5761, 1.1462], device='cuda:1'), covar=tensor([0.0492, 0.0555, 0.0434, 0.0675, 0.0973, 0.0759, 0.1189, 0.0936], device='cuda:1'), in_proj_covar=tensor([0.0349, 0.0346, 0.0347, 0.0376, 0.0446, 0.0381, 0.0326, 0.0335], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 10:49:09,161 INFO [train.py:903] (1/4) Epoch 17, batch 6300, loss[loss=0.1853, simple_loss=0.2584, pruned_loss=0.05617, over 19758.00 frames. ], tot_loss[loss=0.2138, simple_loss=0.2926, pruned_loss=0.06752, over 3837034.89 frames. ], batch size: 47, lr: 4.81e-03, grad_scale: 8.0 2023-04-02 10:49:27,667 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=115563.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 10:50:12,486 INFO [train.py:903] (1/4) Epoch 17, batch 6350, loss[loss=0.228, simple_loss=0.3045, pruned_loss=0.0757, over 19672.00 frames. ], tot_loss[loss=0.2123, simple_loss=0.2912, pruned_loss=0.0667, over 3846453.48 frames. ], batch size: 55, lr: 4.80e-03, grad_scale: 8.0 2023-04-02 10:50:19,854 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.6317, 4.1512, 4.2808, 4.2700, 1.6803, 4.0347, 3.5318, 4.0258], device='cuda:1'), covar=tensor([0.1563, 0.0800, 0.0634, 0.0707, 0.5418, 0.0803, 0.0695, 0.1098], device='cuda:1'), in_proj_covar=tensor([0.0739, 0.0690, 0.0895, 0.0776, 0.0794, 0.0646, 0.0534, 0.0820], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-02 10:50:20,018 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=115604.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 10:50:21,924 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.064e+02 4.833e+02 6.077e+02 8.091e+02 1.466e+03, threshold=1.215e+03, percent-clipped=5.0 2023-04-02 10:50:50,633 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=115629.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 10:51:13,890 INFO [train.py:903] (1/4) Epoch 17, batch 6400, loss[loss=0.2637, simple_loss=0.3374, pruned_loss=0.09494, over 17506.00 frames. ], tot_loss[loss=0.2142, simple_loss=0.2926, pruned_loss=0.06793, over 3829595.40 frames. ], batch size: 102, lr: 4.80e-03, grad_scale: 8.0 2023-04-02 10:52:15,088 INFO [train.py:903] (1/4) Epoch 17, batch 6450, loss[loss=0.1802, simple_loss=0.259, pruned_loss=0.05069, over 16491.00 frames. ], tot_loss[loss=0.2141, simple_loss=0.2922, pruned_loss=0.06795, over 3824134.87 frames. ], batch size: 36, lr: 4.80e-03, grad_scale: 8.0 2023-04-02 10:52:25,104 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.211e+02 4.829e+02 5.862e+02 7.962e+02 1.327e+03, threshold=1.172e+03, percent-clipped=3.0 2023-04-02 10:52:58,107 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=115732.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 10:52:59,220 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=115733.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 10:53:01,396 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9_rvb from training. Duration: 28.638875 2023-04-02 10:53:16,373 INFO [train.py:903] (1/4) Epoch 17, batch 6500, loss[loss=0.2695, simple_loss=0.3317, pruned_loss=0.1036, over 19398.00 frames. ], tot_loss[loss=0.214, simple_loss=0.2922, pruned_loss=0.06795, over 3827351.72 frames. ], batch size: 70, lr: 4.80e-03, grad_scale: 8.0 2023-04-02 10:53:24,001 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1_rvb from training. Duration: 0.836375 2023-04-02 10:54:18,473 INFO [train.py:903] (1/4) Epoch 17, batch 6550, loss[loss=0.221, simple_loss=0.2801, pruned_loss=0.08099, over 19748.00 frames. ], tot_loss[loss=0.2147, simple_loss=0.293, pruned_loss=0.06817, over 3832041.81 frames. ], batch size: 47, lr: 4.80e-03, grad_scale: 8.0 2023-04-02 10:54:28,764 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.723e+02 5.150e+02 6.522e+02 8.804e+02 2.234e+03, threshold=1.304e+03, percent-clipped=7.0 2023-04-02 10:54:41,251 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 2023-04-02 10:54:43,128 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=115819.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 10:55:15,257 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=115844.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 10:55:18,733 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=115847.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 10:55:19,552 INFO [train.py:903] (1/4) Epoch 17, batch 6600, loss[loss=0.1938, simple_loss=0.2648, pruned_loss=0.06144, over 19753.00 frames. ], tot_loss[loss=0.215, simple_loss=0.2935, pruned_loss=0.06826, over 3826171.71 frames. ], batch size: 47, lr: 4.80e-03, grad_scale: 8.0 2023-04-02 10:55:19,947 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=115848.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 10:56:19,810 INFO [train.py:903] (1/4) Epoch 17, batch 6650, loss[loss=0.219, simple_loss=0.2987, pruned_loss=0.06968, over 19651.00 frames. ], tot_loss[loss=0.2144, simple_loss=0.2928, pruned_loss=0.068, over 3830574.11 frames. ], batch size: 60, lr: 4.80e-03, grad_scale: 8.0 2023-04-02 10:56:21,357 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.7744, 2.2352, 2.3127, 2.6253, 2.4182, 2.2822, 2.3236, 2.5916], device='cuda:1'), covar=tensor([0.0817, 0.1668, 0.1323, 0.0988, 0.1376, 0.0484, 0.1109, 0.0610], device='cuda:1'), in_proj_covar=tensor([0.0266, 0.0354, 0.0305, 0.0245, 0.0298, 0.0247, 0.0294, 0.0250], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 10:56:30,891 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.015e+02 4.833e+02 5.946e+02 8.225e+02 1.682e+03, threshold=1.189e+03, percent-clipped=7.0 2023-04-02 10:56:35,578 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=115910.0, num_to_drop=1, layers_to_drop={0} 2023-04-02 10:57:05,031 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.2901, 2.0482, 2.2365, 2.8199, 1.9896, 2.7092, 2.5700, 2.6779], device='cuda:1'), covar=tensor([0.0724, 0.0823, 0.0889, 0.0791, 0.0848, 0.0638, 0.0877, 0.0529], device='cuda:1'), in_proj_covar=tensor([0.0210, 0.0220, 0.0224, 0.0243, 0.0226, 0.0210, 0.0187, 0.0204], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-02 10:57:08,287 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.5051, 4.0363, 4.2125, 4.2227, 1.7095, 3.9748, 3.4869, 3.9021], device='cuda:1'), covar=tensor([0.1659, 0.1001, 0.0658, 0.0682, 0.5309, 0.0953, 0.0684, 0.1177], device='cuda:1'), in_proj_covar=tensor([0.0733, 0.0688, 0.0890, 0.0772, 0.0789, 0.0643, 0.0533, 0.0812], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-02 10:57:21,958 INFO [train.py:903] (1/4) Epoch 17, batch 6700, loss[loss=0.211, simple_loss=0.2938, pruned_loss=0.06408, over 19534.00 frames. ], tot_loss[loss=0.2148, simple_loss=0.2932, pruned_loss=0.06816, over 3829820.07 frames. ], batch size: 56, lr: 4.80e-03, grad_scale: 8.0 2023-04-02 10:58:20,393 INFO [train.py:903] (1/4) Epoch 17, batch 6750, loss[loss=0.2404, simple_loss=0.3239, pruned_loss=0.07849, over 19643.00 frames. ], tot_loss[loss=0.2156, simple_loss=0.2935, pruned_loss=0.06881, over 3828332.34 frames. ], batch size: 58, lr: 4.80e-03, grad_scale: 8.0 2023-04-02 10:58:31,541 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.004e+02 5.304e+02 6.320e+02 7.514e+02 1.971e+03, threshold=1.264e+03, percent-clipped=7.0 2023-04-02 10:58:52,447 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.10 vs. limit=2.0 2023-04-02 10:58:59,757 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=116032.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 10:59:05,990 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.46 vs. limit=2.0 2023-04-02 10:59:12,142 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9512, 2.0399, 2.2132, 2.6410, 1.9743, 2.5261, 2.3007, 2.0307], device='cuda:1'), covar=tensor([0.4009, 0.3616, 0.1819, 0.2323, 0.3947, 0.1935, 0.4409, 0.3207], device='cuda:1'), in_proj_covar=tensor([0.0857, 0.0909, 0.0686, 0.0911, 0.0837, 0.0773, 0.0818, 0.0754], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 10:59:17,211 INFO [train.py:903] (1/4) Epoch 17, batch 6800, loss[loss=0.1965, simple_loss=0.277, pruned_loss=0.05803, over 19780.00 frames. ], tot_loss[loss=0.2166, simple_loss=0.2943, pruned_loss=0.06944, over 3817593.44 frames. ], batch size: 54, lr: 4.80e-03, grad_scale: 8.0 2023-04-02 11:00:03,229 WARNING [train.py:1073] (1/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 2023-04-02 11:00:04,647 WARNING [train.py:1073] (1/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 2023-04-02 11:00:07,138 INFO [train.py:903] (1/4) Epoch 18, batch 0, loss[loss=0.2232, simple_loss=0.3123, pruned_loss=0.06706, over 19520.00 frames. ], tot_loss[loss=0.2232, simple_loss=0.3123, pruned_loss=0.06706, over 19520.00 frames. ], batch size: 56, lr: 4.66e-03, grad_scale: 8.0 2023-04-02 11:00:07,138 INFO [train.py:928] (1/4) Computing validation loss 2023-04-02 11:00:18,781 INFO [train.py:937] (1/4) Epoch 18, validation: loss=0.1712, simple_loss=0.2722, pruned_loss=0.03505, over 944034.00 frames. 2023-04-02 11:00:18,782 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18821MB 2023-04-02 11:00:32,354 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 2023-04-02 11:00:34,761 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.6197, 4.1911, 2.6493, 3.6779, 0.9524, 4.0146, 4.0515, 4.0868], device='cuda:1'), covar=tensor([0.0592, 0.0941, 0.1935, 0.0774, 0.4032, 0.0721, 0.0777, 0.1003], device='cuda:1'), in_proj_covar=tensor([0.0472, 0.0389, 0.0470, 0.0331, 0.0394, 0.0408, 0.0402, 0.0432], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 11:00:51,653 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=116103.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 11:00:52,786 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=116104.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 11:00:55,571 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.457e+02 4.972e+02 6.494e+02 8.085e+02 1.604e+03, threshold=1.299e+03, percent-clipped=1.0 2023-04-02 11:01:15,448 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=116123.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 11:01:18,719 INFO [train.py:903] (1/4) Epoch 18, batch 50, loss[loss=0.2053, simple_loss=0.2876, pruned_loss=0.06148, over 19658.00 frames. ], tot_loss[loss=0.2219, simple_loss=0.3008, pruned_loss=0.07154, over 877091.59 frames. ], batch size: 55, lr: 4.66e-03, grad_scale: 8.0 2023-04-02 11:01:21,430 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=116128.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 11:01:23,499 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=116129.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 11:01:30,196 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=116134.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 11:01:52,476 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1_rvb from training. Duration: 27.0318125 2023-04-02 11:02:00,153 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.5628, 1.1529, 1.3632, 1.1638, 2.2310, 0.9648, 2.1230, 2.4174], device='cuda:1'), covar=tensor([0.0696, 0.2773, 0.2790, 0.1737, 0.0850, 0.2111, 0.0976, 0.0486], device='cuda:1'), in_proj_covar=tensor([0.0386, 0.0354, 0.0370, 0.0339, 0.0359, 0.0343, 0.0357, 0.0381], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 11:02:21,179 INFO [train.py:903] (1/4) Epoch 18, batch 100, loss[loss=0.2227, simple_loss=0.3094, pruned_loss=0.06793, over 19328.00 frames. ], tot_loss[loss=0.2155, simple_loss=0.2951, pruned_loss=0.06792, over 1544577.18 frames. ], batch size: 66, lr: 4.66e-03, grad_scale: 8.0 2023-04-02 11:02:32,225 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 2023-04-02 11:02:41,863 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([6.1958, 5.6329, 3.1766, 4.8830, 1.0807, 5.6877, 5.6044, 5.7888], device='cuda:1'), covar=tensor([0.0384, 0.0823, 0.1778, 0.0654, 0.4175, 0.0498, 0.0713, 0.0905], device='cuda:1'), in_proj_covar=tensor([0.0472, 0.0390, 0.0472, 0.0332, 0.0395, 0.0408, 0.0403, 0.0433], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 11:02:58,309 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.917e+02 4.838e+02 6.090e+02 7.458e+02 2.009e+03, threshold=1.218e+03, percent-clipped=2.0 2023-04-02 11:03:21,607 INFO [train.py:903] (1/4) Epoch 18, batch 150, loss[loss=0.218, simple_loss=0.2881, pruned_loss=0.07392, over 19338.00 frames. ], tot_loss[loss=0.2144, simple_loss=0.2929, pruned_loss=0.06795, over 2050321.51 frames. ], batch size: 44, lr: 4.65e-03, grad_scale: 8.0 2023-04-02 11:03:56,115 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=116254.0, num_to_drop=1, layers_to_drop={0} 2023-04-02 11:04:20,461 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 2023-04-02 11:04:21,616 INFO [train.py:903] (1/4) Epoch 18, batch 200, loss[loss=0.2132, simple_loss=0.3008, pruned_loss=0.06282, over 19665.00 frames. ], tot_loss[loss=0.2143, simple_loss=0.2922, pruned_loss=0.06818, over 2458672.89 frames. ], batch size: 55, lr: 4.65e-03, grad_scale: 8.0 2023-04-02 11:05:01,398 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.017e+02 4.750e+02 5.613e+02 7.153e+02 1.890e+03, threshold=1.123e+03, percent-clipped=2.0 2023-04-02 11:05:24,090 INFO [train.py:903] (1/4) Epoch 18, batch 250, loss[loss=0.2023, simple_loss=0.2831, pruned_loss=0.06073, over 19397.00 frames. ], tot_loss[loss=0.2133, simple_loss=0.2916, pruned_loss=0.06752, over 2766899.31 frames. ], batch size: 48, lr: 4.65e-03, grad_scale: 8.0 2023-04-02 11:05:43,834 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=116341.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 11:05:57,157 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.76 vs. limit=2.0 2023-04-02 11:06:06,000 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4410, 1.6147, 2.0945, 1.8670, 3.1553, 4.7168, 4.5651, 5.1258], device='cuda:1'), covar=tensor([0.1566, 0.3508, 0.2981, 0.1976, 0.0559, 0.0179, 0.0164, 0.0153], device='cuda:1'), in_proj_covar=tensor([0.0264, 0.0312, 0.0340, 0.0258, 0.0232, 0.0176, 0.0210, 0.0237], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-02 11:06:18,033 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=116369.0, num_to_drop=1, layers_to_drop={0} 2023-04-02 11:06:19,007 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4048, 1.0281, 1.2138, 2.0271, 1.6071, 1.5089, 1.6717, 1.3983], device='cuda:1'), covar=tensor([0.1079, 0.1662, 0.1411, 0.0986, 0.1091, 0.1376, 0.1194, 0.1090], device='cuda:1'), in_proj_covar=tensor([0.0208, 0.0220, 0.0223, 0.0240, 0.0225, 0.0209, 0.0186, 0.0202], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-02 11:06:25,430 INFO [train.py:903] (1/4) Epoch 18, batch 300, loss[loss=0.1937, simple_loss=0.294, pruned_loss=0.04673, over 19696.00 frames. ], tot_loss[loss=0.2145, simple_loss=0.2922, pruned_loss=0.0684, over 2997677.55 frames. ], batch size: 59, lr: 4.65e-03, grad_scale: 8.0 2023-04-02 11:06:25,579 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=116376.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 11:07:03,943 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.432e+02 5.241e+02 6.705e+02 8.261e+02 1.478e+03, threshold=1.341e+03, percent-clipped=3.0 2023-04-02 11:07:28,470 INFO [train.py:903] (1/4) Epoch 18, batch 350, loss[loss=0.2045, simple_loss=0.2802, pruned_loss=0.06437, over 19479.00 frames. ], tot_loss[loss=0.215, simple_loss=0.293, pruned_loss=0.06843, over 3179204.94 frames. ], batch size: 49, lr: 4.65e-03, grad_scale: 8.0 2023-04-02 11:07:33,274 WARNING [train.py:1073] (1/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1_rvb from training. Duration: 0.9681875 2023-04-02 11:07:44,272 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 2023-04-02 11:08:16,138 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=116464.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 11:08:19,411 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=116467.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 11:08:29,490 INFO [train.py:903] (1/4) Epoch 18, batch 400, loss[loss=0.2147, simple_loss=0.2935, pruned_loss=0.06791, over 19600.00 frames. ], tot_loss[loss=0.2147, simple_loss=0.293, pruned_loss=0.06823, over 3340844.98 frames. ], batch size: 57, lr: 4.65e-03, grad_scale: 8.0 2023-04-02 11:08:31,845 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=116478.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 11:08:48,738 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=116491.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 11:09:08,777 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.099e+02 4.874e+02 5.859e+02 7.069e+02 1.370e+03, threshold=1.172e+03, percent-clipped=1.0 2023-04-02 11:09:27,696 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=116523.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 11:09:31,058 INFO [train.py:903] (1/4) Epoch 18, batch 450, loss[loss=0.1639, simple_loss=0.2396, pruned_loss=0.04413, over 19731.00 frames. ], tot_loss[loss=0.2143, simple_loss=0.2924, pruned_loss=0.06811, over 3459733.19 frames. ], batch size: 46, lr: 4.65e-03, grad_scale: 8.0 2023-04-02 11:10:06,952 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 2023-04-02 11:10:08,090 WARNING [train.py:1073] (1/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 2023-04-02 11:10:36,075 INFO [train.py:903] (1/4) Epoch 18, batch 500, loss[loss=0.2072, simple_loss=0.2941, pruned_loss=0.06015, over 19773.00 frames. ], tot_loss[loss=0.2141, simple_loss=0.2923, pruned_loss=0.0679, over 3549791.12 frames. ], batch size: 56, lr: 4.65e-03, grad_scale: 8.0 2023-04-02 11:10:43,235 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=116582.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 11:10:57,219 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=116593.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 11:11:13,493 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.162e+02 5.090e+02 6.291e+02 8.243e+02 1.843e+03, threshold=1.258e+03, percent-clipped=5.0 2023-04-02 11:11:38,135 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=116625.0, num_to_drop=1, layers_to_drop={0} 2023-04-02 11:11:38,807 INFO [train.py:903] (1/4) Epoch 18, batch 550, loss[loss=0.2267, simple_loss=0.2947, pruned_loss=0.07939, over 19733.00 frames. ], tot_loss[loss=0.2144, simple_loss=0.2925, pruned_loss=0.06813, over 3604761.34 frames. ], batch size: 45, lr: 4.65e-03, grad_scale: 8.0 2023-04-02 11:11:51,117 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=116636.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 11:12:07,796 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=116650.0, num_to_drop=1, layers_to_drop={1} 2023-04-02 11:12:07,847 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5528, 1.5858, 2.0072, 1.7681, 2.6981, 2.2209, 2.6987, 1.5947], device='cuda:1'), covar=tensor([0.2402, 0.4078, 0.2400, 0.1904, 0.1589, 0.2204, 0.1630, 0.3942], device='cuda:1'), in_proj_covar=tensor([0.0512, 0.0612, 0.0671, 0.0463, 0.0611, 0.0516, 0.0648, 0.0522], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 11:12:41,000 INFO [train.py:903] (1/4) Epoch 18, batch 600, loss[loss=0.2183, simple_loss=0.2989, pruned_loss=0.06885, over 19665.00 frames. ], tot_loss[loss=0.2139, simple_loss=0.2919, pruned_loss=0.06794, over 3648251.88 frames. ], batch size: 55, lr: 4.65e-03, grad_scale: 8.0 2023-04-02 11:12:51,641 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=116685.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 11:13:18,876 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.849e+02 4.880e+02 6.214e+02 8.095e+02 1.532e+03, threshold=1.243e+03, percent-clipped=4.0 2023-04-02 11:13:21,174 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 2023-04-02 11:13:28,790 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.3422, 2.2186, 2.0649, 2.5794, 2.2931, 2.2194, 1.9084, 2.4146], device='cuda:1'), covar=tensor([0.1006, 0.1527, 0.1427, 0.0962, 0.1359, 0.0503, 0.1317, 0.0674], device='cuda:1'), in_proj_covar=tensor([0.0265, 0.0356, 0.0302, 0.0247, 0.0299, 0.0247, 0.0296, 0.0252], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 11:13:37,835 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=116722.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 11:13:42,020 INFO [train.py:903] (1/4) Epoch 18, batch 650, loss[loss=0.2256, simple_loss=0.2997, pruned_loss=0.07578, over 19332.00 frames. ], tot_loss[loss=0.2147, simple_loss=0.2929, pruned_loss=0.06821, over 3695643.24 frames. ], batch size: 44, lr: 4.64e-03, grad_scale: 8.0 2023-04-02 11:13:55,024 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.9333, 1.6384, 1.8046, 1.6887, 4.4522, 1.0022, 2.6878, 4.8435], device='cuda:1'), covar=tensor([0.0438, 0.2625, 0.2628, 0.1990, 0.0707, 0.2741, 0.1329, 0.0178], device='cuda:1'), in_proj_covar=tensor([0.0390, 0.0354, 0.0374, 0.0340, 0.0362, 0.0345, 0.0359, 0.0384], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 11:14:08,712 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=116747.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 11:14:34,419 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.46 vs. limit=5.0 2023-04-02 11:14:38,753 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=116772.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 11:14:43,009 INFO [train.py:903] (1/4) Epoch 18, batch 700, loss[loss=0.2157, simple_loss=0.296, pruned_loss=0.06768, over 19596.00 frames. ], tot_loss[loss=0.215, simple_loss=0.2933, pruned_loss=0.06834, over 3730937.97 frames. ], batch size: 61, lr: 4.64e-03, grad_scale: 8.0 2023-04-02 11:15:15,686 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=116800.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 11:15:23,663 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.148e+02 4.855e+02 5.777e+02 7.000e+02 1.472e+03, threshold=1.155e+03, percent-clipped=1.0 2023-04-02 11:15:24,935 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=116808.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 11:15:27,853 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 2023-04-02 11:15:39,747 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=116820.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 11:15:47,508 INFO [train.py:903] (1/4) Epoch 18, batch 750, loss[loss=0.2232, simple_loss=0.3053, pruned_loss=0.07056, over 19451.00 frames. ], tot_loss[loss=0.215, simple_loss=0.2933, pruned_loss=0.06829, over 3749969.47 frames. ], batch size: 64, lr: 4.64e-03, grad_scale: 8.0 2023-04-02 11:16:03,683 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=116838.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 11:16:16,552 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=116849.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 11:16:35,181 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=116863.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 11:16:39,887 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=116867.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 11:16:50,594 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=116874.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 11:16:52,580 INFO [train.py:903] (1/4) Epoch 18, batch 800, loss[loss=0.2273, simple_loss=0.3059, pruned_loss=0.07436, over 19525.00 frames. ], tot_loss[loss=0.2162, simple_loss=0.2945, pruned_loss=0.06897, over 3765206.80 frames. ], batch size: 54, lr: 4.64e-03, grad_scale: 8.0 2023-04-02 11:17:06,582 WARNING [train.py:1073] (1/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 2023-04-02 11:17:32,326 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.570e+02 5.513e+02 6.326e+02 7.669e+02 1.889e+03, threshold=1.265e+03, percent-clipped=5.0 2023-04-02 11:17:51,817 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=116923.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 11:17:55,032 INFO [train.py:903] (1/4) Epoch 18, batch 850, loss[loss=0.1983, simple_loss=0.2694, pruned_loss=0.06364, over 18146.00 frames. ], tot_loss[loss=0.2166, simple_loss=0.2949, pruned_loss=0.0692, over 3783785.21 frames. ], batch size: 40, lr: 4.64e-03, grad_scale: 4.0 2023-04-02 11:18:12,803 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1301, 1.3991, 1.7017, 1.0806, 2.4985, 3.4228, 3.0932, 3.5914], device='cuda:1'), covar=tensor([0.1571, 0.3469, 0.3069, 0.2399, 0.0555, 0.0173, 0.0226, 0.0245], device='cuda:1'), in_proj_covar=tensor([0.0263, 0.0312, 0.0339, 0.0256, 0.0231, 0.0177, 0.0209, 0.0236], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-02 11:18:48,138 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9_rvb from training. Duration: 25.061125 2023-04-02 11:18:56,389 INFO [train.py:903] (1/4) Epoch 18, batch 900, loss[loss=0.2658, simple_loss=0.3303, pruned_loss=0.1006, over 13277.00 frames. ], tot_loss[loss=0.2162, simple_loss=0.2944, pruned_loss=0.06901, over 3805441.86 frames. ], batch size: 136, lr: 4.64e-03, grad_scale: 4.0 2023-04-02 11:19:01,328 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=116980.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 11:19:03,966 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=116982.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 11:19:38,590 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.269e+02 4.671e+02 5.637e+02 7.276e+02 1.422e+03, threshold=1.127e+03, percent-clipped=2.0 2023-04-02 11:20:00,610 INFO [train.py:903] (1/4) Epoch 18, batch 950, loss[loss=0.2367, simple_loss=0.316, pruned_loss=0.07865, over 19679.00 frames. ], tot_loss[loss=0.2137, simple_loss=0.2918, pruned_loss=0.06775, over 3819848.11 frames. ], batch size: 60, lr: 4.64e-03, grad_scale: 4.0 2023-04-02 11:20:02,910 WARNING [train.py:1073] (1/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9_rvb from training. Duration: 26.32775 2023-04-02 11:20:38,619 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=117056.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 11:20:51,127 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=117066.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 11:21:03,205 INFO [train.py:903] (1/4) Epoch 18, batch 1000, loss[loss=0.2614, simple_loss=0.3248, pruned_loss=0.09901, over 17473.00 frames. ], tot_loss[loss=0.215, simple_loss=0.2932, pruned_loss=0.06842, over 3818516.19 frames. ], batch size: 101, lr: 4.64e-03, grad_scale: 4.0 2023-04-02 11:21:11,172 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=117081.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 11:21:27,337 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=117095.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 11:21:43,056 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.648e+02 4.999e+02 6.181e+02 7.829e+02 2.221e+03, threshold=1.236e+03, percent-clipped=4.0 2023-04-02 11:21:56,747 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 2023-04-02 11:22:07,202 INFO [train.py:903] (1/4) Epoch 18, batch 1050, loss[loss=0.2849, simple_loss=0.3527, pruned_loss=0.1086, over 19716.00 frames. ], tot_loss[loss=0.2144, simple_loss=0.2926, pruned_loss=0.06808, over 3837898.46 frames. ], batch size: 63, lr: 4.64e-03, grad_scale: 4.0 2023-04-02 11:22:40,244 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 2023-04-02 11:22:55,009 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=117164.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 11:23:09,065 INFO [train.py:903] (1/4) Epoch 18, batch 1100, loss[loss=0.1926, simple_loss=0.2777, pruned_loss=0.05375, over 19693.00 frames. ], tot_loss[loss=0.2134, simple_loss=0.2918, pruned_loss=0.06745, over 3839019.19 frames. ], batch size: 59, lr: 4.64e-03, grad_scale: 4.0 2023-04-02 11:23:13,165 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=117179.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 11:23:15,471 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=117181.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 11:23:24,955 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0908, 1.9292, 1.7700, 1.6142, 1.4176, 1.5616, 0.5874, 1.0190], device='cuda:1'), covar=tensor([0.0555, 0.0622, 0.0466, 0.0707, 0.1252, 0.0937, 0.1143, 0.0997], device='cuda:1'), in_proj_covar=tensor([0.0351, 0.0347, 0.0347, 0.0376, 0.0449, 0.0382, 0.0328, 0.0335], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 11:23:44,468 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=117204.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 11:23:49,508 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.961e+02 5.150e+02 6.225e+02 7.900e+02 1.283e+03, threshold=1.245e+03, percent-clipped=2.0 2023-04-02 11:24:11,125 INFO [train.py:903] (1/4) Epoch 18, batch 1150, loss[loss=0.21, simple_loss=0.2934, pruned_loss=0.06332, over 19673.00 frames. ], tot_loss[loss=0.2151, simple_loss=0.2933, pruned_loss=0.06841, over 3830881.09 frames. ], batch size: 60, lr: 4.63e-03, grad_scale: 4.0 2023-04-02 11:24:27,252 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=117238.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 11:24:28,299 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=117239.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 11:24:39,772 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([5.9657, 5.3381, 3.0004, 4.5781, 0.8971, 5.4115, 5.3549, 5.4762], device='cuda:1'), covar=tensor([0.0380, 0.0897, 0.1855, 0.0667, 0.4271, 0.0550, 0.0700, 0.0818], device='cuda:1'), in_proj_covar=tensor([0.0479, 0.0393, 0.0475, 0.0337, 0.0395, 0.0413, 0.0406, 0.0436], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 11:24:58,323 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=117263.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 11:25:14,042 INFO [train.py:903] (1/4) Epoch 18, batch 1200, loss[loss=0.2346, simple_loss=0.305, pruned_loss=0.08209, over 19535.00 frames. ], tot_loss[loss=0.2148, simple_loss=0.293, pruned_loss=0.06831, over 3827250.03 frames. ], batch size: 54, lr: 4.63e-03, grad_scale: 8.0 2023-04-02 11:25:18,957 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=117279.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 11:25:31,264 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.8239, 1.5539, 1.4770, 1.8515, 1.5346, 1.6021, 1.4940, 1.7244], device='cuda:1'), covar=tensor([0.0951, 0.1255, 0.1432, 0.0952, 0.1161, 0.0541, 0.1274, 0.0763], device='cuda:1'), in_proj_covar=tensor([0.0264, 0.0355, 0.0302, 0.0248, 0.0299, 0.0247, 0.0295, 0.0251], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 11:25:50,730 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 2023-04-02 11:25:54,168 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.294e+02 4.775e+02 5.862e+02 7.562e+02 1.280e+03, threshold=1.172e+03, percent-clipped=1.0 2023-04-02 11:26:12,979 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.59 vs. limit=2.0 2023-04-02 11:26:18,175 INFO [train.py:903] (1/4) Epoch 18, batch 1250, loss[loss=0.2235, simple_loss=0.3079, pruned_loss=0.06962, over 18252.00 frames. ], tot_loss[loss=0.2139, simple_loss=0.2927, pruned_loss=0.06756, over 3828660.56 frames. ], batch size: 83, lr: 4.63e-03, grad_scale: 8.0 2023-04-02 11:26:43,786 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=117347.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 11:26:49,489 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=117351.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 11:26:58,465 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.2674, 2.9190, 2.2208, 2.2308, 2.2398, 2.4937, 0.9077, 2.0418], device='cuda:1'), covar=tensor([0.0632, 0.0619, 0.0717, 0.1099, 0.0998, 0.1142, 0.1408, 0.1065], device='cuda:1'), in_proj_covar=tensor([0.0350, 0.0347, 0.0348, 0.0376, 0.0449, 0.0381, 0.0328, 0.0334], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 11:27:20,900 INFO [train.py:903] (1/4) Epoch 18, batch 1300, loss[loss=0.1937, simple_loss=0.2682, pruned_loss=0.05964, over 19772.00 frames. ], tot_loss[loss=0.2142, simple_loss=0.2928, pruned_loss=0.06773, over 3830805.68 frames. ], batch size: 47, lr: 4.63e-03, grad_scale: 8.0 2023-04-02 11:27:21,309 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=117376.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 11:27:48,173 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6235, 1.5375, 1.6124, 2.3225, 1.6801, 2.0366, 2.0625, 1.7992], device='cuda:1'), covar=tensor([0.0872, 0.0974, 0.1009, 0.0718, 0.0869, 0.0724, 0.0857, 0.0715], device='cuda:1'), in_proj_covar=tensor([0.0210, 0.0221, 0.0223, 0.0240, 0.0226, 0.0209, 0.0186, 0.0204], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-02 11:28:01,379 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.655e+02 5.418e+02 6.594e+02 8.357e+02 1.516e+03, threshold=1.319e+03, percent-clipped=5.0 2023-04-02 11:28:22,257 INFO [train.py:903] (1/4) Epoch 18, batch 1350, loss[loss=0.2165, simple_loss=0.3071, pruned_loss=0.06292, over 19659.00 frames. ], tot_loss[loss=0.2139, simple_loss=0.2923, pruned_loss=0.0677, over 3824813.23 frames. ], batch size: 58, lr: 4.63e-03, grad_scale: 8.0 2023-04-02 11:28:37,213 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=117437.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 11:29:07,720 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=117462.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 11:29:21,695 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.75 vs. limit=2.0 2023-04-02 11:29:24,521 INFO [train.py:903] (1/4) Epoch 18, batch 1400, loss[loss=0.2153, simple_loss=0.3045, pruned_loss=0.06301, over 19781.00 frames. ], tot_loss[loss=0.2143, simple_loss=0.2925, pruned_loss=0.06802, over 3824020.76 frames. ], batch size: 56, lr: 4.63e-03, grad_scale: 8.0 2023-04-02 11:30:04,552 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.042e+02 5.443e+02 6.741e+02 8.791e+02 2.167e+03, threshold=1.348e+03, percent-clipped=5.0 2023-04-02 11:30:28,260 INFO [train.py:903] (1/4) Epoch 18, batch 1450, loss[loss=0.2048, simple_loss=0.2876, pruned_loss=0.06097, over 19582.00 frames. ], tot_loss[loss=0.2144, simple_loss=0.2927, pruned_loss=0.06801, over 3822765.05 frames. ], batch size: 52, lr: 4.63e-03, grad_scale: 4.0 2023-04-02 11:30:29,446 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_rvb from training. Duration: 25.85 2023-04-02 11:30:40,106 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=117535.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 11:31:11,203 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=117560.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 11:31:30,945 INFO [train.py:903] (1/4) Epoch 18, batch 1500, loss[loss=0.2047, simple_loss=0.2793, pruned_loss=0.06507, over 19608.00 frames. ], tot_loss[loss=0.2139, simple_loss=0.2925, pruned_loss=0.06758, over 3839298.66 frames. ], batch size: 50, lr: 4.63e-03, grad_scale: 4.0 2023-04-02 11:31:39,057 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=117583.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 11:31:53,638 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.48 vs. limit=2.0 2023-04-02 11:32:11,896 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.089e+02 4.657e+02 6.062e+02 7.787e+02 1.498e+03, threshold=1.212e+03, percent-clipped=2.0 2023-04-02 11:32:32,138 INFO [train.py:903] (1/4) Epoch 18, batch 1550, loss[loss=0.2318, simple_loss=0.3097, pruned_loss=0.07697, over 19774.00 frames. ], tot_loss[loss=0.2148, simple_loss=0.2932, pruned_loss=0.06821, over 3821064.21 frames. ], batch size: 56, lr: 4.63e-03, grad_scale: 4.0 2023-04-02 11:33:34,531 INFO [train.py:903] (1/4) Epoch 18, batch 1600, loss[loss=0.2021, simple_loss=0.2761, pruned_loss=0.06404, over 19597.00 frames. ], tot_loss[loss=0.2148, simple_loss=0.2932, pruned_loss=0.06817, over 3816742.10 frames. ], batch size: 50, lr: 4.63e-03, grad_scale: 8.0 2023-04-02 11:33:54,960 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=117691.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 11:34:01,925 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9_rvb from training. Duration: 30.1555625 2023-04-02 11:34:03,380 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=117698.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 11:34:15,617 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.549e+02 4.964e+02 5.954e+02 7.051e+02 1.393e+03, threshold=1.191e+03, percent-clipped=4.0 2023-04-02 11:34:37,828 INFO [train.py:903] (1/4) Epoch 18, batch 1650, loss[loss=0.2887, simple_loss=0.3571, pruned_loss=0.1102, over 19705.00 frames. ], tot_loss[loss=0.2139, simple_loss=0.2924, pruned_loss=0.0677, over 3825795.83 frames. ], batch size: 59, lr: 4.63e-03, grad_scale: 8.0 2023-04-02 11:34:44,895 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3917, 1.4336, 1.9299, 1.5276, 2.8089, 3.4681, 3.2752, 3.6658], device='cuda:1'), covar=tensor([0.1659, 0.3700, 0.3085, 0.2270, 0.0638, 0.0230, 0.0199, 0.0294], device='cuda:1'), in_proj_covar=tensor([0.0263, 0.0309, 0.0338, 0.0257, 0.0230, 0.0177, 0.0209, 0.0235], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-02 11:35:04,216 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.3620, 2.4244, 2.6043, 3.3075, 2.3686, 3.1678, 2.7163, 2.4032], device='cuda:1'), covar=tensor([0.4000, 0.3929, 0.1720, 0.2213, 0.4240, 0.1828, 0.4213, 0.3160], device='cuda:1'), in_proj_covar=tensor([0.0858, 0.0911, 0.0689, 0.0910, 0.0837, 0.0775, 0.0818, 0.0757], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 11:35:39,488 INFO [train.py:903] (1/4) Epoch 18, batch 1700, loss[loss=0.2307, simple_loss=0.3122, pruned_loss=0.07455, over 18129.00 frames. ], tot_loss[loss=0.2132, simple_loss=0.2919, pruned_loss=0.06732, over 3826480.04 frames. ], batch size: 83, lr: 4.62e-03, grad_scale: 8.0 2023-04-02 11:36:16,633 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=117806.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 11:36:19,947 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.954e+02 5.098e+02 6.258e+02 7.253e+02 1.524e+03, threshold=1.252e+03, percent-clipped=2.0 2023-04-02 11:36:21,124 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 2023-04-02 11:36:40,238 INFO [train.py:903] (1/4) Epoch 18, batch 1750, loss[loss=0.2364, simple_loss=0.3252, pruned_loss=0.07383, over 19719.00 frames. ], tot_loss[loss=0.2149, simple_loss=0.2931, pruned_loss=0.06833, over 3807236.66 frames. ], batch size: 63, lr: 4.62e-03, grad_scale: 8.0 2023-04-02 11:37:14,810 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.14 vs. limit=2.0 2023-04-02 11:37:43,040 INFO [train.py:903] (1/4) Epoch 18, batch 1800, loss[loss=0.1862, simple_loss=0.2732, pruned_loss=0.04957, over 19620.00 frames. ], tot_loss[loss=0.2154, simple_loss=0.2938, pruned_loss=0.06844, over 3808344.33 frames. ], batch size: 50, lr: 4.62e-03, grad_scale: 8.0 2023-04-02 11:38:23,379 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.984e+02 4.803e+02 6.041e+02 7.952e+02 1.877e+03, threshold=1.208e+03, percent-clipped=3.0 2023-04-02 11:38:42,090 WARNING [train.py:1073] (1/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9_rvb from training. Duration: 25.0944375 2023-04-02 11:38:45,250 INFO [train.py:903] (1/4) Epoch 18, batch 1850, loss[loss=0.2115, simple_loss=0.2911, pruned_loss=0.06594, over 19776.00 frames. ], tot_loss[loss=0.2151, simple_loss=0.2935, pruned_loss=0.06834, over 3805680.63 frames. ], batch size: 56, lr: 4.62e-03, grad_scale: 8.0 2023-04-02 11:39:18,903 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9_rvb from training. Duration: 27.8166875 2023-04-02 11:39:19,310 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=117954.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 11:39:47,639 INFO [train.py:903] (1/4) Epoch 18, batch 1900, loss[loss=0.1965, simple_loss=0.2625, pruned_loss=0.06525, over 19001.00 frames. ], tot_loss[loss=0.2148, simple_loss=0.2933, pruned_loss=0.0681, over 3817223.10 frames. ], batch size: 42, lr: 4.62e-03, grad_scale: 8.0 2023-04-02 11:39:51,607 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=117979.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 11:40:03,258 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 2023-04-02 11:40:07,857 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9_rvb from training. Duration: 27.02225 2023-04-02 11:40:27,944 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.102e+02 5.120e+02 6.309e+02 7.973e+02 1.539e+03, threshold=1.262e+03, percent-clipped=6.0 2023-04-02 11:40:34,560 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 2023-04-02 11:40:48,198 INFO [train.py:903] (1/4) Epoch 18, batch 1950, loss[loss=0.2105, simple_loss=0.2993, pruned_loss=0.06083, over 17854.00 frames. ], tot_loss[loss=0.2156, simple_loss=0.2939, pruned_loss=0.06864, over 3813305.10 frames. ], batch size: 83, lr: 4.62e-03, grad_scale: 8.0 2023-04-02 11:41:28,971 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=118058.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 11:41:33,499 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=118062.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 11:41:49,956 INFO [train.py:903] (1/4) Epoch 18, batch 2000, loss[loss=0.2657, simple_loss=0.329, pruned_loss=0.1012, over 18708.00 frames. ], tot_loss[loss=0.2159, simple_loss=0.2944, pruned_loss=0.06867, over 3818366.24 frames. ], batch size: 74, lr: 4.62e-03, grad_scale: 8.0 2023-04-02 11:42:05,421 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=118087.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 11:42:11,305 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.60 vs. limit=5.0 2023-04-02 11:42:31,794 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.263e+02 5.367e+02 6.632e+02 8.072e+02 1.503e+03, threshold=1.326e+03, percent-clipped=5.0 2023-04-02 11:42:47,639 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_rvb from training. Duration: 26.8349375 2023-04-02 11:42:54,016 INFO [train.py:903] (1/4) Epoch 18, batch 2050, loss[loss=0.2062, simple_loss=0.287, pruned_loss=0.06267, over 19683.00 frames. ], tot_loss[loss=0.2153, simple_loss=0.2937, pruned_loss=0.06843, over 3813427.43 frames. ], batch size: 53, lr: 4.62e-03, grad_scale: 8.0 2023-04-02 11:43:06,268 WARNING [train.py:1073] (1/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 2023-04-02 11:43:07,436 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 2023-04-02 11:43:26,099 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 2023-04-02 11:43:55,359 INFO [train.py:903] (1/4) Epoch 18, batch 2100, loss[loss=0.1783, simple_loss=0.2566, pruned_loss=0.04994, over 19764.00 frames. ], tot_loss[loss=0.214, simple_loss=0.2927, pruned_loss=0.06767, over 3815969.08 frames. ], batch size: 47, lr: 4.62e-03, grad_scale: 8.0 2023-04-02 11:44:06,993 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=118186.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 11:44:21,259 WARNING [train.py:1073] (1/4) Exclude cut with ID 453-131332-0000-131866_sp0.9_rvb from training. Duration: 25.3333125 2023-04-02 11:44:36,112 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.283e+02 5.186e+02 6.689e+02 8.493e+02 1.656e+03, threshold=1.338e+03, percent-clipped=6.0 2023-04-02 11:44:44,889 WARNING [train.py:1073] (1/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9_rvb from training. Duration: 26.6166875 2023-04-02 11:44:56,294 INFO [train.py:903] (1/4) Epoch 18, batch 2150, loss[loss=0.2018, simple_loss=0.2703, pruned_loss=0.06666, over 19855.00 frames. ], tot_loss[loss=0.2138, simple_loss=0.2924, pruned_loss=0.06762, over 3815541.27 frames. ], batch size: 52, lr: 4.62e-03, grad_scale: 8.0 2023-04-02 11:45:39,515 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.6101, 2.4819, 2.3068, 2.7687, 2.5741, 2.4666, 2.1281, 2.9880], device='cuda:1'), covar=tensor([0.0943, 0.1613, 0.1410, 0.1038, 0.1351, 0.0471, 0.1346, 0.0534], device='cuda:1'), in_proj_covar=tensor([0.0262, 0.0350, 0.0299, 0.0245, 0.0296, 0.0244, 0.0292, 0.0249], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 11:45:57,807 INFO [train.py:903] (1/4) Epoch 18, batch 2200, loss[loss=0.2013, simple_loss=0.2789, pruned_loss=0.06179, over 19870.00 frames. ], tot_loss[loss=0.2139, simple_loss=0.2922, pruned_loss=0.06777, over 3814441.65 frames. ], batch size: 52, lr: 4.61e-03, grad_scale: 8.0 2023-04-02 11:46:13,139 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=118287.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 11:46:37,312 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=118307.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 11:46:39,291 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.599e+02 5.366e+02 7.280e+02 9.450e+02 2.114e+03, threshold=1.456e+03, percent-clipped=6.0 2023-04-02 11:47:01,200 INFO [train.py:903] (1/4) Epoch 18, batch 2250, loss[loss=0.2033, simple_loss=0.2834, pruned_loss=0.06156, over 19746.00 frames. ], tot_loss[loss=0.2124, simple_loss=0.291, pruned_loss=0.06692, over 3824006.85 frames. ], batch size: 54, lr: 4.61e-03, grad_scale: 8.0 2023-04-02 11:47:27,302 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=118347.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 11:47:47,838 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.1630, 3.7994, 2.6555, 3.4430, 1.1204, 3.7065, 3.6711, 3.7047], device='cuda:1'), covar=tensor([0.0614, 0.1013, 0.1740, 0.0767, 0.3660, 0.0727, 0.0826, 0.1169], device='cuda:1'), in_proj_covar=tensor([0.0478, 0.0395, 0.0479, 0.0340, 0.0399, 0.0416, 0.0411, 0.0441], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 11:48:02,745 INFO [train.py:903] (1/4) Epoch 18, batch 2300, loss[loss=0.1873, simple_loss=0.2752, pruned_loss=0.04964, over 19157.00 frames. ], tot_loss[loss=0.2142, simple_loss=0.2924, pruned_loss=0.06801, over 3803442.33 frames. ], batch size: 69, lr: 4.61e-03, grad_scale: 8.0 2023-04-02 11:48:15,073 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_rvb from training. Duration: 26.205 2023-04-02 11:48:22,236 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.4488, 1.2660, 1.4774, 1.4599, 3.0308, 1.1700, 2.3076, 3.3105], device='cuda:1'), covar=tensor([0.0510, 0.2736, 0.2666, 0.1844, 0.0701, 0.2338, 0.1139, 0.0310], device='cuda:1'), in_proj_covar=tensor([0.0398, 0.0359, 0.0377, 0.0344, 0.0367, 0.0347, 0.0367, 0.0388], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 11:48:35,800 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=118402.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 11:48:44,941 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.375e+02 5.253e+02 6.243e+02 7.561e+02 1.558e+03, threshold=1.249e+03, percent-clipped=2.0 2023-04-02 11:49:05,630 INFO [train.py:903] (1/4) Epoch 18, batch 2350, loss[loss=0.2062, simple_loss=0.2868, pruned_loss=0.06285, over 18276.00 frames. ], tot_loss[loss=0.2119, simple_loss=0.2904, pruned_loss=0.06667, over 3816352.99 frames. ], batch size: 84, lr: 4.61e-03, grad_scale: 8.0 2023-04-02 11:49:12,803 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2657, 2.0094, 1.5490, 1.2907, 1.8397, 1.2367, 1.2517, 1.7300], device='cuda:1'), covar=tensor([0.0947, 0.0821, 0.1059, 0.0871, 0.0519, 0.1215, 0.0713, 0.0460], device='cuda:1'), in_proj_covar=tensor([0.0302, 0.0314, 0.0336, 0.0262, 0.0247, 0.0334, 0.0293, 0.0272], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 11:49:46,252 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_rvb from training. Duration: 25.775 2023-04-02 11:50:02,122 WARNING [train.py:1073] (1/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9_rvb from training. Duration: 25.45 2023-04-02 11:50:06,555 INFO [train.py:903] (1/4) Epoch 18, batch 2400, loss[loss=0.2188, simple_loss=0.3015, pruned_loss=0.06807, over 18754.00 frames. ], tot_loss[loss=0.2143, simple_loss=0.2925, pruned_loss=0.06803, over 3811930.27 frames. ], batch size: 74, lr: 4.61e-03, grad_scale: 8.0 2023-04-02 11:50:48,691 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.320e+02 4.821e+02 5.593e+02 7.730e+02 1.750e+03, threshold=1.119e+03, percent-clipped=3.0 2023-04-02 11:50:58,551 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=118517.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 11:51:10,678 INFO [train.py:903] (1/4) Epoch 18, batch 2450, loss[loss=0.1945, simple_loss=0.2812, pruned_loss=0.0539, over 19539.00 frames. ], tot_loss[loss=0.2145, simple_loss=0.2929, pruned_loss=0.06805, over 3803912.27 frames. ], batch size: 54, lr: 4.61e-03, grad_scale: 8.0 2023-04-02 11:51:16,268 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=118530.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 11:51:57,137 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.0559, 1.4295, 1.7672, 1.1887, 2.7698, 3.7565, 3.4835, 3.9729], device='cuda:1'), covar=tensor([0.1708, 0.3566, 0.3232, 0.2446, 0.0560, 0.0160, 0.0201, 0.0211], device='cuda:1'), in_proj_covar=tensor([0.0263, 0.0310, 0.0340, 0.0257, 0.0232, 0.0178, 0.0210, 0.0237], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-02 11:52:12,595 INFO [train.py:903] (1/4) Epoch 18, batch 2500, loss[loss=0.2167, simple_loss=0.2964, pruned_loss=0.06854, over 19668.00 frames. ], tot_loss[loss=0.2154, simple_loss=0.294, pruned_loss=0.06844, over 3798096.12 frames. ], batch size: 60, lr: 4.61e-03, grad_scale: 8.0 2023-04-02 11:52:53,549 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.320e+02 5.081e+02 6.040e+02 7.266e+02 1.380e+03, threshold=1.208e+03, percent-clipped=1.0 2023-04-02 11:53:13,674 INFO [train.py:903] (1/4) Epoch 18, batch 2550, loss[loss=0.2561, simple_loss=0.3287, pruned_loss=0.09173, over 17209.00 frames. ], tot_loss[loss=0.2165, simple_loss=0.2951, pruned_loss=0.06898, over 3801458.25 frames. ], batch size: 101, lr: 4.61e-03, grad_scale: 8.0 2023-04-02 11:53:19,831 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=118631.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 11:53:37,325 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.51 vs. limit=5.0 2023-04-02 11:53:37,945 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=118645.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 11:53:42,612 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=118649.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 11:53:45,539 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=118651.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 11:54:06,500 WARNING [train.py:1073] (1/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 2023-04-02 11:54:15,893 INFO [train.py:903] (1/4) Epoch 18, batch 2600, loss[loss=0.2155, simple_loss=0.2959, pruned_loss=0.06755, over 17965.00 frames. ], tot_loss[loss=0.2158, simple_loss=0.2943, pruned_loss=0.06861, over 3796214.97 frames. ], batch size: 83, lr: 4.61e-03, grad_scale: 8.0 2023-04-02 11:54:34,614 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=118691.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 11:54:56,096 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.395e+02 4.742e+02 5.749e+02 7.114e+02 1.367e+03, threshold=1.150e+03, percent-clipped=3.0 2023-04-02 11:55:16,582 INFO [train.py:903] (1/4) Epoch 18, batch 2650, loss[loss=0.1977, simple_loss=0.2748, pruned_loss=0.06028, over 19397.00 frames. ], tot_loss[loss=0.2149, simple_loss=0.2934, pruned_loss=0.06813, over 3803552.31 frames. ], batch size: 48, lr: 4.61e-03, grad_scale: 8.0 2023-04-02 11:55:33,760 WARNING [train.py:1073] (1/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9_rvb from training. Duration: 27.25 2023-04-02 11:55:42,221 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=118746.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 11:56:06,194 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=118766.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 11:56:15,279 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=118773.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 11:56:18,250 INFO [train.py:903] (1/4) Epoch 18, batch 2700, loss[loss=0.1857, simple_loss=0.27, pruned_loss=0.05065, over 19592.00 frames. ], tot_loss[loss=0.2153, simple_loss=0.294, pruned_loss=0.06826, over 3810340.16 frames. ], batch size: 52, lr: 4.60e-03, grad_scale: 8.0 2023-04-02 11:56:44,783 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=118798.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 11:56:49,080 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.79 vs. limit=5.0 2023-04-02 11:56:55,766 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=118806.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 11:56:59,040 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.246e+02 5.095e+02 6.154e+02 8.195e+02 1.746e+03, threshold=1.231e+03, percent-clipped=5.0 2023-04-02 11:57:04,047 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([6.1519, 5.4913, 3.3733, 4.9042, 0.8275, 5.6591, 5.5177, 5.6922], device='cuda:1'), covar=tensor([0.0376, 0.0878, 0.1590, 0.0665, 0.4475, 0.0492, 0.0727, 0.0951], device='cuda:1'), in_proj_covar=tensor([0.0483, 0.0394, 0.0480, 0.0341, 0.0402, 0.0419, 0.0413, 0.0445], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 11:57:06,562 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.2552, 2.2024, 1.7339, 2.0937, 2.2747, 1.5848, 1.7983, 2.1267], device='cuda:1'), covar=tensor([0.1061, 0.1685, 0.1887, 0.1426, 0.1565, 0.1098, 0.1766, 0.0938], device='cuda:1'), in_proj_covar=tensor([0.0263, 0.0353, 0.0299, 0.0248, 0.0298, 0.0246, 0.0295, 0.0250], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 11:57:20,472 INFO [train.py:903] (1/4) Epoch 18, batch 2750, loss[loss=0.1954, simple_loss=0.2655, pruned_loss=0.06264, over 19782.00 frames. ], tot_loss[loss=0.2144, simple_loss=0.2933, pruned_loss=0.06771, over 3815953.27 frames. ], batch size: 47, lr: 4.60e-03, grad_scale: 8.0 2023-04-02 11:58:15,376 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2652, 1.2851, 1.5438, 1.4124, 2.3254, 2.0202, 2.4765, 1.0367], device='cuda:1'), covar=tensor([0.2657, 0.4421, 0.2779, 0.2170, 0.1607, 0.2292, 0.1423, 0.4412], device='cuda:1'), in_proj_covar=tensor([0.0519, 0.0622, 0.0679, 0.0468, 0.0615, 0.0521, 0.0655, 0.0530], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 11:58:23,109 INFO [train.py:903] (1/4) Epoch 18, batch 2800, loss[loss=0.2124, simple_loss=0.2936, pruned_loss=0.06557, over 17301.00 frames. ], tot_loss[loss=0.2144, simple_loss=0.2933, pruned_loss=0.06772, over 3826965.64 frames. ], batch size: 101, lr: 4.60e-03, grad_scale: 8.0 2023-04-02 11:58:41,695 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9660, 1.8946, 1.7775, 1.5558, 1.4463, 1.5971, 0.3488, 0.8519], device='cuda:1'), covar=tensor([0.0506, 0.0543, 0.0372, 0.0612, 0.1047, 0.0663, 0.1124, 0.0941], device='cuda:1'), in_proj_covar=tensor([0.0349, 0.0348, 0.0347, 0.0374, 0.0448, 0.0380, 0.0329, 0.0335], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 11:58:43,966 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.9558, 2.0314, 2.0923, 2.0299, 3.6783, 1.6337, 2.9686, 3.7286], device='cuda:1'), covar=tensor([0.0517, 0.2134, 0.2229, 0.1633, 0.0601, 0.2224, 0.1386, 0.0266], device='cuda:1'), in_proj_covar=tensor([0.0399, 0.0357, 0.0378, 0.0342, 0.0365, 0.0345, 0.0366, 0.0389], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 11:58:54,293 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=118901.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 11:59:03,985 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.182e+02 4.804e+02 6.148e+02 8.494e+02 1.418e+03, threshold=1.230e+03, percent-clipped=5.0 2023-04-02 11:59:24,485 INFO [train.py:903] (1/4) Epoch 18, batch 2850, loss[loss=0.2214, simple_loss=0.3087, pruned_loss=0.06701, over 19399.00 frames. ], tot_loss[loss=0.2139, simple_loss=0.2926, pruned_loss=0.06755, over 3834068.38 frames. ], batch size: 70, lr: 4.60e-03, grad_scale: 8.0 2023-04-02 11:59:24,892 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=118926.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 12:00:22,869 WARNING [train.py:1073] (1/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 2023-04-02 12:00:26,009 INFO [train.py:903] (1/4) Epoch 18, batch 2900, loss[loss=0.2056, simple_loss=0.293, pruned_loss=0.05911, over 19671.00 frames. ], tot_loss[loss=0.2129, simple_loss=0.2915, pruned_loss=0.06718, over 3824783.79 frames. ], batch size: 60, lr: 4.60e-03, grad_scale: 8.0 2023-04-02 12:00:27,563 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0902, 1.9920, 1.8447, 1.7011, 1.5878, 1.6943, 0.4470, 1.0148], device='cuda:1'), covar=tensor([0.0536, 0.0578, 0.0411, 0.0671, 0.1065, 0.0831, 0.1258, 0.0953], device='cuda:1'), in_proj_covar=tensor([0.0351, 0.0349, 0.0348, 0.0375, 0.0449, 0.0381, 0.0330, 0.0337], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 12:00:46,052 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=118993.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 12:00:54,378 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=2.00 vs. limit=2.0 2023-04-02 12:00:54,939 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3944, 1.4617, 1.8140, 1.7787, 2.4938, 2.2072, 2.7086, 1.2336], device='cuda:1'), covar=tensor([0.2670, 0.4565, 0.2849, 0.1990, 0.1899, 0.2379, 0.1719, 0.4641], device='cuda:1'), in_proj_covar=tensor([0.0520, 0.0625, 0.0682, 0.0469, 0.0618, 0.0524, 0.0659, 0.0533], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 12:00:56,049 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=119002.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 12:00:58,541 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 2023-04-02 12:01:05,216 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.034e+02 5.127e+02 5.908e+02 8.371e+02 2.467e+03, threshold=1.182e+03, percent-clipped=10.0 2023-04-02 12:01:21,050 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=119022.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 12:01:25,122 INFO [train.py:903] (1/4) Epoch 18, batch 2950, loss[loss=0.211, simple_loss=0.2771, pruned_loss=0.07243, over 19771.00 frames. ], tot_loss[loss=0.214, simple_loss=0.2925, pruned_loss=0.06779, over 3832637.91 frames. ], batch size: 46, lr: 4.60e-03, grad_scale: 8.0 2023-04-02 12:01:26,629 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=119027.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 12:01:50,457 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=119047.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 12:02:08,479 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=119062.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 12:02:24,526 INFO [train.py:903] (1/4) Epoch 18, batch 3000, loss[loss=0.2112, simple_loss=0.2901, pruned_loss=0.0661, over 19756.00 frames. ], tot_loss[loss=0.2136, simple_loss=0.2918, pruned_loss=0.06771, over 3838852.45 frames. ], batch size: 51, lr: 4.60e-03, grad_scale: 8.0 2023-04-02 12:02:24,527 INFO [train.py:928] (1/4) Computing validation loss 2023-04-02 12:02:37,004 INFO [train.py:937] (1/4) Epoch 18, validation: loss=0.1707, simple_loss=0.2711, pruned_loss=0.03521, over 944034.00 frames. 2023-04-02 12:02:37,005 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18821MB 2023-04-02 12:02:37,344 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.9015, 1.5701, 1.7727, 1.6555, 4.4296, 1.0377, 2.5011, 4.7953], device='cuda:1'), covar=tensor([0.0420, 0.2735, 0.2878, 0.1970, 0.0686, 0.2769, 0.1511, 0.0171], device='cuda:1'), in_proj_covar=tensor([0.0398, 0.0359, 0.0379, 0.0343, 0.0366, 0.0347, 0.0367, 0.0390], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 12:02:40,533 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_rvb from training. Duration: 29.735 2023-04-02 12:02:50,762 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=119087.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 12:02:57,635 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=119093.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 12:02:58,893 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6331, 1.6843, 1.4956, 1.2517, 1.1513, 1.3160, 0.2931, 0.5838], device='cuda:1'), covar=tensor([0.0810, 0.0718, 0.0481, 0.0713, 0.1468, 0.0870, 0.1281, 0.1234], device='cuda:1'), in_proj_covar=tensor([0.0351, 0.0348, 0.0349, 0.0374, 0.0449, 0.0381, 0.0329, 0.0338], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 12:03:17,159 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=119108.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 12:03:17,950 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.980e+02 5.504e+02 6.538e+02 8.295e+02 4.074e+03, threshold=1.308e+03, percent-clipped=8.0 2023-04-02 12:03:20,399 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=119111.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 12:03:37,844 INFO [train.py:903] (1/4) Epoch 18, batch 3050, loss[loss=0.2227, simple_loss=0.3067, pruned_loss=0.06932, over 19666.00 frames. ], tot_loss[loss=0.2146, simple_loss=0.293, pruned_loss=0.06814, over 3842579.60 frames. ], batch size: 60, lr: 4.60e-03, grad_scale: 8.0 2023-04-02 12:04:37,424 INFO [train.py:903] (1/4) Epoch 18, batch 3100, loss[loss=0.1849, simple_loss=0.2575, pruned_loss=0.05612, over 19767.00 frames. ], tot_loss[loss=0.2134, simple_loss=0.292, pruned_loss=0.06745, over 3838697.40 frames. ], batch size: 45, lr: 4.60e-03, grad_scale: 8.0 2023-04-02 12:05:18,221 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.373e+02 4.828e+02 5.979e+02 7.400e+02 1.693e+03, threshold=1.196e+03, percent-clipped=2.0 2023-04-02 12:05:39,400 INFO [train.py:903] (1/4) Epoch 18, batch 3150, loss[loss=0.2071, simple_loss=0.2933, pruned_loss=0.06041, over 19737.00 frames. ], tot_loss[loss=0.2133, simple_loss=0.2917, pruned_loss=0.06739, over 3842904.32 frames. ], batch size: 63, lr: 4.60e-03, grad_scale: 8.0 2023-04-02 12:05:57,630 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5054, 1.5809, 1.7936, 1.7373, 2.7148, 2.3173, 2.8849, 1.4435], device='cuda:1'), covar=tensor([0.2305, 0.3931, 0.2490, 0.1797, 0.1362, 0.1987, 0.1275, 0.3812], device='cuda:1'), in_proj_covar=tensor([0.0517, 0.0619, 0.0676, 0.0466, 0.0611, 0.0517, 0.0652, 0.0528], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 12:06:06,969 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 2023-04-02 12:06:35,321 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5807, 1.4818, 1.5057, 1.9938, 1.7079, 1.8401, 1.8572, 1.6993], device='cuda:1'), covar=tensor([0.0792, 0.0897, 0.0916, 0.0663, 0.0784, 0.0716, 0.0796, 0.0633], device='cuda:1'), in_proj_covar=tensor([0.0213, 0.0224, 0.0226, 0.0245, 0.0228, 0.0211, 0.0190, 0.0207], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-04-02 12:06:39,658 INFO [train.py:903] (1/4) Epoch 18, batch 3200, loss[loss=0.2426, simple_loss=0.3198, pruned_loss=0.08266, over 19467.00 frames. ], tot_loss[loss=0.2141, simple_loss=0.2925, pruned_loss=0.06782, over 3842669.77 frames. ], batch size: 64, lr: 4.60e-03, grad_scale: 8.0 2023-04-02 12:06:45,753 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9413, 1.8312, 1.5723, 1.8993, 1.8198, 1.5133, 1.4009, 1.7981], device='cuda:1'), covar=tensor([0.1065, 0.1463, 0.1614, 0.1195, 0.1422, 0.0762, 0.1713, 0.0834], device='cuda:1'), in_proj_covar=tensor([0.0265, 0.0355, 0.0303, 0.0250, 0.0301, 0.0248, 0.0298, 0.0251], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 12:07:18,202 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.241e+02 4.710e+02 5.964e+02 8.137e+02 2.705e+03, threshold=1.193e+03, percent-clipped=4.0 2023-04-02 12:07:39,012 INFO [train.py:903] (1/4) Epoch 18, batch 3250, loss[loss=0.2187, simple_loss=0.3014, pruned_loss=0.06802, over 19664.00 frames. ], tot_loss[loss=0.2131, simple_loss=0.2919, pruned_loss=0.06714, over 3843722.08 frames. ], batch size: 58, lr: 4.59e-03, grad_scale: 8.0 2023-04-02 12:08:24,458 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=119364.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 12:08:37,659 INFO [train.py:903] (1/4) Epoch 18, batch 3300, loss[loss=0.2192, simple_loss=0.2966, pruned_loss=0.07093, over 19581.00 frames. ], tot_loss[loss=0.2139, simple_loss=0.2925, pruned_loss=0.06763, over 3838765.43 frames. ], batch size: 61, lr: 4.59e-03, grad_scale: 8.0 2023-04-02 12:08:42,297 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 2023-04-02 12:08:53,604 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=119389.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 12:09:16,508 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([5.9237, 5.3490, 3.1101, 4.6664, 1.2174, 5.4744, 5.2789, 5.5212], device='cuda:1'), covar=tensor([0.0388, 0.0869, 0.1733, 0.0746, 0.4068, 0.0504, 0.0784, 0.0995], device='cuda:1'), in_proj_covar=tensor([0.0480, 0.0393, 0.0478, 0.0340, 0.0397, 0.0415, 0.0410, 0.0442], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 12:09:17,496 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.159e+02 5.369e+02 6.623e+02 8.148e+02 1.799e+03, threshold=1.325e+03, percent-clipped=5.0 2023-04-02 12:09:37,823 INFO [train.py:903] (1/4) Epoch 18, batch 3350, loss[loss=0.2057, simple_loss=0.2902, pruned_loss=0.06062, over 19672.00 frames. ], tot_loss[loss=0.2152, simple_loss=0.2933, pruned_loss=0.06851, over 3824291.36 frames. ], batch size: 59, lr: 4.59e-03, grad_scale: 8.0 2023-04-02 12:09:50,642 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=119437.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 12:10:11,914 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=119455.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 12:10:37,461 INFO [train.py:903] (1/4) Epoch 18, batch 3400, loss[loss=0.1697, simple_loss=0.2498, pruned_loss=0.04483, over 17648.00 frames. ], tot_loss[loss=0.2164, simple_loss=0.2942, pruned_loss=0.06936, over 3827370.80 frames. ], batch size: 39, lr: 4.59e-03, grad_scale: 4.0 2023-04-02 12:11:18,511 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.251e+02 4.894e+02 5.911e+02 7.983e+02 1.559e+03, threshold=1.182e+03, percent-clipped=2.0 2023-04-02 12:11:37,555 INFO [train.py:903] (1/4) Epoch 18, batch 3450, loss[loss=0.1941, simple_loss=0.28, pruned_loss=0.05412, over 19522.00 frames. ], tot_loss[loss=0.2158, simple_loss=0.2938, pruned_loss=0.06896, over 3829617.16 frames. ], batch size: 54, lr: 4.59e-03, grad_scale: 4.0 2023-04-02 12:11:43,120 WARNING [train.py:1073] (1/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9_rvb from training. Duration: 25.2444375 2023-04-02 12:11:45,616 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=119532.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 12:12:09,596 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=119552.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 12:12:31,765 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=119570.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 12:12:38,201 INFO [train.py:903] (1/4) Epoch 18, batch 3500, loss[loss=0.2005, simple_loss=0.2734, pruned_loss=0.06378, over 19683.00 frames. ], tot_loss[loss=0.2141, simple_loss=0.2919, pruned_loss=0.06818, over 3826218.57 frames. ], batch size: 53, lr: 4.59e-03, grad_scale: 4.0 2023-04-02 12:13:20,028 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.151e+02 4.884e+02 6.516e+02 7.989e+02 1.670e+03, threshold=1.303e+03, percent-clipped=4.0 2023-04-02 12:13:39,059 INFO [train.py:903] (1/4) Epoch 18, batch 3550, loss[loss=0.2017, simple_loss=0.2745, pruned_loss=0.06443, over 19373.00 frames. ], tot_loss[loss=0.214, simple_loss=0.2921, pruned_loss=0.06796, over 3812862.64 frames. ], batch size: 47, lr: 4.59e-03, grad_scale: 4.0 2023-04-02 12:14:39,531 INFO [train.py:903] (1/4) Epoch 18, batch 3600, loss[loss=0.2246, simple_loss=0.3018, pruned_loss=0.07367, over 19528.00 frames. ], tot_loss[loss=0.2143, simple_loss=0.2927, pruned_loss=0.06793, over 3810212.08 frames. ], batch size: 54, lr: 4.59e-03, grad_scale: 8.0 2023-04-02 12:14:47,916 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=119683.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 12:15:20,525 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.297e+02 4.936e+02 6.005e+02 7.350e+02 2.220e+03, threshold=1.201e+03, percent-clipped=3.0 2023-04-02 12:15:39,372 INFO [train.py:903] (1/4) Epoch 18, batch 3650, loss[loss=0.2456, simple_loss=0.3265, pruned_loss=0.08229, over 19774.00 frames. ], tot_loss[loss=0.2161, simple_loss=0.2944, pruned_loss=0.06892, over 3818001.10 frames. ], batch size: 56, lr: 4.59e-03, grad_scale: 8.0 2023-04-02 12:16:14,039 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.49 vs. limit=2.0 2023-04-02 12:16:40,015 INFO [train.py:903] (1/4) Epoch 18, batch 3700, loss[loss=0.2218, simple_loss=0.3038, pruned_loss=0.06996, over 19473.00 frames. ], tot_loss[loss=0.2162, simple_loss=0.2944, pruned_loss=0.06899, over 3816914.86 frames. ], batch size: 49, lr: 4.59e-03, grad_scale: 8.0 2023-04-02 12:17:13,322 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0366, 3.5081, 1.9782, 2.0980, 3.0790, 1.7053, 1.5666, 2.1896], device='cuda:1'), covar=tensor([0.1343, 0.0543, 0.1012, 0.0812, 0.0517, 0.1172, 0.0938, 0.0657], device='cuda:1'), in_proj_covar=tensor([0.0297, 0.0313, 0.0330, 0.0259, 0.0247, 0.0330, 0.0291, 0.0274], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 12:17:15,500 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4262, 1.6065, 1.9704, 1.7391, 3.2511, 2.6753, 3.5006, 1.5972], device='cuda:1'), covar=tensor([0.2453, 0.4175, 0.2692, 0.1838, 0.1400, 0.1972, 0.1484, 0.4067], device='cuda:1'), in_proj_covar=tensor([0.0521, 0.0621, 0.0682, 0.0470, 0.0613, 0.0521, 0.0656, 0.0532], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 12:17:19,627 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=119808.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 12:17:21,520 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.455e+02 5.017e+02 5.995e+02 7.636e+02 1.424e+03, threshold=1.199e+03, percent-clipped=3.0 2023-04-02 12:17:40,474 INFO [train.py:903] (1/4) Epoch 18, batch 3750, loss[loss=0.1855, simple_loss=0.2632, pruned_loss=0.05389, over 19780.00 frames. ], tot_loss[loss=0.2154, simple_loss=0.2938, pruned_loss=0.06851, over 3832390.43 frames. ], batch size: 47, lr: 4.58e-03, grad_scale: 8.0 2023-04-02 12:17:40,887 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=119826.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 12:17:48,789 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=119833.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 12:18:10,611 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=119851.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 12:18:39,735 INFO [train.py:903] (1/4) Epoch 18, batch 3800, loss[loss=0.2165, simple_loss=0.2892, pruned_loss=0.07193, over 19746.00 frames. ], tot_loss[loss=0.2155, simple_loss=0.2937, pruned_loss=0.06862, over 3838315.94 frames. ], batch size: 51, lr: 4.58e-03, grad_scale: 8.0 2023-04-02 12:18:39,893 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=119876.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 12:18:44,344 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6327, 1.4500, 1.4874, 2.2379, 1.7649, 1.9718, 2.0517, 1.9044], device='cuda:1'), covar=tensor([0.0836, 0.0926, 0.0981, 0.0701, 0.0783, 0.0727, 0.0823, 0.0627], device='cuda:1'), in_proj_covar=tensor([0.0210, 0.0221, 0.0223, 0.0242, 0.0226, 0.0208, 0.0187, 0.0204], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-02 12:19:12,057 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9_rvb from training. Duration: 29.1166875 2023-04-02 12:19:16,277 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 2023-04-02 12:19:22,039 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.281e+02 5.499e+02 6.857e+02 8.596e+02 2.059e+03, threshold=1.371e+03, percent-clipped=8.0 2023-04-02 12:19:41,466 INFO [train.py:903] (1/4) Epoch 18, batch 3850, loss[loss=0.25, simple_loss=0.3215, pruned_loss=0.0892, over 13499.00 frames. ], tot_loss[loss=0.2149, simple_loss=0.2936, pruned_loss=0.06814, over 3833695.65 frames. ], batch size: 136, lr: 4.58e-03, grad_scale: 8.0 2023-04-02 12:19:45,479 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0884, 2.0253, 1.8053, 1.6440, 1.4997, 1.7037, 0.4868, 1.1200], device='cuda:1'), covar=tensor([0.0548, 0.0550, 0.0425, 0.0768, 0.1085, 0.0836, 0.1205, 0.0941], device='cuda:1'), in_proj_covar=tensor([0.0347, 0.0345, 0.0345, 0.0372, 0.0446, 0.0378, 0.0325, 0.0333], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 12:20:43,324 INFO [train.py:903] (1/4) Epoch 18, batch 3900, loss[loss=0.2137, simple_loss=0.2988, pruned_loss=0.0643, over 19510.00 frames. ], tot_loss[loss=0.2144, simple_loss=0.2928, pruned_loss=0.06799, over 3817539.88 frames. ], batch size: 54, lr: 4.58e-03, grad_scale: 8.0 2023-04-02 12:20:52,085 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 2023-04-02 12:21:02,810 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=119991.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 12:21:26,389 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.199e+02 4.757e+02 5.722e+02 7.251e+02 1.425e+03, threshold=1.144e+03, percent-clipped=2.0 2023-04-02 12:21:45,629 INFO [train.py:903] (1/4) Epoch 18, batch 3950, loss[loss=0.2163, simple_loss=0.2996, pruned_loss=0.06648, over 18093.00 frames. ], tot_loss[loss=0.2145, simple_loss=0.2931, pruned_loss=0.06796, over 3812378.76 frames. ], batch size: 83, lr: 4.58e-03, grad_scale: 8.0 2023-04-02 12:21:47,756 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=120027.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 12:21:49,995 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 2023-04-02 12:22:47,956 INFO [train.py:903] (1/4) Epoch 18, batch 4000, loss[loss=0.1902, simple_loss=0.2658, pruned_loss=0.05728, over 19383.00 frames. ], tot_loss[loss=0.2133, simple_loss=0.2917, pruned_loss=0.06745, over 3811207.92 frames. ], batch size: 47, lr: 4.58e-03, grad_scale: 8.0 2023-04-02 12:23:02,777 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2070, 1.2535, 1.3458, 1.3440, 1.6807, 1.7481, 1.7801, 0.5828], device='cuda:1'), covar=tensor([0.2451, 0.4094, 0.2570, 0.1951, 0.1620, 0.2216, 0.1315, 0.4466], device='cuda:1'), in_proj_covar=tensor([0.0517, 0.0619, 0.0676, 0.0467, 0.0611, 0.0519, 0.0653, 0.0530], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 12:23:29,303 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.243e+02 4.630e+02 5.764e+02 7.444e+02 1.534e+03, threshold=1.153e+03, percent-clipped=2.0 2023-04-02 12:23:36,025 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1_rvb from training. Duration: 0.7545625 2023-04-02 12:23:49,495 INFO [train.py:903] (1/4) Epoch 18, batch 4050, loss[loss=0.2124, simple_loss=0.2936, pruned_loss=0.06557, over 19672.00 frames. ], tot_loss[loss=0.2135, simple_loss=0.2919, pruned_loss=0.06751, over 3824212.75 frames. ], batch size: 60, lr: 4.58e-03, grad_scale: 8.0 2023-04-02 12:24:08,098 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=120142.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 12:24:28,096 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=4.01 vs. limit=5.0 2023-04-02 12:24:40,869 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.6502, 3.9379, 2.6749, 2.7115, 3.6190, 2.4454, 2.1113, 2.7769], device='cuda:1'), covar=tensor([0.1135, 0.0467, 0.0835, 0.0674, 0.0374, 0.0964, 0.0800, 0.0590], device='cuda:1'), in_proj_covar=tensor([0.0296, 0.0312, 0.0330, 0.0258, 0.0247, 0.0329, 0.0290, 0.0272], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 12:24:49,138 INFO [train.py:903] (1/4) Epoch 18, batch 4100, loss[loss=0.2019, simple_loss=0.2804, pruned_loss=0.06168, over 19831.00 frames. ], tot_loss[loss=0.2139, simple_loss=0.2925, pruned_loss=0.06762, over 3824859.05 frames. ], batch size: 52, lr: 4.58e-03, grad_scale: 8.0 2023-04-02 12:25:20,530 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([5.4630, 5.0234, 3.2653, 4.4049, 1.6521, 4.9063, 4.8073, 4.9883], device='cuda:1'), covar=tensor([0.0422, 0.0830, 0.1535, 0.0745, 0.3315, 0.0565, 0.0766, 0.0907], device='cuda:1'), in_proj_covar=tensor([0.0483, 0.0394, 0.0480, 0.0340, 0.0395, 0.0417, 0.0409, 0.0443], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 12:25:24,886 WARNING [train.py:1073] (1/4) Exclude cut with ID 497-129325-0061-9566_sp1.1_rvb from training. Duration: 0.97725 2023-04-02 12:25:29,250 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.453e+02 5.087e+02 6.190e+02 7.903e+02 1.294e+03, threshold=1.238e+03, percent-clipped=4.0 2023-04-02 12:25:39,648 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3701, 2.0668, 1.5833, 1.3641, 1.8454, 1.3081, 1.3864, 1.8328], device='cuda:1'), covar=tensor([0.0882, 0.0781, 0.1092, 0.0833, 0.0586, 0.1210, 0.0613, 0.0431], device='cuda:1'), in_proj_covar=tensor([0.0295, 0.0311, 0.0329, 0.0258, 0.0246, 0.0328, 0.0289, 0.0272], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 12:25:48,078 INFO [train.py:903] (1/4) Epoch 18, batch 4150, loss[loss=0.2321, simple_loss=0.3123, pruned_loss=0.07599, over 19613.00 frames. ], tot_loss[loss=0.2135, simple_loss=0.2921, pruned_loss=0.06746, over 3825303.88 frames. ], batch size: 57, lr: 4.58e-03, grad_scale: 8.0 2023-04-02 12:26:14,740 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=120247.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 12:26:46,392 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=120272.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 12:26:50,727 INFO [train.py:903] (1/4) Epoch 18, batch 4200, loss[loss=0.2184, simple_loss=0.2979, pruned_loss=0.06942, over 19588.00 frames. ], tot_loss[loss=0.2127, simple_loss=0.2917, pruned_loss=0.06689, over 3810556.88 frames. ], batch size: 52, lr: 4.58e-03, grad_scale: 8.0 2023-04-02 12:26:57,165 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 2023-04-02 12:27:30,830 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.475e+02 4.968e+02 6.385e+02 7.954e+02 1.571e+03, threshold=1.277e+03, percent-clipped=3.0 2023-04-02 12:27:51,828 INFO [train.py:903] (1/4) Epoch 18, batch 4250, loss[loss=0.2414, simple_loss=0.318, pruned_loss=0.08243, over 19537.00 frames. ], tot_loss[loss=0.2129, simple_loss=0.2916, pruned_loss=0.06713, over 3818919.95 frames. ], batch size: 54, lr: 4.57e-03, grad_scale: 8.0 2023-04-02 12:27:56,606 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([5.0930, 5.0949, 5.8735, 5.8906, 1.8928, 5.5839, 4.6092, 5.5384], device='cuda:1'), covar=tensor([0.1540, 0.0847, 0.0487, 0.0600, 0.5952, 0.0684, 0.0577, 0.1053], device='cuda:1'), in_proj_covar=tensor([0.0760, 0.0701, 0.0907, 0.0794, 0.0809, 0.0659, 0.0548, 0.0837], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-02 12:28:09,637 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9_rvb from training. Duration: 29.816625 2023-04-02 12:28:18,891 WARNING [train.py:1073] (1/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 2023-04-02 12:28:47,403 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4804, 2.1668, 1.6406, 1.4134, 2.0142, 1.3451, 1.3602, 1.8819], device='cuda:1'), covar=tensor([0.0991, 0.0852, 0.1019, 0.0849, 0.0541, 0.1225, 0.0698, 0.0462], device='cuda:1'), in_proj_covar=tensor([0.0291, 0.0309, 0.0326, 0.0256, 0.0243, 0.0325, 0.0287, 0.0268], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 12:28:51,543 INFO [train.py:903] (1/4) Epoch 18, batch 4300, loss[loss=0.2037, simple_loss=0.2971, pruned_loss=0.05514, over 19770.00 frames. ], tot_loss[loss=0.2126, simple_loss=0.2912, pruned_loss=0.06699, over 3818134.44 frames. ], batch size: 54, lr: 4.57e-03, grad_scale: 8.0 2023-04-02 12:29:19,817 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=120398.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 12:29:34,133 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.947e+02 5.084e+02 5.969e+02 7.631e+02 1.294e+03, threshold=1.194e+03, percent-clipped=1.0 2023-04-02 12:29:46,145 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 2023-04-02 12:29:49,841 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=120423.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 12:29:52,875 INFO [train.py:903] (1/4) Epoch 18, batch 4350, loss[loss=0.2139, simple_loss=0.2965, pruned_loss=0.06567, over 19678.00 frames. ], tot_loss[loss=0.2143, simple_loss=0.2925, pruned_loss=0.06803, over 3782851.59 frames. ], batch size: 59, lr: 4.57e-03, grad_scale: 8.0 2023-04-02 12:29:57,423 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.5950, 3.9261, 4.3054, 4.3636, 1.6585, 4.0685, 3.4482, 3.7115], device='cuda:1'), covar=tensor([0.2749, 0.1597, 0.1033, 0.1492, 0.7904, 0.2055, 0.1340, 0.2252], device='cuda:1'), in_proj_covar=tensor([0.0753, 0.0697, 0.0901, 0.0786, 0.0801, 0.0654, 0.0543, 0.0832], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-02 12:30:55,181 INFO [train.py:903] (1/4) Epoch 18, batch 4400, loss[loss=0.2165, simple_loss=0.2999, pruned_loss=0.06654, over 19595.00 frames. ], tot_loss[loss=0.2144, simple_loss=0.2925, pruned_loss=0.06811, over 3793421.52 frames. ], batch size: 52, lr: 4.57e-03, grad_scale: 8.0 2023-04-02 12:31:18,742 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_rvb from training. Duration: 25.285 2023-04-02 12:31:28,328 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 2023-04-02 12:31:29,868 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3294, 1.3477, 1.4271, 1.4892, 1.7010, 1.8471, 1.7560, 0.5350], device='cuda:1'), covar=tensor([0.2322, 0.4113, 0.2585, 0.1849, 0.1662, 0.2218, 0.1369, 0.4565], device='cuda:1'), in_proj_covar=tensor([0.0516, 0.0616, 0.0675, 0.0466, 0.0609, 0.0517, 0.0649, 0.0528], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 12:31:35,103 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.222e+02 5.092e+02 6.306e+02 8.223e+02 1.474e+03, threshold=1.261e+03, percent-clipped=7.0 2023-04-02 12:31:55,630 INFO [train.py:903] (1/4) Epoch 18, batch 4450, loss[loss=0.2676, simple_loss=0.3293, pruned_loss=0.103, over 13508.00 frames. ], tot_loss[loss=0.2134, simple_loss=0.2915, pruned_loss=0.06762, over 3793849.00 frames. ], batch size: 136, lr: 4.57e-03, grad_scale: 8.0 2023-04-02 12:32:55,856 INFO [train.py:903] (1/4) Epoch 18, batch 4500, loss[loss=0.2518, simple_loss=0.3363, pruned_loss=0.08364, over 19507.00 frames. ], tot_loss[loss=0.2141, simple_loss=0.2924, pruned_loss=0.06795, over 3805483.39 frames. ], batch size: 64, lr: 4.57e-03, grad_scale: 8.0 2023-04-02 12:33:37,594 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.517e+02 5.202e+02 6.117e+02 7.756e+02 2.072e+03, threshold=1.223e+03, percent-clipped=4.0 2023-04-02 12:33:56,191 INFO [train.py:903] (1/4) Epoch 18, batch 4550, loss[loss=0.2078, simple_loss=0.2901, pruned_loss=0.06275, over 19613.00 frames. ], tot_loss[loss=0.2149, simple_loss=0.293, pruned_loss=0.06836, over 3805636.40 frames. ], batch size: 57, lr: 4.57e-03, grad_scale: 8.0 2023-04-02 12:34:05,862 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9_rvb from training. Duration: 28.72225 2023-04-02 12:34:29,488 WARNING [train.py:1073] (1/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 2023-04-02 12:34:56,266 INFO [train.py:903] (1/4) Epoch 18, batch 4600, loss[loss=0.2075, simple_loss=0.2822, pruned_loss=0.06638, over 19852.00 frames. ], tot_loss[loss=0.2155, simple_loss=0.2932, pruned_loss=0.06891, over 3795055.80 frames. ], batch size: 52, lr: 4.57e-03, grad_scale: 8.0 2023-04-02 12:35:04,405 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.09 vs. limit=5.0 2023-04-02 12:35:09,467 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.5705, 1.1350, 1.3775, 1.2962, 2.2248, 1.0190, 1.9234, 2.4537], device='cuda:1'), covar=tensor([0.0682, 0.2811, 0.2672, 0.1603, 0.0920, 0.2080, 0.1178, 0.0469], device='cuda:1'), in_proj_covar=tensor([0.0397, 0.0356, 0.0376, 0.0340, 0.0366, 0.0345, 0.0367, 0.0387], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 12:35:17,284 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=120694.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 12:35:33,745 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.53 vs. limit=5.0 2023-04-02 12:35:35,195 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.215e+02 4.865e+02 6.169e+02 7.995e+02 1.948e+03, threshold=1.234e+03, percent-clipped=9.0 2023-04-02 12:35:37,843 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.7793, 1.2592, 1.4654, 1.5784, 3.3518, 1.1262, 2.3284, 3.7324], device='cuda:1'), covar=tensor([0.0437, 0.2878, 0.2892, 0.1851, 0.0742, 0.2628, 0.1394, 0.0249], device='cuda:1'), in_proj_covar=tensor([0.0398, 0.0357, 0.0377, 0.0341, 0.0367, 0.0346, 0.0368, 0.0388], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 12:35:55,081 INFO [train.py:903] (1/4) Epoch 18, batch 4650, loss[loss=0.2035, simple_loss=0.295, pruned_loss=0.05605, over 18741.00 frames. ], tot_loss[loss=0.2146, simple_loss=0.2923, pruned_loss=0.0685, over 3803918.59 frames. ], batch size: 74, lr: 4.57e-03, grad_scale: 8.0 2023-04-02 12:36:13,554 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_rvb from training. Duration: 0.92 2023-04-02 12:36:24,599 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_rvb from training. Duration: 0.83 2023-04-02 12:36:55,839 INFO [train.py:903] (1/4) Epoch 18, batch 4700, loss[loss=0.2294, simple_loss=0.3059, pruned_loss=0.07648, over 17223.00 frames. ], tot_loss[loss=0.2148, simple_loss=0.2925, pruned_loss=0.0685, over 3789653.37 frames. ], batch size: 101, lr: 4.57e-03, grad_scale: 8.0 2023-04-02 12:37:18,751 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9_rvb from training. Duration: 0.92225 2023-04-02 12:37:37,344 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.471e+02 5.224e+02 6.544e+02 8.077e+02 2.112e+03, threshold=1.309e+03, percent-clipped=4.0 2023-04-02 12:37:56,116 INFO [train.py:903] (1/4) Epoch 18, batch 4750, loss[loss=0.2093, simple_loss=0.2925, pruned_loss=0.06304, over 19713.00 frames. ], tot_loss[loss=0.2158, simple_loss=0.2932, pruned_loss=0.06918, over 3789682.12 frames. ], batch size: 59, lr: 4.57e-03, grad_scale: 8.0 2023-04-02 12:38:24,376 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.6646, 2.3491, 2.1692, 2.7772, 2.3525, 2.2457, 1.9422, 2.4969], device='cuda:1'), covar=tensor([0.0852, 0.1523, 0.1336, 0.0920, 0.1297, 0.0480, 0.1299, 0.0655], device='cuda:1'), in_proj_covar=tensor([0.0264, 0.0354, 0.0302, 0.0246, 0.0298, 0.0246, 0.0296, 0.0251], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 12:38:57,285 INFO [train.py:903] (1/4) Epoch 18, batch 4800, loss[loss=0.2287, simple_loss=0.3177, pruned_loss=0.0698, over 19491.00 frames. ], tot_loss[loss=0.2149, simple_loss=0.2929, pruned_loss=0.06846, over 3800336.61 frames. ], batch size: 64, lr: 4.56e-03, grad_scale: 8.0 2023-04-02 12:39:38,286 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.835e+02 5.063e+02 6.517e+02 8.294e+02 1.429e+03, threshold=1.303e+03, percent-clipped=1.0 2023-04-02 12:39:57,786 INFO [train.py:903] (1/4) Epoch 18, batch 4850, loss[loss=0.252, simple_loss=0.3227, pruned_loss=0.09068, over 19487.00 frames. ], tot_loss[loss=0.2148, simple_loss=0.2927, pruned_loss=0.06843, over 3810424.74 frames. ], batch size: 64, lr: 4.56e-03, grad_scale: 8.0 2023-04-02 12:40:19,478 WARNING [train.py:1073] (1/4) Exclude cut with ID 774-127930-0014-48411_sp1.1_rvb from training. Duration: 0.95 2023-04-02 12:40:38,932 WARNING [train.py:1073] (1/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1_rvb from training. Duration: 0.9409375 2023-04-02 12:40:44,411 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 2023-04-02 12:40:45,483 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 2023-04-02 12:40:55,308 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1_rvb from training. Duration: 25.3818125 2023-04-02 12:40:57,464 INFO [train.py:903] (1/4) Epoch 18, batch 4900, loss[loss=0.1953, simple_loss=0.2735, pruned_loss=0.05851, over 19355.00 frames. ], tot_loss[loss=0.2151, simple_loss=0.2929, pruned_loss=0.06868, over 3821964.11 frames. ], batch size: 47, lr: 4.56e-03, grad_scale: 8.0 2023-04-02 12:41:15,394 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 2023-04-02 12:41:38,441 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.463e+02 5.379e+02 6.693e+02 8.152e+02 1.326e+03, threshold=1.339e+03, percent-clipped=1.0 2023-04-02 12:41:50,623 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=121021.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 12:41:56,821 INFO [train.py:903] (1/4) Epoch 18, batch 4950, loss[loss=0.1583, simple_loss=0.231, pruned_loss=0.04279, over 19754.00 frames. ], tot_loss[loss=0.2144, simple_loss=0.2922, pruned_loss=0.0683, over 3832909.07 frames. ], batch size: 46, lr: 4.56e-03, grad_scale: 8.0 2023-04-02 12:42:10,901 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=121038.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 12:42:13,931 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 2023-04-02 12:42:36,735 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 2023-04-02 12:42:55,221 INFO [train.py:903] (1/4) Epoch 18, batch 5000, loss[loss=0.2262, simple_loss=0.3077, pruned_loss=0.07232, over 19712.00 frames. ], tot_loss[loss=0.2155, simple_loss=0.2935, pruned_loss=0.06875, over 3840082.99 frames. ], batch size: 59, lr: 4.56e-03, grad_scale: 8.0 2023-04-02 12:43:05,397 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_rvb from training. Duration: 27.14 2023-04-02 12:43:16,151 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 2023-04-02 12:43:36,117 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.358e+02 5.049e+02 6.044e+02 8.198e+02 1.739e+03, threshold=1.209e+03, percent-clipped=8.0 2023-04-02 12:43:52,965 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9583, 3.3512, 1.7849, 1.7897, 3.0934, 1.5774, 1.3628, 2.3614], device='cuda:1'), covar=tensor([0.1215, 0.0515, 0.1107, 0.1017, 0.0503, 0.1238, 0.0994, 0.0566], device='cuda:1'), in_proj_covar=tensor([0.0296, 0.0313, 0.0328, 0.0259, 0.0247, 0.0332, 0.0291, 0.0271], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 12:43:55,696 INFO [train.py:903] (1/4) Epoch 18, batch 5050, loss[loss=0.2313, simple_loss=0.31, pruned_loss=0.07627, over 19522.00 frames. ], tot_loss[loss=0.2154, simple_loss=0.2933, pruned_loss=0.06876, over 3834307.13 frames. ], batch size: 56, lr: 4.56e-03, grad_scale: 8.0 2023-04-02 12:44:04,541 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.0241, 4.3610, 4.7047, 4.7111, 1.8591, 4.4037, 3.8637, 4.4021], device='cuda:1'), covar=tensor([0.1460, 0.0976, 0.0558, 0.0565, 0.5323, 0.0958, 0.0629, 0.1026], device='cuda:1'), in_proj_covar=tensor([0.0758, 0.0702, 0.0906, 0.0788, 0.0806, 0.0657, 0.0548, 0.0833], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-02 12:44:27,562 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=121153.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 12:44:29,504 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9_rvb from training. Duration: 26.62775 2023-04-02 12:44:55,824 INFO [train.py:903] (1/4) Epoch 18, batch 5100, loss[loss=0.2072, simple_loss=0.2942, pruned_loss=0.06009, over 18287.00 frames. ], tot_loss[loss=0.2162, simple_loss=0.2941, pruned_loss=0.06917, over 3826140.63 frames. ], batch size: 83, lr: 4.56e-03, grad_scale: 8.0 2023-04-02 12:45:03,010 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=121182.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 12:45:05,063 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 2023-04-02 12:45:07,234 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_rvb from training. Duration: 25.035 2023-04-02 12:45:12,776 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_rvb from training. Duration: 27.92 2023-04-02 12:45:37,241 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.930e+02 4.928e+02 6.063e+02 8.354e+02 2.244e+03, threshold=1.213e+03, percent-clipped=8.0 2023-04-02 12:45:44,714 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9427, 1.8698, 1.7551, 1.5316, 1.3959, 1.5097, 0.3984, 0.8353], device='cuda:1'), covar=tensor([0.0564, 0.0625, 0.0405, 0.0657, 0.1229, 0.0867, 0.1263, 0.1066], device='cuda:1'), in_proj_covar=tensor([0.0354, 0.0352, 0.0354, 0.0378, 0.0452, 0.0384, 0.0332, 0.0340], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 12:45:56,103 INFO [train.py:903] (1/4) Epoch 18, batch 5150, loss[loss=0.2237, simple_loss=0.3, pruned_loss=0.0737, over 19679.00 frames. ], tot_loss[loss=0.2164, simple_loss=0.2942, pruned_loss=0.06931, over 3796624.16 frames. ], batch size: 53, lr: 4.56e-03, grad_scale: 8.0 2023-04-02 12:46:05,288 WARNING [train.py:1073] (1/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 2023-04-02 12:46:18,093 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.13 vs. limit=2.0 2023-04-02 12:46:40,467 WARNING [train.py:1073] (1/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1_rvb from training. Duration: 0.9681875 2023-04-02 12:46:55,813 INFO [train.py:903] (1/4) Epoch 18, batch 5200, loss[loss=0.2123, simple_loss=0.2962, pruned_loss=0.06418, over 19482.00 frames. ], tot_loss[loss=0.2171, simple_loss=0.2947, pruned_loss=0.0697, over 3805677.60 frames. ], batch size: 64, lr: 4.56e-03, grad_scale: 8.0 2023-04-02 12:47:08,966 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 2023-04-02 12:47:36,015 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.274e+02 5.110e+02 6.332e+02 8.392e+02 3.036e+03, threshold=1.266e+03, percent-clipped=7.0 2023-04-02 12:47:49,053 WARNING [train.py:1073] (1/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 2023-04-02 12:47:54,555 INFO [train.py:903] (1/4) Epoch 18, batch 5250, loss[loss=0.2042, simple_loss=0.2911, pruned_loss=0.05865, over 19256.00 frames. ], tot_loss[loss=0.2163, simple_loss=0.2943, pruned_loss=0.06912, over 3800292.17 frames. ], batch size: 66, lr: 4.56e-03, grad_scale: 8.0 2023-04-02 12:47:59,178 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=121329.0, num_to_drop=1, layers_to_drop={1} 2023-04-02 12:48:35,520 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0773, 1.9179, 1.9999, 2.5597, 1.6810, 2.3845, 2.3880, 2.1527], device='cuda:1'), covar=tensor([0.0759, 0.0829, 0.0898, 0.0878, 0.0992, 0.0690, 0.0857, 0.0635], device='cuda:1'), in_proj_covar=tensor([0.0210, 0.0221, 0.0224, 0.0243, 0.0228, 0.0209, 0.0188, 0.0205], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-02 12:48:40,799 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=121365.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 12:48:54,018 INFO [train.py:903] (1/4) Epoch 18, batch 5300, loss[loss=0.2665, simple_loss=0.3468, pruned_loss=0.09309, over 19332.00 frames. ], tot_loss[loss=0.2156, simple_loss=0.2937, pruned_loss=0.0687, over 3808264.28 frames. ], batch size: 66, lr: 4.56e-03, grad_scale: 4.0 2023-04-02 12:49:10,547 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 2023-04-02 12:49:32,297 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([0.8756, 1.2806, 1.5601, 0.5835, 2.0210, 2.4394, 2.1423, 2.5781], device='cuda:1'), covar=tensor([0.1618, 0.3614, 0.3276, 0.2768, 0.0603, 0.0284, 0.0339, 0.0356], device='cuda:1'), in_proj_covar=tensor([0.0265, 0.0313, 0.0343, 0.0259, 0.0235, 0.0180, 0.0212, 0.0243], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-02 12:49:33,466 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=121409.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 12:49:35,383 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.570e+02 5.404e+02 6.878e+02 8.269e+02 2.518e+03, threshold=1.376e+03, percent-clipped=11.0 2023-04-02 12:49:52,481 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=121425.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 12:49:53,390 INFO [train.py:903] (1/4) Epoch 18, batch 5350, loss[loss=0.2146, simple_loss=0.2975, pruned_loss=0.06581, over 19410.00 frames. ], tot_loss[loss=0.2168, simple_loss=0.2949, pruned_loss=0.06933, over 3807361.88 frames. ], batch size: 70, lr: 4.55e-03, grad_scale: 4.0 2023-04-02 12:50:03,575 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=121434.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 12:50:26,212 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9_rvb from training. Duration: 26.438875 2023-04-02 12:50:53,959 INFO [train.py:903] (1/4) Epoch 18, batch 5400, loss[loss=0.2033, simple_loss=0.2904, pruned_loss=0.05807, over 17417.00 frames. ], tot_loss[loss=0.2168, simple_loss=0.2948, pruned_loss=0.06941, over 3805495.74 frames. ], batch size: 101, lr: 4.55e-03, grad_scale: 4.0 2023-04-02 12:50:59,582 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=121480.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 12:51:35,379 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.204e+02 5.292e+02 6.238e+02 7.901e+02 1.766e+03, threshold=1.248e+03, percent-clipped=4.0 2023-04-02 12:51:54,668 INFO [train.py:903] (1/4) Epoch 18, batch 5450, loss[loss=0.1973, simple_loss=0.2776, pruned_loss=0.05853, over 19730.00 frames. ], tot_loss[loss=0.2166, simple_loss=0.2947, pruned_loss=0.06927, over 3818029.00 frames. ], batch size: 51, lr: 4.55e-03, grad_scale: 4.0 2023-04-02 12:51:54,830 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=121526.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 12:52:20,878 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6661, 1.5483, 1.5296, 2.4475, 1.7821, 2.1479, 2.1663, 1.7795], device='cuda:1'), covar=tensor([0.0789, 0.0928, 0.0984, 0.0640, 0.0862, 0.0632, 0.0752, 0.0669], device='cuda:1'), in_proj_covar=tensor([0.0211, 0.0223, 0.0225, 0.0243, 0.0229, 0.0211, 0.0190, 0.0206], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-04-02 12:52:25,223 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4594, 1.5506, 1.8720, 1.7058, 2.7975, 2.3918, 2.8242, 1.4080], device='cuda:1'), covar=tensor([0.2388, 0.4059, 0.2521, 0.1781, 0.1406, 0.1936, 0.1459, 0.3908], device='cuda:1'), in_proj_covar=tensor([0.0520, 0.0623, 0.0684, 0.0468, 0.0614, 0.0523, 0.0657, 0.0534], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 12:52:54,550 INFO [train.py:903] (1/4) Epoch 18, batch 5500, loss[loss=0.2416, simple_loss=0.3141, pruned_loss=0.08449, over 17167.00 frames. ], tot_loss[loss=0.2161, simple_loss=0.2946, pruned_loss=0.06881, over 3802263.77 frames. ], batch size: 101, lr: 4.55e-03, grad_scale: 4.0 2023-04-02 12:53:19,886 WARNING [train.py:1073] (1/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9_rvb from training. Duration: 27.47775 2023-04-02 12:53:37,491 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.086e+02 4.754e+02 5.891e+02 7.661e+02 1.861e+03, threshold=1.178e+03, percent-clipped=6.0 2023-04-02 12:53:51,009 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.50 vs. limit=2.0 2023-04-02 12:53:55,066 INFO [train.py:903] (1/4) Epoch 18, batch 5550, loss[loss=0.2177, simple_loss=0.3011, pruned_loss=0.06714, over 19666.00 frames. ], tot_loss[loss=0.2155, simple_loss=0.2936, pruned_loss=0.06865, over 3803888.36 frames. ], batch size: 55, lr: 4.55e-03, grad_scale: 4.0 2023-04-02 12:54:02,983 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 2023-04-02 12:54:13,205 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=121641.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 12:54:51,523 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 2023-04-02 12:54:52,547 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=121673.0, num_to_drop=1, layers_to_drop={0} 2023-04-02 12:54:55,722 INFO [train.py:903] (1/4) Epoch 18, batch 5600, loss[loss=0.2235, simple_loss=0.3086, pruned_loss=0.06917, over 19092.00 frames. ], tot_loss[loss=0.2154, simple_loss=0.294, pruned_loss=0.06842, over 3821802.83 frames. ], batch size: 69, lr: 4.55e-03, grad_scale: 8.0 2023-04-02 12:55:37,559 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.280e+02 4.851e+02 5.729e+02 6.791e+02 1.695e+03, threshold=1.146e+03, percent-clipped=1.0 2023-04-02 12:55:56,730 INFO [train.py:903] (1/4) Epoch 18, batch 5650, loss[loss=0.1907, simple_loss=0.2718, pruned_loss=0.05484, over 19845.00 frames. ], tot_loss[loss=0.2148, simple_loss=0.2932, pruned_loss=0.06814, over 3818227.79 frames. ], batch size: 52, lr: 4.55e-03, grad_scale: 8.0 2023-04-02 12:56:08,950 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=121736.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 12:56:37,729 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=121761.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 12:56:44,590 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9_rvb from training. Duration: 28.0944375 2023-04-02 12:56:48,217 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=121769.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 12:56:57,049 INFO [train.py:903] (1/4) Epoch 18, batch 5700, loss[loss=0.1684, simple_loss=0.248, pruned_loss=0.04443, over 19465.00 frames. ], tot_loss[loss=0.214, simple_loss=0.2924, pruned_loss=0.06784, over 3830806.43 frames. ], batch size: 49, lr: 4.55e-03, grad_scale: 8.0 2023-04-02 12:57:11,007 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=121788.0, num_to_drop=1, layers_to_drop={0} 2023-04-02 12:57:39,339 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.986e+02 4.821e+02 5.859e+02 7.608e+02 1.521e+03, threshold=1.172e+03, percent-clipped=4.0 2023-04-02 12:57:54,290 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 2023-04-02 12:57:57,032 INFO [train.py:903] (1/4) Epoch 18, batch 5750, loss[loss=0.2724, simple_loss=0.3342, pruned_loss=0.1053, over 13377.00 frames. ], tot_loss[loss=0.2147, simple_loss=0.2929, pruned_loss=0.0682, over 3820005.39 frames. ], batch size: 136, lr: 4.55e-03, grad_scale: 8.0 2023-04-02 12:57:58,060 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9_rvb from training. Duration: 33.038875 2023-04-02 12:58:05,924 WARNING [train.py:1073] (1/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 2023-04-02 12:58:11,381 WARNING [train.py:1073] (1/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 2023-04-02 12:58:57,512 INFO [train.py:903] (1/4) Epoch 18, batch 5800, loss[loss=0.2088, simple_loss=0.2923, pruned_loss=0.06264, over 19502.00 frames. ], tot_loss[loss=0.2147, simple_loss=0.2926, pruned_loss=0.06833, over 3819366.78 frames. ], batch size: 64, lr: 4.55e-03, grad_scale: 8.0 2023-04-02 12:59:08,068 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=121884.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 12:59:24,568 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=121897.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 12:59:39,869 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.096e+02 5.174e+02 6.386e+02 8.157e+02 2.937e+03, threshold=1.277e+03, percent-clipped=7.0 2023-04-02 12:59:54,647 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=121922.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 12:59:59,626 INFO [train.py:903] (1/4) Epoch 18, batch 5850, loss[loss=0.2282, simple_loss=0.3093, pruned_loss=0.07355, over 18751.00 frames. ], tot_loss[loss=0.2153, simple_loss=0.2932, pruned_loss=0.0687, over 3812088.01 frames. ], batch size: 74, lr: 4.54e-03, grad_scale: 8.0 2023-04-02 13:00:51,281 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=121968.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 13:01:00,155 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9365, 1.6762, 1.6159, 1.9403, 1.6800, 1.7492, 1.6138, 1.8856], device='cuda:1'), covar=tensor([0.0960, 0.1458, 0.1367, 0.0909, 0.1245, 0.0525, 0.1272, 0.0683], device='cuda:1'), in_proj_covar=tensor([0.0266, 0.0356, 0.0304, 0.0248, 0.0300, 0.0247, 0.0298, 0.0251], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 13:01:00,885 INFO [train.py:903] (1/4) Epoch 18, batch 5900, loss[loss=0.2293, simple_loss=0.3087, pruned_loss=0.07493, over 18823.00 frames. ], tot_loss[loss=0.2145, simple_loss=0.2926, pruned_loss=0.06817, over 3819097.25 frames. ], batch size: 74, lr: 4.54e-03, grad_scale: 8.0 2023-04-02 13:01:02,075 WARNING [train.py:1073] (1/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 2023-04-02 13:01:23,312 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9_rvb from training. Duration: 27.511125 2023-04-02 13:01:43,749 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.064e+02 4.950e+02 6.415e+02 8.144e+02 2.513e+03, threshold=1.283e+03, percent-clipped=4.0 2023-04-02 13:02:01,671 INFO [train.py:903] (1/4) Epoch 18, batch 5950, loss[loss=0.2033, simple_loss=0.2847, pruned_loss=0.06094, over 19587.00 frames. ], tot_loss[loss=0.215, simple_loss=0.2932, pruned_loss=0.06843, over 3816064.56 frames. ], batch size: 52, lr: 4.54e-03, grad_scale: 8.0 2023-04-02 13:02:03,446 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.38 vs. limit=5.0 2023-04-02 13:02:24,064 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=122044.0, num_to_drop=1, layers_to_drop={0} 2023-04-02 13:02:42,884 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.41 vs. limit=2.0 2023-04-02 13:02:53,800 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=122069.0, num_to_drop=1, layers_to_drop={1} 2023-04-02 13:03:01,812 INFO [train.py:903] (1/4) Epoch 18, batch 6000, loss[loss=0.2503, simple_loss=0.3261, pruned_loss=0.08727, over 18898.00 frames. ], tot_loss[loss=0.2172, simple_loss=0.2953, pruned_loss=0.06958, over 3810169.54 frames. ], batch size: 74, lr: 4.54e-03, grad_scale: 8.0 2023-04-02 13:03:01,812 INFO [train.py:928] (1/4) Computing validation loss 2023-04-02 13:03:14,294 INFO [train.py:937] (1/4) Epoch 18, validation: loss=0.1702, simple_loss=0.2706, pruned_loss=0.03489, over 944034.00 frames. 2023-04-02 13:03:14,295 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18821MB 2023-04-02 13:03:22,624 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3027, 1.3916, 1.8390, 1.3906, 2.8081, 3.8316, 3.5678, 4.0412], device='cuda:1'), covar=tensor([0.1561, 0.3680, 0.3162, 0.2227, 0.0552, 0.0175, 0.0202, 0.0215], device='cuda:1'), in_proj_covar=tensor([0.0264, 0.0313, 0.0343, 0.0259, 0.0235, 0.0180, 0.0212, 0.0242], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-02 13:03:57,709 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.792e+02 5.055e+02 6.106e+02 7.549e+02 1.634e+03, threshold=1.221e+03, percent-clipped=4.0 2023-04-02 13:04:15,900 INFO [train.py:903] (1/4) Epoch 18, batch 6050, loss[loss=0.2022, simple_loss=0.2874, pruned_loss=0.05848, over 19339.00 frames. ], tot_loss[loss=0.2162, simple_loss=0.2942, pruned_loss=0.06907, over 3820345.82 frames. ], batch size: 66, lr: 4.54e-03, grad_scale: 8.0 2023-04-02 13:04:27,147 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=122135.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 13:04:33,689 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=122140.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 13:05:04,398 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=122165.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 13:05:18,139 INFO [train.py:903] (1/4) Epoch 18, batch 6100, loss[loss=0.2211, simple_loss=0.2879, pruned_loss=0.07718, over 19456.00 frames. ], tot_loss[loss=0.217, simple_loss=0.295, pruned_loss=0.06952, over 3823076.02 frames. ], batch size: 49, lr: 4.54e-03, grad_scale: 8.0 2023-04-02 13:05:59,975 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.364e+02 5.309e+02 6.592e+02 8.010e+02 1.726e+03, threshold=1.318e+03, percent-clipped=1.0 2023-04-02 13:06:03,460 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=122213.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 13:06:18,827 INFO [train.py:903] (1/4) Epoch 18, batch 6150, loss[loss=0.2152, simple_loss=0.287, pruned_loss=0.07167, over 19621.00 frames. ], tot_loss[loss=0.2168, simple_loss=0.2949, pruned_loss=0.0694, over 3829216.44 frames. ], batch size: 50, lr: 4.54e-03, grad_scale: 8.0 2023-04-02 13:06:46,721 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9_rvb from training. Duration: 31.02225 2023-04-02 13:07:20,188 INFO [train.py:903] (1/4) Epoch 18, batch 6200, loss[loss=0.2066, simple_loss=0.2832, pruned_loss=0.06499, over 19689.00 frames. ], tot_loss[loss=0.2163, simple_loss=0.2942, pruned_loss=0.06923, over 3821013.77 frames. ], batch size: 53, lr: 4.54e-03, grad_scale: 8.0 2023-04-02 13:08:04,096 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.932e+02 4.549e+02 5.662e+02 6.707e+02 1.660e+03, threshold=1.132e+03, percent-clipped=3.0 2023-04-02 13:08:05,406 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=122312.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 13:08:17,460 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 2023-04-02 13:08:21,488 INFO [train.py:903] (1/4) Epoch 18, batch 6250, loss[loss=0.217, simple_loss=0.2984, pruned_loss=0.06779, over 18937.00 frames. ], tot_loss[loss=0.216, simple_loss=0.2938, pruned_loss=0.06905, over 3802604.04 frames. ], batch size: 74, lr: 4.54e-03, grad_scale: 8.0 2023-04-02 13:08:35,721 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=122337.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 13:08:53,450 WARNING [train.py:1073] (1/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9_rvb from training. Duration: 25.988875 2023-04-02 13:09:23,685 INFO [train.py:903] (1/4) Epoch 18, batch 6300, loss[loss=0.2508, simple_loss=0.3361, pruned_loss=0.08278, over 19769.00 frames. ], tot_loss[loss=0.2151, simple_loss=0.293, pruned_loss=0.06859, over 3815209.54 frames. ], batch size: 56, lr: 4.54e-03, grad_scale: 8.0 2023-04-02 13:10:06,341 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.668e+02 4.826e+02 5.839e+02 7.420e+02 1.796e+03, threshold=1.168e+03, percent-clipped=3.0 2023-04-02 13:10:25,332 INFO [train.py:903] (1/4) Epoch 18, batch 6350, loss[loss=0.2453, simple_loss=0.3225, pruned_loss=0.08407, over 19574.00 frames. ], tot_loss[loss=0.2126, simple_loss=0.2911, pruned_loss=0.06709, over 3826383.51 frames. ], batch size: 61, lr: 4.54e-03, grad_scale: 8.0 2023-04-02 13:10:26,751 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=122427.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 13:11:25,780 INFO [train.py:903] (1/4) Epoch 18, batch 6400, loss[loss=0.215, simple_loss=0.301, pruned_loss=0.06448, over 19497.00 frames. ], tot_loss[loss=0.2135, simple_loss=0.292, pruned_loss=0.0675, over 3831523.28 frames. ], batch size: 64, lr: 4.53e-03, grad_scale: 8.0 2023-04-02 13:11:29,340 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=122479.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 13:11:30,755 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9394, 1.8997, 1.8494, 1.6125, 1.3996, 1.6356, 0.4562, 0.8648], device='cuda:1'), covar=tensor([0.0551, 0.0541, 0.0362, 0.0593, 0.1150, 0.0718, 0.1150, 0.0988], device='cuda:1'), in_proj_covar=tensor([0.0350, 0.0347, 0.0347, 0.0373, 0.0449, 0.0379, 0.0329, 0.0335], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 13:11:38,325 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=122486.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 13:12:08,686 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.263e+02 4.572e+02 5.352e+02 7.001e+02 2.095e+03, threshold=1.070e+03, percent-clipped=5.0 2023-04-02 13:12:27,162 INFO [train.py:903] (1/4) Epoch 18, batch 6450, loss[loss=0.2534, simple_loss=0.334, pruned_loss=0.08642, over 19291.00 frames. ], tot_loss[loss=0.2127, simple_loss=0.291, pruned_loss=0.06714, over 3817786.37 frames. ], batch size: 66, lr: 4.53e-03, grad_scale: 8.0 2023-04-02 13:12:43,823 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 2023-04-02 13:13:05,515 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=122557.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 13:13:09,885 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9_rvb from training. Duration: 28.638875 2023-04-02 13:13:27,310 INFO [train.py:903] (1/4) Epoch 18, batch 6500, loss[loss=0.2339, simple_loss=0.3085, pruned_loss=0.07966, over 19533.00 frames. ], tot_loss[loss=0.2129, simple_loss=0.2914, pruned_loss=0.06724, over 3823269.31 frames. ], batch size: 54, lr: 4.53e-03, grad_scale: 8.0 2023-04-02 13:13:32,627 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1_rvb from training. Duration: 0.836375 2023-04-02 13:13:41,119 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.45 vs. limit=5.0 2023-04-02 13:13:50,395 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=122594.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 13:14:06,031 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9692, 1.9505, 1.8425, 1.5452, 1.5262, 1.6341, 0.3165, 0.8171], device='cuda:1'), covar=tensor([0.0530, 0.0546, 0.0354, 0.0666, 0.1109, 0.0701, 0.1196, 0.0963], device='cuda:1'), in_proj_covar=tensor([0.0352, 0.0347, 0.0348, 0.0375, 0.0451, 0.0380, 0.0330, 0.0335], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 13:14:10,391 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.928e+02 4.824e+02 5.508e+02 6.898e+02 1.222e+03, threshold=1.102e+03, percent-clipped=3.0 2023-04-02 13:14:28,799 INFO [train.py:903] (1/4) Epoch 18, batch 6550, loss[loss=0.2065, simple_loss=0.2859, pruned_loss=0.0636, over 19383.00 frames. ], tot_loss[loss=0.2123, simple_loss=0.2906, pruned_loss=0.06696, over 3828388.30 frames. ], batch size: 48, lr: 4.53e-03, grad_scale: 8.0 2023-04-02 13:15:24,479 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=122672.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 13:15:29,442 INFO [train.py:903] (1/4) Epoch 18, batch 6600, loss[loss=0.2033, simple_loss=0.2887, pruned_loss=0.05894, over 19603.00 frames. ], tot_loss[loss=0.2126, simple_loss=0.2912, pruned_loss=0.06701, over 3837150.51 frames. ], batch size: 57, lr: 4.53e-03, grad_scale: 8.0 2023-04-02 13:15:35,564 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=122681.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 13:15:37,956 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=122683.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 13:16:08,850 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=122708.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 13:16:11,858 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.339e+02 4.976e+02 6.114e+02 8.137e+02 1.508e+03, threshold=1.223e+03, percent-clipped=10.0 2023-04-02 13:16:29,836 INFO [train.py:903] (1/4) Epoch 18, batch 6650, loss[loss=0.2164, simple_loss=0.2969, pruned_loss=0.06799, over 19686.00 frames. ], tot_loss[loss=0.2121, simple_loss=0.2908, pruned_loss=0.06665, over 3843218.32 frames. ], batch size: 53, lr: 4.53e-03, grad_scale: 8.0 2023-04-02 13:17:30,797 INFO [train.py:903] (1/4) Epoch 18, batch 6700, loss[loss=0.1868, simple_loss=0.2754, pruned_loss=0.04913, over 19525.00 frames. ], tot_loss[loss=0.2116, simple_loss=0.2904, pruned_loss=0.06638, over 3842737.37 frames. ], batch size: 54, lr: 4.53e-03, grad_scale: 8.0 2023-04-02 13:17:55,707 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=122796.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 13:18:12,046 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.354e+02 4.791e+02 5.844e+02 7.603e+02 1.856e+03, threshold=1.169e+03, percent-clipped=2.0 2023-04-02 13:18:28,392 INFO [train.py:903] (1/4) Epoch 18, batch 6750, loss[loss=0.1926, simple_loss=0.2699, pruned_loss=0.05763, over 19752.00 frames. ], tot_loss[loss=0.2118, simple_loss=0.2905, pruned_loss=0.06657, over 3842323.24 frames. ], batch size: 46, lr: 4.53e-03, grad_scale: 8.0 2023-04-02 13:18:33,072 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=122830.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 13:18:55,585 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=122850.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 13:19:24,217 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=122875.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 13:19:25,055 INFO [train.py:903] (1/4) Epoch 18, batch 6800, loss[loss=0.2254, simple_loss=0.3021, pruned_loss=0.07437, over 19301.00 frames. ], tot_loss[loss=0.2134, simple_loss=0.2916, pruned_loss=0.06761, over 3821714.62 frames. ], batch size: 66, lr: 4.53e-03, grad_scale: 8.0 2023-04-02 13:19:40,829 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=122890.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 13:20:08,835 WARNING [train.py:1073] (1/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 2023-04-02 13:20:09,296 WARNING [train.py:1073] (1/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 2023-04-02 13:20:12,595 INFO [train.py:903] (1/4) Epoch 19, batch 0, loss[loss=0.2199, simple_loss=0.3049, pruned_loss=0.06749, over 18769.00 frames. ], tot_loss[loss=0.2199, simple_loss=0.3049, pruned_loss=0.06749, over 18769.00 frames. ], batch size: 74, lr: 4.40e-03, grad_scale: 8.0 2023-04-02 13:20:12,596 INFO [train.py:928] (1/4) Computing validation loss 2023-04-02 13:20:24,051 INFO [train.py:937] (1/4) Epoch 19, validation: loss=0.171, simple_loss=0.2713, pruned_loss=0.03533, over 944034.00 frames. 2023-04-02 13:20:24,052 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18821MB 2023-04-02 13:20:32,696 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.525e+02 4.987e+02 6.075e+02 7.792e+02 1.350e+03, threshold=1.215e+03, percent-clipped=4.0 2023-04-02 13:20:37,462 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 2023-04-02 13:20:53,702 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=122928.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 13:21:13,138 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.52 vs. limit=2.0 2023-04-02 13:21:13,984 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=122945.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 13:21:24,659 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=122953.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 13:21:25,464 INFO [train.py:903] (1/4) Epoch 19, batch 50, loss[loss=0.2162, simple_loss=0.2984, pruned_loss=0.06701, over 19740.00 frames. ], tot_loss[loss=0.2073, simple_loss=0.2861, pruned_loss=0.06425, over 876989.95 frames. ], batch size: 63, lr: 4.40e-03, grad_scale: 8.0 2023-04-02 13:21:51,204 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.30 vs. limit=5.0 2023-04-02 13:22:04,205 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1_rvb from training. Duration: 27.0318125 2023-04-02 13:22:16,582 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([0.8656, 1.3474, 1.0499, 0.9769, 1.1712, 0.9842, 0.9561, 1.2674], device='cuda:1'), covar=tensor([0.0616, 0.0865, 0.1156, 0.0729, 0.0564, 0.1350, 0.0571, 0.0486], device='cuda:1'), in_proj_covar=tensor([0.0299, 0.0314, 0.0331, 0.0259, 0.0247, 0.0334, 0.0291, 0.0273], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 13:22:27,932 INFO [train.py:903] (1/4) Epoch 19, batch 100, loss[loss=0.2076, simple_loss=0.2913, pruned_loss=0.06189, over 18328.00 frames. ], tot_loss[loss=0.2131, simple_loss=0.2908, pruned_loss=0.06773, over 1524087.60 frames. ], batch size: 83, lr: 4.40e-03, grad_scale: 8.0 2023-04-02 13:22:35,845 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.196e+02 4.975e+02 5.926e+02 8.151e+02 1.966e+03, threshold=1.185e+03, percent-clipped=7.0 2023-04-02 13:22:37,410 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2529, 1.3072, 1.2455, 1.0595, 1.1037, 1.1653, 0.0653, 0.4033], device='cuda:1'), covar=tensor([0.0541, 0.0560, 0.0351, 0.0473, 0.1067, 0.0557, 0.1130, 0.0912], device='cuda:1'), in_proj_covar=tensor([0.0350, 0.0347, 0.0347, 0.0374, 0.0450, 0.0380, 0.0328, 0.0333], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 13:22:40,146 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 2023-04-02 13:23:25,700 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=123052.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 13:23:26,583 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.5987, 4.1536, 2.5793, 3.6651, 0.9263, 4.0867, 4.0178, 4.0516], device='cuda:1'), covar=tensor([0.0653, 0.1098, 0.2024, 0.0844, 0.4078, 0.0691, 0.0825, 0.1105], device='cuda:1'), in_proj_covar=tensor([0.0485, 0.0395, 0.0483, 0.0342, 0.0397, 0.0421, 0.0410, 0.0446], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 13:23:27,453 INFO [train.py:903] (1/4) Epoch 19, batch 150, loss[loss=0.2223, simple_loss=0.307, pruned_loss=0.06883, over 19673.00 frames. ], tot_loss[loss=0.2141, simple_loss=0.2919, pruned_loss=0.06816, over 2038952.19 frames. ], batch size: 58, lr: 4.40e-03, grad_scale: 8.0 2023-04-02 13:23:55,483 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=123077.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 13:24:27,288 INFO [train.py:903] (1/4) Epoch 19, batch 200, loss[loss=0.2719, simple_loss=0.3333, pruned_loss=0.1053, over 18305.00 frames. ], tot_loss[loss=0.2147, simple_loss=0.2923, pruned_loss=0.06857, over 2431957.63 frames. ], batch size: 83, lr: 4.40e-03, grad_scale: 8.0 2023-04-02 13:24:29,571 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 2023-04-02 13:24:35,482 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.274e+02 5.223e+02 5.868e+02 7.012e+02 1.944e+03, threshold=1.174e+03, percent-clipped=7.0 2023-04-02 13:24:36,969 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.7813, 1.4060, 1.5817, 1.5865, 3.3096, 1.2048, 2.4891, 3.7711], device='cuda:1'), covar=tensor([0.0487, 0.2733, 0.2744, 0.1847, 0.0708, 0.2490, 0.1166, 0.0240], device='cuda:1'), in_proj_covar=tensor([0.0396, 0.0356, 0.0378, 0.0341, 0.0367, 0.0349, 0.0369, 0.0386], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 13:24:58,739 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([6.2939, 5.6791, 3.2131, 4.9958, 1.1292, 5.7718, 5.6410, 5.8142], device='cuda:1'), covar=tensor([0.0389, 0.0950, 0.1808, 0.0750, 0.4307, 0.0576, 0.0757, 0.0830], device='cuda:1'), in_proj_covar=tensor([0.0485, 0.0396, 0.0483, 0.0342, 0.0396, 0.0420, 0.0410, 0.0445], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 13:25:27,083 INFO [train.py:903] (1/4) Epoch 19, batch 250, loss[loss=0.1794, simple_loss=0.2631, pruned_loss=0.04781, over 19737.00 frames. ], tot_loss[loss=0.2127, simple_loss=0.2907, pruned_loss=0.06736, over 2738241.64 frames. ], batch size: 46, lr: 4.40e-03, grad_scale: 8.0 2023-04-02 13:26:25,147 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=123201.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 13:26:28,234 INFO [train.py:903] (1/4) Epoch 19, batch 300, loss[loss=0.1714, simple_loss=0.2528, pruned_loss=0.04496, over 19727.00 frames. ], tot_loss[loss=0.2116, simple_loss=0.2895, pruned_loss=0.06681, over 2974387.03 frames. ], batch size: 46, lr: 4.40e-03, grad_scale: 8.0 2023-04-02 13:26:37,167 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.064e+02 4.833e+02 6.226e+02 7.852e+02 1.722e+03, threshold=1.245e+03, percent-clipped=4.0 2023-04-02 13:26:55,578 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=123226.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 13:27:04,589 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=123234.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 13:27:27,986 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.43 vs. limit=5.0 2023-04-02 13:27:29,488 INFO [train.py:903] (1/4) Epoch 19, batch 350, loss[loss=0.198, simple_loss=0.2783, pruned_loss=0.05888, over 19507.00 frames. ], tot_loss[loss=0.2112, simple_loss=0.2892, pruned_loss=0.06666, over 3161890.32 frames. ], batch size: 54, lr: 4.40e-03, grad_scale: 8.0 2023-04-02 13:27:35,243 WARNING [train.py:1073] (1/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1_rvb from training. Duration: 0.9681875 2023-04-02 13:27:46,487 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.8460, 1.3412, 1.4894, 1.8106, 3.4131, 1.3252, 2.4474, 3.9083], device='cuda:1'), covar=tensor([0.0489, 0.2755, 0.3010, 0.1759, 0.0726, 0.2529, 0.1349, 0.0216], device='cuda:1'), in_proj_covar=tensor([0.0397, 0.0357, 0.0380, 0.0342, 0.0367, 0.0350, 0.0370, 0.0388], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 13:28:10,780 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.0938, 0.9343, 1.0741, 1.5029, 1.0436, 0.9925, 1.1103, 1.0490], device='cuda:1'), covar=tensor([0.1267, 0.1805, 0.1498, 0.0733, 0.1100, 0.1594, 0.1185, 0.1282], device='cuda:1'), in_proj_covar=tensor([0.0209, 0.0219, 0.0224, 0.0241, 0.0226, 0.0210, 0.0186, 0.0204], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-02 13:28:30,167 INFO [train.py:903] (1/4) Epoch 19, batch 400, loss[loss=0.1866, simple_loss=0.2693, pruned_loss=0.05192, over 19678.00 frames. ], tot_loss[loss=0.2121, simple_loss=0.2902, pruned_loss=0.06706, over 3307460.17 frames. ], batch size: 53, lr: 4.40e-03, grad_scale: 8.0 2023-04-02 13:28:37,932 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.570e+02 5.151e+02 6.306e+02 7.944e+02 1.366e+03, threshold=1.261e+03, percent-clipped=2.0 2023-04-02 13:29:04,108 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=123332.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 13:29:24,965 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=123349.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 13:29:30,457 INFO [train.py:903] (1/4) Epoch 19, batch 450, loss[loss=0.2287, simple_loss=0.3032, pruned_loss=0.0771, over 17379.00 frames. ], tot_loss[loss=0.2127, simple_loss=0.2907, pruned_loss=0.06738, over 3419303.78 frames. ], batch size: 101, lr: 4.40e-03, grad_scale: 8.0 2023-04-02 13:30:04,808 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 2023-04-02 13:30:05,744 WARNING [train.py:1073] (1/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 2023-04-02 13:30:30,933 INFO [train.py:903] (1/4) Epoch 19, batch 500, loss[loss=0.26, simple_loss=0.3258, pruned_loss=0.09705, over 19784.00 frames. ], tot_loss[loss=0.2143, simple_loss=0.292, pruned_loss=0.06834, over 3503893.00 frames. ], batch size: 56, lr: 4.40e-03, grad_scale: 16.0 2023-04-02 13:30:39,751 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.237e+02 5.352e+02 7.180e+02 8.361e+02 2.088e+03, threshold=1.436e+03, percent-clipped=3.0 2023-04-02 13:31:21,143 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=123446.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 13:31:22,360 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.5916, 1.1439, 1.4985, 1.2463, 2.2502, 1.0243, 2.1467, 2.4493], device='cuda:1'), covar=tensor([0.0719, 0.2743, 0.2616, 0.1648, 0.0879, 0.2071, 0.0944, 0.0482], device='cuda:1'), in_proj_covar=tensor([0.0400, 0.0359, 0.0380, 0.0344, 0.0369, 0.0350, 0.0371, 0.0389], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 13:31:30,113 INFO [train.py:903] (1/4) Epoch 19, batch 550, loss[loss=0.2314, simple_loss=0.3025, pruned_loss=0.08014, over 19731.00 frames. ], tot_loss[loss=0.2144, simple_loss=0.2924, pruned_loss=0.06814, over 3587512.07 frames. ], batch size: 51, lr: 4.39e-03, grad_scale: 8.0 2023-04-02 13:32:30,276 INFO [train.py:903] (1/4) Epoch 19, batch 600, loss[loss=0.2013, simple_loss=0.2904, pruned_loss=0.05613, over 17163.00 frames. ], tot_loss[loss=0.2147, simple_loss=0.2929, pruned_loss=0.06828, over 3643085.66 frames. ], batch size: 101, lr: 4.39e-03, grad_scale: 8.0 2023-04-02 13:32:39,907 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.253e+02 4.960e+02 5.982e+02 8.370e+02 1.865e+03, threshold=1.196e+03, percent-clipped=4.0 2023-04-02 13:32:49,059 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.8189, 4.3325, 2.5517, 3.8601, 1.1349, 4.2849, 4.1740, 4.2604], device='cuda:1'), covar=tensor([0.0589, 0.1013, 0.2041, 0.0767, 0.3689, 0.0684, 0.0791, 0.1010], device='cuda:1'), in_proj_covar=tensor([0.0480, 0.0396, 0.0480, 0.0339, 0.0394, 0.0418, 0.0407, 0.0443], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 13:33:14,016 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 2023-04-02 13:33:30,843 INFO [train.py:903] (1/4) Epoch 19, batch 650, loss[loss=0.246, simple_loss=0.3177, pruned_loss=0.08718, over 17999.00 frames. ], tot_loss[loss=0.2142, simple_loss=0.2925, pruned_loss=0.06799, over 3691986.87 frames. ], batch size: 83, lr: 4.39e-03, grad_scale: 8.0 2023-04-02 13:34:07,483 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4908, 2.0529, 1.6393, 1.5540, 1.9985, 1.4414, 1.4484, 1.8277], device='cuda:1'), covar=tensor([0.0883, 0.0728, 0.0908, 0.0608, 0.0469, 0.1012, 0.0580, 0.0418], device='cuda:1'), in_proj_covar=tensor([0.0301, 0.0316, 0.0335, 0.0261, 0.0248, 0.0337, 0.0293, 0.0274], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 13:34:30,416 INFO [train.py:903] (1/4) Epoch 19, batch 700, loss[loss=0.2193, simple_loss=0.2976, pruned_loss=0.07046, over 19660.00 frames. ], tot_loss[loss=0.2138, simple_loss=0.2924, pruned_loss=0.06764, over 3734086.21 frames. ], batch size: 58, lr: 4.39e-03, grad_scale: 8.0 2023-04-02 13:34:31,969 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=123605.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 13:34:41,232 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.354e+02 5.460e+02 6.000e+02 7.674e+02 1.249e+03, threshold=1.200e+03, percent-clipped=2.0 2023-04-02 13:35:02,794 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=123630.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 13:35:31,844 INFO [train.py:903] (1/4) Epoch 19, batch 750, loss[loss=0.2019, simple_loss=0.2909, pruned_loss=0.05642, over 19662.00 frames. ], tot_loss[loss=0.2138, simple_loss=0.2927, pruned_loss=0.06746, over 3747588.35 frames. ], batch size: 55, lr: 4.39e-03, grad_scale: 8.0 2023-04-02 13:35:59,275 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=123676.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 13:36:34,036 INFO [train.py:903] (1/4) Epoch 19, batch 800, loss[loss=0.1855, simple_loss=0.258, pruned_loss=0.05652, over 19283.00 frames. ], tot_loss[loss=0.2125, simple_loss=0.2916, pruned_loss=0.06671, over 3768897.97 frames. ], batch size: 44, lr: 4.39e-03, grad_scale: 8.0 2023-04-02 13:36:37,827 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0228, 1.8436, 1.7115, 2.0184, 1.8716, 1.7549, 1.6916, 1.9799], device='cuda:1'), covar=tensor([0.0995, 0.1414, 0.1331, 0.0993, 0.1216, 0.0543, 0.1297, 0.0659], device='cuda:1'), in_proj_covar=tensor([0.0266, 0.0352, 0.0303, 0.0249, 0.0298, 0.0248, 0.0297, 0.0251], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 13:36:43,873 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.634e+02 4.695e+02 6.255e+02 7.449e+02 1.390e+03, threshold=1.251e+03, percent-clipped=2.0 2023-04-02 13:36:47,178 WARNING [train.py:1073] (1/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 2023-04-02 13:37:20,978 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0414, 2.9205, 1.8361, 1.9063, 2.6811, 1.5883, 1.4823, 2.2294], device='cuda:1'), covar=tensor([0.1242, 0.0684, 0.0984, 0.0751, 0.0458, 0.1192, 0.0907, 0.0616], device='cuda:1'), in_proj_covar=tensor([0.0298, 0.0313, 0.0332, 0.0259, 0.0245, 0.0335, 0.0291, 0.0271], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 13:37:34,010 INFO [train.py:903] (1/4) Epoch 19, batch 850, loss[loss=0.2202, simple_loss=0.3021, pruned_loss=0.06913, over 19799.00 frames. ], tot_loss[loss=0.2134, simple_loss=0.2923, pruned_loss=0.06727, over 3786985.89 frames. ], batch size: 63, lr: 4.39e-03, grad_scale: 8.0 2023-04-02 13:38:17,258 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=123790.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 13:38:18,584 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=123791.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 13:38:23,845 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9_rvb from training. Duration: 25.061125 2023-04-02 13:38:34,410 INFO [train.py:903] (1/4) Epoch 19, batch 900, loss[loss=0.202, simple_loss=0.292, pruned_loss=0.05598, over 19530.00 frames. ], tot_loss[loss=0.2124, simple_loss=0.2915, pruned_loss=0.06666, over 3804285.52 frames. ], batch size: 54, lr: 4.39e-03, grad_scale: 8.0 2023-04-02 13:38:38,092 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.3426, 3.0757, 2.2819, 2.7497, 0.6326, 3.0550, 2.8843, 2.9950], device='cuda:1'), covar=tensor([0.1126, 0.1349, 0.1944, 0.1028, 0.4034, 0.0897, 0.1050, 0.1342], device='cuda:1'), in_proj_covar=tensor([0.0481, 0.0394, 0.0477, 0.0338, 0.0394, 0.0417, 0.0406, 0.0442], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 13:38:44,952 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.405e+02 4.993e+02 6.523e+02 8.112e+02 2.572e+03, threshold=1.305e+03, percent-clipped=9.0 2023-04-02 13:39:14,822 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6915, 1.5927, 1.5292, 2.0039, 1.6460, 1.8219, 1.9754, 1.7463], device='cuda:1'), covar=tensor([0.0802, 0.0883, 0.0999, 0.0737, 0.0779, 0.0728, 0.0790, 0.0684], device='cuda:1'), in_proj_covar=tensor([0.0209, 0.0221, 0.0226, 0.0242, 0.0227, 0.0211, 0.0188, 0.0205], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-02 13:39:35,378 INFO [train.py:903] (1/4) Epoch 19, batch 950, loss[loss=0.2156, simple_loss=0.2814, pruned_loss=0.07488, over 19805.00 frames. ], tot_loss[loss=0.2122, simple_loss=0.2912, pruned_loss=0.06654, over 3803292.61 frames. ], batch size: 48, lr: 4.39e-03, grad_scale: 8.0 2023-04-02 13:39:35,392 WARNING [train.py:1073] (1/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9_rvb from training. Duration: 26.32775 2023-04-02 13:39:38,942 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.3589, 3.9743, 2.6260, 3.5701, 0.8818, 3.9429, 3.7859, 3.8927], device='cuda:1'), covar=tensor([0.0663, 0.1015, 0.1823, 0.0833, 0.3992, 0.0654, 0.0866, 0.1078], device='cuda:1'), in_proj_covar=tensor([0.0483, 0.0395, 0.0479, 0.0340, 0.0395, 0.0418, 0.0408, 0.0444], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 13:40:18,681 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=123890.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 13:40:35,886 INFO [train.py:903] (1/4) Epoch 19, batch 1000, loss[loss=0.2044, simple_loss=0.2941, pruned_loss=0.05737, over 19519.00 frames. ], tot_loss[loss=0.2135, simple_loss=0.2921, pruned_loss=0.06742, over 3792846.32 frames. ], batch size: 56, lr: 4.39e-03, grad_scale: 8.0 2023-04-02 13:40:37,276 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=123905.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 13:40:44,459 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.613e+02 5.533e+02 6.903e+02 9.244e+02 2.435e+03, threshold=1.381e+03, percent-clipped=8.0 2023-04-02 13:41:23,916 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 2023-04-02 13:41:33,150 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1885, 1.2660, 1.7992, 1.3479, 2.6260, 3.5169, 3.3386, 3.7947], device='cuda:1'), covar=tensor([0.1637, 0.3847, 0.3228, 0.2361, 0.0665, 0.0210, 0.0219, 0.0247], device='cuda:1'), in_proj_covar=tensor([0.0264, 0.0311, 0.0343, 0.0259, 0.0236, 0.0179, 0.0211, 0.0242], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-02 13:41:33,874 INFO [train.py:903] (1/4) Epoch 19, batch 1050, loss[loss=0.2282, simple_loss=0.3082, pruned_loss=0.07409, over 18667.00 frames. ], tot_loss[loss=0.2137, simple_loss=0.292, pruned_loss=0.06777, over 3788676.35 frames. ], batch size: 74, lr: 4.39e-03, grad_scale: 8.0 2023-04-02 13:42:03,302 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 2023-04-02 13:42:36,015 INFO [train.py:903] (1/4) Epoch 19, batch 1100, loss[loss=0.216, simple_loss=0.2922, pruned_loss=0.06983, over 19606.00 frames. ], tot_loss[loss=0.2138, simple_loss=0.2921, pruned_loss=0.0678, over 3800916.80 frames. ], batch size: 52, lr: 4.38e-03, grad_scale: 8.0 2023-04-02 13:42:45,318 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.024e+02 4.737e+02 5.703e+02 7.685e+02 1.238e+03, threshold=1.141e+03, percent-clipped=0.0 2023-04-02 13:43:28,385 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=124047.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 13:43:36,791 INFO [train.py:903] (1/4) Epoch 19, batch 1150, loss[loss=0.1824, simple_loss=0.2634, pruned_loss=0.05068, over 17816.00 frames. ], tot_loss[loss=0.2135, simple_loss=0.2916, pruned_loss=0.06765, over 3804929.22 frames. ], batch size: 39, lr: 4.38e-03, grad_scale: 8.0 2023-04-02 13:43:37,070 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=124054.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 13:43:59,193 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=124072.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 13:44:07,956 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4117, 2.1292, 1.6580, 1.3836, 2.0104, 1.2059, 1.3001, 1.8771], device='cuda:1'), covar=tensor([0.1011, 0.0810, 0.0968, 0.0867, 0.0536, 0.1245, 0.0698, 0.0426], device='cuda:1'), in_proj_covar=tensor([0.0296, 0.0311, 0.0331, 0.0258, 0.0243, 0.0332, 0.0288, 0.0269], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 13:44:08,935 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=124080.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 13:44:37,474 INFO [train.py:903] (1/4) Epoch 19, batch 1200, loss[loss=0.2212, simple_loss=0.3064, pruned_loss=0.06803, over 19710.00 frames. ], tot_loss[loss=0.2133, simple_loss=0.2913, pruned_loss=0.06767, over 3821574.13 frames. ], batch size: 59, lr: 4.38e-03, grad_scale: 8.0 2023-04-02 13:44:40,172 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([0.9806, 1.3484, 1.0647, 0.9594, 1.1805, 1.0063, 1.0283, 1.2317], device='cuda:1'), covar=tensor([0.0545, 0.0882, 0.1139, 0.0708, 0.0556, 0.1231, 0.0557, 0.0478], device='cuda:1'), in_proj_covar=tensor([0.0295, 0.0310, 0.0329, 0.0257, 0.0241, 0.0331, 0.0287, 0.0268], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 13:44:48,187 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.573e+02 5.076e+02 6.121e+02 8.217e+02 1.455e+03, threshold=1.224e+03, percent-clipped=6.0 2023-04-02 13:45:09,226 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 2023-04-02 13:45:38,203 INFO [train.py:903] (1/4) Epoch 19, batch 1250, loss[loss=0.2301, simple_loss=0.3102, pruned_loss=0.07503, over 19788.00 frames. ], tot_loss[loss=0.2136, simple_loss=0.2916, pruned_loss=0.06782, over 3820080.13 frames. ], batch size: 56, lr: 4.38e-03, grad_scale: 8.0 2023-04-02 13:45:47,190 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=124161.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 13:46:16,805 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=124186.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 13:46:38,208 INFO [train.py:903] (1/4) Epoch 19, batch 1300, loss[loss=0.2159, simple_loss=0.2883, pruned_loss=0.07179, over 19391.00 frames. ], tot_loss[loss=0.2134, simple_loss=0.2916, pruned_loss=0.06766, over 3811952.92 frames. ], batch size: 48, lr: 4.38e-03, grad_scale: 8.0 2023-04-02 13:46:47,590 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.9749, 1.5967, 1.8030, 1.8693, 4.5284, 1.2421, 2.5973, 4.7980], device='cuda:1'), covar=tensor([0.0397, 0.2778, 0.2729, 0.1914, 0.0640, 0.2645, 0.1446, 0.0189], device='cuda:1'), in_proj_covar=tensor([0.0396, 0.0355, 0.0373, 0.0340, 0.0365, 0.0348, 0.0366, 0.0386], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 13:46:48,375 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.504e+02 4.873e+02 5.866e+02 7.986e+02 1.872e+03, threshold=1.173e+03, percent-clipped=5.0 2023-04-02 13:47:15,082 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=124234.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 13:47:37,770 INFO [train.py:903] (1/4) Epoch 19, batch 1350, loss[loss=0.2086, simple_loss=0.2916, pruned_loss=0.06283, over 19615.00 frames. ], tot_loss[loss=0.2136, simple_loss=0.2918, pruned_loss=0.06764, over 3807687.28 frames. ], batch size: 61, lr: 4.38e-03, grad_scale: 8.0 2023-04-02 13:47:58,311 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2186, 1.2155, 1.2276, 1.4159, 1.0680, 1.2520, 1.2793, 1.2712], device='cuda:1'), covar=tensor([0.0864, 0.0931, 0.1049, 0.0650, 0.0909, 0.0896, 0.0902, 0.0770], device='cuda:1'), in_proj_covar=tensor([0.0210, 0.0220, 0.0226, 0.0242, 0.0227, 0.0211, 0.0189, 0.0205], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-02 13:48:22,191 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=124290.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 13:48:22,381 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9798, 2.0709, 2.3718, 2.7030, 1.9659, 2.5862, 2.3559, 2.1070], device='cuda:1'), covar=tensor([0.4454, 0.4152, 0.1872, 0.2487, 0.4257, 0.2186, 0.4693, 0.3497], device='cuda:1'), in_proj_covar=tensor([0.0871, 0.0933, 0.0699, 0.0919, 0.0855, 0.0792, 0.0826, 0.0766], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 13:48:33,040 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9353, 2.0095, 2.2962, 2.6553, 1.8965, 2.5104, 2.3550, 2.0928], device='cuda:1'), covar=tensor([0.4120, 0.3955, 0.1862, 0.2239, 0.4097, 0.2065, 0.4668, 0.3258], device='cuda:1'), in_proj_covar=tensor([0.0872, 0.0933, 0.0699, 0.0920, 0.0856, 0.0792, 0.0827, 0.0767], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 13:48:39,027 INFO [train.py:903] (1/4) Epoch 19, batch 1400, loss[loss=0.2103, simple_loss=0.2821, pruned_loss=0.06922, over 19733.00 frames. ], tot_loss[loss=0.2134, simple_loss=0.2918, pruned_loss=0.0675, over 3805031.97 frames. ], batch size: 45, lr: 4.38e-03, grad_scale: 8.0 2023-04-02 13:48:48,019 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.3321, 3.9173, 2.3934, 3.5395, 0.9243, 3.8754, 3.7496, 3.8747], device='cuda:1'), covar=tensor([0.0689, 0.1121, 0.2192, 0.0879, 0.4005, 0.0696, 0.0894, 0.1039], device='cuda:1'), in_proj_covar=tensor([0.0488, 0.0394, 0.0482, 0.0339, 0.0396, 0.0421, 0.0409, 0.0443], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 13:48:48,825 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.237e+02 5.508e+02 6.829e+02 9.566e+02 2.163e+03, threshold=1.366e+03, percent-clipped=9.0 2023-04-02 13:49:32,985 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=124349.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 13:49:36,287 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.5503, 4.1108, 2.6182, 3.7064, 0.8929, 4.0182, 3.9280, 4.0540], device='cuda:1'), covar=tensor([0.0678, 0.1118, 0.2146, 0.0883, 0.4260, 0.0803, 0.0916, 0.1217], device='cuda:1'), in_proj_covar=tensor([0.0489, 0.0395, 0.0485, 0.0340, 0.0397, 0.0423, 0.0411, 0.0446], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 13:49:38,314 INFO [train.py:903] (1/4) Epoch 19, batch 1450, loss[loss=0.1858, simple_loss=0.2695, pruned_loss=0.0511, over 19663.00 frames. ], tot_loss[loss=0.2149, simple_loss=0.2933, pruned_loss=0.06828, over 3791982.20 frames. ], batch size: 53, lr: 4.38e-03, grad_scale: 8.0 2023-04-02 13:49:40,279 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_rvb from training. Duration: 25.85 2023-04-02 13:49:44,391 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 2023-04-02 13:49:53,701 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=124366.0, num_to_drop=1, layers_to_drop={1} 2023-04-02 13:50:32,822 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=124398.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 13:50:40,521 INFO [train.py:903] (1/4) Epoch 19, batch 1500, loss[loss=0.2504, simple_loss=0.3188, pruned_loss=0.09102, over 13606.00 frames. ], tot_loss[loss=0.2157, simple_loss=0.294, pruned_loss=0.06872, over 3775977.53 frames. ], batch size: 135, lr: 4.38e-03, grad_scale: 8.0 2023-04-02 13:50:50,230 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.987e+02 4.822e+02 6.235e+02 8.378e+02 1.519e+03, threshold=1.247e+03, percent-clipped=2.0 2023-04-02 13:51:04,414 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=124424.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 13:51:35,992 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=124450.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 13:51:39,959 INFO [train.py:903] (1/4) Epoch 19, batch 1550, loss[loss=0.2001, simple_loss=0.2696, pruned_loss=0.06526, over 19726.00 frames. ], tot_loss[loss=0.2152, simple_loss=0.2938, pruned_loss=0.06832, over 3789428.14 frames. ], batch size: 46, lr: 4.38e-03, grad_scale: 8.0 2023-04-02 13:52:40,758 INFO [train.py:903] (1/4) Epoch 19, batch 1600, loss[loss=0.2298, simple_loss=0.3087, pruned_loss=0.07545, over 19711.00 frames. ], tot_loss[loss=0.2143, simple_loss=0.2928, pruned_loss=0.06787, over 3799869.62 frames. ], batch size: 59, lr: 4.38e-03, grad_scale: 8.0 2023-04-02 13:52:51,838 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.803e+02 4.800e+02 6.281e+02 8.115e+02 1.566e+03, threshold=1.256e+03, percent-clipped=2.0 2023-04-02 13:52:52,207 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=124513.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 13:53:06,488 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9_rvb from training. Duration: 30.1555625 2023-04-02 13:53:23,361 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=124539.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 13:53:40,464 INFO [train.py:903] (1/4) Epoch 19, batch 1650, loss[loss=0.221, simple_loss=0.2908, pruned_loss=0.07554, over 19471.00 frames. ], tot_loss[loss=0.2157, simple_loss=0.2939, pruned_loss=0.06874, over 3798861.49 frames. ], batch size: 49, lr: 4.37e-03, grad_scale: 8.0 2023-04-02 13:54:43,097 INFO [train.py:903] (1/4) Epoch 19, batch 1700, loss[loss=0.2213, simple_loss=0.3028, pruned_loss=0.06984, over 19529.00 frames. ], tot_loss[loss=0.2157, simple_loss=0.294, pruned_loss=0.0687, over 3805540.35 frames. ], batch size: 56, lr: 4.37e-03, grad_scale: 8.0 2023-04-02 13:54:44,641 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=124605.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 13:54:53,185 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.721e+02 4.942e+02 5.736e+02 7.226e+02 1.444e+03, threshold=1.147e+03, percent-clipped=3.0 2023-04-02 13:55:14,580 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=124630.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 13:55:18,826 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=124634.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 13:55:21,973 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 2023-04-02 13:55:42,973 INFO [train.py:903] (1/4) Epoch 19, batch 1750, loss[loss=0.2102, simple_loss=0.2859, pruned_loss=0.06726, over 18715.00 frames. ], tot_loss[loss=0.2153, simple_loss=0.2938, pruned_loss=0.06842, over 3807677.95 frames. ], batch size: 74, lr: 4.37e-03, grad_scale: 8.0 2023-04-02 13:55:46,449 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=124656.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 13:56:16,352 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=124681.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 13:56:20,585 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4945, 1.3417, 1.3898, 1.7731, 1.5715, 1.6016, 1.6625, 1.4957], device='cuda:1'), covar=tensor([0.0661, 0.0763, 0.0808, 0.0615, 0.0867, 0.0688, 0.0837, 0.0614], device='cuda:1'), in_proj_covar=tensor([0.0209, 0.0220, 0.0225, 0.0243, 0.0227, 0.0211, 0.0189, 0.0205], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-02 13:56:44,197 INFO [train.py:903] (1/4) Epoch 19, batch 1800, loss[loss=0.1958, simple_loss=0.2609, pruned_loss=0.06531, over 19725.00 frames. ], tot_loss[loss=0.2159, simple_loss=0.2945, pruned_loss=0.06863, over 3793639.27 frames. ], batch size: 45, lr: 4.37e-03, grad_scale: 8.0 2023-04-02 13:56:51,811 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=124710.0, num_to_drop=1, layers_to_drop={0} 2023-04-02 13:56:54,991 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.267e+02 5.111e+02 6.286e+02 7.731e+02 1.656e+03, threshold=1.257e+03, percent-clipped=2.0 2023-04-02 13:57:38,759 WARNING [train.py:1073] (1/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9_rvb from training. Duration: 25.0944375 2023-04-02 13:57:39,035 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=124749.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 13:57:40,117 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([5.0841, 1.9694, 2.0781, 1.7118, 4.6411, 1.1935, 2.8530, 5.0400], device='cuda:1'), covar=tensor([0.0402, 0.2482, 0.2489, 0.1949, 0.0679, 0.2665, 0.1240, 0.0149], device='cuda:1'), in_proj_covar=tensor([0.0398, 0.0357, 0.0377, 0.0342, 0.0366, 0.0349, 0.0369, 0.0388], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 13:57:44,335 INFO [train.py:903] (1/4) Epoch 19, batch 1850, loss[loss=0.1942, simple_loss=0.2752, pruned_loss=0.05666, over 19860.00 frames. ], tot_loss[loss=0.2146, simple_loss=0.2931, pruned_loss=0.06809, over 3799532.53 frames. ], batch size: 52, lr: 4.37e-03, grad_scale: 8.0 2023-04-02 13:58:03,198 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=124769.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 13:58:17,475 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9_rvb from training. Duration: 27.8166875 2023-04-02 13:58:33,195 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=124794.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 13:58:33,383 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=124794.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 13:58:34,536 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=124795.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 13:58:46,482 INFO [train.py:903] (1/4) Epoch 19, batch 1900, loss[loss=0.2366, simple_loss=0.3133, pruned_loss=0.07996, over 19493.00 frames. ], tot_loss[loss=0.2144, simple_loss=0.293, pruned_loss=0.06787, over 3798537.95 frames. ], batch size: 64, lr: 4.37e-03, grad_scale: 8.0 2023-04-02 13:58:56,687 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.301e+02 4.862e+02 5.969e+02 7.822e+02 1.490e+03, threshold=1.194e+03, percent-clipped=1.0 2023-04-02 13:59:01,911 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 2023-04-02 13:59:05,754 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=124820.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 13:59:07,778 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9_rvb from training. Duration: 27.02225 2023-04-02 13:59:11,238 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=124825.0, num_to_drop=1, layers_to_drop={0} 2023-04-02 13:59:26,846 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=124838.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 13:59:32,012 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 2023-04-02 13:59:46,593 INFO [train.py:903] (1/4) Epoch 19, batch 1950, loss[loss=0.1867, simple_loss=0.2714, pruned_loss=0.05098, over 18000.00 frames. ], tot_loss[loss=0.2134, simple_loss=0.2923, pruned_loss=0.06726, over 3811101.18 frames. ], batch size: 83, lr: 4.37e-03, grad_scale: 8.0 2023-04-02 14:00:47,544 INFO [train.py:903] (1/4) Epoch 19, batch 2000, loss[loss=0.1949, simple_loss=0.2751, pruned_loss=0.05735, over 19846.00 frames. ], tot_loss[loss=0.2133, simple_loss=0.2924, pruned_loss=0.06709, over 3816682.56 frames. ], batch size: 52, lr: 4.37e-03, grad_scale: 8.0 2023-04-02 14:00:54,354 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=124909.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 14:00:58,644 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.950e+02 4.950e+02 6.179e+02 7.428e+02 1.573e+03, threshold=1.236e+03, percent-clipped=5.0 2023-04-02 14:01:07,234 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.12 vs. limit=2.0 2023-04-02 14:01:45,078 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_rvb from training. Duration: 26.8349375 2023-04-02 14:01:48,486 INFO [train.py:903] (1/4) Epoch 19, batch 2050, loss[loss=0.1926, simple_loss=0.2765, pruned_loss=0.05441, over 19815.00 frames. ], tot_loss[loss=0.2132, simple_loss=0.2921, pruned_loss=0.06713, over 3824330.48 frames. ], batch size: 49, lr: 4.37e-03, grad_scale: 8.0 2023-04-02 14:02:06,442 WARNING [train.py:1073] (1/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 2023-04-02 14:02:07,323 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 2023-04-02 14:02:25,195 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 2023-04-02 14:02:44,755 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=125000.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 14:02:50,765 INFO [train.py:903] (1/4) Epoch 19, batch 2100, loss[loss=0.2149, simple_loss=0.302, pruned_loss=0.06392, over 19602.00 frames. ], tot_loss[loss=0.2132, simple_loss=0.2919, pruned_loss=0.06724, over 3832314.71 frames. ], batch size: 61, lr: 4.37e-03, grad_scale: 8.0 2023-04-02 14:02:52,314 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=125005.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 14:02:52,529 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.51 vs. limit=2.0 2023-04-02 14:03:00,992 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.657e+02 4.808e+02 5.766e+02 7.881e+02 2.968e+03, threshold=1.153e+03, percent-clipped=4.0 2023-04-02 14:03:15,326 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=125025.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 14:03:15,688 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9067, 1.9983, 2.3062, 2.6263, 1.8705, 2.4510, 2.3880, 2.1448], device='cuda:1'), covar=tensor([0.4243, 0.3850, 0.1792, 0.2086, 0.3909, 0.2027, 0.4382, 0.3194], device='cuda:1'), in_proj_covar=tensor([0.0869, 0.0927, 0.0696, 0.0912, 0.0853, 0.0786, 0.0824, 0.0763], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 14:03:18,741 WARNING [train.py:1073] (1/4) Exclude cut with ID 453-131332-0000-131866_sp0.9_rvb from training. Duration: 25.3333125 2023-04-02 14:03:19,124 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.6835, 2.4274, 2.2525, 2.8921, 2.3339, 2.2980, 2.1236, 2.6537], device='cuda:1'), covar=tensor([0.0848, 0.1512, 0.1364, 0.0963, 0.1384, 0.0491, 0.1294, 0.0614], device='cuda:1'), in_proj_covar=tensor([0.0268, 0.0356, 0.0307, 0.0250, 0.0299, 0.0249, 0.0300, 0.0251], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 14:03:22,181 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=125030.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 14:03:30,063 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6359, 1.7275, 2.0651, 1.8662, 3.0369, 2.4077, 3.2090, 1.9469], device='cuda:1'), covar=tensor([0.2346, 0.4045, 0.2682, 0.1892, 0.1609, 0.2147, 0.1597, 0.3679], device='cuda:1'), in_proj_covar=tensor([0.0515, 0.0624, 0.0684, 0.0467, 0.0614, 0.0518, 0.0653, 0.0532], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 14:03:39,548 WARNING [train.py:1073] (1/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9_rvb from training. Duration: 26.6166875 2023-04-02 14:03:49,805 INFO [train.py:903] (1/4) Epoch 19, batch 2150, loss[loss=0.1956, simple_loss=0.2663, pruned_loss=0.06248, over 19401.00 frames. ], tot_loss[loss=0.2119, simple_loss=0.2906, pruned_loss=0.06653, over 3840487.53 frames. ], batch size: 48, lr: 4.37e-03, grad_scale: 8.0 2023-04-02 14:04:22,809 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=125081.0, num_to_drop=1, layers_to_drop={0} 2023-04-02 14:04:50,319 INFO [train.py:903] (1/4) Epoch 19, batch 2200, loss[loss=0.2401, simple_loss=0.3149, pruned_loss=0.08258, over 19512.00 frames. ], tot_loss[loss=0.2118, simple_loss=0.291, pruned_loss=0.06635, over 3837356.28 frames. ], batch size: 56, lr: 4.37e-03, grad_scale: 8.0 2023-04-02 14:04:53,031 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=125106.0, num_to_drop=1, layers_to_drop={0} 2023-04-02 14:05:01,344 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.808e+02 4.810e+02 5.592e+02 6.977e+02 1.826e+03, threshold=1.118e+03, percent-clipped=4.0 2023-04-02 14:05:04,007 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=125115.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 14:05:34,535 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=125140.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 14:05:40,924 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.42 vs. limit=2.0 2023-04-02 14:05:41,574 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=125146.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 14:05:51,113 INFO [train.py:903] (1/4) Epoch 19, batch 2250, loss[loss=0.2453, simple_loss=0.3232, pruned_loss=0.08374, over 19526.00 frames. ], tot_loss[loss=0.2114, simple_loss=0.2906, pruned_loss=0.06607, over 3842894.92 frames. ], batch size: 56, lr: 4.36e-03, grad_scale: 8.0 2023-04-02 14:05:52,656 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.4088, 2.3800, 2.1737, 2.7620, 2.2340, 2.2321, 2.2163, 2.4477], device='cuda:1'), covar=tensor([0.1007, 0.1644, 0.1406, 0.0996, 0.1465, 0.0503, 0.1246, 0.0677], device='cuda:1'), in_proj_covar=tensor([0.0267, 0.0356, 0.0307, 0.0249, 0.0299, 0.0249, 0.0300, 0.0251], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 14:06:05,782 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=125165.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 14:06:25,692 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=125182.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 14:06:35,059 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=125190.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 14:06:52,194 INFO [train.py:903] (1/4) Epoch 19, batch 2300, loss[loss=0.192, simple_loss=0.2667, pruned_loss=0.05871, over 19298.00 frames. ], tot_loss[loss=0.2107, simple_loss=0.29, pruned_loss=0.06563, over 3845445.64 frames. ], batch size: 44, lr: 4.36e-03, grad_scale: 4.0 2023-04-02 14:07:04,356 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.756e+02 5.218e+02 6.176e+02 8.185e+02 2.110e+03, threshold=1.235e+03, percent-clipped=6.0 2023-04-02 14:07:06,714 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_rvb from training. Duration: 26.205 2023-04-02 14:07:52,837 INFO [train.py:903] (1/4) Epoch 19, batch 2350, loss[loss=0.188, simple_loss=0.2615, pruned_loss=0.05731, over 19719.00 frames. ], tot_loss[loss=0.2104, simple_loss=0.2898, pruned_loss=0.06553, over 3845139.04 frames. ], batch size: 46, lr: 4.36e-03, grad_scale: 4.0 2023-04-02 14:08:32,711 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_rvb from training. Duration: 25.775 2023-04-02 14:08:45,069 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=125297.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 14:08:49,231 WARNING [train.py:1073] (1/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9_rvb from training. Duration: 25.45 2023-04-02 14:08:52,806 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9988, 2.0315, 2.2352, 2.6397, 2.0271, 2.6432, 2.3242, 2.0413], device='cuda:1'), covar=tensor([0.3872, 0.3748, 0.1781, 0.2371, 0.4034, 0.1981, 0.4331, 0.3219], device='cuda:1'), in_proj_covar=tensor([0.0874, 0.0934, 0.0700, 0.0921, 0.0859, 0.0789, 0.0831, 0.0768], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 14:08:53,424 INFO [train.py:903] (1/4) Epoch 19, batch 2400, loss[loss=0.2207, simple_loss=0.3002, pruned_loss=0.07065, over 19766.00 frames. ], tot_loss[loss=0.2118, simple_loss=0.2907, pruned_loss=0.06647, over 3836078.60 frames. ], batch size: 56, lr: 4.36e-03, grad_scale: 8.0 2023-04-02 14:08:54,884 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([5.0817, 5.1199, 5.8793, 5.9186, 1.9018, 5.5430, 4.7198, 5.4852], device='cuda:1'), covar=tensor([0.1522, 0.0805, 0.0575, 0.0558, 0.6243, 0.0831, 0.0596, 0.1151], device='cuda:1'), in_proj_covar=tensor([0.0765, 0.0710, 0.0916, 0.0806, 0.0816, 0.0666, 0.0554, 0.0849], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-02 14:09:05,368 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.876e+02 4.925e+02 6.150e+02 7.515e+02 1.529e+03, threshold=1.230e+03, percent-clipped=3.0 2023-04-02 14:09:45,467 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.57 vs. limit=2.0 2023-04-02 14:09:53,256 INFO [train.py:903] (1/4) Epoch 19, batch 2450, loss[loss=0.1947, simple_loss=0.2745, pruned_loss=0.05743, over 19619.00 frames. ], tot_loss[loss=0.2131, simple_loss=0.2919, pruned_loss=0.06716, over 3840646.41 frames. ], batch size: 50, lr: 4.36e-03, grad_scale: 8.0 2023-04-02 14:10:14,719 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=125371.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 14:10:17,760 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=125374.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 14:10:41,865 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=125394.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 14:10:44,222 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=125396.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 14:10:44,251 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=125396.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 14:10:53,818 INFO [train.py:903] (1/4) Epoch 19, batch 2500, loss[loss=0.2135, simple_loss=0.2877, pruned_loss=0.06969, over 19585.00 frames. ], tot_loss[loss=0.2134, simple_loss=0.2921, pruned_loss=0.06737, over 3829493.69 frames. ], batch size: 52, lr: 4.36e-03, grad_scale: 8.0 2023-04-02 14:11:05,670 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.142e+02 4.898e+02 6.460e+02 8.264e+02 2.020e+03, threshold=1.292e+03, percent-clipped=4.0 2023-04-02 14:11:13,948 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=125421.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 14:11:35,789 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=125439.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 14:11:54,215 INFO [train.py:903] (1/4) Epoch 19, batch 2550, loss[loss=0.1568, simple_loss=0.2384, pruned_loss=0.03765, over 19404.00 frames. ], tot_loss[loss=0.2125, simple_loss=0.2915, pruned_loss=0.06678, over 3835614.31 frames. ], batch size: 47, lr: 4.36e-03, grad_scale: 8.0 2023-04-02 14:11:56,651 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([6.0729, 5.4963, 3.2397, 4.7894, 0.9702, 5.5288, 5.4546, 5.6384], device='cuda:1'), covar=tensor([0.0415, 0.0965, 0.1800, 0.0730, 0.4193, 0.0539, 0.0768, 0.0969], device='cuda:1'), in_proj_covar=tensor([0.0486, 0.0396, 0.0480, 0.0337, 0.0394, 0.0420, 0.0411, 0.0445], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 14:12:38,049 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=125490.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 14:12:47,107 WARNING [train.py:1073] (1/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 2023-04-02 14:12:53,797 INFO [train.py:903] (1/4) Epoch 19, batch 2600, loss[loss=0.2531, simple_loss=0.3228, pruned_loss=0.09167, over 19530.00 frames. ], tot_loss[loss=0.2118, simple_loss=0.2908, pruned_loss=0.06643, over 3833585.15 frames. ], batch size: 54, lr: 4.36e-03, grad_scale: 8.0 2023-04-02 14:13:05,879 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.683e+02 4.815e+02 5.841e+02 7.665e+02 1.339e+03, threshold=1.168e+03, percent-clipped=2.0 2023-04-02 14:13:21,465 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1888, 1.4526, 1.8375, 1.3151, 2.6467, 3.5788, 3.2857, 3.7701], device='cuda:1'), covar=tensor([0.1589, 0.3469, 0.3086, 0.2313, 0.0535, 0.0168, 0.0209, 0.0230], device='cuda:1'), in_proj_covar=tensor([0.0266, 0.0311, 0.0343, 0.0260, 0.0237, 0.0179, 0.0212, 0.0244], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-02 14:13:48,578 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=1.97 vs. limit=5.0 2023-04-02 14:13:52,895 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=125553.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 14:13:53,673 INFO [train.py:903] (1/4) Epoch 19, batch 2650, loss[loss=0.1708, simple_loss=0.2557, pruned_loss=0.0429, over 19850.00 frames. ], tot_loss[loss=0.2113, simple_loss=0.2905, pruned_loss=0.06602, over 3846332.79 frames. ], batch size: 52, lr: 4.36e-03, grad_scale: 8.0 2023-04-02 14:14:15,458 WARNING [train.py:1073] (1/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9_rvb from training. Duration: 27.25 2023-04-02 14:14:23,622 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=125578.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 14:14:54,493 INFO [train.py:903] (1/4) Epoch 19, batch 2700, loss[loss=0.2099, simple_loss=0.2917, pruned_loss=0.06406, over 19768.00 frames. ], tot_loss[loss=0.2114, simple_loss=0.2903, pruned_loss=0.06619, over 3829982.58 frames. ], batch size: 54, lr: 4.36e-03, grad_scale: 8.0 2023-04-02 14:14:55,976 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=125605.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 14:15:07,180 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.594e+02 4.817e+02 5.964e+02 7.468e+02 1.608e+03, threshold=1.193e+03, percent-clipped=5.0 2023-04-02 14:15:15,690 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=125621.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 14:15:31,469 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5952, 1.6371, 1.5438, 1.3818, 1.2742, 1.3888, 0.3397, 0.7112], device='cuda:1'), covar=tensor([0.0584, 0.0551, 0.0355, 0.0519, 0.0933, 0.0618, 0.1064, 0.0885], device='cuda:1'), in_proj_covar=tensor([0.0354, 0.0349, 0.0351, 0.0374, 0.0451, 0.0385, 0.0331, 0.0339], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 14:15:35,556 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.7336, 1.2396, 1.5642, 1.6038, 3.3161, 1.0192, 2.1746, 3.6965], device='cuda:1'), covar=tensor([0.0471, 0.2830, 0.2807, 0.1743, 0.0645, 0.2606, 0.1467, 0.0247], device='cuda:1'), in_proj_covar=tensor([0.0397, 0.0358, 0.0376, 0.0342, 0.0365, 0.0349, 0.0368, 0.0387], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 14:15:56,193 INFO [train.py:903] (1/4) Epoch 19, batch 2750, loss[loss=0.2308, simple_loss=0.3079, pruned_loss=0.07685, over 19692.00 frames. ], tot_loss[loss=0.2114, simple_loss=0.2903, pruned_loss=0.06631, over 3836510.86 frames. ], batch size: 59, lr: 4.36e-03, grad_scale: 8.0 2023-04-02 14:16:05,831 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6249, 1.6544, 1.9944, 1.8640, 2.9304, 2.4114, 2.9408, 1.5968], device='cuda:1'), covar=tensor([0.2500, 0.4355, 0.2699, 0.1940, 0.1446, 0.2150, 0.1503, 0.4034], device='cuda:1'), in_proj_covar=tensor([0.0518, 0.0625, 0.0688, 0.0470, 0.0614, 0.0521, 0.0655, 0.0534], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 14:16:55,538 INFO [train.py:903] (1/4) Epoch 19, batch 2800, loss[loss=0.1749, simple_loss=0.2529, pruned_loss=0.04847, over 19743.00 frames. ], tot_loss[loss=0.2119, simple_loss=0.291, pruned_loss=0.06641, over 3835653.61 frames. ], batch size: 46, lr: 4.35e-03, grad_scale: 8.0 2023-04-02 14:17:08,456 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.000e+02 5.188e+02 6.338e+02 7.813e+02 1.733e+03, threshold=1.268e+03, percent-clipped=8.0 2023-04-02 14:17:12,856 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=125718.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 14:17:37,018 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=125738.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 14:17:39,791 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 2023-04-02 14:17:42,813 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=125743.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 14:17:55,253 INFO [train.py:903] (1/4) Epoch 19, batch 2850, loss[loss=0.2135, simple_loss=0.2953, pruned_loss=0.06583, over 19661.00 frames. ], tot_loss[loss=0.2124, simple_loss=0.2914, pruned_loss=0.06666, over 3812992.92 frames. ], batch size: 58, lr: 4.35e-03, grad_scale: 8.0 2023-04-02 14:18:25,401 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([0.9729, 1.1546, 1.4676, 0.6180, 1.8974, 2.1574, 1.9586, 2.3183], device='cuda:1'), covar=tensor([0.1431, 0.3464, 0.3000, 0.2686, 0.0775, 0.0389, 0.0343, 0.0408], device='cuda:1'), in_proj_covar=tensor([0.0266, 0.0312, 0.0343, 0.0260, 0.0238, 0.0179, 0.0212, 0.0243], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-02 14:18:30,689 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=125783.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 14:18:56,093 INFO [train.py:903] (1/4) Epoch 19, batch 2900, loss[loss=0.2416, simple_loss=0.3171, pruned_loss=0.08301, over 19463.00 frames. ], tot_loss[loss=0.2114, simple_loss=0.2905, pruned_loss=0.06616, over 3807153.27 frames. ], batch size: 70, lr: 4.35e-03, grad_scale: 8.0 2023-04-02 14:18:56,106 WARNING [train.py:1073] (1/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 2023-04-02 14:19:01,449 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.9449, 4.4709, 2.6632, 3.9348, 0.9501, 4.4756, 4.3682, 4.4455], device='cuda:1'), covar=tensor([0.0591, 0.0978, 0.2193, 0.0880, 0.4081, 0.0655, 0.0806, 0.1109], device='cuda:1'), in_proj_covar=tensor([0.0487, 0.0398, 0.0481, 0.0338, 0.0395, 0.0419, 0.0410, 0.0446], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 14:19:09,043 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.234e+02 4.625e+02 5.511e+02 7.350e+02 1.619e+03, threshold=1.102e+03, percent-clipped=3.0 2023-04-02 14:19:31,524 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=125833.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 14:19:55,779 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=125853.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 14:19:56,475 INFO [train.py:903] (1/4) Epoch 19, batch 2950, loss[loss=0.2239, simple_loss=0.3007, pruned_loss=0.07353, over 19520.00 frames. ], tot_loss[loss=0.2107, simple_loss=0.29, pruned_loss=0.06571, over 3813013.58 frames. ], batch size: 54, lr: 4.35e-03, grad_scale: 8.0 2023-04-02 14:20:05,693 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=125861.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 14:20:35,694 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=125886.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 14:20:50,759 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=125898.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 14:20:57,388 INFO [train.py:903] (1/4) Epoch 19, batch 3000, loss[loss=0.1897, simple_loss=0.277, pruned_loss=0.0512, over 18344.00 frames. ], tot_loss[loss=0.2113, simple_loss=0.2902, pruned_loss=0.06623, over 3802721.37 frames. ], batch size: 84, lr: 4.35e-03, grad_scale: 8.0 2023-04-02 14:20:57,388 INFO [train.py:928] (1/4) Computing validation loss 2023-04-02 14:21:10,733 INFO [train.py:937] (1/4) Epoch 19, validation: loss=0.1696, simple_loss=0.2702, pruned_loss=0.03451, over 944034.00 frames. 2023-04-02 14:21:10,733 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18821MB 2023-04-02 14:21:10,808 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_rvb from training. Duration: 29.735 2023-04-02 14:21:24,049 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.456e+02 4.999e+02 6.816e+02 8.693e+02 1.814e+03, threshold=1.363e+03, percent-clipped=12.0 2023-04-02 14:22:11,565 INFO [train.py:903] (1/4) Epoch 19, batch 3050, loss[loss=0.1901, simple_loss=0.2675, pruned_loss=0.05636, over 19477.00 frames. ], tot_loss[loss=0.2115, simple_loss=0.2906, pruned_loss=0.06621, over 3814568.48 frames. ], batch size: 49, lr: 4.35e-03, grad_scale: 8.0 2023-04-02 14:22:24,978 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=125965.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 14:23:13,567 INFO [train.py:903] (1/4) Epoch 19, batch 3100, loss[loss=0.1846, simple_loss=0.266, pruned_loss=0.05163, over 19750.00 frames. ], tot_loss[loss=0.2109, simple_loss=0.2899, pruned_loss=0.06599, over 3816619.43 frames. ], batch size: 51, lr: 4.35e-03, grad_scale: 8.0 2023-04-02 14:23:14,310 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 2023-04-02 14:23:26,817 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.089e+02 4.651e+02 5.591e+02 6.916e+02 1.279e+03, threshold=1.118e+03, percent-clipped=0.0 2023-04-02 14:23:28,010 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=126016.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 14:24:13,739 INFO [train.py:903] (1/4) Epoch 19, batch 3150, loss[loss=0.1903, simple_loss=0.2757, pruned_loss=0.05243, over 19777.00 frames. ], tot_loss[loss=0.2122, simple_loss=0.2908, pruned_loss=0.06681, over 3815616.65 frames. ], batch size: 54, lr: 4.35e-03, grad_scale: 8.0 2023-04-02 14:24:40,340 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 2023-04-02 14:24:45,987 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=126080.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 14:24:48,473 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 2023-04-02 14:24:53,524 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=126087.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 14:24:55,925 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=126089.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 14:25:14,240 INFO [train.py:903] (1/4) Epoch 19, batch 3200, loss[loss=0.2306, simple_loss=0.3146, pruned_loss=0.07327, over 19667.00 frames. ], tot_loss[loss=0.2124, simple_loss=0.2908, pruned_loss=0.06698, over 3792220.71 frames. ], batch size: 58, lr: 4.35e-03, grad_scale: 8.0 2023-04-02 14:25:21,256 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=126109.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 14:25:26,849 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=126114.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 14:25:27,594 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.283e+02 4.817e+02 6.226e+02 7.515e+02 1.545e+03, threshold=1.245e+03, percent-clipped=7.0 2023-04-02 14:25:51,111 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=126134.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 14:26:15,152 INFO [train.py:903] (1/4) Epoch 19, batch 3250, loss[loss=0.2489, simple_loss=0.3288, pruned_loss=0.08451, over 19366.00 frames. ], tot_loss[loss=0.2123, simple_loss=0.2911, pruned_loss=0.06674, over 3799276.60 frames. ], batch size: 66, lr: 4.35e-03, grad_scale: 8.0 2023-04-02 14:26:15,591 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=126154.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 14:26:32,437 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.7148, 1.5588, 1.6538, 2.1956, 1.7090, 1.9862, 2.0208, 1.8582], device='cuda:1'), covar=tensor([0.0792, 0.0887, 0.0964, 0.0751, 0.0839, 0.0745, 0.0875, 0.0638], device='cuda:1'), in_proj_covar=tensor([0.0211, 0.0221, 0.0227, 0.0245, 0.0227, 0.0212, 0.0189, 0.0205], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-02 14:26:46,076 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=126179.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 14:27:14,828 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=126202.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 14:27:16,668 INFO [train.py:903] (1/4) Epoch 19, batch 3300, loss[loss=0.2273, simple_loss=0.3095, pruned_loss=0.0725, over 19548.00 frames. ], tot_loss[loss=0.2121, simple_loss=0.291, pruned_loss=0.06665, over 3808634.51 frames. ], batch size: 56, lr: 4.35e-03, grad_scale: 8.0 2023-04-02 14:27:20,145 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 2023-04-02 14:27:30,249 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.439e+02 5.162e+02 6.410e+02 7.971e+02 2.422e+03, threshold=1.282e+03, percent-clipped=4.0 2023-04-02 14:27:36,976 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.9801, 4.5234, 2.8110, 3.9059, 0.9807, 4.3896, 4.3193, 4.4680], device='cuda:1'), covar=tensor([0.0559, 0.0871, 0.1896, 0.0896, 0.4105, 0.0720, 0.0884, 0.0984], device='cuda:1'), in_proj_covar=tensor([0.0488, 0.0394, 0.0479, 0.0337, 0.0394, 0.0420, 0.0410, 0.0446], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 14:28:17,444 INFO [train.py:903] (1/4) Epoch 19, batch 3350, loss[loss=0.2263, simple_loss=0.3064, pruned_loss=0.07312, over 19318.00 frames. ], tot_loss[loss=0.2115, simple_loss=0.2904, pruned_loss=0.06629, over 3826906.12 frames. ], batch size: 66, lr: 4.35e-03, grad_scale: 8.0 2023-04-02 14:29:08,294 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.2732, 2.1349, 1.9514, 1.7679, 1.6037, 1.7678, 0.5443, 1.2796], device='cuda:1'), covar=tensor([0.0601, 0.0606, 0.0482, 0.0844, 0.1140, 0.0942, 0.1300, 0.0981], device='cuda:1'), in_proj_covar=tensor([0.0357, 0.0350, 0.0352, 0.0377, 0.0453, 0.0385, 0.0332, 0.0341], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 14:29:17,950 INFO [train.py:903] (1/4) Epoch 19, batch 3400, loss[loss=0.2504, simple_loss=0.3208, pruned_loss=0.09002, over 17464.00 frames. ], tot_loss[loss=0.2125, simple_loss=0.2909, pruned_loss=0.06712, over 3812886.79 frames. ], batch size: 101, lr: 4.34e-03, grad_scale: 8.0 2023-04-02 14:29:25,161 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3798, 1.4403, 1.6410, 1.5972, 2.2015, 2.0594, 2.2459, 0.9439], device='cuda:1'), covar=tensor([0.2391, 0.4101, 0.2559, 0.1914, 0.1526, 0.2096, 0.1403, 0.4397], device='cuda:1'), in_proj_covar=tensor([0.0519, 0.0627, 0.0689, 0.0474, 0.0617, 0.0521, 0.0656, 0.0536], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 14:29:31,307 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.222e+02 5.180e+02 6.014e+02 8.021e+02 1.733e+03, threshold=1.203e+03, percent-clipped=4.0 2023-04-02 14:29:57,168 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=126336.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 14:30:18,213 INFO [train.py:903] (1/4) Epoch 19, batch 3450, loss[loss=0.2759, simple_loss=0.3449, pruned_loss=0.1034, over 19282.00 frames. ], tot_loss[loss=0.2115, simple_loss=0.29, pruned_loss=0.06652, over 3818593.32 frames. ], batch size: 66, lr: 4.34e-03, grad_scale: 4.0 2023-04-02 14:30:22,535 WARNING [train.py:1073] (1/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9_rvb from training. Duration: 25.2444375 2023-04-02 14:30:27,185 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=126360.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 14:30:28,589 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=126361.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 14:31:13,300 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=126398.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 14:31:20,718 INFO [train.py:903] (1/4) Epoch 19, batch 3500, loss[loss=0.1877, simple_loss=0.2756, pruned_loss=0.04991, over 19768.00 frames. ], tot_loss[loss=0.2107, simple_loss=0.2897, pruned_loss=0.06588, over 3830163.78 frames. ], batch size: 54, lr: 4.34e-03, grad_scale: 4.0 2023-04-02 14:31:34,958 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.238e+02 4.908e+02 6.053e+02 7.325e+02 1.346e+03, threshold=1.211e+03, percent-clipped=1.0 2023-04-02 14:32:21,707 INFO [train.py:903] (1/4) Epoch 19, batch 3550, loss[loss=0.2063, simple_loss=0.2859, pruned_loss=0.06333, over 19674.00 frames. ], tot_loss[loss=0.2124, simple_loss=0.2911, pruned_loss=0.06685, over 3817436.16 frames. ], batch size: 53, lr: 4.34e-03, grad_scale: 4.0 2023-04-02 14:32:23,061 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1381, 1.4365, 1.8820, 1.2963, 2.7364, 3.6996, 3.4340, 3.8867], device='cuda:1'), covar=tensor([0.1686, 0.3627, 0.2997, 0.2308, 0.0560, 0.0166, 0.0204, 0.0250], device='cuda:1'), in_proj_covar=tensor([0.0267, 0.0313, 0.0344, 0.0260, 0.0238, 0.0181, 0.0214, 0.0245], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-02 14:32:26,679 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=126458.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 14:32:38,974 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=126468.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 14:32:48,151 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=126475.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 14:32:57,924 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=126483.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 14:33:21,951 INFO [train.py:903] (1/4) Epoch 19, batch 3600, loss[loss=0.2554, simple_loss=0.3289, pruned_loss=0.09094, over 19521.00 frames. ], tot_loss[loss=0.2123, simple_loss=0.2911, pruned_loss=0.06679, over 3830028.05 frames. ], batch size: 54, lr: 4.34e-03, grad_scale: 8.0 2023-04-02 14:33:37,204 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.060e+02 4.926e+02 5.826e+02 7.456e+02 2.258e+03, threshold=1.165e+03, percent-clipped=2.0 2023-04-02 14:34:04,272 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6134, 2.3851, 1.6836, 1.5688, 2.1929, 1.2920, 1.4560, 2.0291], device='cuda:1'), covar=tensor([0.1030, 0.0740, 0.1121, 0.0836, 0.0528, 0.1320, 0.0734, 0.0470], device='cuda:1'), in_proj_covar=tensor([0.0300, 0.0311, 0.0332, 0.0260, 0.0243, 0.0333, 0.0289, 0.0271], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 14:34:23,091 INFO [train.py:903] (1/4) Epoch 19, batch 3650, loss[loss=0.2062, simple_loss=0.2823, pruned_loss=0.06506, over 19680.00 frames. ], tot_loss[loss=0.2126, simple_loss=0.2913, pruned_loss=0.06692, over 3832232.41 frames. ], batch size: 53, lr: 4.34e-03, grad_scale: 8.0 2023-04-02 14:35:24,546 INFO [train.py:903] (1/4) Epoch 19, batch 3700, loss[loss=0.2363, simple_loss=0.2998, pruned_loss=0.08635, over 19762.00 frames. ], tot_loss[loss=0.2123, simple_loss=0.2908, pruned_loss=0.0669, over 3835084.01 frames. ], batch size: 47, lr: 4.34e-03, grad_scale: 8.0 2023-04-02 14:35:38,476 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.032e+02 5.326e+02 6.409e+02 8.349e+02 1.648e+03, threshold=1.282e+03, percent-clipped=7.0 2023-04-02 14:36:23,995 INFO [train.py:903] (1/4) Epoch 19, batch 3750, loss[loss=0.2463, simple_loss=0.3296, pruned_loss=0.08155, over 18803.00 frames. ], tot_loss[loss=0.2116, simple_loss=0.2903, pruned_loss=0.06647, over 3836175.62 frames. ], batch size: 74, lr: 4.34e-03, grad_scale: 8.0 2023-04-02 14:37:25,130 INFO [train.py:903] (1/4) Epoch 19, batch 3800, loss[loss=0.1999, simple_loss=0.2918, pruned_loss=0.05398, over 19653.00 frames. ], tot_loss[loss=0.2127, simple_loss=0.2915, pruned_loss=0.06694, over 3812060.71 frames. ], batch size: 55, lr: 4.34e-03, grad_scale: 8.0 2023-04-02 14:37:40,984 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.076e+02 4.757e+02 5.693e+02 7.302e+02 1.543e+03, threshold=1.139e+03, percent-clipped=1.0 2023-04-02 14:37:57,317 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9_rvb from training. Duration: 29.1166875 2023-04-02 14:37:58,809 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=126731.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 14:38:12,086 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=126742.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 14:38:13,551 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.1812, 2.2377, 2.4039, 3.0140, 2.2051, 2.9595, 2.5075, 2.1096], device='cuda:1'), covar=tensor([0.4368, 0.3945, 0.1874, 0.2509, 0.4385, 0.2077, 0.4754, 0.3465], device='cuda:1'), in_proj_covar=tensor([0.0873, 0.0932, 0.0699, 0.0918, 0.0856, 0.0789, 0.0825, 0.0766], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 14:38:26,703 INFO [train.py:903] (1/4) Epoch 19, batch 3850, loss[loss=0.2263, simple_loss=0.3061, pruned_loss=0.07327, over 17530.00 frames. ], tot_loss[loss=0.2116, simple_loss=0.2907, pruned_loss=0.06631, over 3812259.37 frames. ], batch size: 101, lr: 4.34e-03, grad_scale: 8.0 2023-04-02 14:38:30,342 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=126756.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 14:38:46,100 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9767, 1.9503, 1.7844, 1.6133, 1.4701, 1.5746, 0.4409, 0.9196], device='cuda:1'), covar=tensor([0.0557, 0.0532, 0.0387, 0.0619, 0.1122, 0.0757, 0.1179, 0.0997], device='cuda:1'), in_proj_covar=tensor([0.0355, 0.0347, 0.0353, 0.0375, 0.0453, 0.0386, 0.0331, 0.0340], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 14:39:28,455 INFO [train.py:903] (1/4) Epoch 19, batch 3900, loss[loss=0.2057, simple_loss=0.2904, pruned_loss=0.06051, over 19613.00 frames. ], tot_loss[loss=0.2112, simple_loss=0.2905, pruned_loss=0.06594, over 3817487.89 frames. ], batch size: 57, lr: 4.34e-03, grad_scale: 8.0 2023-04-02 14:39:37,773 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=126812.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 14:39:40,355 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 2023-04-02 14:39:42,910 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.374e+02 5.082e+02 6.454e+02 7.734e+02 3.345e+03, threshold=1.291e+03, percent-clipped=6.0 2023-04-02 14:40:04,522 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=126834.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 14:40:21,639 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9911, 2.0535, 2.3363, 2.6228, 1.8521, 2.4868, 2.3130, 2.1060], device='cuda:1'), covar=tensor([0.4134, 0.3996, 0.1921, 0.2468, 0.4249, 0.2119, 0.4725, 0.3391], device='cuda:1'), in_proj_covar=tensor([0.0871, 0.0929, 0.0696, 0.0915, 0.0853, 0.0787, 0.0822, 0.0763], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 14:40:29,132 INFO [train.py:903] (1/4) Epoch 19, batch 3950, loss[loss=0.1817, simple_loss=0.2566, pruned_loss=0.05342, over 19369.00 frames. ], tot_loss[loss=0.2097, simple_loss=0.2893, pruned_loss=0.06508, over 3823221.38 frames. ], batch size: 47, lr: 4.34e-03, grad_scale: 8.0 2023-04-02 14:40:33,527 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=126857.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 14:40:35,262 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 2023-04-02 14:41:13,516 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3750, 1.4215, 1.8810, 1.6186, 2.4326, 2.0615, 2.5163, 1.0375], device='cuda:1'), covar=tensor([0.2671, 0.4563, 0.2636, 0.2155, 0.1786, 0.2459, 0.1680, 0.4773], device='cuda:1'), in_proj_covar=tensor([0.0518, 0.0626, 0.0690, 0.0473, 0.0616, 0.0520, 0.0656, 0.0535], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 14:41:29,517 INFO [train.py:903] (1/4) Epoch 19, batch 4000, loss[loss=0.2379, simple_loss=0.3218, pruned_loss=0.07702, over 19737.00 frames. ], tot_loss[loss=0.2099, simple_loss=0.2895, pruned_loss=0.06515, over 3815342.71 frames. ], batch size: 63, lr: 4.33e-03, grad_scale: 8.0 2023-04-02 14:41:43,557 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.579e+02 4.980e+02 6.258e+02 9.023e+02 1.716e+03, threshold=1.252e+03, percent-clipped=4.0 2023-04-02 14:41:47,754 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.0272, 3.6730, 2.5924, 3.2341, 0.8820, 3.5605, 3.5073, 3.5610], device='cuda:1'), covar=tensor([0.0776, 0.1070, 0.1852, 0.0972, 0.3971, 0.0781, 0.0995, 0.1284], device='cuda:1'), in_proj_covar=tensor([0.0486, 0.0394, 0.0480, 0.0338, 0.0396, 0.0419, 0.0409, 0.0444], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 14:41:57,631 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=126927.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 14:42:16,842 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1_rvb from training. Duration: 0.7545625 2023-04-02 14:42:30,357 INFO [train.py:903] (1/4) Epoch 19, batch 4050, loss[loss=0.2158, simple_loss=0.2941, pruned_loss=0.06878, over 19587.00 frames. ], tot_loss[loss=0.2103, simple_loss=0.2894, pruned_loss=0.06564, over 3825492.93 frames. ], batch size: 61, lr: 4.33e-03, grad_scale: 8.0 2023-04-02 14:43:26,561 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1790, 1.1216, 1.4860, 1.2800, 2.6695, 3.5519, 3.3547, 3.7822], device='cuda:1'), covar=tensor([0.1729, 0.4053, 0.3681, 0.2451, 0.0631, 0.0192, 0.0225, 0.0280], device='cuda:1'), in_proj_covar=tensor([0.0268, 0.0315, 0.0345, 0.0261, 0.0238, 0.0181, 0.0214, 0.0245], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-02 14:43:30,605 INFO [train.py:903] (1/4) Epoch 19, batch 4100, loss[loss=0.2055, simple_loss=0.291, pruned_loss=0.06003, over 17518.00 frames. ], tot_loss[loss=0.2096, simple_loss=0.2887, pruned_loss=0.06518, over 3828075.40 frames. ], batch size: 101, lr: 4.33e-03, grad_scale: 8.0 2023-04-02 14:43:45,830 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.326e+02 4.699e+02 5.681e+02 7.096e+02 1.300e+03, threshold=1.136e+03, percent-clipped=1.0 2023-04-02 14:44:06,976 WARNING [train.py:1073] (1/4) Exclude cut with ID 497-129325-0061-9566_sp1.1_rvb from training. Duration: 0.97725 2023-04-02 14:44:31,634 INFO [train.py:903] (1/4) Epoch 19, batch 4150, loss[loss=0.2197, simple_loss=0.2971, pruned_loss=0.07116, over 19672.00 frames. ], tot_loss[loss=0.2115, simple_loss=0.2908, pruned_loss=0.06613, over 3833626.47 frames. ], batch size: 60, lr: 4.33e-03, grad_scale: 8.0 2023-04-02 14:44:42,403 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.81 vs. limit=5.0 2023-04-02 14:45:04,799 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.7316, 4.1945, 4.4512, 4.4402, 1.7003, 4.1971, 3.6938, 4.1489], device='cuda:1'), covar=tensor([0.1611, 0.1018, 0.0587, 0.0655, 0.5968, 0.0797, 0.0631, 0.1091], device='cuda:1'), in_proj_covar=tensor([0.0770, 0.0716, 0.0921, 0.0808, 0.0821, 0.0675, 0.0556, 0.0858], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-02 14:45:23,652 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3866, 1.4822, 1.8798, 1.7047, 2.6834, 2.0708, 2.7367, 1.2315], device='cuda:1'), covar=tensor([0.2699, 0.4554, 0.2881, 0.2066, 0.1588, 0.2551, 0.1714, 0.4605], device='cuda:1'), in_proj_covar=tensor([0.0520, 0.0627, 0.0691, 0.0472, 0.0616, 0.0520, 0.0659, 0.0536], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 14:45:32,545 INFO [train.py:903] (1/4) Epoch 19, batch 4200, loss[loss=0.2057, simple_loss=0.2937, pruned_loss=0.05884, over 19469.00 frames. ], tot_loss[loss=0.2112, simple_loss=0.2904, pruned_loss=0.06599, over 3831137.96 frames. ], batch size: 49, lr: 4.33e-03, grad_scale: 8.0 2023-04-02 14:45:35,885 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 2023-04-02 14:45:43,054 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=127113.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 14:45:46,906 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.729e+02 4.915e+02 5.762e+02 6.825e+02 1.362e+03, threshold=1.152e+03, percent-clipped=8.0 2023-04-02 14:46:14,867 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=127138.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 14:46:32,718 INFO [train.py:903] (1/4) Epoch 19, batch 4250, loss[loss=0.1773, simple_loss=0.252, pruned_loss=0.05124, over 19745.00 frames. ], tot_loss[loss=0.2106, simple_loss=0.2901, pruned_loss=0.06553, over 3831990.44 frames. ], batch size: 46, lr: 4.33e-03, grad_scale: 8.0 2023-04-02 14:46:33,001 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.8154, 0.8334, 1.0098, 1.0316, 1.6062, 0.8363, 1.5483, 1.7656], device='cuda:1'), covar=tensor([0.0576, 0.2229, 0.2110, 0.1237, 0.0651, 0.1630, 0.1182, 0.0516], device='cuda:1'), in_proj_covar=tensor([0.0393, 0.0357, 0.0375, 0.0340, 0.0365, 0.0346, 0.0366, 0.0385], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 14:46:50,065 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9_rvb from training. Duration: 29.816625 2023-04-02 14:47:01,529 WARNING [train.py:1073] (1/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 2023-04-02 14:47:03,645 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=127178.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 14:47:09,518 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=127183.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 14:47:34,698 INFO [train.py:903] (1/4) Epoch 19, batch 4300, loss[loss=0.1669, simple_loss=0.2452, pruned_loss=0.04432, over 19744.00 frames. ], tot_loss[loss=0.2106, simple_loss=0.2899, pruned_loss=0.06566, over 3814785.22 frames. ], batch size: 46, lr: 4.33e-03, grad_scale: 8.0 2023-04-02 14:47:40,369 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=127208.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 14:47:50,092 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.800e+02 4.893e+02 5.914e+02 7.996e+02 1.682e+03, threshold=1.183e+03, percent-clipped=7.0 2023-04-02 14:48:17,679 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.1318, 1.9977, 1.8449, 1.6669, 1.5101, 1.6011, 0.5469, 1.0717], device='cuda:1'), covar=tensor([0.0524, 0.0594, 0.0435, 0.0734, 0.1165, 0.0884, 0.1243, 0.0989], device='cuda:1'), in_proj_covar=tensor([0.0353, 0.0345, 0.0350, 0.0374, 0.0452, 0.0381, 0.0328, 0.0336], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 14:48:28,059 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 2023-04-02 14:48:35,537 INFO [train.py:903] (1/4) Epoch 19, batch 4350, loss[loss=0.2324, simple_loss=0.3131, pruned_loss=0.07582, over 18921.00 frames. ], tot_loss[loss=0.2118, simple_loss=0.2912, pruned_loss=0.06619, over 3802571.87 frames. ], batch size: 74, lr: 4.33e-03, grad_scale: 8.0 2023-04-02 14:48:51,129 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=127267.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 14:49:19,646 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.8378, 1.8932, 1.4531, 1.8235, 1.8246, 1.4707, 1.4487, 1.6937], device='cuda:1'), covar=tensor([0.1214, 0.1392, 0.1759, 0.1217, 0.1341, 0.0956, 0.1795, 0.0999], device='cuda:1'), in_proj_covar=tensor([0.0263, 0.0352, 0.0304, 0.0248, 0.0296, 0.0245, 0.0297, 0.0248], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 14:49:23,164 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=127293.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 14:49:26,472 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.5061, 1.0907, 1.3215, 1.0892, 2.1109, 0.9321, 1.9719, 2.3892], device='cuda:1'), covar=tensor([0.0918, 0.3092, 0.3056, 0.2086, 0.1186, 0.2417, 0.1255, 0.0565], device='cuda:1'), in_proj_covar=tensor([0.0393, 0.0357, 0.0375, 0.0341, 0.0366, 0.0348, 0.0367, 0.0385], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 14:49:35,821 INFO [train.py:903] (1/4) Epoch 19, batch 4400, loss[loss=0.1919, simple_loss=0.2749, pruned_loss=0.05447, over 19725.00 frames. ], tot_loss[loss=0.2117, simple_loss=0.2911, pruned_loss=0.06611, over 3824820.70 frames. ], batch size: 51, lr: 4.33e-03, grad_scale: 8.0 2023-04-02 14:49:36,053 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.3601, 3.9192, 2.6812, 3.5665, 0.8268, 3.8188, 3.7498, 3.8528], device='cuda:1'), covar=tensor([0.0696, 0.1017, 0.1888, 0.0862, 0.4076, 0.0772, 0.0912, 0.1134], device='cuda:1'), in_proj_covar=tensor([0.0487, 0.0393, 0.0484, 0.0337, 0.0397, 0.0420, 0.0410, 0.0446], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 14:49:49,568 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.177e+02 5.239e+02 6.175e+02 6.853e+02 1.670e+03, threshold=1.235e+03, percent-clipped=2.0 2023-04-02 14:50:02,113 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_rvb from training. Duration: 25.285 2023-04-02 14:50:12,008 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 2023-04-02 14:50:36,316 INFO [train.py:903] (1/4) Epoch 19, batch 4450, loss[loss=0.1952, simple_loss=0.271, pruned_loss=0.05968, over 19484.00 frames. ], tot_loss[loss=0.2121, simple_loss=0.2909, pruned_loss=0.06663, over 3815156.01 frames. ], batch size: 49, lr: 4.33e-03, grad_scale: 8.0 2023-04-02 14:51:03,362 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.3737, 3.9660, 2.7199, 3.5693, 0.8405, 3.8675, 3.7890, 3.8777], device='cuda:1'), covar=tensor([0.0679, 0.0973, 0.1816, 0.0800, 0.4137, 0.0728, 0.0912, 0.1110], device='cuda:1'), in_proj_covar=tensor([0.0487, 0.0393, 0.0482, 0.0337, 0.0397, 0.0419, 0.0410, 0.0445], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 14:51:38,004 INFO [train.py:903] (1/4) Epoch 19, batch 4500, loss[loss=0.2543, simple_loss=0.3292, pruned_loss=0.08969, over 17249.00 frames. ], tot_loss[loss=0.212, simple_loss=0.2911, pruned_loss=0.06648, over 3813540.42 frames. ], batch size: 101, lr: 4.33e-03, grad_scale: 8.0 2023-04-02 14:51:52,896 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.369e+02 5.116e+02 6.133e+02 7.767e+02 1.446e+03, threshold=1.227e+03, percent-clipped=3.0 2023-04-02 14:51:57,330 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.8177, 1.3657, 1.5743, 1.5486, 3.4155, 1.2705, 2.5228, 3.7979], device='cuda:1'), covar=tensor([0.0447, 0.2672, 0.2735, 0.1886, 0.0682, 0.2388, 0.1106, 0.0221], device='cuda:1'), in_proj_covar=tensor([0.0393, 0.0357, 0.0376, 0.0341, 0.0367, 0.0347, 0.0368, 0.0385], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 14:52:16,014 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6025, 1.7206, 1.9306, 2.0110, 1.5334, 1.9223, 1.9640, 1.7987], device='cuda:1'), covar=tensor([0.3887, 0.3304, 0.1817, 0.2088, 0.3448, 0.1896, 0.4582, 0.3137], device='cuda:1'), in_proj_covar=tensor([0.0873, 0.0930, 0.0698, 0.0916, 0.0855, 0.0790, 0.0826, 0.0763], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 14:52:28,080 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=127445.0, num_to_drop=1, layers_to_drop={1} 2023-04-02 14:52:39,720 INFO [train.py:903] (1/4) Epoch 19, batch 4550, loss[loss=0.2186, simple_loss=0.3059, pruned_loss=0.06564, over 19665.00 frames. ], tot_loss[loss=0.2122, simple_loss=0.2914, pruned_loss=0.06651, over 3821221.90 frames. ], batch size: 60, lr: 4.32e-03, grad_scale: 8.0 2023-04-02 14:52:48,345 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9_rvb from training. Duration: 28.72225 2023-04-02 14:53:11,995 WARNING [train.py:1073] (1/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 2023-04-02 14:53:40,574 INFO [train.py:903] (1/4) Epoch 19, batch 4600, loss[loss=0.2176, simple_loss=0.3005, pruned_loss=0.06739, over 19746.00 frames. ], tot_loss[loss=0.2126, simple_loss=0.2913, pruned_loss=0.06697, over 3825130.04 frames. ], batch size: 63, lr: 4.32e-03, grad_scale: 8.0 2023-04-02 14:53:48,019 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.13 vs. limit=2.0 2023-04-02 14:53:52,430 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.1096, 1.7300, 1.7537, 2.7370, 2.1440, 2.2698, 2.4087, 2.0498], device='cuda:1'), covar=tensor([0.0827, 0.0957, 0.1059, 0.0808, 0.0864, 0.0792, 0.0881, 0.0709], device='cuda:1'), in_proj_covar=tensor([0.0213, 0.0223, 0.0227, 0.0246, 0.0230, 0.0213, 0.0190, 0.0206], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-04-02 14:53:54,252 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.388e+02 5.018e+02 6.286e+02 8.427e+02 2.189e+03, threshold=1.257e+03, percent-clipped=8.0 2023-04-02 14:54:32,408 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=127547.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 14:54:34,722 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=127549.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 14:54:39,906 INFO [train.py:903] (1/4) Epoch 19, batch 4650, loss[loss=0.2311, simple_loss=0.3119, pruned_loss=0.07515, over 19522.00 frames. ], tot_loss[loss=0.2127, simple_loss=0.2915, pruned_loss=0.06691, over 3826596.88 frames. ], batch size: 56, lr: 4.32e-03, grad_scale: 8.0 2023-04-02 14:54:55,864 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_rvb from training. Duration: 0.92 2023-04-02 14:55:02,163 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9077, 2.0688, 2.2438, 2.6523, 1.9286, 2.4468, 2.2443, 1.9079], device='cuda:1'), covar=tensor([0.4546, 0.4034, 0.2072, 0.2535, 0.4335, 0.2351, 0.5112, 0.3767], device='cuda:1'), in_proj_covar=tensor([0.0879, 0.0937, 0.0702, 0.0924, 0.0861, 0.0796, 0.0831, 0.0768], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 14:55:05,384 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=127574.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 14:55:07,389 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_rvb from training. Duration: 0.83 2023-04-02 14:55:40,716 INFO [train.py:903] (1/4) Epoch 19, batch 4700, loss[loss=0.1981, simple_loss=0.2827, pruned_loss=0.05669, over 19791.00 frames. ], tot_loss[loss=0.2122, simple_loss=0.291, pruned_loss=0.06667, over 3830054.99 frames. ], batch size: 56, lr: 4.32e-03, grad_scale: 8.0 2023-04-02 14:55:41,052 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4035, 1.0942, 1.2787, 2.2463, 1.6513, 1.4049, 1.6396, 1.4499], device='cuda:1'), covar=tensor([0.1195, 0.1794, 0.1421, 0.0970, 0.1149, 0.1475, 0.1366, 0.1105], device='cuda:1'), in_proj_covar=tensor([0.0212, 0.0222, 0.0226, 0.0245, 0.0229, 0.0211, 0.0189, 0.0204], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-02 14:55:50,499 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=127611.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 14:55:55,863 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.385e+02 5.036e+02 6.203e+02 8.078e+02 1.735e+03, threshold=1.241e+03, percent-clipped=3.0 2023-04-02 14:56:02,614 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9_rvb from training. Duration: 0.92225 2023-04-02 14:56:41,689 INFO [train.py:903] (1/4) Epoch 19, batch 4750, loss[loss=0.1629, simple_loss=0.2432, pruned_loss=0.04127, over 19795.00 frames. ], tot_loss[loss=0.2126, simple_loss=0.2914, pruned_loss=0.06688, over 3829714.94 frames. ], batch size: 48, lr: 4.32e-03, grad_scale: 8.0 2023-04-02 14:57:13,617 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=127680.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 14:57:33,552 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.6032, 1.3335, 1.4683, 1.6040, 3.1956, 1.1185, 2.3003, 3.5770], device='cuda:1'), covar=tensor([0.0516, 0.2771, 0.2824, 0.1755, 0.0688, 0.2547, 0.1369, 0.0248], device='cuda:1'), in_proj_covar=tensor([0.0395, 0.0359, 0.0377, 0.0341, 0.0367, 0.0347, 0.0369, 0.0387], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 14:57:41,808 INFO [train.py:903] (1/4) Epoch 19, batch 4800, loss[loss=0.2177, simple_loss=0.2931, pruned_loss=0.0712, over 18130.00 frames. ], tot_loss[loss=0.2111, simple_loss=0.2899, pruned_loss=0.06615, over 3829761.70 frames. ], batch size: 83, lr: 4.32e-03, grad_scale: 8.0 2023-04-02 14:57:55,391 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.353e+02 4.990e+02 6.329e+02 8.030e+02 1.437e+03, threshold=1.266e+03, percent-clipped=1.0 2023-04-02 14:58:07,614 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=127726.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 14:58:40,155 INFO [train.py:903] (1/4) Epoch 19, batch 4850, loss[loss=0.1747, simple_loss=0.2595, pruned_loss=0.04495, over 19606.00 frames. ], tot_loss[loss=0.2125, simple_loss=0.291, pruned_loss=0.06698, over 3821019.93 frames. ], batch size: 50, lr: 4.32e-03, grad_scale: 8.0 2023-04-02 14:59:04,710 WARNING [train.py:1073] (1/4) Exclude cut with ID 774-127930-0014-48411_sp1.1_rvb from training. Duration: 0.95 2023-04-02 14:59:22,986 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=127789.0, num_to_drop=1, layers_to_drop={0} 2023-04-02 14:59:25,076 WARNING [train.py:1073] (1/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1_rvb from training. Duration: 0.9409375 2023-04-02 14:59:30,829 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 2023-04-02 14:59:30,850 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 2023-04-02 14:59:32,258 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=127797.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 14:59:35,937 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.8220, 1.5593, 1.4181, 1.8894, 1.4536, 1.6080, 1.4484, 1.6668], device='cuda:1'), covar=tensor([0.1027, 0.1271, 0.1444, 0.0854, 0.1253, 0.0531, 0.1307, 0.0733], device='cuda:1'), in_proj_covar=tensor([0.0262, 0.0351, 0.0305, 0.0248, 0.0296, 0.0246, 0.0296, 0.0248], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 14:59:40,800 INFO [train.py:903] (1/4) Epoch 19, batch 4900, loss[loss=0.196, simple_loss=0.283, pruned_loss=0.05454, over 19488.00 frames. ], tot_loss[loss=0.2117, simple_loss=0.2907, pruned_loss=0.0664, over 3827278.38 frames. ], batch size: 64, lr: 4.32e-03, grad_scale: 8.0 2023-04-02 14:59:40,834 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1_rvb from training. Duration: 25.3818125 2023-04-02 14:59:55,907 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.219e+02 4.848e+02 5.865e+02 7.992e+02 2.664e+03, threshold=1.173e+03, percent-clipped=3.0 2023-04-02 15:00:01,798 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 2023-04-02 15:00:41,588 INFO [train.py:903] (1/4) Epoch 19, batch 4950, loss[loss=0.2281, simple_loss=0.3118, pruned_loss=0.07217, over 19588.00 frames. ], tot_loss[loss=0.2124, simple_loss=0.291, pruned_loss=0.06684, over 3821008.52 frames. ], batch size: 61, lr: 4.32e-03, grad_scale: 8.0 2023-04-02 15:00:58,818 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=127868.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 15:00:59,648 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 2023-04-02 15:01:22,180 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 2023-04-02 15:01:23,545 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2035, 1.1622, 1.5974, 1.0654, 2.3681, 3.3300, 3.0502, 3.5075], device='cuda:1'), covar=tensor([0.1645, 0.3988, 0.3481, 0.2584, 0.0664, 0.0186, 0.0236, 0.0305], device='cuda:1'), in_proj_covar=tensor([0.0267, 0.0315, 0.0344, 0.0260, 0.0237, 0.0181, 0.0213, 0.0245], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-02 15:01:24,529 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=127890.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 15:01:26,340 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=127891.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 15:01:41,516 INFO [train.py:903] (1/4) Epoch 19, batch 5000, loss[loss=0.186, simple_loss=0.2609, pruned_loss=0.05551, over 19409.00 frames. ], tot_loss[loss=0.2112, simple_loss=0.2901, pruned_loss=0.06615, over 3827976.37 frames. ], batch size: 48, lr: 4.32e-03, grad_scale: 8.0 2023-04-02 15:01:41,905 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=127904.0, num_to_drop=1, layers_to_drop={1} 2023-04-02 15:01:51,189 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_rvb from training. Duration: 27.14 2023-04-02 15:01:55,669 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.309e+02 4.675e+02 5.614e+02 6.818e+02 2.294e+03, threshold=1.123e+03, percent-clipped=3.0 2023-04-02 15:02:03,228 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 2023-04-02 15:02:04,134 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 2023-04-02 15:02:22,569 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.50 vs. limit=2.0 2023-04-02 15:02:41,852 INFO [train.py:903] (1/4) Epoch 19, batch 5050, loss[loss=0.241, simple_loss=0.3229, pruned_loss=0.07952, over 19568.00 frames. ], tot_loss[loss=0.2113, simple_loss=0.2902, pruned_loss=0.06616, over 3827457.45 frames. ], batch size: 61, lr: 4.32e-03, grad_scale: 8.0 2023-04-02 15:03:16,224 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=127982.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 15:03:18,092 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9_rvb from training. Duration: 26.62775 2023-04-02 15:03:43,134 INFO [train.py:903] (1/4) Epoch 19, batch 5100, loss[loss=0.224, simple_loss=0.2977, pruned_loss=0.07509, over 19586.00 frames. ], tot_loss[loss=0.2106, simple_loss=0.2898, pruned_loss=0.06569, over 3836516.75 frames. ], batch size: 52, lr: 4.32e-03, grad_scale: 8.0 2023-04-02 15:03:45,749 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=128006.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 15:03:47,025 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=128007.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 15:03:47,355 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.12 vs. limit=2.0 2023-04-02 15:03:56,504 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 2023-04-02 15:03:58,283 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.139e+02 4.818e+02 5.706e+02 8.227e+02 1.561e+03, threshold=1.141e+03, percent-clipped=7.0 2023-04-02 15:04:00,665 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_rvb from training. Duration: 25.035 2023-04-02 15:04:04,160 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_rvb from training. Duration: 27.92 2023-04-02 15:04:07,801 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=128024.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 15:04:43,092 INFO [train.py:903] (1/4) Epoch 19, batch 5150, loss[loss=0.2505, simple_loss=0.3193, pruned_loss=0.09081, over 13277.00 frames. ], tot_loss[loss=0.2117, simple_loss=0.2904, pruned_loss=0.06645, over 3818970.26 frames. ], batch size: 136, lr: 4.31e-03, grad_scale: 8.0 2023-04-02 15:04:51,936 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.7135, 1.6966, 1.5958, 1.3529, 1.3525, 1.4246, 0.2240, 0.6741], device='cuda:1'), covar=tensor([0.0560, 0.0577, 0.0364, 0.0575, 0.1095, 0.0677, 0.1174, 0.0984], device='cuda:1'), in_proj_covar=tensor([0.0352, 0.0346, 0.0348, 0.0372, 0.0447, 0.0380, 0.0328, 0.0336], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 15:04:57,225 WARNING [train.py:1073] (1/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 2023-04-02 15:05:31,583 WARNING [train.py:1073] (1/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1_rvb from training. Duration: 0.9681875 2023-04-02 15:05:44,863 INFO [train.py:903] (1/4) Epoch 19, batch 5200, loss[loss=0.2225, simple_loss=0.2822, pruned_loss=0.08144, over 19758.00 frames. ], tot_loss[loss=0.2135, simple_loss=0.292, pruned_loss=0.06755, over 3805959.14 frames. ], batch size: 47, lr: 4.31e-03, grad_scale: 8.0 2023-04-02 15:05:59,011 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.969e+02 4.920e+02 6.208e+02 7.921e+02 1.726e+03, threshold=1.242e+03, percent-clipped=7.0 2023-04-02 15:05:59,072 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 2023-04-02 15:06:16,652 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 2023-04-02 15:06:28,682 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=128139.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 15:06:29,678 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.1199, 1.3124, 1.4332, 1.2740, 2.7307, 1.0026, 2.1802, 3.0150], device='cuda:1'), covar=tensor([0.0572, 0.2670, 0.2779, 0.1938, 0.0773, 0.2479, 0.1137, 0.0357], device='cuda:1'), in_proj_covar=tensor([0.0397, 0.0358, 0.0378, 0.0344, 0.0369, 0.0349, 0.0371, 0.0390], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 15:06:30,662 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=128141.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 15:06:41,127 WARNING [train.py:1073] (1/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 2023-04-02 15:06:45,758 INFO [train.py:903] (1/4) Epoch 19, batch 5250, loss[loss=0.2022, simple_loss=0.2874, pruned_loss=0.05848, over 19527.00 frames. ], tot_loss[loss=0.2138, simple_loss=0.2923, pruned_loss=0.06761, over 3797714.36 frames. ], batch size: 56, lr: 4.31e-03, grad_scale: 8.0 2023-04-02 15:06:53,640 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=128160.0, num_to_drop=1, layers_to_drop={1} 2023-04-02 15:07:23,135 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=128185.0, num_to_drop=1, layers_to_drop={0} 2023-04-02 15:07:45,449 INFO [train.py:903] (1/4) Epoch 19, batch 5300, loss[loss=0.2334, simple_loss=0.3132, pruned_loss=0.07686, over 19526.00 frames. ], tot_loss[loss=0.2121, simple_loss=0.2911, pruned_loss=0.06658, over 3811768.69 frames. ], batch size: 56, lr: 4.31e-03, grad_scale: 8.0 2023-04-02 15:07:54,706 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=128212.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 15:07:59,668 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.291e+02 5.001e+02 6.088e+02 7.600e+02 1.403e+03, threshold=1.218e+03, percent-clipped=2.0 2023-04-02 15:08:00,870 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 2023-04-02 15:08:21,807 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=128234.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 15:08:34,536 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3198, 1.3460, 1.4672, 1.4505, 1.7019, 1.7960, 1.6607, 0.5737], device='cuda:1'), covar=tensor([0.2352, 0.4056, 0.2517, 0.1894, 0.1601, 0.2178, 0.1406, 0.4654], device='cuda:1'), in_proj_covar=tensor([0.0522, 0.0629, 0.0692, 0.0475, 0.0615, 0.0521, 0.0660, 0.0536], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 15:08:46,734 INFO [train.py:903] (1/4) Epoch 19, batch 5350, loss[loss=0.1765, simple_loss=0.2546, pruned_loss=0.04918, over 19749.00 frames. ], tot_loss[loss=0.2115, simple_loss=0.2908, pruned_loss=0.06614, over 3812410.99 frames. ], batch size: 46, lr: 4.31e-03, grad_scale: 8.0 2023-04-02 15:08:50,307 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1328, 1.1697, 1.6777, 1.3133, 2.6393, 3.6876, 3.4448, 3.9557], device='cuda:1'), covar=tensor([0.1759, 0.4006, 0.3530, 0.2479, 0.0652, 0.0210, 0.0212, 0.0248], device='cuda:1'), in_proj_covar=tensor([0.0267, 0.0315, 0.0346, 0.0260, 0.0238, 0.0181, 0.0213, 0.0246], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-02 15:08:50,319 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=128256.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 15:08:57,142 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=128262.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 15:09:20,023 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9_rvb from training. Duration: 26.438875 2023-04-02 15:09:26,910 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=128287.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 15:09:47,392 INFO [train.py:903] (1/4) Epoch 19, batch 5400, loss[loss=0.2313, simple_loss=0.3123, pruned_loss=0.07513, over 19701.00 frames. ], tot_loss[loss=0.2119, simple_loss=0.2912, pruned_loss=0.06629, over 3822376.76 frames. ], batch size: 59, lr: 4.31e-03, grad_scale: 8.0 2023-04-02 15:10:00,870 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([5.6623, 5.1218, 2.9800, 4.6012, 1.2828, 5.1352, 5.0454, 5.2701], device='cuda:1'), covar=tensor([0.0420, 0.0714, 0.1928, 0.0615, 0.3839, 0.0540, 0.0748, 0.1035], device='cuda:1'), in_proj_covar=tensor([0.0489, 0.0394, 0.0482, 0.0340, 0.0396, 0.0420, 0.0410, 0.0446], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 15:10:01,764 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.288e+02 4.631e+02 5.571e+02 7.152e+02 1.493e+03, threshold=1.114e+03, percent-clipped=2.0 2023-04-02 15:10:15,374 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=128327.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 15:10:29,852 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=128338.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 15:10:42,711 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=128349.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 15:10:48,541 INFO [train.py:903] (1/4) Epoch 19, batch 5450, loss[loss=0.303, simple_loss=0.3622, pruned_loss=0.1219, over 19080.00 frames. ], tot_loss[loss=0.2123, simple_loss=0.2912, pruned_loss=0.06666, over 3820912.45 frames. ], batch size: 69, lr: 4.31e-03, grad_scale: 16.0 2023-04-02 15:11:39,531 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=128395.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 15:11:50,173 INFO [train.py:903] (1/4) Epoch 19, batch 5500, loss[loss=0.2095, simple_loss=0.2979, pruned_loss=0.06053, over 19531.00 frames. ], tot_loss[loss=0.2106, simple_loss=0.2897, pruned_loss=0.0657, over 3830900.93 frames. ], batch size: 54, lr: 4.31e-03, grad_scale: 8.0 2023-04-02 15:12:06,852 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.088e+02 5.166e+02 6.121e+02 7.872e+02 1.632e+03, threshold=1.224e+03, percent-clipped=5.0 2023-04-02 15:12:10,674 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=128420.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 15:12:12,700 WARNING [train.py:1073] (1/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9_rvb from training. Duration: 27.47775 2023-04-02 15:12:16,416 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2110, 1.5339, 2.0621, 1.6165, 2.8274, 4.6071, 4.5226, 5.0793], device='cuda:1'), covar=tensor([0.1647, 0.3638, 0.3109, 0.2159, 0.0680, 0.0180, 0.0161, 0.0167], device='cuda:1'), in_proj_covar=tensor([0.0267, 0.0315, 0.0346, 0.0260, 0.0238, 0.0181, 0.0214, 0.0246], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-02 15:12:23,762 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=128431.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 15:12:50,275 INFO [train.py:903] (1/4) Epoch 19, batch 5550, loss[loss=0.2272, simple_loss=0.3005, pruned_loss=0.07694, over 19399.00 frames. ], tot_loss[loss=0.2104, simple_loss=0.2895, pruned_loss=0.06567, over 3825632.36 frames. ], batch size: 48, lr: 4.31e-03, grad_scale: 8.0 2023-04-02 15:12:56,490 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 2023-04-02 15:13:44,858 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 2023-04-02 15:13:51,307 INFO [train.py:903] (1/4) Epoch 19, batch 5600, loss[loss=0.2642, simple_loss=0.3351, pruned_loss=0.09663, over 18466.00 frames. ], tot_loss[loss=0.2126, simple_loss=0.2913, pruned_loss=0.06699, over 3819622.85 frames. ], batch size: 84, lr: 4.31e-03, grad_scale: 8.0 2023-04-02 15:14:01,870 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=128512.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 15:14:07,037 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.064e+02 4.843e+02 5.877e+02 7.578e+02 1.194e+03, threshold=1.175e+03, percent-clipped=0.0 2023-04-02 15:14:32,645 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=128537.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 15:14:52,040 INFO [train.py:903] (1/4) Epoch 19, batch 5650, loss[loss=0.2638, simple_loss=0.329, pruned_loss=0.0993, over 19472.00 frames. ], tot_loss[loss=0.2125, simple_loss=0.2912, pruned_loss=0.06695, over 3825730.88 frames. ], batch size: 64, lr: 4.31e-03, grad_scale: 8.0 2023-04-02 15:15:27,811 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=128583.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 15:15:37,262 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9_rvb from training. Duration: 28.0944375 2023-04-02 15:15:53,142 INFO [train.py:903] (1/4) Epoch 19, batch 5700, loss[loss=0.1982, simple_loss=0.2898, pruned_loss=0.05331, over 19618.00 frames. ], tot_loss[loss=0.2116, simple_loss=0.2907, pruned_loss=0.06628, over 3826186.56 frames. ], batch size: 57, lr: 4.31e-03, grad_scale: 8.0 2023-04-02 15:15:54,811 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=128605.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 15:15:58,249 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=128608.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 15:16:08,027 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.931e+02 5.156e+02 6.108e+02 7.232e+02 1.309e+03, threshold=1.222e+03, percent-clipped=4.0 2023-04-02 15:16:24,887 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=128630.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 15:16:53,482 INFO [train.py:903] (1/4) Epoch 19, batch 5750, loss[loss=0.2345, simple_loss=0.3059, pruned_loss=0.08152, over 19539.00 frames. ], tot_loss[loss=0.2115, simple_loss=0.2902, pruned_loss=0.06646, over 3808565.00 frames. ], batch size: 54, lr: 4.30e-03, grad_scale: 8.0 2023-04-02 15:16:55,732 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9_rvb from training. Duration: 33.038875 2023-04-02 15:17:05,285 WARNING [train.py:1073] (1/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 2023-04-02 15:17:09,596 WARNING [train.py:1073] (1/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 2023-04-02 15:17:26,653 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.50 vs. limit=2.0 2023-04-02 15:17:27,291 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=128682.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 15:17:42,188 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4321, 1.4859, 1.6906, 1.7033, 1.2812, 1.6813, 1.7300, 1.5696], device='cuda:1'), covar=tensor([0.3419, 0.2902, 0.1547, 0.1884, 0.3165, 0.1711, 0.4195, 0.2797], device='cuda:1'), in_proj_covar=tensor([0.0874, 0.0935, 0.0700, 0.0923, 0.0859, 0.0794, 0.0833, 0.0767], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 15:17:55,080 INFO [train.py:903] (1/4) Epoch 19, batch 5800, loss[loss=0.183, simple_loss=0.2625, pruned_loss=0.05178, over 19788.00 frames. ], tot_loss[loss=0.2122, simple_loss=0.2905, pruned_loss=0.06695, over 3808938.18 frames. ], batch size: 49, lr: 4.30e-03, grad_scale: 8.0 2023-04-02 15:18:10,461 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.713e+02 4.671e+02 6.414e+02 7.787e+02 1.302e+03, threshold=1.283e+03, percent-clipped=2.0 2023-04-02 15:18:24,046 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.7562, 2.1106, 1.7315, 1.5030, 2.0507, 1.4933, 1.5629, 1.9941], device='cuda:1'), covar=tensor([0.0758, 0.0626, 0.0756, 0.0792, 0.0427, 0.0980, 0.0615, 0.0392], device='cuda:1'), in_proj_covar=tensor([0.0299, 0.0309, 0.0327, 0.0259, 0.0242, 0.0332, 0.0289, 0.0271], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 15:18:55,624 INFO [train.py:903] (1/4) Epoch 19, batch 5850, loss[loss=0.216, simple_loss=0.2986, pruned_loss=0.0667, over 19656.00 frames. ], tot_loss[loss=0.2126, simple_loss=0.2912, pruned_loss=0.06701, over 3821966.50 frames. ], batch size: 58, lr: 4.30e-03, grad_scale: 8.0 2023-04-02 15:19:20,624 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=128775.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 15:19:48,469 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=128797.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 15:19:50,613 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=128799.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 15:19:55,909 INFO [train.py:903] (1/4) Epoch 19, batch 5900, loss[loss=0.1771, simple_loss=0.2523, pruned_loss=0.05098, over 19743.00 frames. ], tot_loss[loss=0.213, simple_loss=0.2915, pruned_loss=0.06723, over 3821249.40 frames. ], batch size: 46, lr: 4.30e-03, grad_scale: 8.0 2023-04-02 15:20:02,588 WARNING [train.py:1073] (1/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 2023-04-02 15:20:11,628 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.258e+02 4.778e+02 5.849e+02 7.721e+02 1.320e+03, threshold=1.170e+03, percent-clipped=1.0 2023-04-02 15:20:22,220 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9_rvb from training. Duration: 27.511125 2023-04-02 15:20:56,175 INFO [train.py:903] (1/4) Epoch 19, batch 5950, loss[loss=0.2087, simple_loss=0.2947, pruned_loss=0.06138, over 19044.00 frames. ], tot_loss[loss=0.2128, simple_loss=0.2914, pruned_loss=0.06713, over 3812925.38 frames. ], batch size: 69, lr: 4.30e-03, grad_scale: 8.0 2023-04-02 15:20:57,899 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.14 vs. limit=2.0 2023-04-02 15:21:41,422 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=128890.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 15:21:57,516 INFO [train.py:903] (1/4) Epoch 19, batch 6000, loss[loss=0.2132, simple_loss=0.3008, pruned_loss=0.06284, over 18267.00 frames. ], tot_loss[loss=0.2133, simple_loss=0.2921, pruned_loss=0.06728, over 3821404.13 frames. ], batch size: 83, lr: 4.30e-03, grad_scale: 8.0 2023-04-02 15:21:57,516 INFO [train.py:928] (1/4) Computing validation loss 2023-04-02 15:22:07,158 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.8083, 1.5689, 1.4396, 1.7960, 1.4710, 1.5894, 1.4028, 1.6334], device='cuda:1'), covar=tensor([0.1128, 0.1278, 0.1571, 0.1071, 0.1424, 0.0575, 0.1639, 0.0830], device='cuda:1'), in_proj_covar=tensor([0.0265, 0.0355, 0.0308, 0.0249, 0.0298, 0.0248, 0.0300, 0.0251], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 15:22:12,608 INFO [train.py:937] (1/4) Epoch 19, validation: loss=0.1702, simple_loss=0.2702, pruned_loss=0.03514, over 944034.00 frames. 2023-04-02 15:22:12,609 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18821MB 2023-04-02 15:22:17,233 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=128908.0, num_to_drop=1, layers_to_drop={1} 2023-04-02 15:22:28,003 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.370e+02 5.136e+02 6.485e+02 9.043e+02 2.174e+03, threshold=1.297e+03, percent-clipped=7.0 2023-04-02 15:23:13,564 INFO [train.py:903] (1/4) Epoch 19, batch 6050, loss[loss=0.2001, simple_loss=0.2816, pruned_loss=0.05928, over 19831.00 frames. ], tot_loss[loss=0.2127, simple_loss=0.2915, pruned_loss=0.06693, over 3815244.19 frames. ], batch size: 52, lr: 4.30e-03, grad_scale: 8.0 2023-04-02 15:23:33,154 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.8673, 1.9761, 2.1929, 2.5063, 1.8558, 2.3933, 2.2093, 2.0030], device='cuda:1'), covar=tensor([0.3985, 0.3540, 0.1728, 0.2131, 0.3739, 0.1924, 0.4480, 0.3196], device='cuda:1'), in_proj_covar=tensor([0.0875, 0.0934, 0.0699, 0.0923, 0.0859, 0.0793, 0.0831, 0.0766], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 15:24:14,706 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4502, 1.6049, 1.9319, 1.7253, 3.3058, 2.7341, 3.5938, 1.7168], device='cuda:1'), covar=tensor([0.2453, 0.4237, 0.2724, 0.1995, 0.1421, 0.1896, 0.1343, 0.3910], device='cuda:1'), in_proj_covar=tensor([0.0523, 0.0632, 0.0695, 0.0476, 0.0620, 0.0524, 0.0664, 0.0540], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 15:24:15,355 INFO [train.py:903] (1/4) Epoch 19, batch 6100, loss[loss=0.2214, simple_loss=0.3036, pruned_loss=0.06962, over 19684.00 frames. ], tot_loss[loss=0.2122, simple_loss=0.2908, pruned_loss=0.06678, over 3808849.19 frames. ], batch size: 53, lr: 4.30e-03, grad_scale: 8.0 2023-04-02 15:24:30,769 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.431e+02 4.994e+02 6.076e+02 7.380e+02 1.472e+03, threshold=1.215e+03, percent-clipped=4.0 2023-04-02 15:25:14,919 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.3704, 3.9787, 2.5247, 3.5362, 0.6691, 3.8548, 3.7285, 3.8716], device='cuda:1'), covar=tensor([0.0668, 0.0974, 0.2048, 0.0849, 0.4299, 0.0732, 0.0927, 0.1347], device='cuda:1'), in_proj_covar=tensor([0.0493, 0.0397, 0.0486, 0.0343, 0.0399, 0.0422, 0.0413, 0.0450], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 15:25:15,134 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=129053.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 15:25:15,830 INFO [train.py:903] (1/4) Epoch 19, batch 6150, loss[loss=0.2149, simple_loss=0.2992, pruned_loss=0.06526, over 19688.00 frames. ], tot_loss[loss=0.2129, simple_loss=0.2914, pruned_loss=0.06723, over 3795922.15 frames. ], batch size: 59, lr: 4.30e-03, grad_scale: 8.0 2023-04-02 15:25:37,316 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 2023-04-02 15:25:44,279 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9_rvb from training. Duration: 31.02225 2023-04-02 15:25:44,623 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=129078.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 15:26:16,319 INFO [train.py:903] (1/4) Epoch 19, batch 6200, loss[loss=0.1917, simple_loss=0.2708, pruned_loss=0.0563, over 19731.00 frames. ], tot_loss[loss=0.2119, simple_loss=0.2904, pruned_loss=0.0667, over 3806189.53 frames. ], batch size: 51, lr: 4.30e-03, grad_scale: 8.0 2023-04-02 15:26:32,067 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.839e+02 4.815e+02 6.250e+02 7.621e+02 1.523e+03, threshold=1.250e+03, percent-clipped=7.0 2023-04-02 15:26:53,154 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.8082, 1.9239, 2.1151, 2.3570, 1.6905, 2.2562, 2.1351, 1.9658], device='cuda:1'), covar=tensor([0.3940, 0.3223, 0.1778, 0.2012, 0.3502, 0.1860, 0.4375, 0.3078], device='cuda:1'), in_proj_covar=tensor([0.0874, 0.0934, 0.0701, 0.0922, 0.0858, 0.0793, 0.0829, 0.0766], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 15:27:04,016 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=129143.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 15:27:07,707 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=129146.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 15:27:17,154 INFO [train.py:903] (1/4) Epoch 19, batch 6250, loss[loss=0.2167, simple_loss=0.3018, pruned_loss=0.06578, over 19293.00 frames. ], tot_loss[loss=0.2118, simple_loss=0.2903, pruned_loss=0.06665, over 3801675.72 frames. ], batch size: 66, lr: 4.30e-03, grad_scale: 8.0 2023-04-02 15:27:38,408 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=129171.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 15:27:47,709 WARNING [train.py:1073] (1/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9_rvb from training. Duration: 25.988875 2023-04-02 15:28:17,840 INFO [train.py:903] (1/4) Epoch 19, batch 6300, loss[loss=0.2044, simple_loss=0.2935, pruned_loss=0.05766, over 19519.00 frames. ], tot_loss[loss=0.2115, simple_loss=0.2902, pruned_loss=0.0664, over 3807593.55 frames. ], batch size: 56, lr: 4.30e-03, grad_scale: 8.0 2023-04-02 15:28:33,725 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.168e+02 5.181e+02 6.373e+02 8.503e+02 1.874e+03, threshold=1.275e+03, percent-clipped=7.0 2023-04-02 15:29:17,495 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=129252.0, num_to_drop=1, layers_to_drop={1} 2023-04-02 15:29:19,488 INFO [train.py:903] (1/4) Epoch 19, batch 6350, loss[loss=0.2089, simple_loss=0.2872, pruned_loss=0.06526, over 19852.00 frames. ], tot_loss[loss=0.2113, simple_loss=0.2896, pruned_loss=0.06643, over 3815022.86 frames. ], batch size: 52, lr: 4.29e-03, grad_scale: 8.0 2023-04-02 15:29:25,234 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=129258.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 15:30:20,708 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.13 vs. limit=2.0 2023-04-02 15:30:21,271 INFO [train.py:903] (1/4) Epoch 19, batch 6400, loss[loss=0.2324, simple_loss=0.3079, pruned_loss=0.07845, over 19278.00 frames. ], tot_loss[loss=0.2113, simple_loss=0.2901, pruned_loss=0.06622, over 3821910.79 frames. ], batch size: 66, lr: 4.29e-03, grad_scale: 8.0 2023-04-02 15:30:36,859 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.331e+02 4.991e+02 6.008e+02 7.727e+02 1.608e+03, threshold=1.202e+03, percent-clipped=4.0 2023-04-02 15:30:58,516 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.1916, 3.6629, 2.1082, 2.3038, 3.1820, 1.7712, 1.5315, 2.3160], device='cuda:1'), covar=tensor([0.1277, 0.0489, 0.1084, 0.0795, 0.0486, 0.1204, 0.0947, 0.0682], device='cuda:1'), in_proj_covar=tensor([0.0298, 0.0310, 0.0328, 0.0259, 0.0241, 0.0331, 0.0288, 0.0270], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 15:31:22,261 INFO [train.py:903] (1/4) Epoch 19, batch 6450, loss[loss=0.1839, simple_loss=0.2618, pruned_loss=0.053, over 19369.00 frames. ], tot_loss[loss=0.2104, simple_loss=0.2893, pruned_loss=0.06572, over 3829469.36 frames. ], batch size: 48, lr: 4.29e-03, grad_scale: 8.0 2023-04-02 15:31:38,204 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=129367.0, num_to_drop=1, layers_to_drop={1} 2023-04-02 15:31:48,937 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6231, 2.4558, 1.7925, 1.6827, 2.1652, 1.4055, 1.5501, 2.0463], device='cuda:1'), covar=tensor([0.1002, 0.0713, 0.1013, 0.0784, 0.0519, 0.1200, 0.0657, 0.0470], device='cuda:1'), in_proj_covar=tensor([0.0299, 0.0310, 0.0329, 0.0260, 0.0242, 0.0332, 0.0289, 0.0270], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 15:32:06,812 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9_rvb from training. Duration: 28.638875 2023-04-02 15:32:07,090 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.7697, 1.4312, 1.5545, 1.5321, 3.3136, 1.0267, 2.3668, 3.7602], device='cuda:1'), covar=tensor([0.0436, 0.2600, 0.2772, 0.1843, 0.0702, 0.2647, 0.1308, 0.0246], device='cuda:1'), in_proj_covar=tensor([0.0396, 0.0358, 0.0377, 0.0340, 0.0367, 0.0348, 0.0370, 0.0391], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 15:32:22,358 INFO [train.py:903] (1/4) Epoch 19, batch 6500, loss[loss=0.1678, simple_loss=0.2482, pruned_loss=0.04363, over 19382.00 frames. ], tot_loss[loss=0.2109, simple_loss=0.2899, pruned_loss=0.06594, over 3823729.17 frames. ], batch size: 47, lr: 4.29e-03, grad_scale: 8.0 2023-04-02 15:32:29,762 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1_rvb from training. Duration: 0.836375 2023-04-02 15:32:38,768 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.146e+02 5.340e+02 6.897e+02 8.888e+02 1.987e+03, threshold=1.379e+03, percent-clipped=7.0 2023-04-02 15:32:47,821 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.5838, 1.2551, 1.4611, 1.2423, 2.2234, 1.0719, 2.0186, 2.5150], device='cuda:1'), covar=tensor([0.0646, 0.2556, 0.2567, 0.1618, 0.0851, 0.2034, 0.1049, 0.0418], device='cuda:1'), in_proj_covar=tensor([0.0396, 0.0357, 0.0377, 0.0340, 0.0367, 0.0348, 0.0370, 0.0391], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 15:32:56,844 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=129431.0, num_to_drop=1, layers_to_drop={0} 2023-04-02 15:33:24,203 INFO [train.py:903] (1/4) Epoch 19, batch 6550, loss[loss=0.1878, simple_loss=0.2658, pruned_loss=0.05493, over 19487.00 frames. ], tot_loss[loss=0.2115, simple_loss=0.29, pruned_loss=0.06647, over 3820842.10 frames. ], batch size: 49, lr: 4.29e-03, grad_scale: 8.0 2023-04-02 15:34:07,474 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4937, 2.1152, 1.6340, 1.5333, 1.9804, 1.3262, 1.4098, 1.8416], device='cuda:1'), covar=tensor([0.0883, 0.0664, 0.0965, 0.0719, 0.0473, 0.1122, 0.0628, 0.0436], device='cuda:1'), in_proj_covar=tensor([0.0301, 0.0312, 0.0331, 0.0261, 0.0243, 0.0335, 0.0291, 0.0272], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 15:34:24,936 INFO [train.py:903] (1/4) Epoch 19, batch 6600, loss[loss=0.1974, simple_loss=0.2816, pruned_loss=0.05659, over 19650.00 frames. ], tot_loss[loss=0.2121, simple_loss=0.2907, pruned_loss=0.06675, over 3820418.47 frames. ], batch size: 55, lr: 4.29e-03, grad_scale: 8.0 2023-04-02 15:34:37,628 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=129514.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 15:34:40,453 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.539e+02 5.847e+02 6.807e+02 8.552e+02 1.538e+03, threshold=1.361e+03, percent-clipped=4.0 2023-04-02 15:35:03,129 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.73 vs. limit=5.0 2023-04-02 15:35:03,895 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6354, 1.5367, 1.5784, 2.0994, 1.7108, 2.0299, 1.9868, 1.7069], device='cuda:1'), covar=tensor([0.0834, 0.0935, 0.1030, 0.0820, 0.0843, 0.0736, 0.0908, 0.0739], device='cuda:1'), in_proj_covar=tensor([0.0209, 0.0219, 0.0225, 0.0245, 0.0227, 0.0209, 0.0189, 0.0204], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-02 15:35:07,266 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=129539.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 15:35:25,975 INFO [train.py:903] (1/4) Epoch 19, batch 6650, loss[loss=0.1775, simple_loss=0.2529, pruned_loss=0.05101, over 19362.00 frames. ], tot_loss[loss=0.2121, simple_loss=0.2906, pruned_loss=0.06684, over 3813606.30 frames. ], batch size: 47, lr: 4.29e-03, grad_scale: 8.0 2023-04-02 15:35:32,242 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.4685, 2.4314, 2.6975, 3.3667, 2.5388, 3.1785, 2.8569, 2.4505], device='cuda:1'), covar=tensor([0.3893, 0.3929, 0.1678, 0.2180, 0.3988, 0.1882, 0.4110, 0.3100], device='cuda:1'), in_proj_covar=tensor([0.0868, 0.0927, 0.0697, 0.0917, 0.0854, 0.0787, 0.0826, 0.0761], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 15:36:26,340 INFO [train.py:903] (1/4) Epoch 19, batch 6700, loss[loss=0.2241, simple_loss=0.3114, pruned_loss=0.06839, over 19694.00 frames. ], tot_loss[loss=0.2109, simple_loss=0.2898, pruned_loss=0.06602, over 3821364.25 frames. ], batch size: 59, lr: 4.29e-03, grad_scale: 8.0 2023-04-02 15:36:42,872 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.775e+02 4.923e+02 5.649e+02 7.598e+02 1.428e+03, threshold=1.130e+03, percent-clipped=1.0 2023-04-02 15:36:51,015 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=129623.0, num_to_drop=1, layers_to_drop={1} 2023-04-02 15:36:58,737 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=129630.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 15:37:03,804 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.14 vs. limit=2.0 2023-04-02 15:37:04,580 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9284, 1.9029, 1.7724, 2.9411, 1.9685, 2.8683, 1.9541, 1.4346], device='cuda:1'), covar=tensor([0.4690, 0.4150, 0.2867, 0.2922, 0.4283, 0.2038, 0.6126, 0.5248], device='cuda:1'), in_proj_covar=tensor([0.0869, 0.0929, 0.0697, 0.0918, 0.0855, 0.0788, 0.0826, 0.0762], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 15:37:18,806 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=129648.0, num_to_drop=1, layers_to_drop={0} 2023-04-02 15:37:25,089 INFO [train.py:903] (1/4) Epoch 19, batch 6750, loss[loss=0.1873, simple_loss=0.2601, pruned_loss=0.05728, over 19738.00 frames. ], tot_loss[loss=0.2121, simple_loss=0.2909, pruned_loss=0.06667, over 3817876.18 frames. ], batch size: 45, lr: 4.29e-03, grad_scale: 8.0 2023-04-02 15:38:20,254 INFO [train.py:903] (1/4) Epoch 19, batch 6800, loss[loss=0.2058, simple_loss=0.2809, pruned_loss=0.06537, over 19747.00 frames. ], tot_loss[loss=0.2121, simple_loss=0.291, pruned_loss=0.06661, over 3832884.19 frames. ], batch size: 51, lr: 4.29e-03, grad_scale: 8.0 2023-04-02 15:38:34,410 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.328e+02 4.884e+02 6.226e+02 8.201e+02 1.689e+03, threshold=1.245e+03, percent-clipped=11.0 2023-04-02 15:39:04,981 WARNING [train.py:1073] (1/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 2023-04-02 15:39:05,441 WARNING [train.py:1073] (1/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 2023-04-02 15:39:08,363 INFO [train.py:903] (1/4) Epoch 20, batch 0, loss[loss=0.2519, simple_loss=0.3371, pruned_loss=0.08333, over 17179.00 frames. ], tot_loss[loss=0.2519, simple_loss=0.3371, pruned_loss=0.08333, over 17179.00 frames. ], batch size: 101, lr: 4.18e-03, grad_scale: 8.0 2023-04-02 15:39:08,363 INFO [train.py:928] (1/4) Computing validation loss 2023-04-02 15:39:19,739 INFO [train.py:937] (1/4) Epoch 20, validation: loss=0.1695, simple_loss=0.2703, pruned_loss=0.03432, over 944034.00 frames. 2023-04-02 15:39:19,739 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18821MB 2023-04-02 15:39:31,866 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 2023-04-02 15:40:12,666 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=129775.0, num_to_drop=1, layers_to_drop={0} 2023-04-02 15:40:20,204 INFO [train.py:903] (1/4) Epoch 20, batch 50, loss[loss=0.2201, simple_loss=0.2921, pruned_loss=0.07403, over 18181.00 frames. ], tot_loss[loss=0.2161, simple_loss=0.2952, pruned_loss=0.06847, over 847815.54 frames. ], batch size: 40, lr: 4.18e-03, grad_scale: 8.0 2023-04-02 15:40:51,322 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=129809.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 15:40:54,556 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1_rvb from training. Duration: 27.0318125 2023-04-02 15:41:03,010 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.371e+02 5.543e+02 6.891e+02 8.835e+02 1.770e+03, threshold=1.378e+03, percent-clipped=8.0 2023-04-02 15:41:05,588 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.7434, 1.2727, 1.4913, 1.7005, 3.2931, 1.2293, 2.2863, 3.8045], device='cuda:1'), covar=tensor([0.0500, 0.2847, 0.2856, 0.1707, 0.0733, 0.2437, 0.1373, 0.0217], device='cuda:1'), in_proj_covar=tensor([0.0394, 0.0355, 0.0375, 0.0337, 0.0365, 0.0345, 0.0369, 0.0388], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 15:41:20,211 INFO [train.py:903] (1/4) Epoch 20, batch 100, loss[loss=0.2465, simple_loss=0.3244, pruned_loss=0.08429, over 19527.00 frames. ], tot_loss[loss=0.2165, simple_loss=0.2955, pruned_loss=0.06873, over 1510798.61 frames. ], batch size: 54, lr: 4.17e-03, grad_scale: 8.0 2023-04-02 15:41:31,359 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 2023-04-02 15:42:14,112 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3033, 1.4629, 1.9026, 1.4416, 2.7133, 3.3987, 3.1930, 3.6007], device='cuda:1'), covar=tensor([0.1744, 0.3831, 0.3359, 0.2537, 0.0776, 0.0277, 0.0267, 0.0346], device='cuda:1'), in_proj_covar=tensor([0.0267, 0.0316, 0.0346, 0.0261, 0.0237, 0.0181, 0.0213, 0.0247], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-02 15:42:21,224 INFO [train.py:903] (1/4) Epoch 20, batch 150, loss[loss=0.2585, simple_loss=0.343, pruned_loss=0.08698, over 19541.00 frames. ], tot_loss[loss=0.2117, simple_loss=0.2914, pruned_loss=0.066, over 2029734.47 frames. ], batch size: 56, lr: 4.17e-03, grad_scale: 8.0 2023-04-02 15:42:30,221 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=129890.0, num_to_drop=1, layers_to_drop={1} 2023-04-02 15:43:03,459 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.452e+02 4.799e+02 5.935e+02 7.467e+02 3.197e+03, threshold=1.187e+03, percent-clipped=3.0 2023-04-02 15:43:21,684 INFO [train.py:903] (1/4) Epoch 20, batch 200, loss[loss=0.2184, simple_loss=0.2937, pruned_loss=0.07153, over 19656.00 frames. ], tot_loss[loss=0.2091, simple_loss=0.2885, pruned_loss=0.06487, over 2432447.22 frames. ], batch size: 60, lr: 4.17e-03, grad_scale: 8.0 2023-04-02 15:43:22,857 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 2023-04-02 15:43:36,816 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0 2023-04-02 15:44:13,731 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=129974.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 15:44:23,549 INFO [train.py:903] (1/4) Epoch 20, batch 250, loss[loss=0.196, simple_loss=0.2704, pruned_loss=0.06076, over 19611.00 frames. ], tot_loss[loss=0.2078, simple_loss=0.2866, pruned_loss=0.06451, over 2744469.43 frames. ], batch size: 50, lr: 4.17e-03, grad_scale: 8.0 2023-04-02 15:45:06,934 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.070e+02 5.298e+02 6.354e+02 8.098e+02 1.543e+03, threshold=1.271e+03, percent-clipped=7.0 2023-04-02 15:45:08,329 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=130018.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 15:45:10,537 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=130020.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 15:45:25,487 INFO [train.py:903] (1/4) Epoch 20, batch 300, loss[loss=0.2232, simple_loss=0.3034, pruned_loss=0.07146, over 19464.00 frames. ], tot_loss[loss=0.2093, simple_loss=0.2875, pruned_loss=0.0655, over 2987963.97 frames. ], batch size: 64, lr: 4.17e-03, grad_scale: 8.0 2023-04-02 15:45:57,464 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.2578, 3.7470, 3.8855, 3.9041, 1.6488, 3.6947, 3.2293, 3.6317], device='cuda:1'), covar=tensor([0.1642, 0.1363, 0.0720, 0.0741, 0.5659, 0.1062, 0.0735, 0.1176], device='cuda:1'), in_proj_covar=tensor([0.0766, 0.0719, 0.0923, 0.0805, 0.0817, 0.0679, 0.0555, 0.0856], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-02 15:46:25,937 INFO [train.py:903] (1/4) Epoch 20, batch 350, loss[loss=0.2175, simple_loss=0.305, pruned_loss=0.06498, over 19531.00 frames. ], tot_loss[loss=0.2083, simple_loss=0.2869, pruned_loss=0.06483, over 3183249.95 frames. ], batch size: 54, lr: 4.17e-03, grad_scale: 8.0 2023-04-02 15:46:35,011 WARNING [train.py:1073] (1/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1_rvb from training. Duration: 0.9681875 2023-04-02 15:46:35,361 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=130089.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 15:46:38,879 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.4069, 1.3903, 1.6474, 1.3752, 2.9923, 1.0708, 2.3569, 3.4023], device='cuda:1'), covar=tensor([0.0535, 0.2668, 0.2678, 0.1934, 0.0779, 0.2520, 0.1143, 0.0272], device='cuda:1'), in_proj_covar=tensor([0.0399, 0.0359, 0.0378, 0.0341, 0.0369, 0.0349, 0.0373, 0.0393], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 15:46:50,440 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4648, 1.3575, 1.3459, 1.8283, 1.4019, 1.6172, 1.6872, 1.4945], device='cuda:1'), covar=tensor([0.0837, 0.0941, 0.1073, 0.0654, 0.0829, 0.0764, 0.0807, 0.0724], device='cuda:1'), in_proj_covar=tensor([0.0207, 0.0219, 0.0224, 0.0242, 0.0225, 0.0209, 0.0187, 0.0203], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-02 15:47:08,654 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.196e+02 5.230e+02 6.397e+02 7.792e+02 1.393e+03, threshold=1.279e+03, percent-clipped=3.0 2023-04-02 15:47:26,635 INFO [train.py:903] (1/4) Epoch 20, batch 400, loss[loss=0.1776, simple_loss=0.2459, pruned_loss=0.05465, over 19755.00 frames. ], tot_loss[loss=0.2083, simple_loss=0.2869, pruned_loss=0.06487, over 3341905.53 frames. ], batch size: 46, lr: 4.17e-03, grad_scale: 8.0 2023-04-02 15:47:41,009 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.34 vs. limit=5.0 2023-04-02 15:47:42,982 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=130146.0, num_to_drop=1, layers_to_drop={0} 2023-04-02 15:47:53,127 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=130153.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 15:48:14,973 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=130171.0, num_to_drop=1, layers_to_drop={1} 2023-04-02 15:48:27,090 INFO [train.py:903] (1/4) Epoch 20, batch 450, loss[loss=0.2013, simple_loss=0.291, pruned_loss=0.05583, over 19574.00 frames. ], tot_loss[loss=0.2082, simple_loss=0.2868, pruned_loss=0.06475, over 3458248.70 frames. ], batch size: 61, lr: 4.17e-03, grad_scale: 8.0 2023-04-02 15:48:41,201 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.3357, 2.2474, 2.0443, 1.8878, 1.7235, 1.8894, 0.7823, 1.3177], device='cuda:1'), covar=tensor([0.0573, 0.0604, 0.0493, 0.0776, 0.1144, 0.1009, 0.1208, 0.0990], device='cuda:1'), in_proj_covar=tensor([0.0358, 0.0350, 0.0353, 0.0376, 0.0451, 0.0383, 0.0333, 0.0338], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 15:49:03,624 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 2023-04-02 15:49:04,559 WARNING [train.py:1073] (1/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 2023-04-02 15:49:09,165 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.114e+02 5.003e+02 6.443e+02 8.043e+02 1.786e+03, threshold=1.289e+03, percent-clipped=5.0 2023-04-02 15:49:18,460 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3296, 1.3565, 1.4611, 1.4442, 1.7644, 1.8697, 1.7738, 0.6020], device='cuda:1'), covar=tensor([0.2238, 0.3872, 0.2553, 0.1805, 0.1546, 0.2165, 0.1461, 0.4334], device='cuda:1'), in_proj_covar=tensor([0.0526, 0.0632, 0.0697, 0.0475, 0.0618, 0.0526, 0.0661, 0.0541], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 15:49:27,172 INFO [train.py:903] (1/4) Epoch 20, batch 500, loss[loss=0.1773, simple_loss=0.2615, pruned_loss=0.0465, over 19847.00 frames. ], tot_loss[loss=0.209, simple_loss=0.2874, pruned_loss=0.06524, over 3539638.55 frames. ], batch size: 52, lr: 4.17e-03, grad_scale: 8.0 2023-04-02 15:49:50,923 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.67 vs. limit=2.0 2023-04-02 15:50:10,039 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=130268.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 15:50:27,980 INFO [train.py:903] (1/4) Epoch 20, batch 550, loss[loss=0.2487, simple_loss=0.3214, pruned_loss=0.08798, over 19538.00 frames. ], tot_loss[loss=0.21, simple_loss=0.2886, pruned_loss=0.06565, over 3606199.20 frames. ], batch size: 54, lr: 4.17e-03, grad_scale: 8.0 2023-04-02 15:51:11,221 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.373e+02 5.062e+02 6.327e+02 8.479e+02 2.088e+03, threshold=1.265e+03, percent-clipped=6.0 2023-04-02 15:51:28,461 INFO [train.py:903] (1/4) Epoch 20, batch 600, loss[loss=0.2284, simple_loss=0.3093, pruned_loss=0.07377, over 19361.00 frames. ], tot_loss[loss=0.2119, simple_loss=0.2905, pruned_loss=0.06666, over 3649822.96 frames. ], batch size: 70, lr: 4.17e-03, grad_scale: 8.0 2023-04-02 15:51:44,829 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=130345.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 15:52:06,341 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=130362.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 15:52:08,702 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=130364.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 15:52:15,670 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 2023-04-02 15:52:16,042 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=130370.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 15:52:30,044 INFO [train.py:903] (1/4) Epoch 20, batch 650, loss[loss=0.1993, simple_loss=0.2885, pruned_loss=0.05503, over 19542.00 frames. ], tot_loss[loss=0.2116, simple_loss=0.2902, pruned_loss=0.06647, over 3694883.10 frames. ], batch size: 56, lr: 4.17e-03, grad_scale: 8.0 2023-04-02 15:52:40,950 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1560, 1.7205, 1.3595, 1.1111, 1.5576, 1.0450, 1.1669, 1.6075], device='cuda:1'), covar=tensor([0.0760, 0.0724, 0.0970, 0.0763, 0.0476, 0.1198, 0.0580, 0.0387], device='cuda:1'), in_proj_covar=tensor([0.0304, 0.0313, 0.0330, 0.0261, 0.0244, 0.0336, 0.0290, 0.0274], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 15:52:56,051 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=130402.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 15:53:13,834 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.680e+02 5.222e+02 6.307e+02 8.250e+02 2.391e+03, threshold=1.261e+03, percent-clipped=5.0 2023-04-02 15:53:31,369 INFO [train.py:903] (1/4) Epoch 20, batch 700, loss[loss=0.2318, simple_loss=0.3062, pruned_loss=0.07873, over 19797.00 frames. ], tot_loss[loss=0.211, simple_loss=0.2898, pruned_loss=0.0661, over 3722534.60 frames. ], batch size: 48, lr: 4.17e-03, grad_scale: 8.0 2023-04-02 15:53:55,079 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=130450.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 15:54:02,062 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=130456.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 15:54:28,644 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=130477.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 15:54:32,041 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=130479.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 15:54:35,652 INFO [train.py:903] (1/4) Epoch 20, batch 750, loss[loss=0.2528, simple_loss=0.3185, pruned_loss=0.09355, over 12853.00 frames. ], tot_loss[loss=0.2116, simple_loss=0.29, pruned_loss=0.06654, over 3746515.48 frames. ], batch size: 135, lr: 4.16e-03, grad_scale: 8.0 2023-04-02 15:54:58,171 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 2023-04-02 15:55:19,217 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.433e+02 4.790e+02 6.042e+02 7.311e+02 1.890e+03, threshold=1.208e+03, percent-clipped=4.0 2023-04-02 15:55:28,441 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=130524.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 15:55:37,192 INFO [train.py:903] (1/4) Epoch 20, batch 800, loss[loss=0.2828, simple_loss=0.3368, pruned_loss=0.1143, over 13213.00 frames. ], tot_loss[loss=0.2121, simple_loss=0.2908, pruned_loss=0.06671, over 3754613.14 frames. ], batch size: 135, lr: 4.16e-03, grad_scale: 8.0 2023-04-02 15:55:53,534 WARNING [train.py:1073] (1/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 2023-04-02 15:55:58,692 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=130549.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 15:55:59,928 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1889, 1.3565, 1.8261, 1.4421, 2.9228, 4.4343, 4.3584, 4.9561], device='cuda:1'), covar=tensor([0.1691, 0.3914, 0.3439, 0.2433, 0.0651, 0.0259, 0.0194, 0.0194], device='cuda:1'), in_proj_covar=tensor([0.0268, 0.0316, 0.0347, 0.0262, 0.0238, 0.0182, 0.0213, 0.0248], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-02 15:56:21,753 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=130566.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 15:56:27,694 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4077, 1.4030, 1.7247, 1.6420, 2.5214, 2.3039, 2.6862, 1.0675], device='cuda:1'), covar=tensor([0.2456, 0.4419, 0.2676, 0.1947, 0.1580, 0.2063, 0.1469, 0.4510], device='cuda:1'), in_proj_covar=tensor([0.0526, 0.0633, 0.0696, 0.0476, 0.0616, 0.0526, 0.0659, 0.0540], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 15:56:40,732 INFO [train.py:903] (1/4) Epoch 20, batch 850, loss[loss=0.2148, simple_loss=0.3007, pruned_loss=0.06448, over 19785.00 frames. ], tot_loss[loss=0.2109, simple_loss=0.29, pruned_loss=0.06591, over 3786993.33 frames. ], batch size: 56, lr: 4.16e-03, grad_scale: 8.0 2023-04-02 15:56:41,930 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=130583.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 15:57:07,826 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.7326, 1.5319, 1.6100, 2.2274, 1.6511, 1.9684, 2.0123, 1.7539], device='cuda:1'), covar=tensor([0.0808, 0.0956, 0.1008, 0.0682, 0.0823, 0.0738, 0.0833, 0.0729], device='cuda:1'), in_proj_covar=tensor([0.0209, 0.0220, 0.0226, 0.0243, 0.0227, 0.0209, 0.0188, 0.0205], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-02 15:57:25,283 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.130e+02 4.920e+02 5.760e+02 7.852e+02 1.760e+03, threshold=1.152e+03, percent-clipped=6.0 2023-04-02 15:57:33,253 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9_rvb from training. Duration: 25.061125 2023-04-02 15:57:40,813 INFO [train.py:903] (1/4) Epoch 20, batch 900, loss[loss=0.2248, simple_loss=0.301, pruned_loss=0.07433, over 19624.00 frames. ], tot_loss[loss=0.212, simple_loss=0.2906, pruned_loss=0.06669, over 3802387.45 frames. ], batch size: 61, lr: 4.16e-03, grad_scale: 8.0 2023-04-02 15:58:44,423 INFO [train.py:903] (1/4) Epoch 20, batch 950, loss[loss=0.24, simple_loss=0.3199, pruned_loss=0.08006, over 19306.00 frames. ], tot_loss[loss=0.2115, simple_loss=0.2908, pruned_loss=0.06609, over 3814951.85 frames. ], batch size: 66, lr: 4.16e-03, grad_scale: 8.0 2023-04-02 15:58:47,640 WARNING [train.py:1073] (1/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9_rvb from training. Duration: 26.32775 2023-04-02 15:59:28,725 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.021e+02 4.928e+02 5.917e+02 7.294e+02 1.421e+03, threshold=1.183e+03, percent-clipped=3.0 2023-04-02 15:59:46,665 INFO [train.py:903] (1/4) Epoch 20, batch 1000, loss[loss=0.2219, simple_loss=0.3125, pruned_loss=0.06562, over 19683.00 frames. ], tot_loss[loss=0.2115, simple_loss=0.2909, pruned_loss=0.06605, over 3823154.55 frames. ], batch size: 58, lr: 4.16e-03, grad_scale: 8.0 2023-04-02 15:59:48,251 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=130733.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 15:59:50,495 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=130735.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 16:00:03,656 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=130746.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 16:00:16,572 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.13 vs. limit=2.0 2023-04-02 16:00:18,688 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=130758.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 16:00:20,900 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=130760.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 16:00:38,544 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 2023-04-02 16:00:48,485 INFO [train.py:903] (1/4) Epoch 20, batch 1050, loss[loss=0.229, simple_loss=0.3106, pruned_loss=0.07366, over 19539.00 frames. ], tot_loss[loss=0.2132, simple_loss=0.2921, pruned_loss=0.06712, over 3814874.61 frames. ], batch size: 56, lr: 4.16e-03, grad_scale: 8.0 2023-04-02 16:01:02,854 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=130794.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 16:01:10,716 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=130800.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 16:01:20,583 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 2023-04-02 16:01:33,077 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.731e+02 5.562e+02 6.742e+02 8.268e+02 2.102e+03, threshold=1.348e+03, percent-clipped=2.0 2023-04-02 16:01:49,837 INFO [train.py:903] (1/4) Epoch 20, batch 1100, loss[loss=0.2553, simple_loss=0.3192, pruned_loss=0.09567, over 13393.00 frames. ], tot_loss[loss=0.2136, simple_loss=0.2922, pruned_loss=0.06749, over 3806442.35 frames. ], batch size: 137, lr: 4.16e-03, grad_scale: 4.0 2023-04-02 16:02:27,183 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=130861.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 16:02:36,683 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.09 vs. limit=2.0 2023-04-02 16:02:47,563 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.59 vs. limit=2.0 2023-04-02 16:02:52,369 INFO [train.py:903] (1/4) Epoch 20, batch 1150, loss[loss=0.1655, simple_loss=0.2403, pruned_loss=0.0453, over 19704.00 frames. ], tot_loss[loss=0.2127, simple_loss=0.2915, pruned_loss=0.06696, over 3804674.81 frames. ], batch size: 45, lr: 4.16e-03, grad_scale: 4.0 2023-04-02 16:03:26,707 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=130909.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 16:03:27,635 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=130910.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 16:03:33,827 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=130915.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 16:03:39,054 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.314e+02 5.063e+02 6.056e+02 7.993e+02 1.743e+03, threshold=1.211e+03, percent-clipped=5.0 2023-04-02 16:03:50,400 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=130927.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 16:03:55,680 INFO [train.py:903] (1/4) Epoch 20, batch 1200, loss[loss=0.2324, simple_loss=0.3047, pruned_loss=0.08009, over 17286.00 frames. ], tot_loss[loss=0.2124, simple_loss=0.2913, pruned_loss=0.06678, over 3808974.91 frames. ], batch size: 101, lr: 4.16e-03, grad_scale: 8.0 2023-04-02 16:04:23,988 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 2023-04-02 16:04:56,110 INFO [train.py:903] (1/4) Epoch 20, batch 1250, loss[loss=0.1904, simple_loss=0.2663, pruned_loss=0.05725, over 19851.00 frames. ], tot_loss[loss=0.2123, simple_loss=0.291, pruned_loss=0.06681, over 3805417.38 frames. ], batch size: 52, lr: 4.16e-03, grad_scale: 8.0 2023-04-02 16:05:42,763 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.304e+02 5.121e+02 6.297e+02 7.673e+02 2.016e+03, threshold=1.259e+03, percent-clipped=4.0 2023-04-02 16:05:50,221 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=131025.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 16:05:52,263 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1475, 1.2459, 1.6630, 1.0660, 2.4528, 3.3433, 3.0120, 3.5282], device='cuda:1'), covar=tensor([0.1627, 0.3770, 0.3353, 0.2557, 0.0611, 0.0191, 0.0222, 0.0239], device='cuda:1'), in_proj_covar=tensor([0.0268, 0.0317, 0.0348, 0.0262, 0.0238, 0.0183, 0.0213, 0.0249], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-02 16:05:58,785 INFO [train.py:903] (1/4) Epoch 20, batch 1300, loss[loss=0.1875, simple_loss=0.2807, pruned_loss=0.04713, over 19657.00 frames. ], tot_loss[loss=0.2127, simple_loss=0.2917, pruned_loss=0.06689, over 3801452.93 frames. ], batch size: 53, lr: 4.16e-03, grad_scale: 8.0 2023-04-02 16:06:12,045 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=131042.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 16:06:32,914 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.7872, 4.2979, 4.4788, 4.4941, 1.5728, 4.1504, 3.6451, 4.1624], device='cuda:1'), covar=tensor([0.1705, 0.0866, 0.0643, 0.0682, 0.6209, 0.0912, 0.0680, 0.1214], device='cuda:1'), in_proj_covar=tensor([0.0758, 0.0716, 0.0917, 0.0799, 0.0815, 0.0675, 0.0551, 0.0856], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-02 16:06:59,469 INFO [train.py:903] (1/4) Epoch 20, batch 1350, loss[loss=0.2295, simple_loss=0.3116, pruned_loss=0.07363, over 19665.00 frames. ], tot_loss[loss=0.2117, simple_loss=0.2902, pruned_loss=0.06658, over 3812771.02 frames. ], batch size: 58, lr: 4.15e-03, grad_scale: 8.0 2023-04-02 16:07:19,071 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.12 vs. limit=2.0 2023-04-02 16:07:43,038 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=131117.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 16:07:44,787 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.358e+02 5.090e+02 6.517e+02 8.267e+02 2.193e+03, threshold=1.303e+03, percent-clipped=8.0 2023-04-02 16:08:02,288 INFO [train.py:903] (1/4) Epoch 20, batch 1400, loss[loss=0.2196, simple_loss=0.2852, pruned_loss=0.07697, over 19737.00 frames. ], tot_loss[loss=0.2112, simple_loss=0.2897, pruned_loss=0.06635, over 3814347.74 frames. ], batch size: 45, lr: 4.15e-03, grad_scale: 8.0 2023-04-02 16:08:15,129 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2688, 1.1937, 1.2345, 1.3512, 1.0290, 1.3608, 1.3660, 1.2598], device='cuda:1'), covar=tensor([0.0875, 0.0986, 0.1063, 0.0678, 0.0854, 0.0807, 0.0794, 0.0791], device='cuda:1'), in_proj_covar=tensor([0.0209, 0.0219, 0.0225, 0.0243, 0.0226, 0.0209, 0.0188, 0.0204], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-02 16:08:15,156 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=131142.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 16:08:42,763 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=131165.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 16:08:50,688 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=131171.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 16:09:03,834 INFO [train.py:903] (1/4) Epoch 20, batch 1450, loss[loss=0.1777, simple_loss=0.2533, pruned_loss=0.05105, over 19767.00 frames. ], tot_loss[loss=0.211, simple_loss=0.2898, pruned_loss=0.06609, over 3822276.14 frames. ], batch size: 46, lr: 4.15e-03, grad_scale: 8.0 2023-04-02 16:09:06,068 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_rvb from training. Duration: 25.85 2023-04-02 16:09:14,494 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=131190.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 16:09:16,703 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=131192.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 16:09:21,595 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=131196.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 16:09:50,840 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.524e+02 5.028e+02 6.181e+02 7.641e+02 1.699e+03, threshold=1.236e+03, percent-clipped=6.0 2023-04-02 16:10:06,715 INFO [train.py:903] (1/4) Epoch 20, batch 1500, loss[loss=0.1931, simple_loss=0.2747, pruned_loss=0.05576, over 19532.00 frames. ], tot_loss[loss=0.2103, simple_loss=0.2893, pruned_loss=0.06571, over 3828604.59 frames. ], batch size: 54, lr: 4.15e-03, grad_scale: 8.0 2023-04-02 16:10:21,878 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.6115, 4.1873, 2.7055, 3.7268, 1.2713, 4.1274, 4.0070, 4.1739], device='cuda:1'), covar=tensor([0.0581, 0.0972, 0.1921, 0.0822, 0.3531, 0.0714, 0.0803, 0.1005], device='cuda:1'), in_proj_covar=tensor([0.0494, 0.0397, 0.0485, 0.0345, 0.0400, 0.0423, 0.0416, 0.0447], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 16:10:49,079 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.29 vs. limit=5.0 2023-04-02 16:11:07,004 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=131281.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 16:11:07,712 INFO [train.py:903] (1/4) Epoch 20, batch 1550, loss[loss=0.2397, simple_loss=0.3145, pruned_loss=0.08239, over 19539.00 frames. ], tot_loss[loss=0.21, simple_loss=0.2889, pruned_loss=0.06553, over 3843539.61 frames. ], batch size: 56, lr: 4.15e-03, grad_scale: 8.0 2023-04-02 16:11:29,260 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=131298.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 16:11:38,410 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=131306.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 16:11:53,762 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.770e+02 5.087e+02 6.243e+02 7.473e+02 1.350e+03, threshold=1.249e+03, percent-clipped=1.0 2023-04-02 16:11:58,738 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=131323.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 16:12:08,235 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.5550, 1.1978, 1.3030, 1.2078, 2.2178, 0.9612, 2.0724, 2.4910], device='cuda:1'), covar=tensor([0.0664, 0.2691, 0.2966, 0.1672, 0.0860, 0.2120, 0.1035, 0.0455], device='cuda:1'), in_proj_covar=tensor([0.0399, 0.0357, 0.0377, 0.0339, 0.0368, 0.0348, 0.0371, 0.0393], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 16:12:10,215 INFO [train.py:903] (1/4) Epoch 20, batch 1600, loss[loss=0.1652, simple_loss=0.2488, pruned_loss=0.04076, over 19469.00 frames. ], tot_loss[loss=0.2108, simple_loss=0.2897, pruned_loss=0.06591, over 3831072.46 frames. ], batch size: 49, lr: 4.15e-03, grad_scale: 8.0 2023-04-02 16:12:36,115 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9_rvb from training. Duration: 30.1555625 2023-04-02 16:13:12,822 INFO [train.py:903] (1/4) Epoch 20, batch 1650, loss[loss=0.2017, simple_loss=0.2842, pruned_loss=0.05961, over 19713.00 frames. ], tot_loss[loss=0.2101, simple_loss=0.2891, pruned_loss=0.0656, over 3833449.33 frames. ], batch size: 63, lr: 4.15e-03, grad_scale: 8.0 2023-04-02 16:13:59,215 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.697e+02 5.218e+02 6.304e+02 8.075e+02 1.501e+03, threshold=1.261e+03, percent-clipped=5.0 2023-04-02 16:14:08,599 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=131427.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 16:14:15,058 INFO [train.py:903] (1/4) Epoch 20, batch 1700, loss[loss=0.196, simple_loss=0.2727, pruned_loss=0.05968, over 19793.00 frames. ], tot_loss[loss=0.2119, simple_loss=0.2904, pruned_loss=0.06674, over 3836519.51 frames. ], batch size: 48, lr: 4.15e-03, grad_scale: 8.0 2023-04-02 16:14:17,648 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=131434.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 16:14:55,857 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 2023-04-02 16:15:16,165 INFO [train.py:903] (1/4) Epoch 20, batch 1750, loss[loss=0.2234, simple_loss=0.3046, pruned_loss=0.07114, over 19777.00 frames. ], tot_loss[loss=0.2121, simple_loss=0.2905, pruned_loss=0.06682, over 3835280.01 frames. ], batch size: 56, lr: 4.15e-03, grad_scale: 8.0 2023-04-02 16:15:52,703 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=131510.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 16:16:02,698 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.468e+02 5.359e+02 6.412e+02 7.691e+02 1.507e+03, threshold=1.282e+03, percent-clipped=3.0 2023-04-02 16:16:18,696 INFO [train.py:903] (1/4) Epoch 20, batch 1800, loss[loss=0.1991, simple_loss=0.2848, pruned_loss=0.05671, over 19612.00 frames. ], tot_loss[loss=0.2124, simple_loss=0.2913, pruned_loss=0.06678, over 3829863.67 frames. ], batch size: 57, lr: 4.15e-03, grad_scale: 4.0 2023-04-02 16:16:24,467 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=131536.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 16:17:16,067 WARNING [train.py:1073] (1/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9_rvb from training. Duration: 25.0944375 2023-04-02 16:17:22,070 INFO [train.py:903] (1/4) Epoch 20, batch 1850, loss[loss=0.2115, simple_loss=0.2855, pruned_loss=0.06876, over 19482.00 frames. ], tot_loss[loss=0.2131, simple_loss=0.2921, pruned_loss=0.06703, over 3824415.01 frames. ], batch size: 49, lr: 4.15e-03, grad_scale: 4.0 2023-04-02 16:17:33,624 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.9078, 1.5675, 1.8834, 1.7224, 4.3968, 1.0256, 2.5175, 4.8205], device='cuda:1'), covar=tensor([0.0443, 0.2790, 0.2636, 0.1900, 0.0784, 0.2740, 0.1463, 0.0171], device='cuda:1'), in_proj_covar=tensor([0.0396, 0.0355, 0.0375, 0.0337, 0.0366, 0.0346, 0.0370, 0.0390], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 16:17:54,261 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9_rvb from training. Duration: 27.8166875 2023-04-02 16:18:09,351 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3101, 1.3529, 1.7984, 1.5735, 2.5823, 2.0545, 2.6842, 1.1311], device='cuda:1'), covar=tensor([0.2732, 0.4625, 0.2827, 0.2127, 0.1662, 0.2523, 0.1714, 0.4764], device='cuda:1'), in_proj_covar=tensor([0.0526, 0.0631, 0.0694, 0.0474, 0.0612, 0.0526, 0.0658, 0.0539], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 16:18:09,980 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.335e+02 4.826e+02 6.634e+02 9.045e+02 2.049e+03, threshold=1.327e+03, percent-clipped=6.0 2023-04-02 16:18:25,161 INFO [train.py:903] (1/4) Epoch 20, batch 1900, loss[loss=0.2333, simple_loss=0.3098, pruned_loss=0.07842, over 19493.00 frames. ], tot_loss[loss=0.213, simple_loss=0.2918, pruned_loss=0.06712, over 3819963.59 frames. ], batch size: 64, lr: 4.15e-03, grad_scale: 4.0 2023-04-02 16:18:40,113 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 2023-04-02 16:18:45,485 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9_rvb from training. Duration: 27.02225 2023-04-02 16:18:47,959 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=131651.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 16:19:10,725 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 2023-04-02 16:19:25,446 INFO [train.py:903] (1/4) Epoch 20, batch 1950, loss[loss=0.1962, simple_loss=0.2847, pruned_loss=0.05383, over 19517.00 frames. ], tot_loss[loss=0.2136, simple_loss=0.2925, pruned_loss=0.06737, over 3827553.28 frames. ], batch size: 64, lr: 4.15e-03, grad_scale: 4.0 2023-04-02 16:19:49,191 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([0.9414, 1.3089, 1.0532, 0.9264, 1.1933, 0.8839, 0.9742, 1.2089], device='cuda:1'), covar=tensor([0.0719, 0.0673, 0.0788, 0.0646, 0.0439, 0.1029, 0.0502, 0.0401], device='cuda:1'), in_proj_covar=tensor([0.0306, 0.0316, 0.0335, 0.0263, 0.0248, 0.0337, 0.0293, 0.0277], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 16:20:03,223 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([0.8630, 1.3329, 1.0375, 0.9156, 1.1364, 0.9354, 0.8897, 1.2063], device='cuda:1'), covar=tensor([0.0651, 0.0824, 0.1134, 0.0769, 0.0582, 0.1319, 0.0648, 0.0497], device='cuda:1'), in_proj_covar=tensor([0.0305, 0.0315, 0.0335, 0.0263, 0.0247, 0.0336, 0.0292, 0.0277], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 16:20:13,300 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.458e+02 5.129e+02 6.435e+02 8.344e+02 2.370e+03, threshold=1.287e+03, percent-clipped=2.0 2023-04-02 16:20:28,633 INFO [train.py:903] (1/4) Epoch 20, batch 2000, loss[loss=0.2009, simple_loss=0.2781, pruned_loss=0.06188, over 19840.00 frames. ], tot_loss[loss=0.2141, simple_loss=0.2927, pruned_loss=0.06775, over 3813152.00 frames. ], batch size: 52, lr: 4.14e-03, grad_scale: 8.0 2023-04-02 16:21:18,590 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=131771.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 16:21:25,583 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_rvb from training. Duration: 26.8349375 2023-04-02 16:21:28,782 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=131778.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 16:21:33,329 INFO [train.py:903] (1/4) Epoch 20, batch 2050, loss[loss=0.2025, simple_loss=0.2915, pruned_loss=0.05673, over 18811.00 frames. ], tot_loss[loss=0.2126, simple_loss=0.2914, pruned_loss=0.0669, over 3812367.74 frames. ], batch size: 74, lr: 4.14e-03, grad_scale: 8.0 2023-04-02 16:21:45,617 WARNING [train.py:1073] (1/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 2023-04-02 16:21:46,760 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 2023-04-02 16:22:07,707 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 2023-04-02 16:22:22,558 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.239e+02 4.827e+02 6.005e+02 7.777e+02 1.829e+03, threshold=1.201e+03, percent-clipped=4.0 2023-04-02 16:22:35,844 INFO [train.py:903] (1/4) Epoch 20, batch 2100, loss[loss=0.1822, simple_loss=0.2574, pruned_loss=0.05344, over 19796.00 frames. ], tot_loss[loss=0.2123, simple_loss=0.291, pruned_loss=0.0668, over 3815412.02 frames. ], batch size: 45, lr: 4.14e-03, grad_scale: 4.0 2023-04-02 16:22:39,934 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.7629, 1.8652, 2.1298, 2.3293, 1.7569, 2.2258, 2.1985, 1.9095], device='cuda:1'), covar=tensor([0.3860, 0.3326, 0.1756, 0.2164, 0.3592, 0.1921, 0.4304, 0.3285], device='cuda:1'), in_proj_covar=tensor([0.0881, 0.0943, 0.0703, 0.0925, 0.0863, 0.0796, 0.0828, 0.0773], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 16:23:03,392 WARNING [train.py:1073] (1/4) Exclude cut with ID 453-131332-0000-131866_sp0.9_rvb from training. Duration: 25.3333125 2023-04-02 16:23:03,509 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=131854.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 16:23:26,825 WARNING [train.py:1073] (1/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9_rvb from training. Duration: 26.6166875 2023-04-02 16:23:27,226 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5910, 2.3284, 1.6153, 1.6158, 2.1443, 1.3082, 1.4791, 1.9727], device='cuda:1'), covar=tensor([0.1128, 0.0795, 0.1104, 0.0809, 0.0542, 0.1346, 0.0748, 0.0525], device='cuda:1'), in_proj_covar=tensor([0.0306, 0.0316, 0.0335, 0.0262, 0.0247, 0.0339, 0.0294, 0.0278], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 16:23:29,482 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=131875.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 16:23:37,620 INFO [train.py:903] (1/4) Epoch 20, batch 2150, loss[loss=0.2498, simple_loss=0.3256, pruned_loss=0.08699, over 18182.00 frames. ], tot_loss[loss=0.212, simple_loss=0.2907, pruned_loss=0.06662, over 3814938.11 frames. ], batch size: 83, lr: 4.14e-03, grad_scale: 4.0 2023-04-02 16:23:42,652 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=131886.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 16:23:51,985 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=131893.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 16:24:09,630 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=131907.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 16:24:26,290 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.293e+02 5.118e+02 6.039e+02 8.265e+02 1.505e+03, threshold=1.208e+03, percent-clipped=4.0 2023-04-02 16:24:39,694 INFO [train.py:903] (1/4) Epoch 20, batch 2200, loss[loss=0.2319, simple_loss=0.3127, pruned_loss=0.07555, over 19371.00 frames. ], tot_loss[loss=0.2117, simple_loss=0.2906, pruned_loss=0.06641, over 3821327.44 frames. ], batch size: 70, lr: 4.14e-03, grad_scale: 4.0 2023-04-02 16:24:40,112 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=131932.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 16:25:10,258 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.7501, 4.2655, 4.4836, 4.4855, 1.7393, 4.2176, 3.6421, 4.2071], device='cuda:1'), covar=tensor([0.1602, 0.0738, 0.0570, 0.0597, 0.5485, 0.0846, 0.0669, 0.1022], device='cuda:1'), in_proj_covar=tensor([0.0762, 0.0714, 0.0920, 0.0807, 0.0814, 0.0680, 0.0553, 0.0855], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-02 16:25:26,188 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=131969.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 16:25:42,868 INFO [train.py:903] (1/4) Epoch 20, batch 2250, loss[loss=0.173, simple_loss=0.2464, pruned_loss=0.04975, over 19739.00 frames. ], tot_loss[loss=0.2104, simple_loss=0.2893, pruned_loss=0.06572, over 3827648.15 frames. ], batch size: 46, lr: 4.14e-03, grad_scale: 4.0 2023-04-02 16:25:54,526 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.8015, 3.2911, 3.3292, 3.3483, 1.3206, 3.1867, 2.8159, 3.0934], device='cuda:1'), covar=tensor([0.1743, 0.0992, 0.0789, 0.0956, 0.5429, 0.1069, 0.0769, 0.1332], device='cuda:1'), in_proj_covar=tensor([0.0762, 0.0714, 0.0918, 0.0806, 0.0813, 0.0678, 0.0553, 0.0855], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-02 16:26:31,961 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.368e+02 4.978e+02 6.272e+02 7.965e+02 1.499e+03, threshold=1.254e+03, percent-clipped=2.0 2023-04-02 16:26:44,538 INFO [train.py:903] (1/4) Epoch 20, batch 2300, loss[loss=0.2037, simple_loss=0.2839, pruned_loss=0.06175, over 19530.00 frames. ], tot_loss[loss=0.2114, simple_loss=0.29, pruned_loss=0.06645, over 3819794.31 frames. ], batch size: 54, lr: 4.14e-03, grad_scale: 4.0 2023-04-02 16:26:52,921 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.1223, 1.7675, 1.8748, 2.6215, 2.1657, 2.3104, 2.4238, 2.0369], device='cuda:1'), covar=tensor([0.0791, 0.0915, 0.0996, 0.0884, 0.0824, 0.0795, 0.0824, 0.0692], device='cuda:1'), in_proj_covar=tensor([0.0211, 0.0221, 0.0225, 0.0243, 0.0228, 0.0211, 0.0187, 0.0205], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-02 16:26:58,039 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_rvb from training. Duration: 26.205 2023-04-02 16:27:05,365 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=132049.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 16:27:34,690 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=132072.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 16:27:47,302 INFO [train.py:903] (1/4) Epoch 20, batch 2350, loss[loss=0.2646, simple_loss=0.3285, pruned_loss=0.1004, over 12716.00 frames. ], tot_loss[loss=0.2119, simple_loss=0.2905, pruned_loss=0.06659, over 3810230.52 frames. ], batch size: 136, lr: 4.14e-03, grad_scale: 4.0 2023-04-02 16:28:26,805 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_rvb from training. Duration: 25.775 2023-04-02 16:28:35,982 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.260e+02 4.795e+02 5.776e+02 7.778e+02 1.972e+03, threshold=1.155e+03, percent-clipped=8.0 2023-04-02 16:28:42,737 WARNING [train.py:1073] (1/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9_rvb from training. Duration: 25.45 2023-04-02 16:28:49,775 INFO [train.py:903] (1/4) Epoch 20, batch 2400, loss[loss=0.2024, simple_loss=0.2863, pruned_loss=0.05925, over 19652.00 frames. ], tot_loss[loss=0.2117, simple_loss=0.2905, pruned_loss=0.06644, over 3816150.43 frames. ], batch size: 55, lr: 4.14e-03, grad_scale: 8.0 2023-04-02 16:29:03,470 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=132142.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 16:29:11,445 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=132149.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 16:29:32,890 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=132167.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 16:29:41,688 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=132174.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 16:29:51,981 INFO [train.py:903] (1/4) Epoch 20, batch 2450, loss[loss=0.2348, simple_loss=0.311, pruned_loss=0.07935, over 18122.00 frames. ], tot_loss[loss=0.2127, simple_loss=0.2916, pruned_loss=0.06693, over 3817634.23 frames. ], batch size: 83, lr: 4.14e-03, grad_scale: 8.0 2023-04-02 16:30:07,508 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.47 vs. limit=5.0 2023-04-02 16:30:38,900 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=132219.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 16:30:41,020 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.150e+02 4.995e+02 6.263e+02 8.063e+02 1.363e+03, threshold=1.253e+03, percent-clipped=5.0 2023-04-02 16:30:46,976 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=132225.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 16:30:54,279 INFO [train.py:903] (1/4) Epoch 20, batch 2500, loss[loss=0.1845, simple_loss=0.2694, pruned_loss=0.04984, over 19619.00 frames. ], tot_loss[loss=0.2144, simple_loss=0.2931, pruned_loss=0.0678, over 3803533.68 frames. ], batch size: 50, lr: 4.14e-03, grad_scale: 8.0 2023-04-02 16:31:11,670 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.70 vs. limit=5.0 2023-04-02 16:31:15,980 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=132250.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 16:31:56,367 INFO [train.py:903] (1/4) Epoch 20, batch 2550, loss[loss=0.2387, simple_loss=0.3155, pruned_loss=0.08096, over 19741.00 frames. ], tot_loss[loss=0.2141, simple_loss=0.2933, pruned_loss=0.06749, over 3808530.72 frames. ], batch size: 63, lr: 4.14e-03, grad_scale: 8.0 2023-04-02 16:32:45,610 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.513e+02 4.985e+02 5.787e+02 8.066e+02 1.995e+03, threshold=1.157e+03, percent-clipped=4.0 2023-04-02 16:32:52,686 WARNING [train.py:1073] (1/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 2023-04-02 16:32:58,515 INFO [train.py:903] (1/4) Epoch 20, batch 2600, loss[loss=0.2247, simple_loss=0.302, pruned_loss=0.07371, over 19469.00 frames. ], tot_loss[loss=0.2129, simple_loss=0.2921, pruned_loss=0.06688, over 3800101.11 frames. ], batch size: 49, lr: 4.14e-03, grad_scale: 8.0 2023-04-02 16:33:02,434 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=132334.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 16:33:43,880 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.4284, 1.3575, 1.4770, 1.5193, 2.9942, 1.2055, 2.3036, 3.4289], device='cuda:1'), covar=tensor([0.0516, 0.2734, 0.2912, 0.1844, 0.0742, 0.2441, 0.1295, 0.0267], device='cuda:1'), in_proj_covar=tensor([0.0397, 0.0356, 0.0376, 0.0340, 0.0367, 0.0347, 0.0371, 0.0392], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 16:33:44,023 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3998, 1.5043, 1.7935, 1.6038, 2.3879, 2.1901, 2.4768, 1.0247], device='cuda:1'), covar=tensor([0.2509, 0.4226, 0.2550, 0.1956, 0.1599, 0.2090, 0.1527, 0.4453], device='cuda:1'), in_proj_covar=tensor([0.0528, 0.0632, 0.0698, 0.0477, 0.0616, 0.0526, 0.0660, 0.0541], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 16:33:45,144 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9620, 1.7346, 1.5860, 1.8931, 1.6635, 1.7293, 1.6335, 1.8093], device='cuda:1'), covar=tensor([0.1059, 0.1494, 0.1521, 0.1141, 0.1404, 0.0532, 0.1325, 0.0763], device='cuda:1'), in_proj_covar=tensor([0.0267, 0.0357, 0.0310, 0.0251, 0.0301, 0.0249, 0.0305, 0.0253], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 16:34:01,396 INFO [train.py:903] (1/4) Epoch 20, batch 2650, loss[loss=0.2388, simple_loss=0.3222, pruned_loss=0.07775, over 18740.00 frames. ], tot_loss[loss=0.2138, simple_loss=0.2931, pruned_loss=0.06732, over 3796494.18 frames. ], batch size: 74, lr: 4.13e-03, grad_scale: 8.0 2023-04-02 16:34:15,342 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=132393.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 16:34:23,152 WARNING [train.py:1073] (1/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9_rvb from training. Duration: 27.25 2023-04-02 16:34:26,516 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0 2023-04-02 16:34:44,196 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=132416.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 16:34:50,448 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.050e+02 4.865e+02 6.149e+02 7.368e+02 1.585e+03, threshold=1.230e+03, percent-clipped=4.0 2023-04-02 16:35:04,062 INFO [train.py:903] (1/4) Epoch 20, batch 2700, loss[loss=0.1917, simple_loss=0.2714, pruned_loss=0.05595, over 19832.00 frames. ], tot_loss[loss=0.2139, simple_loss=0.2928, pruned_loss=0.06744, over 3813343.40 frames. ], batch size: 52, lr: 4.13e-03, grad_scale: 8.0 2023-04-02 16:35:33,672 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6638, 1.4122, 1.4934, 2.1871, 1.6346, 1.8948, 1.8824, 1.7310], device='cuda:1'), covar=tensor([0.0861, 0.0977, 0.1020, 0.0776, 0.0908, 0.0786, 0.0874, 0.0687], device='cuda:1'), in_proj_covar=tensor([0.0209, 0.0219, 0.0223, 0.0241, 0.0227, 0.0210, 0.0186, 0.0203], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-02 16:36:06,698 INFO [train.py:903] (1/4) Epoch 20, batch 2750, loss[loss=0.2265, simple_loss=0.3069, pruned_loss=0.07303, over 18583.00 frames. ], tot_loss[loss=0.2135, simple_loss=0.2926, pruned_loss=0.06722, over 3818975.39 frames. ], batch size: 74, lr: 4.13e-03, grad_scale: 8.0 2023-04-02 16:36:39,128 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=132508.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 16:36:55,655 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.965e+02 5.186e+02 6.180e+02 7.968e+02 1.505e+03, threshold=1.236e+03, percent-clipped=2.0 2023-04-02 16:37:07,844 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=132531.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 16:37:08,656 INFO [train.py:903] (1/4) Epoch 20, batch 2800, loss[loss=0.2101, simple_loss=0.298, pruned_loss=0.06112, over 18131.00 frames. ], tot_loss[loss=0.2125, simple_loss=0.2917, pruned_loss=0.06667, over 3820795.16 frames. ], batch size: 83, lr: 4.13e-03, grad_scale: 8.0 2023-04-02 16:37:39,113 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.7278, 1.6829, 1.5641, 1.4008, 1.3011, 1.4076, 0.2840, 0.6211], device='cuda:1'), covar=tensor([0.0594, 0.0620, 0.0392, 0.0615, 0.1204, 0.0711, 0.1209, 0.1084], device='cuda:1'), in_proj_covar=tensor([0.0358, 0.0353, 0.0356, 0.0381, 0.0456, 0.0385, 0.0333, 0.0338], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 16:37:50,883 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=4.12 vs. limit=5.0 2023-04-02 16:38:13,010 INFO [train.py:903] (1/4) Epoch 20, batch 2850, loss[loss=0.1979, simple_loss=0.2779, pruned_loss=0.05899, over 19628.00 frames. ], tot_loss[loss=0.2126, simple_loss=0.2916, pruned_loss=0.0668, over 3815500.41 frames. ], batch size: 50, lr: 4.13e-03, grad_scale: 4.0 2023-04-02 16:38:22,443 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=132590.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 16:38:29,077 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.1278, 1.3227, 1.4393, 1.2993, 2.7299, 1.0953, 2.1166, 3.0736], device='cuda:1'), covar=tensor([0.0538, 0.2561, 0.2747, 0.1800, 0.0728, 0.2314, 0.1186, 0.0333], device='cuda:1'), in_proj_covar=tensor([0.0399, 0.0356, 0.0376, 0.0339, 0.0369, 0.0347, 0.0371, 0.0392], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 16:38:46,039 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.8080, 1.3590, 1.5735, 1.4402, 3.3399, 1.1900, 2.5436, 3.7816], device='cuda:1'), covar=tensor([0.0456, 0.2670, 0.2683, 0.1933, 0.0732, 0.2462, 0.1124, 0.0241], device='cuda:1'), in_proj_covar=tensor([0.0399, 0.0357, 0.0376, 0.0339, 0.0369, 0.0347, 0.0371, 0.0392], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 16:38:52,867 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=132615.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 16:39:01,716 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.639e+02 4.861e+02 5.815e+02 7.642e+02 3.357e+03, threshold=1.163e+03, percent-clipped=7.0 2023-04-02 16:39:14,635 INFO [train.py:903] (1/4) Epoch 20, batch 2900, loss[loss=0.2534, simple_loss=0.3319, pruned_loss=0.08749, over 19617.00 frames. ], tot_loss[loss=0.2119, simple_loss=0.2911, pruned_loss=0.06638, over 3824775.06 frames. ], batch size: 61, lr: 4.13e-03, grad_scale: 4.0 2023-04-02 16:39:14,672 WARNING [train.py:1073] (1/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 2023-04-02 16:40:13,481 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=132678.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 16:40:18,627 INFO [train.py:903] (1/4) Epoch 20, batch 2950, loss[loss=0.253, simple_loss=0.3244, pruned_loss=0.09079, over 19583.00 frames. ], tot_loss[loss=0.2108, simple_loss=0.2902, pruned_loss=0.06568, over 3825686.48 frames. ], batch size: 61, lr: 4.13e-03, grad_scale: 4.0 2023-04-02 16:41:09,138 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.010e+02 4.627e+02 5.679e+02 7.371e+02 2.153e+03, threshold=1.136e+03, percent-clipped=3.0 2023-04-02 16:41:20,817 INFO [train.py:903] (1/4) Epoch 20, batch 3000, loss[loss=0.2249, simple_loss=0.3025, pruned_loss=0.07362, over 19107.00 frames. ], tot_loss[loss=0.2119, simple_loss=0.2911, pruned_loss=0.06636, over 3828751.85 frames. ], batch size: 69, lr: 4.13e-03, grad_scale: 4.0 2023-04-02 16:41:20,818 INFO [train.py:928] (1/4) Computing validation loss 2023-04-02 16:41:34,264 INFO [train.py:937] (1/4) Epoch 20, validation: loss=0.1695, simple_loss=0.2697, pruned_loss=0.03462, over 944034.00 frames. 2023-04-02 16:41:34,266 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18821MB 2023-04-02 16:41:40,250 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_rvb from training. Duration: 29.735 2023-04-02 16:42:10,873 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.3344, 3.7676, 3.9089, 3.9274, 1.5405, 3.7154, 3.2817, 3.6406], device='cuda:1'), covar=tensor([0.1657, 0.1028, 0.0712, 0.0770, 0.5886, 0.0993, 0.0677, 0.1257], device='cuda:1'), in_proj_covar=tensor([0.0765, 0.0723, 0.0919, 0.0806, 0.0819, 0.0682, 0.0555, 0.0859], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-02 16:42:12,905 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=132764.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 16:42:13,046 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=132764.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 16:42:15,795 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.09 vs. limit=2.0 2023-04-02 16:42:35,225 INFO [train.py:903] (1/4) Epoch 20, batch 3050, loss[loss=0.2228, simple_loss=0.2984, pruned_loss=0.07361, over 18840.00 frames. ], tot_loss[loss=0.212, simple_loss=0.2911, pruned_loss=0.06648, over 3831051.97 frames. ], batch size: 74, lr: 4.13e-03, grad_scale: 4.0 2023-04-02 16:42:41,468 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=132787.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 16:42:43,550 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=132789.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 16:43:13,096 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=132812.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 16:43:24,234 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.615e+02 4.785e+02 6.187e+02 7.720e+02 1.879e+03, threshold=1.237e+03, percent-clipped=7.0 2023-04-02 16:43:28,321 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9589, 2.0479, 2.2557, 2.6215, 1.8898, 2.4034, 2.2780, 2.0783], device='cuda:1'), covar=tensor([0.4317, 0.4107, 0.1909, 0.2523, 0.4256, 0.2341, 0.4792, 0.3438], device='cuda:1'), in_proj_covar=tensor([0.0886, 0.0948, 0.0708, 0.0931, 0.0866, 0.0800, 0.0837, 0.0776], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 16:43:37,016 INFO [train.py:903] (1/4) Epoch 20, batch 3100, loss[loss=0.1858, simple_loss=0.2755, pruned_loss=0.04803, over 19676.00 frames. ], tot_loss[loss=0.213, simple_loss=0.2922, pruned_loss=0.06696, over 3827057.55 frames. ], batch size: 53, lr: 4.13e-03, grad_scale: 4.0 2023-04-02 16:43:39,678 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.7443, 1.7445, 1.5899, 1.3953, 1.4362, 1.3791, 0.1391, 0.6074], device='cuda:1'), covar=tensor([0.0655, 0.0604, 0.0420, 0.0647, 0.1240, 0.0779, 0.1224, 0.1115], device='cuda:1'), in_proj_covar=tensor([0.0360, 0.0352, 0.0357, 0.0383, 0.0457, 0.0386, 0.0334, 0.0339], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 16:44:33,821 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.7707, 4.2833, 4.4998, 4.5034, 1.6943, 4.2115, 3.6609, 4.2334], device='cuda:1'), covar=tensor([0.1721, 0.0833, 0.0655, 0.0697, 0.6138, 0.0853, 0.0715, 0.1263], device='cuda:1'), in_proj_covar=tensor([0.0766, 0.0722, 0.0921, 0.0807, 0.0821, 0.0682, 0.0556, 0.0859], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-02 16:44:40,220 INFO [train.py:903] (1/4) Epoch 20, batch 3150, loss[loss=0.2979, simple_loss=0.3653, pruned_loss=0.1153, over 19260.00 frames. ], tot_loss[loss=0.2127, simple_loss=0.2917, pruned_loss=0.06685, over 3827656.18 frames. ], batch size: 66, lr: 4.13e-03, grad_scale: 4.0 2023-04-02 16:45:07,825 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 2023-04-02 16:45:11,718 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5400, 1.6097, 1.8454, 1.7845, 2.5894, 2.2185, 2.6485, 1.3759], device='cuda:1'), covar=tensor([0.2041, 0.3585, 0.2203, 0.1630, 0.1219, 0.1830, 0.1254, 0.3741], device='cuda:1'), in_proj_covar=tensor([0.0524, 0.0630, 0.0694, 0.0476, 0.0612, 0.0523, 0.0656, 0.0540], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 16:45:29,826 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.089e+02 5.025e+02 5.951e+02 7.011e+02 1.371e+03, threshold=1.190e+03, percent-clipped=2.0 2023-04-02 16:45:42,514 INFO [train.py:903] (1/4) Epoch 20, batch 3200, loss[loss=0.2901, simple_loss=0.3431, pruned_loss=0.1185, over 13651.00 frames. ], tot_loss[loss=0.213, simple_loss=0.2919, pruned_loss=0.067, over 3812556.08 frames. ], batch size: 136, lr: 4.13e-03, grad_scale: 8.0 2023-04-02 16:46:13,784 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.1674, 2.9189, 2.3278, 2.1496, 2.0475, 2.5153, 0.9261, 2.0482], device='cuda:1'), covar=tensor([0.0644, 0.0557, 0.0679, 0.1101, 0.1134, 0.1062, 0.1448, 0.1096], device='cuda:1'), in_proj_covar=tensor([0.0360, 0.0352, 0.0358, 0.0383, 0.0458, 0.0386, 0.0334, 0.0339], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 16:46:14,344 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.42 vs. limit=2.0 2023-04-02 16:46:46,021 INFO [train.py:903] (1/4) Epoch 20, batch 3250, loss[loss=0.2817, simple_loss=0.3569, pruned_loss=0.1032, over 19662.00 frames. ], tot_loss[loss=0.2123, simple_loss=0.2914, pruned_loss=0.06666, over 3812346.20 frames. ], batch size: 55, lr: 4.13e-03, grad_scale: 8.0 2023-04-02 16:47:37,648 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.383e+02 4.867e+02 6.333e+02 8.818e+02 1.782e+03, threshold=1.267e+03, percent-clipped=7.0 2023-04-02 16:47:37,829 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=133022.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 16:47:49,136 INFO [train.py:903] (1/4) Epoch 20, batch 3300, loss[loss=0.1845, simple_loss=0.2543, pruned_loss=0.05737, over 19730.00 frames. ], tot_loss[loss=0.2126, simple_loss=0.2918, pruned_loss=0.0667, over 3803739.58 frames. ], batch size: 46, lr: 4.12e-03, grad_scale: 8.0 2023-04-02 16:47:57,174 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 2023-04-02 16:48:03,423 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5273, 1.5858, 2.0736, 1.7239, 3.1703, 4.0588, 3.9409, 4.4400], device='cuda:1'), covar=tensor([0.1616, 0.3609, 0.3125, 0.2266, 0.0620, 0.0313, 0.0195, 0.0228], device='cuda:1'), in_proj_covar=tensor([0.0270, 0.0319, 0.0349, 0.0263, 0.0240, 0.0184, 0.0215, 0.0252], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-02 16:48:54,782 INFO [train.py:903] (1/4) Epoch 20, batch 3350, loss[loss=0.1865, simple_loss=0.2663, pruned_loss=0.05335, over 19288.00 frames. ], tot_loss[loss=0.2104, simple_loss=0.2901, pruned_loss=0.06533, over 3817268.31 frames. ], batch size: 44, lr: 4.12e-03, grad_scale: 8.0 2023-04-02 16:49:27,434 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=133108.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 16:49:45,519 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.359e+02 4.956e+02 6.093e+02 7.175e+02 1.819e+03, threshold=1.219e+03, percent-clipped=1.0 2023-04-02 16:49:57,527 INFO [train.py:903] (1/4) Epoch 20, batch 3400, loss[loss=0.2328, simple_loss=0.3102, pruned_loss=0.0777, over 19671.00 frames. ], tot_loss[loss=0.2113, simple_loss=0.2903, pruned_loss=0.0661, over 3820412.96 frames. ], batch size: 58, lr: 4.12e-03, grad_scale: 8.0 2023-04-02 16:50:06,137 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=133137.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 16:51:02,179 INFO [train.py:903] (1/4) Epoch 20, batch 3450, loss[loss=0.2122, simple_loss=0.2959, pruned_loss=0.06425, over 19590.00 frames. ], tot_loss[loss=0.2112, simple_loss=0.29, pruned_loss=0.06619, over 3815283.49 frames. ], batch size: 52, lr: 4.12e-03, grad_scale: 8.0 2023-04-02 16:51:08,031 WARNING [train.py:1073] (1/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9_rvb from training. Duration: 25.2444375 2023-04-02 16:51:41,631 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4117, 2.1095, 1.6851, 1.2535, 2.0350, 1.2807, 1.2493, 1.9612], device='cuda:1'), covar=tensor([0.1043, 0.0751, 0.1021, 0.1045, 0.0527, 0.1278, 0.0790, 0.0434], device='cuda:1'), in_proj_covar=tensor([0.0300, 0.0313, 0.0333, 0.0259, 0.0245, 0.0335, 0.0291, 0.0274], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 16:51:52,697 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.981e+02 4.746e+02 5.634e+02 7.504e+02 1.582e+03, threshold=1.127e+03, percent-clipped=2.0 2023-04-02 16:51:54,124 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=133223.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 16:52:04,224 INFO [train.py:903] (1/4) Epoch 20, batch 3500, loss[loss=0.2149, simple_loss=0.2994, pruned_loss=0.06524, over 19541.00 frames. ], tot_loss[loss=0.2115, simple_loss=0.2903, pruned_loss=0.06636, over 3815065.24 frames. ], batch size: 54, lr: 4.12e-03, grad_scale: 8.0 2023-04-02 16:53:08,045 INFO [train.py:903] (1/4) Epoch 20, batch 3550, loss[loss=0.1814, simple_loss=0.2556, pruned_loss=0.05358, over 19782.00 frames. ], tot_loss[loss=0.211, simple_loss=0.2899, pruned_loss=0.06609, over 3815980.36 frames. ], batch size: 47, lr: 4.12e-03, grad_scale: 8.0 2023-04-02 16:53:15,828 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.09 vs. limit=2.0 2023-04-02 16:53:52,739 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 2023-04-02 16:53:58,716 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.076e+02 5.017e+02 5.933e+02 7.980e+02 2.795e+03, threshold=1.187e+03, percent-clipped=11.0 2023-04-02 16:54:10,372 INFO [train.py:903] (1/4) Epoch 20, batch 3600, loss[loss=0.1868, simple_loss=0.2733, pruned_loss=0.05011, over 19682.00 frames. ], tot_loss[loss=0.2103, simple_loss=0.2892, pruned_loss=0.0657, over 3818948.17 frames. ], batch size: 53, lr: 4.12e-03, grad_scale: 8.0 2023-04-02 16:54:48,395 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9943, 2.0688, 2.3680, 2.7022, 2.0331, 2.5967, 2.4216, 2.1146], device='cuda:1'), covar=tensor([0.4252, 0.3841, 0.1700, 0.2349, 0.4102, 0.2030, 0.4733, 0.3244], device='cuda:1'), in_proj_covar=tensor([0.0883, 0.0945, 0.0705, 0.0928, 0.0864, 0.0799, 0.0834, 0.0771], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 16:55:15,910 INFO [train.py:903] (1/4) Epoch 20, batch 3650, loss[loss=0.22, simple_loss=0.3036, pruned_loss=0.06825, over 19749.00 frames. ], tot_loss[loss=0.2106, simple_loss=0.2896, pruned_loss=0.06584, over 3818113.37 frames. ], batch size: 51, lr: 4.12e-03, grad_scale: 8.0 2023-04-02 16:55:28,125 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=133392.0, num_to_drop=1, layers_to_drop={1} 2023-04-02 16:55:29,479 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=133393.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 16:55:30,478 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=133394.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 16:55:33,868 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.1612, 1.9601, 1.8892, 2.8048, 1.8299, 2.4260, 2.2433, 2.1726], device='cuda:1'), covar=tensor([0.0858, 0.0938, 0.1040, 0.0865, 0.0981, 0.0745, 0.0998, 0.0687], device='cuda:1'), in_proj_covar=tensor([0.0212, 0.0222, 0.0226, 0.0243, 0.0228, 0.0211, 0.0187, 0.0205], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-04-02 16:56:02,469 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=133418.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 16:56:06,666 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.934e+02 5.113e+02 6.456e+02 7.889e+02 1.610e+03, threshold=1.291e+03, percent-clipped=2.0 2023-04-02 16:56:18,442 INFO [train.py:903] (1/4) Epoch 20, batch 3700, loss[loss=0.2152, simple_loss=0.2951, pruned_loss=0.06767, over 19660.00 frames. ], tot_loss[loss=0.2111, simple_loss=0.2901, pruned_loss=0.06602, over 3830617.50 frames. ], batch size: 53, lr: 4.12e-03, grad_scale: 8.0 2023-04-02 16:57:02,959 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 2023-04-02 16:57:19,846 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=133479.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 16:57:23,014 INFO [train.py:903] (1/4) Epoch 20, batch 3750, loss[loss=0.1796, simple_loss=0.2674, pruned_loss=0.04594, over 19854.00 frames. ], tot_loss[loss=0.2122, simple_loss=0.2911, pruned_loss=0.06661, over 3814688.43 frames. ], batch size: 52, lr: 4.12e-03, grad_scale: 8.0 2023-04-02 16:57:34,899 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=133492.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 16:57:39,349 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6397, 1.7453, 2.0038, 2.0318, 1.5596, 1.9456, 2.0394, 1.8866], device='cuda:1'), covar=tensor([0.3926, 0.3366, 0.1816, 0.2152, 0.3543, 0.1976, 0.4604, 0.3216], device='cuda:1'), in_proj_covar=tensor([0.0881, 0.0942, 0.0703, 0.0923, 0.0861, 0.0794, 0.0832, 0.0768], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 16:57:50,929 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=133504.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 16:58:13,501 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.259e+02 5.077e+02 5.935e+02 8.206e+02 1.595e+03, threshold=1.187e+03, percent-clipped=3.0 2023-04-02 16:58:24,812 INFO [train.py:903] (1/4) Epoch 20, batch 3800, loss[loss=0.3309, simple_loss=0.3765, pruned_loss=0.1427, over 17583.00 frames. ], tot_loss[loss=0.2136, simple_loss=0.2922, pruned_loss=0.06749, over 3817519.80 frames. ], batch size: 101, lr: 4.12e-03, grad_scale: 8.0 2023-04-02 16:58:58,190 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9_rvb from training. Duration: 29.1166875 2023-04-02 16:59:30,204 INFO [train.py:903] (1/4) Epoch 20, batch 3850, loss[loss=0.2004, simple_loss=0.2796, pruned_loss=0.06063, over 19674.00 frames. ], tot_loss[loss=0.2135, simple_loss=0.292, pruned_loss=0.06746, over 3816062.05 frames. ], batch size: 60, lr: 4.12e-03, grad_scale: 8.0 2023-04-02 16:59:31,747 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.7293, 1.6437, 1.5964, 2.1465, 1.8342, 1.9616, 2.0407, 1.7686], device='cuda:1'), covar=tensor([0.0821, 0.0859, 0.0969, 0.0732, 0.0816, 0.0732, 0.0817, 0.0684], device='cuda:1'), in_proj_covar=tensor([0.0210, 0.0221, 0.0224, 0.0242, 0.0227, 0.0210, 0.0187, 0.0204], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-02 16:59:48,046 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.59 vs. limit=5.0 2023-04-02 17:00:20,968 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.481e+02 5.079e+02 6.219e+02 7.261e+02 1.808e+03, threshold=1.244e+03, percent-clipped=5.0 2023-04-02 17:00:32,687 INFO [train.py:903] (1/4) Epoch 20, batch 3900, loss[loss=0.1768, simple_loss=0.2548, pruned_loss=0.04945, over 19758.00 frames. ], tot_loss[loss=0.2126, simple_loss=0.2914, pruned_loss=0.06691, over 3822876.08 frames. ], batch size: 47, lr: 4.12e-03, grad_scale: 8.0 2023-04-02 17:01:21,169 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=133670.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 17:01:37,398 INFO [train.py:903] (1/4) Epoch 20, batch 3950, loss[loss=0.2029, simple_loss=0.2886, pruned_loss=0.05863, over 17422.00 frames. ], tot_loss[loss=0.2132, simple_loss=0.2923, pruned_loss=0.06705, over 3822887.86 frames. ], batch size: 101, lr: 4.11e-03, grad_scale: 8.0 2023-04-02 17:01:42,243 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 2023-04-02 17:01:50,354 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0 2023-04-02 17:02:26,843 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.611e+02 5.167e+02 6.244e+02 7.613e+02 1.189e+03, threshold=1.249e+03, percent-clipped=0.0 2023-04-02 17:02:38,777 INFO [train.py:903] (1/4) Epoch 20, batch 4000, loss[loss=0.1942, simple_loss=0.2812, pruned_loss=0.05359, over 19683.00 frames. ], tot_loss[loss=0.2121, simple_loss=0.2917, pruned_loss=0.06625, over 3835675.58 frames. ], batch size: 60, lr: 4.11e-03, grad_scale: 8.0 2023-04-02 17:02:43,760 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=133736.0, num_to_drop=1, layers_to_drop={0} 2023-04-02 17:02:46,991 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=133738.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 17:02:53,752 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 2023-04-02 17:03:26,438 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1_rvb from training. Duration: 0.7545625 2023-04-02 17:03:41,981 INFO [train.py:903] (1/4) Epoch 20, batch 4050, loss[loss=0.2297, simple_loss=0.2915, pruned_loss=0.08398, over 19480.00 frames. ], tot_loss[loss=0.2113, simple_loss=0.291, pruned_loss=0.06577, over 3833905.65 frames. ], batch size: 49, lr: 4.11e-03, grad_scale: 8.0 2023-04-02 17:04:30,941 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.559e+02 4.971e+02 6.413e+02 8.150e+02 1.897e+03, threshold=1.283e+03, percent-clipped=7.0 2023-04-02 17:04:42,290 INFO [train.py:903] (1/4) Epoch 20, batch 4100, loss[loss=0.1961, simple_loss=0.2841, pruned_loss=0.05404, over 19670.00 frames. ], tot_loss[loss=0.2115, simple_loss=0.2913, pruned_loss=0.06581, over 3836699.46 frames. ], batch size: 58, lr: 4.11e-03, grad_scale: 8.0 2023-04-02 17:04:46,021 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.7689, 4.3751, 2.9813, 3.8248, 1.2829, 4.3007, 4.1624, 4.3382], device='cuda:1'), covar=tensor([0.0611, 0.0883, 0.1751, 0.0748, 0.3728, 0.0607, 0.0844, 0.1062], device='cuda:1'), in_proj_covar=tensor([0.0489, 0.0398, 0.0480, 0.0340, 0.0398, 0.0422, 0.0414, 0.0447], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 17:04:47,152 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=133836.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 17:05:03,576 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.37 vs. limit=5.0 2023-04-02 17:05:04,943 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.14 vs. limit=2.0 2023-04-02 17:05:06,757 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=133851.0, num_to_drop=1, layers_to_drop={1} 2023-04-02 17:05:09,045 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=133853.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 17:05:18,885 WARNING [train.py:1073] (1/4) Exclude cut with ID 497-129325-0061-9566_sp1.1_rvb from training. Duration: 0.97725 2023-04-02 17:05:35,467 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9259, 2.0044, 2.2890, 2.5994, 1.8902, 2.3637, 2.3107, 2.0409], device='cuda:1'), covar=tensor([0.4231, 0.3879, 0.1840, 0.2435, 0.4150, 0.2219, 0.4555, 0.3321], device='cuda:1'), in_proj_covar=tensor([0.0887, 0.0948, 0.0707, 0.0931, 0.0868, 0.0801, 0.0836, 0.0773], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 17:05:45,970 INFO [train.py:903] (1/4) Epoch 20, batch 4150, loss[loss=0.1973, simple_loss=0.2771, pruned_loss=0.05876, over 19782.00 frames. ], tot_loss[loss=0.2125, simple_loss=0.2922, pruned_loss=0.06638, over 3833038.44 frames. ], batch size: 54, lr: 4.11e-03, grad_scale: 8.0 2023-04-02 17:06:00,763 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.79 vs. limit=2.0 2023-04-02 17:06:19,872 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3478, 1.3896, 1.5610, 1.5014, 2.1482, 1.9380, 2.2612, 1.0400], device='cuda:1'), covar=tensor([0.1950, 0.3513, 0.2229, 0.1537, 0.1242, 0.1761, 0.1144, 0.3620], device='cuda:1'), in_proj_covar=tensor([0.0525, 0.0633, 0.0698, 0.0477, 0.0613, 0.0524, 0.0658, 0.0541], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 17:06:35,678 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.109e+02 4.757e+02 5.877e+02 6.653e+02 1.329e+03, threshold=1.175e+03, percent-clipped=1.0 2023-04-02 17:06:47,888 INFO [train.py:903] (1/4) Epoch 20, batch 4200, loss[loss=0.174, simple_loss=0.2539, pruned_loss=0.04708, over 19772.00 frames. ], tot_loss[loss=0.2109, simple_loss=0.2907, pruned_loss=0.06556, over 3832244.90 frames. ], batch size: 47, lr: 4.11e-03, grad_scale: 8.0 2023-04-02 17:06:51,422 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 2023-04-02 17:07:12,196 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=133951.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 17:07:30,719 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.3116, 3.0566, 2.2797, 2.7426, 0.9722, 3.0297, 2.9048, 2.9807], device='cuda:1'), covar=tensor([0.1128, 0.1319, 0.2035, 0.1064, 0.3550, 0.0951, 0.1092, 0.1431], device='cuda:1'), in_proj_covar=tensor([0.0492, 0.0398, 0.0483, 0.0342, 0.0399, 0.0423, 0.0415, 0.0449], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 17:07:31,262 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 2023-04-02 17:07:50,931 INFO [train.py:903] (1/4) Epoch 20, batch 4250, loss[loss=0.1971, simple_loss=0.2862, pruned_loss=0.05403, over 19384.00 frames. ], tot_loss[loss=0.2098, simple_loss=0.2898, pruned_loss=0.06493, over 3837560.38 frames. ], batch size: 70, lr: 4.11e-03, grad_scale: 8.0 2023-04-02 17:08:08,172 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9_rvb from training. Duration: 29.816625 2023-04-02 17:08:20,529 WARNING [train.py:1073] (1/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 2023-04-02 17:08:32,197 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=134014.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 17:08:33,406 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=134015.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 17:08:36,363 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.50 vs. limit=2.0 2023-04-02 17:08:41,180 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.039e+02 5.008e+02 5.790e+02 6.980e+02 1.679e+03, threshold=1.158e+03, percent-clipped=4.0 2023-04-02 17:08:54,654 INFO [train.py:903] (1/4) Epoch 20, batch 4300, loss[loss=0.2178, simple_loss=0.2981, pruned_loss=0.06874, over 19785.00 frames. ], tot_loss[loss=0.2096, simple_loss=0.2897, pruned_loss=0.06475, over 3840330.46 frames. ], batch size: 56, lr: 4.11e-03, grad_scale: 8.0 2023-04-02 17:09:27,693 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 2023-04-02 17:09:50,630 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 2023-04-02 17:09:56,295 INFO [train.py:903] (1/4) Epoch 20, batch 4350, loss[loss=0.2195, simple_loss=0.2944, pruned_loss=0.07233, over 16026.00 frames. ], tot_loss[loss=0.2096, simple_loss=0.2897, pruned_loss=0.06478, over 3819254.30 frames. ], batch size: 35, lr: 4.11e-03, grad_scale: 4.0 2023-04-02 17:10:30,274 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=134107.0, num_to_drop=1, layers_to_drop={0} 2023-04-02 17:10:32,506 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=134109.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 17:10:48,203 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.336e+02 5.216e+02 6.315e+02 8.361e+02 2.012e+03, threshold=1.263e+03, percent-clipped=10.0 2023-04-02 17:10:57,690 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=134129.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 17:11:00,857 INFO [train.py:903] (1/4) Epoch 20, batch 4400, loss[loss=0.2582, simple_loss=0.3269, pruned_loss=0.0948, over 19455.00 frames. ], tot_loss[loss=0.2096, simple_loss=0.2896, pruned_loss=0.06481, over 3829224.41 frames. ], batch size: 64, lr: 4.11e-03, grad_scale: 8.0 2023-04-02 17:11:01,228 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=134132.0, num_to_drop=1, layers_to_drop={1} 2023-04-02 17:11:03,602 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=134134.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 17:11:18,416 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.76 vs. limit=2.0 2023-04-02 17:11:29,115 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_rvb from training. Duration: 25.285 2023-04-02 17:11:37,430 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 2023-04-02 17:12:05,046 INFO [train.py:903] (1/4) Epoch 20, batch 4450, loss[loss=0.1952, simple_loss=0.2794, pruned_loss=0.05552, over 19521.00 frames. ], tot_loss[loss=0.2101, simple_loss=0.2901, pruned_loss=0.06499, over 3820062.01 frames. ], batch size: 54, lr: 4.11e-03, grad_scale: 8.0 2023-04-02 17:12:36,938 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=134207.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 17:12:56,171 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.329e+02 4.942e+02 5.989e+02 7.537e+02 1.405e+03, threshold=1.198e+03, percent-clipped=2.0 2023-04-02 17:13:08,112 INFO [train.py:903] (1/4) Epoch 20, batch 4500, loss[loss=0.1927, simple_loss=0.2827, pruned_loss=0.05138, over 19750.00 frames. ], tot_loss[loss=0.2103, simple_loss=0.2905, pruned_loss=0.06504, over 3825988.80 frames. ], batch size: 54, lr: 4.11e-03, grad_scale: 8.0 2023-04-02 17:13:08,535 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=134232.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 17:14:10,220 INFO [train.py:903] (1/4) Epoch 20, batch 4550, loss[loss=0.2217, simple_loss=0.3007, pruned_loss=0.0714, over 17309.00 frames. ], tot_loss[loss=0.2111, simple_loss=0.2911, pruned_loss=0.06558, over 3815411.97 frames. ], batch size: 101, lr: 4.11e-03, grad_scale: 4.0 2023-04-02 17:14:19,195 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9_rvb from training. Duration: 28.72225 2023-04-02 17:14:42,992 WARNING [train.py:1073] (1/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 2023-04-02 17:15:02,206 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=134322.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 17:15:04,302 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.805e+02 4.924e+02 5.886e+02 8.898e+02 2.816e+03, threshold=1.177e+03, percent-clipped=9.0 2023-04-02 17:15:09,040 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=134327.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 17:15:10,307 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5580, 1.4381, 1.3999, 1.9521, 1.5191, 1.7229, 1.8953, 1.6130], device='cuda:1'), covar=tensor([0.0934, 0.0982, 0.1074, 0.0736, 0.0868, 0.0824, 0.0813, 0.0730], device='cuda:1'), in_proj_covar=tensor([0.0212, 0.0222, 0.0226, 0.0243, 0.0227, 0.0211, 0.0187, 0.0205], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-02 17:15:14,419 INFO [train.py:903] (1/4) Epoch 20, batch 4600, loss[loss=0.183, simple_loss=0.2572, pruned_loss=0.05436, over 19798.00 frames. ], tot_loss[loss=0.2112, simple_loss=0.2912, pruned_loss=0.06564, over 3823743.02 frames. ], batch size: 49, lr: 4.10e-03, grad_scale: 4.0 2023-04-02 17:15:47,462 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=134359.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 17:16:15,900 INFO [train.py:903] (1/4) Epoch 20, batch 4650, loss[loss=0.213, simple_loss=0.2928, pruned_loss=0.06655, over 19537.00 frames. ], tot_loss[loss=0.2117, simple_loss=0.2919, pruned_loss=0.06577, over 3821715.40 frames. ], batch size: 54, lr: 4.10e-03, grad_scale: 4.0 2023-04-02 17:16:20,717 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=134385.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 17:16:32,930 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_rvb from training. Duration: 0.92 2023-04-02 17:16:44,600 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_rvb from training. Duration: 0.83 2023-04-02 17:16:52,759 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=134410.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 17:17:09,009 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.247e+02 4.575e+02 5.939e+02 8.134e+02 1.295e+03, threshold=1.188e+03, percent-clipped=3.0 2023-04-02 17:17:14,338 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.37 vs. limit=5.0 2023-04-02 17:17:19,254 INFO [train.py:903] (1/4) Epoch 20, batch 4700, loss[loss=0.1941, simple_loss=0.2667, pruned_loss=0.06074, over 19352.00 frames. ], tot_loss[loss=0.2109, simple_loss=0.2909, pruned_loss=0.06541, over 3837309.03 frames. ], batch size: 47, lr: 4.10e-03, grad_scale: 4.0 2023-04-02 17:17:23,257 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4432, 2.2555, 1.7345, 1.5113, 2.0945, 1.3937, 1.2530, 1.9001], device='cuda:1'), covar=tensor([0.1039, 0.0807, 0.1067, 0.0823, 0.0517, 0.1285, 0.0802, 0.0499], device='cuda:1'), in_proj_covar=tensor([0.0294, 0.0310, 0.0331, 0.0256, 0.0244, 0.0334, 0.0286, 0.0273], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 17:17:41,477 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9_rvb from training. Duration: 0.92225 2023-04-02 17:18:11,931 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=134474.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 17:18:20,719 INFO [train.py:903] (1/4) Epoch 20, batch 4750, loss[loss=0.2635, simple_loss=0.3219, pruned_loss=0.1026, over 19683.00 frames. ], tot_loss[loss=0.2127, simple_loss=0.292, pruned_loss=0.06669, over 3830993.18 frames. ], batch size: 53, lr: 4.10e-03, grad_scale: 4.0 2023-04-02 17:18:58,067 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 2023-04-02 17:19:09,029 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 2023-04-02 17:19:14,952 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.087e+02 5.001e+02 5.955e+02 7.090e+02 1.974e+03, threshold=1.191e+03, percent-clipped=7.0 2023-04-02 17:19:15,472 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4083, 1.4618, 1.6577, 1.6576, 2.4151, 2.1711, 2.5368, 0.9494], device='cuda:1'), covar=tensor([0.2421, 0.4310, 0.2632, 0.1916, 0.1570, 0.2168, 0.1459, 0.4686], device='cuda:1'), in_proj_covar=tensor([0.0525, 0.0635, 0.0697, 0.0477, 0.0615, 0.0523, 0.0658, 0.0543], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 17:19:25,461 INFO [train.py:903] (1/4) Epoch 20, batch 4800, loss[loss=0.184, simple_loss=0.2692, pruned_loss=0.04941, over 19476.00 frames. ], tot_loss[loss=0.2113, simple_loss=0.2908, pruned_loss=0.0659, over 3820586.73 frames. ], batch size: 49, lr: 4.10e-03, grad_scale: 8.0 2023-04-02 17:20:15,873 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4222, 1.3794, 1.3835, 1.8074, 1.3828, 1.6609, 1.5638, 1.4258], device='cuda:1'), covar=tensor([0.0881, 0.0975, 0.1043, 0.0679, 0.0830, 0.0797, 0.0892, 0.0797], device='cuda:1'), in_proj_covar=tensor([0.0210, 0.0220, 0.0222, 0.0241, 0.0224, 0.0208, 0.0185, 0.0202], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-02 17:20:26,815 INFO [train.py:903] (1/4) Epoch 20, batch 4850, loss[loss=0.2089, simple_loss=0.2883, pruned_loss=0.06475, over 19762.00 frames. ], tot_loss[loss=0.2114, simple_loss=0.2909, pruned_loss=0.06597, over 3820930.75 frames. ], batch size: 54, lr: 4.10e-03, grad_scale: 4.0 2023-04-02 17:20:49,982 WARNING [train.py:1073] (1/4) Exclude cut with ID 774-127930-0014-48411_sp1.1_rvb from training. Duration: 0.95 2023-04-02 17:21:13,058 WARNING [train.py:1073] (1/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1_rvb from training. Duration: 0.9409375 2023-04-02 17:21:18,741 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 2023-04-02 17:21:18,771 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 2023-04-02 17:21:21,956 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.496e+02 4.787e+02 5.710e+02 7.760e+02 1.554e+03, threshold=1.142e+03, percent-clipped=3.0 2023-04-02 17:21:31,371 INFO [train.py:903] (1/4) Epoch 20, batch 4900, loss[loss=0.1957, simple_loss=0.2618, pruned_loss=0.06475, over 16430.00 frames. ], tot_loss[loss=0.2109, simple_loss=0.2903, pruned_loss=0.06577, over 3828355.11 frames. ], batch size: 36, lr: 4.10e-03, grad_scale: 4.0 2023-04-02 17:21:31,379 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1_rvb from training. Duration: 25.3818125 2023-04-02 17:21:51,449 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 2023-04-02 17:21:53,084 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.2582, 2.2854, 2.5548, 3.0169, 2.2544, 2.9521, 2.5629, 2.2941], device='cuda:1'), covar=tensor([0.4085, 0.3752, 0.1667, 0.2568, 0.4361, 0.2095, 0.4659, 0.3147], device='cuda:1'), in_proj_covar=tensor([0.0883, 0.0947, 0.0707, 0.0926, 0.0866, 0.0801, 0.0834, 0.0773], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 17:22:14,374 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=134666.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 17:22:21,056 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=134671.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 17:22:33,362 INFO [train.py:903] (1/4) Epoch 20, batch 4950, loss[loss=0.2162, simple_loss=0.2991, pruned_loss=0.06667, over 19335.00 frames. ], tot_loss[loss=0.2104, simple_loss=0.2899, pruned_loss=0.06548, over 3820040.42 frames. ], batch size: 66, lr: 4.10e-03, grad_scale: 4.0 2023-04-02 17:22:49,210 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 2023-04-02 17:23:15,473 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 2023-04-02 17:23:28,331 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.953e+02 4.876e+02 5.783e+02 7.285e+02 1.244e+03, threshold=1.157e+03, percent-clipped=2.0 2023-04-02 17:23:34,558 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=134730.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 17:23:37,527 INFO [train.py:903] (1/4) Epoch 20, batch 5000, loss[loss=0.2419, simple_loss=0.3188, pruned_loss=0.08245, over 19494.00 frames. ], tot_loss[loss=0.2104, simple_loss=0.29, pruned_loss=0.0654, over 3827243.07 frames. ], batch size: 64, lr: 4.10e-03, grad_scale: 4.0 2023-04-02 17:23:45,512 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_rvb from training. Duration: 27.14 2023-04-02 17:23:56,643 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 2023-04-02 17:24:06,096 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=134755.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 17:24:37,396 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.4169, 1.1544, 1.4524, 1.4949, 2.9841, 1.0090, 2.1788, 3.3234], device='cuda:1'), covar=tensor([0.0498, 0.3033, 0.2946, 0.1755, 0.0731, 0.2575, 0.1273, 0.0313], device='cuda:1'), in_proj_covar=tensor([0.0400, 0.0361, 0.0381, 0.0342, 0.0371, 0.0346, 0.0371, 0.0396], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 17:24:37,440 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=134781.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 17:24:38,200 INFO [train.py:903] (1/4) Epoch 20, batch 5050, loss[loss=0.2038, simple_loss=0.2838, pruned_loss=0.06188, over 19681.00 frames. ], tot_loss[loss=0.2098, simple_loss=0.2893, pruned_loss=0.06517, over 3838969.21 frames. ], batch size: 53, lr: 4.10e-03, grad_scale: 4.0 2023-04-02 17:24:44,151 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=134786.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 17:24:56,610 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3430, 1.3937, 1.5200, 1.5286, 1.8668, 1.8862, 1.7823, 0.5627], device='cuda:1'), covar=tensor([0.2653, 0.4599, 0.2881, 0.2132, 0.1636, 0.2445, 0.1441, 0.5162], device='cuda:1'), in_proj_covar=tensor([0.0523, 0.0631, 0.0693, 0.0476, 0.0612, 0.0521, 0.0654, 0.0540], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 17:25:13,080 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9_rvb from training. Duration: 26.62775 2023-04-02 17:25:31,564 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.532e+02 4.915e+02 6.341e+02 8.226e+02 2.739e+03, threshold=1.268e+03, percent-clipped=9.0 2023-04-02 17:25:41,323 INFO [train.py:903] (1/4) Epoch 20, batch 5100, loss[loss=0.2153, simple_loss=0.2886, pruned_loss=0.07103, over 19787.00 frames. ], tot_loss[loss=0.2101, simple_loss=0.2893, pruned_loss=0.06543, over 3820680.32 frames. ], batch size: 47, lr: 4.10e-03, grad_scale: 4.0 2023-04-02 17:25:42,924 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6296, 1.7450, 1.6566, 2.6565, 1.7634, 2.3675, 1.8371, 1.4002], device='cuda:1'), covar=tensor([0.4987, 0.4615, 0.2902, 0.2693, 0.4697, 0.2460, 0.6218, 0.5395], device='cuda:1'), in_proj_covar=tensor([0.0882, 0.0946, 0.0707, 0.0926, 0.0865, 0.0799, 0.0834, 0.0773], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 17:25:50,464 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 2023-04-02 17:25:53,843 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_rvb from training. Duration: 25.035 2023-04-02 17:25:58,168 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_rvb from training. Duration: 27.92 2023-04-02 17:26:26,074 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.7100, 3.4494, 2.5973, 3.0871, 1.3687, 3.3497, 3.2183, 3.3007], device='cuda:1'), covar=tensor([0.0898, 0.1042, 0.1881, 0.0876, 0.2943, 0.0810, 0.0947, 0.1301], device='cuda:1'), in_proj_covar=tensor([0.0490, 0.0399, 0.0486, 0.0341, 0.0398, 0.0422, 0.0415, 0.0450], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 17:26:41,262 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=134880.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 17:26:43,293 INFO [train.py:903] (1/4) Epoch 20, batch 5150, loss[loss=0.1971, simple_loss=0.2854, pruned_loss=0.05446, over 19659.00 frames. ], tot_loss[loss=0.2087, simple_loss=0.2881, pruned_loss=0.06465, over 3824444.97 frames. ], batch size: 60, lr: 4.10e-03, grad_scale: 4.0 2023-04-02 17:26:55,956 WARNING [train.py:1073] (1/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 2023-04-02 17:27:07,089 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.13 vs. limit=2.0 2023-04-02 17:27:32,559 WARNING [train.py:1073] (1/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1_rvb from training. Duration: 0.9681875 2023-04-02 17:27:37,090 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.618e+02 4.964e+02 6.321e+02 8.056e+02 1.479e+03, threshold=1.264e+03, percent-clipped=3.0 2023-04-02 17:27:46,172 INFO [train.py:903] (1/4) Epoch 20, batch 5200, loss[loss=0.2083, simple_loss=0.2914, pruned_loss=0.0626, over 19595.00 frames. ], tot_loss[loss=0.2091, simple_loss=0.2888, pruned_loss=0.06474, over 3820002.39 frames. ], batch size: 52, lr: 4.10e-03, grad_scale: 8.0 2023-04-02 17:28:00,851 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 2023-04-02 17:28:33,749 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 2023-04-02 17:28:46,969 WARNING [train.py:1073] (1/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 2023-04-02 17:28:49,186 INFO [train.py:903] (1/4) Epoch 20, batch 5250, loss[loss=0.2083, simple_loss=0.2965, pruned_loss=0.06002, over 19682.00 frames. ], tot_loss[loss=0.2094, simple_loss=0.289, pruned_loss=0.06487, over 3825282.25 frames. ], batch size: 58, lr: 4.09e-03, grad_scale: 8.0 2023-04-02 17:29:42,806 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.326e+02 4.845e+02 5.899e+02 8.450e+02 1.811e+03, threshold=1.180e+03, percent-clipped=4.0 2023-04-02 17:29:51,988 INFO [train.py:903] (1/4) Epoch 20, batch 5300, loss[loss=0.2082, simple_loss=0.2982, pruned_loss=0.05912, over 19612.00 frames. ], tot_loss[loss=0.2081, simple_loss=0.2876, pruned_loss=0.06432, over 3813022.75 frames. ], batch size: 57, lr: 4.09e-03, grad_scale: 8.0 2023-04-02 17:29:59,113 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=135037.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 17:30:04,856 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=135042.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 17:30:11,581 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 2023-04-02 17:30:24,963 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.2644, 2.2749, 2.5443, 3.1128, 2.3135, 2.9682, 2.6209, 2.3444], device='cuda:1'), covar=tensor([0.4102, 0.3855, 0.1724, 0.2328, 0.4275, 0.1987, 0.4453, 0.3056], device='cuda:1'), in_proj_covar=tensor([0.0885, 0.0947, 0.0708, 0.0928, 0.0867, 0.0800, 0.0837, 0.0773], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 17:30:30,713 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=135062.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 17:30:36,480 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=135067.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 17:30:54,697 INFO [train.py:903] (1/4) Epoch 20, batch 5350, loss[loss=0.2494, simple_loss=0.3148, pruned_loss=0.09203, over 19603.00 frames. ], tot_loss[loss=0.2101, simple_loss=0.2894, pruned_loss=0.06543, over 3804348.65 frames. ], batch size: 52, lr: 4.09e-03, grad_scale: 8.0 2023-04-02 17:31:29,771 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9_rvb from training. Duration: 26.438875 2023-04-02 17:31:45,417 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0961, 1.9232, 1.7990, 2.1191, 1.8552, 1.7870, 1.7524, 1.9685], device='cuda:1'), covar=tensor([0.0991, 0.1440, 0.1330, 0.1016, 0.1359, 0.0523, 0.1240, 0.0672], device='cuda:1'), in_proj_covar=tensor([0.0262, 0.0350, 0.0304, 0.0247, 0.0296, 0.0247, 0.0302, 0.0250], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 17:31:48,522 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.364e+02 5.248e+02 6.375e+02 8.534e+02 1.946e+03, threshold=1.275e+03, percent-clipped=9.0 2023-04-02 17:31:57,764 INFO [train.py:903] (1/4) Epoch 20, batch 5400, loss[loss=0.2257, simple_loss=0.3051, pruned_loss=0.07313, over 19492.00 frames. ], tot_loss[loss=0.2104, simple_loss=0.2896, pruned_loss=0.06555, over 3824253.72 frames. ], batch size: 64, lr: 4.09e-03, grad_scale: 8.0 2023-04-02 17:33:00,878 INFO [train.py:903] (1/4) Epoch 20, batch 5450, loss[loss=0.2028, simple_loss=0.2871, pruned_loss=0.05925, over 19679.00 frames. ], tot_loss[loss=0.2105, simple_loss=0.2898, pruned_loss=0.06558, over 3816657.55 frames. ], batch size: 55, lr: 4.09e-03, grad_scale: 8.0 2023-04-02 17:33:09,027 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=135189.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 17:33:36,848 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.0977, 1.2461, 1.6228, 0.9445, 2.3804, 2.9977, 2.7304, 3.2028], device='cuda:1'), covar=tensor([0.1652, 0.3720, 0.3314, 0.2639, 0.0582, 0.0239, 0.0255, 0.0308], device='cuda:1'), in_proj_covar=tensor([0.0265, 0.0316, 0.0344, 0.0262, 0.0237, 0.0183, 0.0213, 0.0249], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-02 17:33:52,671 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=135224.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 17:33:54,892 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.649e+02 5.255e+02 6.010e+02 7.558e+02 1.824e+03, threshold=1.202e+03, percent-clipped=3.0 2023-04-02 17:34:03,097 INFO [train.py:903] (1/4) Epoch 20, batch 5500, loss[loss=0.2268, simple_loss=0.3169, pruned_loss=0.06836, over 19080.00 frames. ], tot_loss[loss=0.2113, simple_loss=0.2907, pruned_loss=0.06594, over 3821288.69 frames. ], batch size: 69, lr: 4.09e-03, grad_scale: 4.0 2023-04-02 17:34:29,198 WARNING [train.py:1073] (1/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9_rvb from training. Duration: 27.47775 2023-04-02 17:35:05,520 INFO [train.py:903] (1/4) Epoch 20, batch 5550, loss[loss=0.2107, simple_loss=0.2779, pruned_loss=0.07177, over 19805.00 frames. ], tot_loss[loss=0.2108, simple_loss=0.2903, pruned_loss=0.06559, over 3818845.31 frames. ], batch size: 48, lr: 4.09e-03, grad_scale: 4.0 2023-04-02 17:35:13,926 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 2023-04-02 17:35:47,229 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.2923, 3.0712, 2.1751, 2.7782, 0.7601, 2.9974, 2.8747, 2.9343], device='cuda:1'), covar=tensor([0.1164, 0.1253, 0.2048, 0.1060, 0.3728, 0.1012, 0.1142, 0.1505], device='cuda:1'), in_proj_covar=tensor([0.0492, 0.0401, 0.0488, 0.0344, 0.0401, 0.0424, 0.0419, 0.0454], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 17:36:00,656 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.142e+02 4.884e+02 6.418e+02 8.039e+02 2.322e+03, threshold=1.284e+03, percent-clipped=8.0 2023-04-02 17:36:04,003 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 2023-04-02 17:36:07,368 INFO [train.py:903] (1/4) Epoch 20, batch 5600, loss[loss=0.2163, simple_loss=0.2958, pruned_loss=0.06841, over 19774.00 frames. ], tot_loss[loss=0.2105, simple_loss=0.2897, pruned_loss=0.06568, over 3826640.75 frames. ], batch size: 56, lr: 4.09e-03, grad_scale: 8.0 2023-04-02 17:36:18,132 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=135339.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 17:36:29,879 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3305, 1.3534, 1.5755, 1.5034, 2.2480, 2.0110, 2.3377, 0.9126], device='cuda:1'), covar=tensor([0.2392, 0.4274, 0.2651, 0.1957, 0.1523, 0.2170, 0.1340, 0.4612], device='cuda:1'), in_proj_covar=tensor([0.0526, 0.0633, 0.0697, 0.0477, 0.0616, 0.0523, 0.0657, 0.0543], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 17:37:11,264 INFO [train.py:903] (1/4) Epoch 20, batch 5650, loss[loss=0.2023, simple_loss=0.2803, pruned_loss=0.06221, over 19586.00 frames. ], tot_loss[loss=0.2102, simple_loss=0.2894, pruned_loss=0.06551, over 3829976.13 frames. ], batch size: 52, lr: 4.09e-03, grad_scale: 8.0 2023-04-02 17:38:01,372 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9_rvb from training. Duration: 28.0944375 2023-04-02 17:38:05,717 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.932e+02 4.892e+02 5.590e+02 6.975e+02 1.698e+03, threshold=1.118e+03, percent-clipped=3.0 2023-04-02 17:38:12,543 INFO [train.py:903] (1/4) Epoch 20, batch 5700, loss[loss=0.1737, simple_loss=0.2502, pruned_loss=0.0486, over 19751.00 frames. ], tot_loss[loss=0.2104, simple_loss=0.2896, pruned_loss=0.0656, over 3828695.19 frames. ], batch size: 46, lr: 4.09e-03, grad_scale: 8.0 2023-04-02 17:38:47,545 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3658, 1.0747, 1.2708, 1.9841, 1.3990, 1.4096, 1.5420, 1.3131], device='cuda:1'), covar=tensor([0.1255, 0.1790, 0.1383, 0.0866, 0.1165, 0.1399, 0.1327, 0.1117], device='cuda:1'), in_proj_covar=tensor([0.0213, 0.0222, 0.0225, 0.0243, 0.0227, 0.0211, 0.0186, 0.0204], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-02 17:38:48,713 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1709, 1.7896, 1.4652, 1.0505, 1.5927, 1.1091, 1.1908, 1.7729], device='cuda:1'), covar=tensor([0.0799, 0.0739, 0.0982, 0.0952, 0.0507, 0.1297, 0.0625, 0.0376], device='cuda:1'), in_proj_covar=tensor([0.0298, 0.0311, 0.0331, 0.0257, 0.0244, 0.0333, 0.0289, 0.0273], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 17:39:14,115 INFO [train.py:903] (1/4) Epoch 20, batch 5750, loss[loss=0.2319, simple_loss=0.3062, pruned_loss=0.07877, over 19052.00 frames. ], tot_loss[loss=0.2113, simple_loss=0.2904, pruned_loss=0.0661, over 3808187.37 frames. ], batch size: 69, lr: 4.09e-03, grad_scale: 8.0 2023-04-02 17:39:17,231 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9_rvb from training. Duration: 33.038875 2023-04-02 17:39:25,433 WARNING [train.py:1073] (1/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 2023-04-02 17:39:31,253 WARNING [train.py:1073] (1/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 2023-04-02 17:40:10,279 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.984e+02 5.062e+02 5.931e+02 7.729e+02 1.708e+03, threshold=1.186e+03, percent-clipped=6.0 2023-04-02 17:40:18,212 INFO [train.py:903] (1/4) Epoch 20, batch 5800, loss[loss=0.2044, simple_loss=0.2934, pruned_loss=0.05774, over 19704.00 frames. ], tot_loss[loss=0.212, simple_loss=0.2909, pruned_loss=0.06652, over 3782114.61 frames. ], batch size: 59, lr: 4.09e-03, grad_scale: 8.0 2023-04-02 17:40:19,525 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=135533.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 17:40:21,359 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 2023-04-02 17:40:56,840 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 2023-04-02 17:41:20,580 INFO [train.py:903] (1/4) Epoch 20, batch 5850, loss[loss=0.1638, simple_loss=0.2564, pruned_loss=0.03563, over 19841.00 frames. ], tot_loss[loss=0.2113, simple_loss=0.2901, pruned_loss=0.06623, over 3796255.30 frames. ], batch size: 52, lr: 4.09e-03, grad_scale: 8.0 2023-04-02 17:41:35,405 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=135595.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 17:42:03,219 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.54 vs. limit=2.0 2023-04-02 17:42:09,020 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=135620.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 17:42:15,863 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.765e+02 5.164e+02 6.346e+02 8.557e+02 2.066e+03, threshold=1.269e+03, percent-clipped=9.0 2023-04-02 17:42:23,028 INFO [train.py:903] (1/4) Epoch 20, batch 5900, loss[loss=0.1763, simple_loss=0.2627, pruned_loss=0.04492, over 19778.00 frames. ], tot_loss[loss=0.2108, simple_loss=0.2893, pruned_loss=0.06613, over 3790040.67 frames. ], batch size: 51, lr: 4.08e-03, grad_scale: 8.0 2023-04-02 17:42:25,435 WARNING [train.py:1073] (1/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 2023-04-02 17:42:43,057 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=135648.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 17:42:43,129 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=135648.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 17:42:46,334 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9_rvb from training. Duration: 27.511125 2023-04-02 17:43:19,368 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2944, 1.1902, 1.2453, 1.3398, 1.0849, 1.3643, 1.3617, 1.2516], device='cuda:1'), covar=tensor([0.0935, 0.0999, 0.1084, 0.0677, 0.0851, 0.0816, 0.0826, 0.0785], device='cuda:1'), in_proj_covar=tensor([0.0212, 0.0223, 0.0226, 0.0244, 0.0227, 0.0212, 0.0187, 0.0204], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-02 17:43:24,694 INFO [train.py:903] (1/4) Epoch 20, batch 5950, loss[loss=0.2066, simple_loss=0.289, pruned_loss=0.06214, over 19546.00 frames. ], tot_loss[loss=0.2125, simple_loss=0.2909, pruned_loss=0.06701, over 3785305.16 frames. ], batch size: 56, lr: 4.08e-03, grad_scale: 8.0 2023-04-02 17:44:00,301 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.3184, 2.3563, 2.5721, 3.1345, 2.4506, 2.9867, 2.6453, 2.3192], device='cuda:1'), covar=tensor([0.3988, 0.3818, 0.1754, 0.2299, 0.3967, 0.1995, 0.4172, 0.3153], device='cuda:1'), in_proj_covar=tensor([0.0881, 0.0945, 0.0703, 0.0925, 0.0863, 0.0794, 0.0831, 0.0770], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 17:44:19,053 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.081e+02 5.126e+02 6.708e+02 1.004e+03 2.382e+03, threshold=1.342e+03, percent-clipped=11.0 2023-04-02 17:44:19,670 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.14 vs. limit=2.0 2023-04-02 17:44:27,257 INFO [train.py:903] (1/4) Epoch 20, batch 6000, loss[loss=0.2106, simple_loss=0.2962, pruned_loss=0.06249, over 19748.00 frames. ], tot_loss[loss=0.212, simple_loss=0.2905, pruned_loss=0.06672, over 3805369.83 frames. ], batch size: 63, lr: 4.08e-03, grad_scale: 8.0 2023-04-02 17:44:27,257 INFO [train.py:928] (1/4) Computing validation loss 2023-04-02 17:44:39,927 INFO [train.py:937] (1/4) Epoch 20, validation: loss=0.1697, simple_loss=0.2697, pruned_loss=0.0349, over 944034.00 frames. 2023-04-02 17:44:39,928 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18821MB 2023-04-02 17:45:41,586 INFO [train.py:903] (1/4) Epoch 20, batch 6050, loss[loss=0.2246, simple_loss=0.2987, pruned_loss=0.07529, over 19674.00 frames. ], tot_loss[loss=0.2117, simple_loss=0.2901, pruned_loss=0.06666, over 3812377.53 frames. ], batch size: 53, lr: 4.08e-03, grad_scale: 8.0 2023-04-02 17:46:36,914 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.594e+02 4.873e+02 5.806e+02 7.519e+02 1.887e+03, threshold=1.161e+03, percent-clipped=2.0 2023-04-02 17:46:43,930 INFO [train.py:903] (1/4) Epoch 20, batch 6100, loss[loss=0.2064, simple_loss=0.291, pruned_loss=0.06089, over 19517.00 frames. ], tot_loss[loss=0.2124, simple_loss=0.2904, pruned_loss=0.06721, over 3811477.98 frames. ], batch size: 54, lr: 4.08e-03, grad_scale: 8.0 2023-04-02 17:47:46,926 INFO [train.py:903] (1/4) Epoch 20, batch 6150, loss[loss=0.2194, simple_loss=0.2857, pruned_loss=0.07656, over 19756.00 frames. ], tot_loss[loss=0.2121, simple_loss=0.2906, pruned_loss=0.06679, over 3814945.02 frames. ], batch size: 47, lr: 4.08e-03, grad_scale: 8.0 2023-04-02 17:48:15,340 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=135904.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 17:48:16,095 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9_rvb from training. Duration: 31.02225 2023-04-02 17:48:41,891 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.532e+02 4.701e+02 6.211e+02 7.118e+02 1.417e+03, threshold=1.242e+03, percent-clipped=3.0 2023-04-02 17:48:45,710 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=135929.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 17:48:49,677 INFO [train.py:903] (1/4) Epoch 20, batch 6200, loss[loss=0.2365, simple_loss=0.3173, pruned_loss=0.07791, over 19306.00 frames. ], tot_loss[loss=0.2123, simple_loss=0.2909, pruned_loss=0.06688, over 3817747.98 frames. ], batch size: 66, lr: 4.08e-03, grad_scale: 8.0 2023-04-02 17:49:18,736 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.7721, 1.5606, 1.6009, 2.3281, 1.6604, 1.9973, 2.0758, 1.7849], device='cuda:1'), covar=tensor([0.0885, 0.1007, 0.1053, 0.0743, 0.0916, 0.0835, 0.0925, 0.0739], device='cuda:1'), in_proj_covar=tensor([0.0211, 0.0221, 0.0225, 0.0243, 0.0226, 0.0211, 0.0186, 0.0202], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-02 17:49:52,113 INFO [train.py:903] (1/4) Epoch 20, batch 6250, loss[loss=0.2137, simple_loss=0.2982, pruned_loss=0.06464, over 19649.00 frames. ], tot_loss[loss=0.2124, simple_loss=0.2911, pruned_loss=0.06688, over 3827168.24 frames. ], batch size: 58, lr: 4.08e-03, grad_scale: 8.0 2023-04-02 17:50:04,690 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=135992.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 17:50:15,862 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=136000.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 17:50:24,790 WARNING [train.py:1073] (1/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9_rvb from training. Duration: 25.988875 2023-04-02 17:50:30,669 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=136012.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 17:50:48,357 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.357e+02 5.185e+02 6.758e+02 8.208e+02 2.074e+03, threshold=1.352e+03, percent-clipped=8.0 2023-04-02 17:50:55,356 INFO [train.py:903] (1/4) Epoch 20, batch 6300, loss[loss=0.217, simple_loss=0.3003, pruned_loss=0.06681, over 19696.00 frames. ], tot_loss[loss=0.2123, simple_loss=0.2905, pruned_loss=0.06709, over 3822703.03 frames. ], batch size: 60, lr: 4.08e-03, grad_scale: 8.0 2023-04-02 17:51:26,752 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.7250, 4.2952, 4.4797, 4.4554, 1.6568, 4.1813, 3.6470, 4.1791], device='cuda:1'), covar=tensor([0.1691, 0.0707, 0.0572, 0.0652, 0.6140, 0.0877, 0.0735, 0.1128], device='cuda:1'), in_proj_covar=tensor([0.0770, 0.0724, 0.0927, 0.0810, 0.0820, 0.0685, 0.0557, 0.0856], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-02 17:51:58,465 INFO [train.py:903] (1/4) Epoch 20, batch 6350, loss[loss=0.2484, simple_loss=0.3227, pruned_loss=0.087, over 19332.00 frames. ], tot_loss[loss=0.2109, simple_loss=0.2896, pruned_loss=0.06608, over 3820669.24 frames. ], batch size: 66, lr: 4.08e-03, grad_scale: 8.0 2023-04-02 17:52:30,869 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=136107.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 17:52:53,613 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.408e+02 5.045e+02 5.932e+02 7.156e+02 1.987e+03, threshold=1.186e+03, percent-clipped=3.0 2023-04-02 17:53:01,247 INFO [train.py:903] (1/4) Epoch 20, batch 6400, loss[loss=0.2133, simple_loss=0.3091, pruned_loss=0.05874, over 19722.00 frames. ], tot_loss[loss=0.2097, simple_loss=0.2885, pruned_loss=0.06545, over 3823096.88 frames. ], batch size: 59, lr: 4.08e-03, grad_scale: 8.0 2023-04-02 17:53:36,374 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=136160.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 17:54:02,569 INFO [train.py:903] (1/4) Epoch 20, batch 6450, loss[loss=0.2466, simple_loss=0.3204, pruned_loss=0.08637, over 19712.00 frames. ], tot_loss[loss=0.2108, simple_loss=0.2896, pruned_loss=0.06604, over 3829794.65 frames. ], batch size: 63, lr: 4.08e-03, grad_scale: 8.0 2023-04-02 17:54:51,296 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9_rvb from training. Duration: 28.638875 2023-04-02 17:54:59,003 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.358e+02 5.012e+02 6.099e+02 8.089e+02 3.011e+03, threshold=1.220e+03, percent-clipped=7.0 2023-04-02 17:55:07,145 INFO [train.py:903] (1/4) Epoch 20, batch 6500, loss[loss=0.2408, simple_loss=0.3109, pruned_loss=0.08532, over 13677.00 frames. ], tot_loss[loss=0.2107, simple_loss=0.2896, pruned_loss=0.06593, over 3827725.57 frames. ], batch size: 136, lr: 4.08e-03, grad_scale: 8.0 2023-04-02 17:55:12,970 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1_rvb from training. Duration: 0.836375 2023-04-02 17:55:44,588 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.50 vs. limit=2.0 2023-04-02 17:56:09,977 INFO [train.py:903] (1/4) Epoch 20, batch 6550, loss[loss=0.1638, simple_loss=0.2418, pruned_loss=0.04292, over 19726.00 frames. ], tot_loss[loss=0.2103, simple_loss=0.2893, pruned_loss=0.06566, over 3818615.98 frames. ], batch size: 45, lr: 4.07e-03, grad_scale: 8.0 2023-04-02 17:57:06,361 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.205e+02 5.319e+02 6.594e+02 8.030e+02 1.579e+03, threshold=1.319e+03, percent-clipped=2.0 2023-04-02 17:57:14,440 INFO [train.py:903] (1/4) Epoch 20, batch 6600, loss[loss=0.2072, simple_loss=0.2834, pruned_loss=0.06543, over 18652.00 frames. ], tot_loss[loss=0.2098, simple_loss=0.2888, pruned_loss=0.06536, over 3819721.36 frames. ], batch size: 41, lr: 4.07e-03, grad_scale: 8.0 2023-04-02 17:57:28,427 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=136344.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 17:57:41,122 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=136353.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 17:57:45,655 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=136356.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 17:57:53,972 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=136363.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 17:58:16,829 INFO [train.py:903] (1/4) Epoch 20, batch 6650, loss[loss=0.2192, simple_loss=0.3059, pruned_loss=0.06625, over 19680.00 frames. ], tot_loss[loss=0.2111, simple_loss=0.29, pruned_loss=0.06612, over 3809922.96 frames. ], batch size: 60, lr: 4.07e-03, grad_scale: 8.0 2023-04-02 17:58:26,160 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=136388.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 17:58:56,619 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0 2023-04-02 17:59:14,402 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.147e+02 4.763e+02 5.879e+02 7.861e+02 2.647e+03, threshold=1.176e+03, percent-clipped=6.0 2023-04-02 17:59:22,072 INFO [train.py:903] (1/4) Epoch 20, batch 6700, loss[loss=0.2094, simple_loss=0.2937, pruned_loss=0.0625, over 19523.00 frames. ], tot_loss[loss=0.2126, simple_loss=0.291, pruned_loss=0.06705, over 3809940.51 frames. ], batch size: 56, lr: 4.07e-03, grad_scale: 8.0 2023-04-02 17:59:55,278 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=136459.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 18:00:08,568 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=136471.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 18:00:20,886 INFO [train.py:903] (1/4) Epoch 20, batch 6750, loss[loss=0.2413, simple_loss=0.3146, pruned_loss=0.08399, over 19111.00 frames. ], tot_loss[loss=0.213, simple_loss=0.2915, pruned_loss=0.06728, over 3810853.29 frames. ], batch size: 69, lr: 4.07e-03, grad_scale: 8.0 2023-04-02 18:00:45,732 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=136504.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 18:01:11,794 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.251e+02 5.547e+02 6.932e+02 9.039e+02 1.788e+03, threshold=1.386e+03, percent-clipped=9.0 2023-04-02 18:01:19,069 INFO [train.py:903] (1/4) Epoch 20, batch 6800, loss[loss=0.2003, simple_loss=0.2772, pruned_loss=0.06173, over 19471.00 frames. ], tot_loss[loss=0.2119, simple_loss=0.2907, pruned_loss=0.06652, over 3810486.72 frames. ], batch size: 49, lr: 4.07e-03, grad_scale: 8.0 2023-04-02 18:02:04,317 WARNING [train.py:1073] (1/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 2023-04-02 18:02:04,769 WARNING [train.py:1073] (1/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 2023-04-02 18:02:07,962 INFO [train.py:903] (1/4) Epoch 21, batch 0, loss[loss=0.2235, simple_loss=0.2955, pruned_loss=0.07576, over 19471.00 frames. ], tot_loss[loss=0.2235, simple_loss=0.2955, pruned_loss=0.07576, over 19471.00 frames. ], batch size: 49, lr: 3.97e-03, grad_scale: 8.0 2023-04-02 18:02:07,962 INFO [train.py:928] (1/4) Computing validation loss 2023-04-02 18:02:18,720 INFO [train.py:937] (1/4) Epoch 21, validation: loss=0.1691, simple_loss=0.2696, pruned_loss=0.03427, over 944034.00 frames. 2023-04-02 18:02:18,721 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18821MB 2023-04-02 18:02:30,981 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 2023-04-02 18:03:20,704 INFO [train.py:903] (1/4) Epoch 21, batch 50, loss[loss=0.2047, simple_loss=0.2908, pruned_loss=0.0593, over 19542.00 frames. ], tot_loss[loss=0.2088, simple_loss=0.2888, pruned_loss=0.06441, over 864816.88 frames. ], batch size: 54, lr: 3.97e-03, grad_scale: 4.0 2023-04-02 18:03:33,296 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=136619.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 18:03:43,287 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.449e+02 4.947e+02 6.173e+02 6.953e+02 1.295e+03, threshold=1.235e+03, percent-clipped=0.0 2023-04-02 18:03:54,611 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1_rvb from training. Duration: 27.0318125 2023-04-02 18:04:02,462 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=4.08 vs. limit=5.0 2023-04-02 18:04:22,858 INFO [train.py:903] (1/4) Epoch 21, batch 100, loss[loss=0.2083, simple_loss=0.2827, pruned_loss=0.06696, over 19728.00 frames. ], tot_loss[loss=0.2094, simple_loss=0.2893, pruned_loss=0.06469, over 1519734.26 frames. ], batch size: 51, lr: 3.97e-03, grad_scale: 4.0 2023-04-02 18:04:25,237 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=136661.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 18:04:34,143 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 2023-04-02 18:05:09,044 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=136697.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 18:05:25,011 INFO [train.py:903] (1/4) Epoch 21, batch 150, loss[loss=0.1549, simple_loss=0.2368, pruned_loss=0.03651, over 19783.00 frames. ], tot_loss[loss=0.2087, simple_loss=0.2886, pruned_loss=0.06441, over 2044514.02 frames. ], batch size: 47, lr: 3.97e-03, grad_scale: 4.0 2023-04-02 18:05:31,954 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=136715.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 18:05:45,589 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.079e+02 5.031e+02 5.954e+02 7.665e+02 1.668e+03, threshold=1.191e+03, percent-clipped=3.0 2023-04-02 18:05:46,012 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=136727.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 18:06:02,663 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=136740.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 18:06:03,826 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6487, 1.5366, 1.5543, 2.0338, 1.5584, 1.8840, 1.8725, 1.6771], device='cuda:1'), covar=tensor([0.0819, 0.0896, 0.0936, 0.0664, 0.0815, 0.0690, 0.0831, 0.0693], device='cuda:1'), in_proj_covar=tensor([0.0211, 0.0222, 0.0225, 0.0242, 0.0226, 0.0212, 0.0187, 0.0203], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-02 18:06:18,640 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=136752.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 18:06:25,319 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 2023-04-02 18:06:27,571 INFO [train.py:903] (1/4) Epoch 21, batch 200, loss[loss=0.221, simple_loss=0.303, pruned_loss=0.06945, over 19523.00 frames. ], tot_loss[loss=0.2093, simple_loss=0.2887, pruned_loss=0.06498, over 2423996.81 frames. ], batch size: 64, lr: 3.97e-03, grad_scale: 4.0 2023-04-02 18:06:29,122 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.3654, 1.8540, 1.8957, 2.8979, 2.1735, 2.6603, 2.5205, 2.0821], device='cuda:1'), covar=tensor([0.0747, 0.0958, 0.0979, 0.0769, 0.0810, 0.0693, 0.0876, 0.0697], device='cuda:1'), in_proj_covar=tensor([0.0212, 0.0223, 0.0226, 0.0243, 0.0226, 0.0213, 0.0188, 0.0204], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-02 18:06:35,833 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.9282, 3.5659, 2.5123, 3.1773, 0.8177, 3.4851, 3.3669, 3.5417], device='cuda:1'), covar=tensor([0.0869, 0.1180, 0.2100, 0.0975, 0.4231, 0.0918, 0.1094, 0.1319], device='cuda:1'), in_proj_covar=tensor([0.0496, 0.0403, 0.0491, 0.0343, 0.0403, 0.0426, 0.0422, 0.0458], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 18:06:43,875 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4172, 1.2527, 1.2779, 1.7337, 1.3930, 1.5911, 1.6255, 1.3364], device='cuda:1'), covar=tensor([0.0923, 0.1066, 0.1106, 0.0719, 0.0783, 0.0777, 0.0860, 0.0830], device='cuda:1'), in_proj_covar=tensor([0.0211, 0.0222, 0.0226, 0.0243, 0.0226, 0.0213, 0.0188, 0.0204], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-02 18:07:12,671 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 2023-04-02 18:07:29,906 INFO [train.py:903] (1/4) Epoch 21, batch 250, loss[loss=0.2181, simple_loss=0.302, pruned_loss=0.06713, over 19762.00 frames. ], tot_loss[loss=0.2103, simple_loss=0.2897, pruned_loss=0.06539, over 2739568.26 frames. ], batch size: 54, lr: 3.97e-03, grad_scale: 4.0 2023-04-02 18:07:32,610 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=136812.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 18:07:47,492 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=136824.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 18:07:52,057 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.077e+02 4.815e+02 6.209e+02 8.068e+02 1.278e+03, threshold=1.242e+03, percent-clipped=1.0 2023-04-02 18:08:31,167 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2980, 2.0138, 1.6887, 1.2856, 1.8197, 1.3082, 1.2764, 1.9110], device='cuda:1'), covar=tensor([0.0947, 0.0760, 0.0970, 0.0870, 0.0540, 0.1211, 0.0681, 0.0424], device='cuda:1'), in_proj_covar=tensor([0.0303, 0.0314, 0.0338, 0.0260, 0.0247, 0.0336, 0.0292, 0.0275], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 18:08:33,027 INFO [train.py:903] (1/4) Epoch 21, batch 300, loss[loss=0.2149, simple_loss=0.2943, pruned_loss=0.0678, over 19577.00 frames. ], tot_loss[loss=0.2095, simple_loss=0.2891, pruned_loss=0.06501, over 2983935.33 frames. ], batch size: 52, lr: 3.97e-03, grad_scale: 4.0 2023-04-02 18:08:53,791 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=136875.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 18:09:23,344 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=136900.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 18:09:36,421 INFO [train.py:903] (1/4) Epoch 21, batch 350, loss[loss=0.2082, simple_loss=0.2898, pruned_loss=0.0633, over 18755.00 frames. ], tot_loss[loss=0.21, simple_loss=0.2888, pruned_loss=0.06563, over 3172305.65 frames. ], batch size: 74, lr: 3.97e-03, grad_scale: 4.0 2023-04-02 18:09:38,653 WARNING [train.py:1073] (1/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1_rvb from training. Duration: 0.9681875 2023-04-02 18:09:56,984 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.395e+02 4.835e+02 5.943e+02 7.281e+02 1.741e+03, threshold=1.189e+03, percent-clipped=3.0 2023-04-02 18:10:39,358 INFO [train.py:903] (1/4) Epoch 21, batch 400, loss[loss=0.2023, simple_loss=0.2894, pruned_loss=0.05759, over 19606.00 frames. ], tot_loss[loss=0.2091, simple_loss=0.2878, pruned_loss=0.06514, over 3323219.99 frames. ], batch size: 57, lr: 3.97e-03, grad_scale: 8.0 2023-04-02 18:11:02,227 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.44 vs. limit=2.0 2023-04-02 18:11:35,387 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=137005.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 18:11:41,221 INFO [train.py:903] (1/4) Epoch 21, batch 450, loss[loss=0.2325, simple_loss=0.3032, pruned_loss=0.08088, over 18252.00 frames. ], tot_loss[loss=0.2098, simple_loss=0.2883, pruned_loss=0.06567, over 3440324.86 frames. ], batch size: 83, lr: 3.96e-03, grad_scale: 4.0 2023-04-02 18:11:59,611 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.8139, 4.0766, 4.5784, 4.6177, 1.9042, 4.2294, 3.5335, 4.0130], device='cuda:1'), covar=tensor([0.2072, 0.1335, 0.0815, 0.1047, 0.6834, 0.1820, 0.1302, 0.1756], device='cuda:1'), in_proj_covar=tensor([0.0767, 0.0719, 0.0928, 0.0817, 0.0819, 0.0687, 0.0560, 0.0860], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-02 18:12:00,879 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5541, 1.6875, 2.0096, 1.8221, 3.0717, 2.7642, 3.4105, 1.6200], device='cuda:1'), covar=tensor([0.2504, 0.4367, 0.2851, 0.1938, 0.1681, 0.2025, 0.1634, 0.4118], device='cuda:1'), in_proj_covar=tensor([0.0526, 0.0634, 0.0700, 0.0479, 0.0618, 0.0525, 0.0657, 0.0540], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 18:12:03,759 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.868e+02 4.613e+02 5.973e+02 7.527e+02 1.521e+03, threshold=1.195e+03, percent-clipped=6.0 2023-04-02 18:12:12,953 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 2023-04-02 18:12:14,109 WARNING [train.py:1073] (1/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 2023-04-02 18:12:43,485 INFO [train.py:903] (1/4) Epoch 21, batch 500, loss[loss=0.2188, simple_loss=0.3009, pruned_loss=0.0684, over 19691.00 frames. ], tot_loss[loss=0.2097, simple_loss=0.2883, pruned_loss=0.0655, over 3537121.62 frames. ], batch size: 59, lr: 3.96e-03, grad_scale: 4.0 2023-04-02 18:12:53,189 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=137068.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 18:13:25,205 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=137093.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 18:13:43,791 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.3026, 3.8298, 3.9268, 3.9374, 1.5560, 3.7025, 3.2255, 3.6671], device='cuda:1'), covar=tensor([0.1660, 0.0882, 0.0672, 0.0754, 0.5635, 0.1041, 0.0713, 0.1194], device='cuda:1'), in_proj_covar=tensor([0.0769, 0.0720, 0.0929, 0.0817, 0.0820, 0.0688, 0.0560, 0.0860], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-02 18:13:45,871 INFO [train.py:903] (1/4) Epoch 21, batch 550, loss[loss=0.1939, simple_loss=0.2859, pruned_loss=0.05098, over 19620.00 frames. ], tot_loss[loss=0.2098, simple_loss=0.2886, pruned_loss=0.06555, over 3582173.81 frames. ], batch size: 61, lr: 3.96e-03, grad_scale: 4.0 2023-04-02 18:13:59,933 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=137120.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 18:14:09,847 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.012e+02 5.365e+02 6.823e+02 8.347e+02 2.113e+03, threshold=1.365e+03, percent-clipped=7.0 2023-04-02 18:14:33,025 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.7255, 1.6068, 1.7375, 1.7346, 3.3179, 1.4100, 2.5754, 3.6669], device='cuda:1'), covar=tensor([0.0460, 0.2491, 0.2589, 0.1810, 0.0646, 0.2349, 0.1213, 0.0244], device='cuda:1'), in_proj_covar=tensor([0.0404, 0.0361, 0.0382, 0.0344, 0.0372, 0.0347, 0.0373, 0.0398], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 18:14:49,009 INFO [train.py:903] (1/4) Epoch 21, batch 600, loss[loss=0.2502, simple_loss=0.3184, pruned_loss=0.09105, over 19675.00 frames. ], tot_loss[loss=0.2101, simple_loss=0.2888, pruned_loss=0.0657, over 3619539.66 frames. ], batch size: 58, lr: 3.96e-03, grad_scale: 4.0 2023-04-02 18:15:00,539 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=137168.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 18:15:29,953 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 2023-04-02 18:15:47,618 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3821, 1.1684, 1.2684, 1.7815, 1.4680, 1.4001, 1.3828, 1.3456], device='cuda:1'), covar=tensor([0.0892, 0.1345, 0.1005, 0.0695, 0.1064, 0.1087, 0.1212, 0.0871], device='cuda:1'), in_proj_covar=tensor([0.0210, 0.0220, 0.0224, 0.0241, 0.0224, 0.0210, 0.0186, 0.0202], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-02 18:15:53,981 INFO [train.py:903] (1/4) Epoch 21, batch 650, loss[loss=0.1893, simple_loss=0.27, pruned_loss=0.0543, over 19674.00 frames. ], tot_loss[loss=0.2112, simple_loss=0.2896, pruned_loss=0.06645, over 3658962.13 frames. ], batch size: 53, lr: 3.96e-03, grad_scale: 4.0 2023-04-02 18:16:16,603 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.985e+02 4.775e+02 5.968e+02 8.002e+02 1.696e+03, threshold=1.194e+03, percent-clipped=7.0 2023-04-02 18:16:56,145 INFO [train.py:903] (1/4) Epoch 21, batch 700, loss[loss=0.1951, simple_loss=0.2734, pruned_loss=0.05838, over 19872.00 frames. ], tot_loss[loss=0.2124, simple_loss=0.2906, pruned_loss=0.06708, over 3686176.60 frames. ], batch size: 52, lr: 3.96e-03, grad_scale: 4.0 2023-04-02 18:17:03,350 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.6397, 2.1298, 2.2162, 2.6088, 2.2340, 2.0533, 2.1638, 2.6425], device='cuda:1'), covar=tensor([0.0842, 0.1617, 0.1274, 0.1005, 0.1329, 0.0541, 0.1231, 0.0581], device='cuda:1'), in_proj_covar=tensor([0.0267, 0.0354, 0.0309, 0.0249, 0.0298, 0.0251, 0.0307, 0.0253], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 18:17:26,131 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=137283.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 18:18:00,859 INFO [train.py:903] (1/4) Epoch 21, batch 750, loss[loss=0.1949, simple_loss=0.2764, pruned_loss=0.05669, over 19767.00 frames. ], tot_loss[loss=0.2119, simple_loss=0.2904, pruned_loss=0.06671, over 3694012.76 frames. ], batch size: 54, lr: 3.96e-03, grad_scale: 4.0 2023-04-02 18:18:18,217 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9688, 2.0395, 2.2765, 2.6089, 1.9778, 2.5313, 2.2886, 2.0551], device='cuda:1'), covar=tensor([0.4240, 0.3918, 0.1874, 0.2418, 0.4141, 0.2080, 0.4722, 0.3302], device='cuda:1'), in_proj_covar=tensor([0.0888, 0.0950, 0.0709, 0.0927, 0.0870, 0.0800, 0.0835, 0.0775], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 18:18:22,898 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.306e+02 4.851e+02 6.236e+02 7.648e+02 2.101e+03, threshold=1.247e+03, percent-clipped=5.0 2023-04-02 18:18:42,013 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.14 vs. limit=2.0 2023-04-02 18:19:01,635 INFO [train.py:903] (1/4) Epoch 21, batch 800, loss[loss=0.2487, simple_loss=0.3223, pruned_loss=0.08755, over 19058.00 frames. ], tot_loss[loss=0.2128, simple_loss=0.2916, pruned_loss=0.06705, over 3727354.34 frames. ], batch size: 69, lr: 3.96e-03, grad_scale: 8.0 2023-04-02 18:19:07,853 WARNING [train.py:1073] (1/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 2023-04-02 18:19:22,493 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=137376.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 18:19:53,548 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=137401.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 18:20:05,543 INFO [train.py:903] (1/4) Epoch 21, batch 850, loss[loss=0.1938, simple_loss=0.2822, pruned_loss=0.05267, over 19660.00 frames. ], tot_loss[loss=0.2126, simple_loss=0.2914, pruned_loss=0.06684, over 3736164.20 frames. ], batch size: 55, lr: 3.96e-03, grad_scale: 8.0 2023-04-02 18:20:27,121 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.313e+02 5.583e+02 6.647e+02 8.818e+02 2.027e+03, threshold=1.329e+03, percent-clipped=5.0 2023-04-02 18:20:48,405 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9_rvb from training. Duration: 25.061125 2023-04-02 18:21:06,665 INFO [train.py:903] (1/4) Epoch 21, batch 900, loss[loss=0.1701, simple_loss=0.2562, pruned_loss=0.04198, over 19740.00 frames. ], tot_loss[loss=0.2124, simple_loss=0.2912, pruned_loss=0.06681, over 3754399.68 frames. ], batch size: 51, lr: 3.96e-03, grad_scale: 8.0 2023-04-02 18:21:59,598 WARNING [train.py:1073] (1/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9_rvb from training. Duration: 26.32775 2023-04-02 18:22:08,627 INFO [train.py:903] (1/4) Epoch 21, batch 950, loss[loss=0.1933, simple_loss=0.2812, pruned_loss=0.05271, over 19767.00 frames. ], tot_loss[loss=0.2124, simple_loss=0.2911, pruned_loss=0.06685, over 3754177.07 frames. ], batch size: 54, lr: 3.96e-03, grad_scale: 8.0 2023-04-02 18:22:31,547 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.353e+02 5.256e+02 6.264e+02 7.895e+02 1.664e+03, threshold=1.253e+03, percent-clipped=2.0 2023-04-02 18:22:45,763 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=137539.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 18:23:10,890 INFO [train.py:903] (1/4) Epoch 21, batch 1000, loss[loss=0.2258, simple_loss=0.3069, pruned_loss=0.07234, over 18372.00 frames. ], tot_loss[loss=0.2115, simple_loss=0.2904, pruned_loss=0.06629, over 3772216.01 frames. ], batch size: 84, lr: 3.96e-03, grad_scale: 8.0 2023-04-02 18:23:15,635 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=137564.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 18:23:55,433 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.13 vs. limit=2.0 2023-04-02 18:23:56,778 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 2023-04-02 18:24:14,001 INFO [train.py:903] (1/4) Epoch 21, batch 1050, loss[loss=0.2017, simple_loss=0.2705, pruned_loss=0.06646, over 19303.00 frames. ], tot_loss[loss=0.2107, simple_loss=0.2898, pruned_loss=0.06579, over 3784265.61 frames. ], batch size: 44, lr: 3.96e-03, grad_scale: 8.0 2023-04-02 18:24:35,337 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.928e+02 5.318e+02 6.486e+02 8.521e+02 3.216e+03, threshold=1.297e+03, percent-clipped=7.0 2023-04-02 18:24:36,541 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 2023-04-02 18:25:17,799 INFO [train.py:903] (1/4) Epoch 21, batch 1100, loss[loss=0.184, simple_loss=0.2697, pruned_loss=0.04918, over 19662.00 frames. ], tot_loss[loss=0.2105, simple_loss=0.2896, pruned_loss=0.06573, over 3789442.55 frames. ], batch size: 53, lr: 3.96e-03, grad_scale: 8.0 2023-04-02 18:26:19,700 INFO [train.py:903] (1/4) Epoch 21, batch 1150, loss[loss=0.1862, simple_loss=0.2757, pruned_loss=0.0483, over 19524.00 frames. ], tot_loss[loss=0.2093, simple_loss=0.2885, pruned_loss=0.06506, over 3805784.55 frames. ], batch size: 54, lr: 3.95e-03, grad_scale: 4.0 2023-04-02 18:26:43,797 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.088e+02 5.127e+02 6.263e+02 7.580e+02 1.245e+03, threshold=1.253e+03, percent-clipped=0.0 2023-04-02 18:27:22,180 INFO [train.py:903] (1/4) Epoch 21, batch 1200, loss[loss=0.2122, simple_loss=0.2937, pruned_loss=0.06537, over 19365.00 frames. ], tot_loss[loss=0.2102, simple_loss=0.2895, pruned_loss=0.0655, over 3821332.94 frames. ], batch size: 66, lr: 3.95e-03, grad_scale: 8.0 2023-04-02 18:27:46,785 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 2023-04-02 18:28:10,912 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 2023-04-02 18:28:25,534 INFO [train.py:903] (1/4) Epoch 21, batch 1250, loss[loss=0.2006, simple_loss=0.2853, pruned_loss=0.05799, over 19529.00 frames. ], tot_loss[loss=0.2098, simple_loss=0.2894, pruned_loss=0.06513, over 3825353.42 frames. ], batch size: 64, lr: 3.95e-03, grad_scale: 8.0 2023-04-02 18:28:48,827 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.413e+02 5.084e+02 5.991e+02 7.123e+02 1.423e+03, threshold=1.198e+03, percent-clipped=5.0 2023-04-02 18:29:27,990 INFO [train.py:903] (1/4) Epoch 21, batch 1300, loss[loss=0.206, simple_loss=0.2855, pruned_loss=0.06322, over 19672.00 frames. ], tot_loss[loss=0.2094, simple_loss=0.2885, pruned_loss=0.06513, over 3831464.40 frames. ], batch size: 53, lr: 3.95e-03, grad_scale: 8.0 2023-04-02 18:29:41,258 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0479, 1.9091, 1.9475, 2.7577, 1.9132, 2.4691, 2.3815, 2.1431], device='cuda:1'), covar=tensor([0.0811, 0.0870, 0.0937, 0.0774, 0.0869, 0.0680, 0.0891, 0.0669], device='cuda:1'), in_proj_covar=tensor([0.0213, 0.0225, 0.0227, 0.0245, 0.0227, 0.0214, 0.0190, 0.0206], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-04-02 18:30:19,740 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.69 vs. limit=5.0 2023-04-02 18:30:30,607 INFO [train.py:903] (1/4) Epoch 21, batch 1350, loss[loss=0.2332, simple_loss=0.3116, pruned_loss=0.07739, over 18843.00 frames. ], tot_loss[loss=0.2098, simple_loss=0.289, pruned_loss=0.06527, over 3821302.88 frames. ], batch size: 74, lr: 3.95e-03, grad_scale: 8.0 2023-04-02 18:30:30,920 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=137910.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 18:30:34,463 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=137913.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 18:30:54,787 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.119e+02 5.098e+02 6.340e+02 8.452e+02 2.491e+03, threshold=1.268e+03, percent-clipped=6.0 2023-04-02 18:31:33,372 INFO [train.py:903] (1/4) Epoch 21, batch 1400, loss[loss=0.1577, simple_loss=0.2365, pruned_loss=0.03941, over 19752.00 frames. ], tot_loss[loss=0.2095, simple_loss=0.2888, pruned_loss=0.06503, over 3826909.07 frames. ], batch size: 47, lr: 3.95e-03, grad_scale: 8.0 2023-04-02 18:32:08,998 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.2867, 2.0668, 2.2205, 2.8628, 1.9561, 2.5733, 2.3871, 2.3199], device='cuda:1'), covar=tensor([0.0745, 0.0845, 0.0834, 0.0788, 0.0875, 0.0695, 0.0912, 0.0632], device='cuda:1'), in_proj_covar=tensor([0.0213, 0.0224, 0.0227, 0.0244, 0.0227, 0.0213, 0.0189, 0.0206], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-04-02 18:32:28,137 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_rvb from training. Duration: 25.85 2023-04-02 18:32:37,221 INFO [train.py:903] (1/4) Epoch 21, batch 1450, loss[loss=0.2064, simple_loss=0.2881, pruned_loss=0.0623, over 18085.00 frames. ], tot_loss[loss=0.2114, simple_loss=0.2904, pruned_loss=0.06616, over 3807813.16 frames. ], batch size: 83, lr: 3.95e-03, grad_scale: 8.0 2023-04-02 18:32:41,599 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.31 vs. limit=2.0 2023-04-02 18:32:45,900 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6814, 1.6911, 1.7138, 2.1823, 1.6760, 1.9968, 2.0359, 1.8267], device='cuda:1'), covar=tensor([0.0842, 0.0837, 0.0928, 0.0775, 0.0850, 0.0741, 0.0845, 0.0674], device='cuda:1'), in_proj_covar=tensor([0.0212, 0.0224, 0.0228, 0.0244, 0.0227, 0.0213, 0.0189, 0.0206], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-04-02 18:33:01,290 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.354e+02 4.676e+02 5.543e+02 6.978e+02 2.034e+03, threshold=1.109e+03, percent-clipped=2.0 2023-04-02 18:33:39,042 INFO [train.py:903] (1/4) Epoch 21, batch 1500, loss[loss=0.1976, simple_loss=0.2853, pruned_loss=0.05491, over 19539.00 frames. ], tot_loss[loss=0.209, simple_loss=0.2882, pruned_loss=0.06496, over 3825385.02 frames. ], batch size: 54, lr: 3.95e-03, grad_scale: 8.0 2023-04-02 18:34:17,952 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2171, 1.6108, 2.0988, 1.6835, 3.2227, 4.9004, 4.6635, 5.2131], device='cuda:1'), covar=tensor([0.1688, 0.3461, 0.2944, 0.2152, 0.0541, 0.0160, 0.0149, 0.0153], device='cuda:1'), in_proj_covar=tensor([0.0269, 0.0320, 0.0349, 0.0264, 0.0241, 0.0185, 0.0215, 0.0252], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-02 18:34:42,047 INFO [train.py:903] (1/4) Epoch 21, batch 1550, loss[loss=0.2309, simple_loss=0.3123, pruned_loss=0.07478, over 19143.00 frames. ], tot_loss[loss=0.2112, simple_loss=0.2903, pruned_loss=0.06602, over 3829406.51 frames. ], batch size: 69, lr: 3.95e-03, grad_scale: 8.0 2023-04-02 18:34:43,696 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.2024, 2.2477, 2.5128, 3.0206, 2.2010, 2.9529, 2.5405, 2.2088], device='cuda:1'), covar=tensor([0.4310, 0.4116, 0.1814, 0.2430, 0.4443, 0.2018, 0.4834, 0.3403], device='cuda:1'), in_proj_covar=tensor([0.0888, 0.0952, 0.0711, 0.0927, 0.0869, 0.0800, 0.0833, 0.0777], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 18:34:56,479 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.9220, 1.5998, 1.9032, 1.4538, 4.4456, 1.0239, 2.3503, 4.8813], device='cuda:1'), covar=tensor([0.0510, 0.2827, 0.2686, 0.2112, 0.0739, 0.2879, 0.1639, 0.0165], device='cuda:1'), in_proj_covar=tensor([0.0408, 0.0365, 0.0385, 0.0346, 0.0374, 0.0349, 0.0377, 0.0400], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 18:35:05,480 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.935e+02 5.280e+02 6.228e+02 7.726e+02 2.313e+03, threshold=1.246e+03, percent-clipped=5.0 2023-04-02 18:35:38,367 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.9941, 5.0161, 5.7572, 5.7913, 2.0838, 5.4829, 4.7069, 5.3530], device='cuda:1'), covar=tensor([0.1599, 0.0910, 0.0554, 0.0547, 0.6005, 0.0665, 0.0559, 0.1290], device='cuda:1'), in_proj_covar=tensor([0.0770, 0.0721, 0.0929, 0.0816, 0.0817, 0.0687, 0.0561, 0.0862], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-02 18:35:44,861 INFO [train.py:903] (1/4) Epoch 21, batch 1600, loss[loss=0.2134, simple_loss=0.2754, pruned_loss=0.07567, over 19783.00 frames. ], tot_loss[loss=0.211, simple_loss=0.2905, pruned_loss=0.06577, over 3835446.64 frames. ], batch size: 48, lr: 3.95e-03, grad_scale: 8.0 2023-04-02 18:36:07,172 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9_rvb from training. Duration: 30.1555625 2023-04-02 18:36:48,266 INFO [train.py:903] (1/4) Epoch 21, batch 1650, loss[loss=0.1753, simple_loss=0.2537, pruned_loss=0.04846, over 19405.00 frames. ], tot_loss[loss=0.2097, simple_loss=0.2893, pruned_loss=0.06501, over 3840979.77 frames. ], batch size: 48, lr: 3.95e-03, grad_scale: 8.0 2023-04-02 18:36:50,157 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 2023-04-02 18:37:12,981 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.826e+02 4.591e+02 5.832e+02 7.172e+02 1.632e+03, threshold=1.166e+03, percent-clipped=1.0 2023-04-02 18:37:43,169 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=138254.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 18:37:46,577 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=138257.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 18:37:49,827 INFO [train.py:903] (1/4) Epoch 21, batch 1700, loss[loss=0.2224, simple_loss=0.3038, pruned_loss=0.07054, over 19357.00 frames. ], tot_loss[loss=0.2103, simple_loss=0.29, pruned_loss=0.06535, over 3839989.84 frames. ], batch size: 66, lr: 3.95e-03, grad_scale: 8.0 2023-04-02 18:38:24,209 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=138288.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 18:38:28,566 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 2023-04-02 18:38:53,714 INFO [train.py:903] (1/4) Epoch 21, batch 1750, loss[loss=0.2173, simple_loss=0.3035, pruned_loss=0.0656, over 19315.00 frames. ], tot_loss[loss=0.2104, simple_loss=0.2901, pruned_loss=0.06535, over 3848826.51 frames. ], batch size: 70, lr: 3.95e-03, grad_scale: 8.0 2023-04-02 18:39:16,068 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.511e+02 4.819e+02 5.989e+02 8.047e+02 2.111e+03, threshold=1.198e+03, percent-clipped=8.0 2023-04-02 18:39:30,916 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=138340.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 18:39:55,287 INFO [train.py:903] (1/4) Epoch 21, batch 1800, loss[loss=0.2104, simple_loss=0.2937, pruned_loss=0.0635, over 17464.00 frames. ], tot_loss[loss=0.2101, simple_loss=0.2901, pruned_loss=0.06509, over 3843025.36 frames. ], batch size: 101, lr: 3.95e-03, grad_scale: 8.0 2023-04-02 18:40:06,007 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=138369.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 18:40:10,880 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=138372.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 18:40:54,356 WARNING [train.py:1073] (1/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9_rvb from training. Duration: 25.0944375 2023-04-02 18:40:57,827 INFO [train.py:903] (1/4) Epoch 21, batch 1850, loss[loss=0.2436, simple_loss=0.3173, pruned_loss=0.08488, over 19378.00 frames. ], tot_loss[loss=0.21, simple_loss=0.2898, pruned_loss=0.06511, over 3841443.12 frames. ], batch size: 70, lr: 3.94e-03, grad_scale: 8.0 2023-04-02 18:41:22,804 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.428e+02 4.784e+02 5.677e+02 7.800e+02 1.333e+03, threshold=1.135e+03, percent-clipped=2.0 2023-04-02 18:41:34,203 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9_rvb from training. Duration: 27.8166875 2023-04-02 18:42:01,842 INFO [train.py:903] (1/4) Epoch 21, batch 1900, loss[loss=0.1831, simple_loss=0.2662, pruned_loss=0.05, over 19622.00 frames. ], tot_loss[loss=0.2088, simple_loss=0.2885, pruned_loss=0.06454, over 3852161.78 frames. ], batch size: 50, lr: 3.94e-03, grad_scale: 8.0 2023-04-02 18:42:18,816 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 2023-04-02 18:42:19,482 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.13 vs. limit=2.0 2023-04-02 18:42:23,569 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9_rvb from training. Duration: 27.02225 2023-04-02 18:42:48,373 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 2023-04-02 18:43:04,462 INFO [train.py:903] (1/4) Epoch 21, batch 1950, loss[loss=0.1882, simple_loss=0.2721, pruned_loss=0.05221, over 19765.00 frames. ], tot_loss[loss=0.2091, simple_loss=0.2887, pruned_loss=0.06475, over 3837729.40 frames. ], batch size: 54, lr: 3.94e-03, grad_scale: 8.0 2023-04-02 18:43:27,762 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.334e+02 4.783e+02 6.007e+02 7.405e+02 3.008e+03, threshold=1.201e+03, percent-clipped=4.0 2023-04-02 18:44:06,919 INFO [train.py:903] (1/4) Epoch 21, batch 2000, loss[loss=0.2091, simple_loss=0.2934, pruned_loss=0.06244, over 19475.00 frames. ], tot_loss[loss=0.2098, simple_loss=0.2891, pruned_loss=0.06523, over 3832825.21 frames. ], batch size: 49, lr: 3.94e-03, grad_scale: 8.0 2023-04-02 18:44:14,585 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 2023-04-02 18:45:03,882 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_rvb from training. Duration: 26.8349375 2023-04-02 18:45:08,469 INFO [train.py:903] (1/4) Epoch 21, batch 2050, loss[loss=0.2249, simple_loss=0.3062, pruned_loss=0.07176, over 19282.00 frames. ], tot_loss[loss=0.2106, simple_loss=0.29, pruned_loss=0.06561, over 3828195.55 frames. ], batch size: 66, lr: 3.94e-03, grad_scale: 8.0 2023-04-02 18:45:22,122 WARNING [train.py:1073] (1/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 2023-04-02 18:45:23,289 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 2023-04-02 18:45:28,235 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=138625.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 18:45:32,780 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=138628.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 18:45:33,537 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.887e+02 4.818e+02 6.151e+02 8.119e+02 1.355e+03, threshold=1.230e+03, percent-clipped=5.0 2023-04-02 18:45:38,230 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=138632.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 18:45:45,234 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 2023-04-02 18:46:00,045 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=138650.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 18:46:03,546 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=138653.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 18:46:12,221 INFO [train.py:903] (1/4) Epoch 21, batch 2100, loss[loss=0.1644, simple_loss=0.2431, pruned_loss=0.04289, over 19772.00 frames. ], tot_loss[loss=0.2086, simple_loss=0.2878, pruned_loss=0.0647, over 3830538.56 frames. ], batch size: 48, lr: 3.94e-03, grad_scale: 8.0 2023-04-02 18:46:13,855 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3329, 2.0334, 1.5902, 1.3542, 1.8604, 1.3006, 1.3234, 1.8335], device='cuda:1'), covar=tensor([0.0900, 0.0801, 0.1070, 0.0890, 0.0553, 0.1309, 0.0632, 0.0439], device='cuda:1'), in_proj_covar=tensor([0.0302, 0.0313, 0.0334, 0.0260, 0.0245, 0.0336, 0.0290, 0.0275], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 18:46:39,336 WARNING [train.py:1073] (1/4) Exclude cut with ID 453-131332-0000-131866_sp0.9_rvb from training. Duration: 25.3333125 2023-04-02 18:46:41,753 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=138684.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 18:47:00,962 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=138699.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 18:47:02,055 WARNING [train.py:1073] (1/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9_rvb from training. Duration: 26.6166875 2023-04-02 18:47:06,972 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=138704.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 18:47:14,541 INFO [train.py:903] (1/4) Epoch 21, batch 2150, loss[loss=0.2422, simple_loss=0.3287, pruned_loss=0.07792, over 19755.00 frames. ], tot_loss[loss=0.2099, simple_loss=0.2888, pruned_loss=0.06554, over 3830631.78 frames. ], batch size: 63, lr: 3.94e-03, grad_scale: 8.0 2023-04-02 18:47:21,627 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=138715.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 18:47:37,621 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.976e+02 5.033e+02 5.895e+02 7.227e+02 1.459e+03, threshold=1.179e+03, percent-clipped=3.0 2023-04-02 18:48:00,964 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=138747.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 18:48:03,253 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.5417, 4.0905, 4.2583, 4.2502, 1.6721, 3.9880, 3.4478, 3.9629], device='cuda:1'), covar=tensor([0.1665, 0.0799, 0.0637, 0.0698, 0.5694, 0.0832, 0.0747, 0.1202], device='cuda:1'), in_proj_covar=tensor([0.0781, 0.0735, 0.0941, 0.0824, 0.0829, 0.0696, 0.0570, 0.0870], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-02 18:48:18,135 INFO [train.py:903] (1/4) Epoch 21, batch 2200, loss[loss=0.2406, simple_loss=0.3278, pruned_loss=0.07671, over 18012.00 frames. ], tot_loss[loss=0.21, simple_loss=0.2888, pruned_loss=0.06556, over 3830745.64 frames. ], batch size: 83, lr: 3.94e-03, grad_scale: 8.0 2023-04-02 18:48:39,543 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.6988, 2.2957, 2.2663, 2.7286, 2.3773, 2.2217, 2.3440, 2.7464], device='cuda:1'), covar=tensor([0.0858, 0.1689, 0.1338, 0.1080, 0.1373, 0.0486, 0.1206, 0.0614], device='cuda:1'), in_proj_covar=tensor([0.0268, 0.0358, 0.0311, 0.0250, 0.0301, 0.0251, 0.0309, 0.0254], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 18:49:08,071 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=138799.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 18:49:20,496 INFO [train.py:903] (1/4) Epoch 21, batch 2250, loss[loss=0.2472, simple_loss=0.317, pruned_loss=0.08869, over 19747.00 frames. ], tot_loss[loss=0.2099, simple_loss=0.2891, pruned_loss=0.06539, over 3831143.38 frames. ], batch size: 51, lr: 3.94e-03, grad_scale: 8.0 2023-04-02 18:49:41,395 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2011, 1.3502, 1.8716, 1.3241, 2.7748, 3.7310, 3.4866, 3.8741], device='cuda:1'), covar=tensor([0.1641, 0.3703, 0.3073, 0.2403, 0.0602, 0.0192, 0.0192, 0.0250], device='cuda:1'), in_proj_covar=tensor([0.0270, 0.0321, 0.0350, 0.0264, 0.0241, 0.0185, 0.0216, 0.0252], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-02 18:49:44,481 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.352e+02 5.230e+02 6.732e+02 8.271e+02 2.316e+03, threshold=1.346e+03, percent-clipped=7.0 2023-04-02 18:50:12,143 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.5665, 2.9911, 3.1525, 3.2008, 1.2943, 2.9569, 2.5966, 2.7073], device='cuda:1'), covar=tensor([0.3223, 0.1932, 0.1550, 0.2089, 0.7388, 0.2410, 0.1681, 0.2879], device='cuda:1'), in_proj_covar=tensor([0.0780, 0.0733, 0.0942, 0.0823, 0.0826, 0.0694, 0.0570, 0.0870], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-02 18:50:23,831 INFO [train.py:903] (1/4) Epoch 21, batch 2300, loss[loss=0.2159, simple_loss=0.2972, pruned_loss=0.06727, over 19586.00 frames. ], tot_loss[loss=0.2102, simple_loss=0.2896, pruned_loss=0.06541, over 3824585.71 frames. ], batch size: 61, lr: 3.94e-03, grad_scale: 8.0 2023-04-02 18:50:38,391 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_rvb from training. Duration: 26.205 2023-04-02 18:51:27,174 INFO [train.py:903] (1/4) Epoch 21, batch 2350, loss[loss=0.1939, simple_loss=0.279, pruned_loss=0.05442, over 19786.00 frames. ], tot_loss[loss=0.2093, simple_loss=0.2887, pruned_loss=0.06495, over 3821007.88 frames. ], batch size: 56, lr: 3.94e-03, grad_scale: 8.0 2023-04-02 18:51:48,934 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=138927.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 18:51:50,724 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.927e+02 4.807e+02 6.646e+02 8.268e+02 1.737e+03, threshold=1.329e+03, percent-clipped=3.0 2023-04-02 18:52:08,336 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_rvb from training. Duration: 25.775 2023-04-02 18:52:24,468 WARNING [train.py:1073] (1/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9_rvb from training. Duration: 25.45 2023-04-02 18:52:31,448 INFO [train.py:903] (1/4) Epoch 21, batch 2400, loss[loss=0.2259, simple_loss=0.3106, pruned_loss=0.07063, over 19768.00 frames. ], tot_loss[loss=0.2099, simple_loss=0.2898, pruned_loss=0.06501, over 3831379.90 frames. ], batch size: 56, lr: 3.94e-03, grad_scale: 8.0 2023-04-02 18:53:26,788 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=139003.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 18:53:34,433 INFO [train.py:903] (1/4) Epoch 21, batch 2450, loss[loss=0.2082, simple_loss=0.2943, pruned_loss=0.06101, over 19456.00 frames. ], tot_loss[loss=0.2089, simple_loss=0.2888, pruned_loss=0.06447, over 3830591.07 frames. ], batch size: 64, lr: 3.94e-03, grad_scale: 8.0 2023-04-02 18:53:58,416 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=139028.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 18:53:59,947 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.106e+02 5.027e+02 6.383e+02 8.090e+02 1.476e+03, threshold=1.277e+03, percent-clipped=1.0 2023-04-02 18:54:17,225 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=139043.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 18:54:22,831 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=139048.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 18:54:31,501 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=139055.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 18:54:38,304 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=139059.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 18:54:39,196 INFO [train.py:903] (1/4) Epoch 21, batch 2500, loss[loss=0.269, simple_loss=0.3379, pruned_loss=0.1001, over 13562.00 frames. ], tot_loss[loss=0.2095, simple_loss=0.2891, pruned_loss=0.06491, over 3819880.12 frames. ], batch size: 136, lr: 3.94e-03, grad_scale: 8.0 2023-04-02 18:55:05,366 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=139080.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 18:55:20,462 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.47 vs. limit=2.0 2023-04-02 18:55:34,451 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.66 vs. limit=5.0 2023-04-02 18:55:42,980 INFO [train.py:903] (1/4) Epoch 21, batch 2550, loss[loss=0.1877, simple_loss=0.2618, pruned_loss=0.05676, over 19778.00 frames. ], tot_loss[loss=0.209, simple_loss=0.2885, pruned_loss=0.06472, over 3826031.29 frames. ], batch size: 48, lr: 3.93e-03, grad_scale: 8.0 2023-04-02 18:55:53,889 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.2514, 2.0821, 1.9574, 1.8113, 1.4894, 1.7976, 0.6564, 1.2638], device='cuda:1'), covar=tensor([0.0679, 0.0695, 0.0514, 0.0927, 0.1404, 0.1029, 0.1442, 0.1080], device='cuda:1'), in_proj_covar=tensor([0.0358, 0.0352, 0.0357, 0.0381, 0.0458, 0.0386, 0.0333, 0.0339], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 18:56:06,420 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.656e+02 4.926e+02 6.170e+02 7.459e+02 2.294e+03, threshold=1.234e+03, percent-clipped=3.0 2023-04-02 18:56:14,911 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1158, 1.3007, 1.7953, 1.3202, 2.7673, 3.7061, 3.4174, 3.9195], device='cuda:1'), covar=tensor([0.1782, 0.3996, 0.3331, 0.2499, 0.0623, 0.0223, 0.0224, 0.0248], device='cuda:1'), in_proj_covar=tensor([0.0271, 0.0322, 0.0351, 0.0264, 0.0242, 0.0186, 0.0217, 0.0253], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-02 18:56:35,499 WARNING [train.py:1073] (1/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 2023-04-02 18:56:43,493 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=139158.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 18:56:45,341 INFO [train.py:903] (1/4) Epoch 21, batch 2600, loss[loss=0.1989, simple_loss=0.282, pruned_loss=0.05792, over 18215.00 frames. ], tot_loss[loss=0.2094, simple_loss=0.2887, pruned_loss=0.06504, over 3839320.56 frames. ], batch size: 84, lr: 3.93e-03, grad_scale: 8.0 2023-04-02 18:56:49,249 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=139163.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 18:57:02,177 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=139174.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 18:57:47,159 INFO [train.py:903] (1/4) Epoch 21, batch 2650, loss[loss=0.2419, simple_loss=0.3167, pruned_loss=0.08353, over 19370.00 frames. ], tot_loss[loss=0.2096, simple_loss=0.2891, pruned_loss=0.06505, over 3852179.38 frames. ], batch size: 66, lr: 3.93e-03, grad_scale: 8.0 2023-04-02 18:58:08,400 WARNING [train.py:1073] (1/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9_rvb from training. Duration: 27.25 2023-04-02 18:58:13,098 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.645e+02 4.962e+02 5.879e+02 7.846e+02 2.263e+03, threshold=1.176e+03, percent-clipped=6.0 2023-04-02 18:58:45,340 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=139256.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 18:58:49,507 INFO [train.py:903] (1/4) Epoch 21, batch 2700, loss[loss=0.2042, simple_loss=0.2821, pruned_loss=0.06312, over 19677.00 frames. ], tot_loss[loss=0.2096, simple_loss=0.2892, pruned_loss=0.06503, over 3839447.52 frames. ], batch size: 53, lr: 3.93e-03, grad_scale: 4.0 2023-04-02 18:59:04,250 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=139271.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 18:59:26,351 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 2023-04-02 18:59:33,184 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=139294.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 18:59:49,918 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=139307.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 18:59:52,995 INFO [train.py:903] (1/4) Epoch 21, batch 2750, loss[loss=0.2171, simple_loss=0.2985, pruned_loss=0.06788, over 19502.00 frames. ], tot_loss[loss=0.2104, simple_loss=0.2897, pruned_loss=0.0655, over 3831190.14 frames. ], batch size: 64, lr: 3.93e-03, grad_scale: 4.0 2023-04-02 19:00:18,050 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.927e+02 4.705e+02 5.796e+02 7.407e+02 1.811e+03, threshold=1.159e+03, percent-clipped=2.0 2023-04-02 19:00:55,534 INFO [train.py:903] (1/4) Epoch 21, batch 2800, loss[loss=0.1821, simple_loss=0.2684, pruned_loss=0.04792, over 19582.00 frames. ], tot_loss[loss=0.21, simple_loss=0.2895, pruned_loss=0.0653, over 3841768.41 frames. ], batch size: 52, lr: 3.93e-03, grad_scale: 8.0 2023-04-02 19:01:08,569 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.12 vs. limit=2.0 2023-04-02 19:01:28,017 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=139386.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:01:45,344 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 2023-04-02 19:01:58,450 INFO [train.py:903] (1/4) Epoch 21, batch 2850, loss[loss=0.2603, simple_loss=0.3322, pruned_loss=0.09416, over 19669.00 frames. ], tot_loss[loss=0.2089, simple_loss=0.2886, pruned_loss=0.06458, over 3837036.04 frames. ], batch size: 55, lr: 3.93e-03, grad_scale: 8.0 2023-04-02 19:02:03,776 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=139414.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:02:09,514 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=139419.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:02:23,967 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.945e+02 5.282e+02 6.184e+02 7.725e+02 1.552e+03, threshold=1.237e+03, percent-clipped=4.0 2023-04-02 19:02:24,385 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=139430.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:02:35,353 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=139439.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:02:42,188 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=139444.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:02:54,861 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=139455.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:02:57,821 WARNING [train.py:1073] (1/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 2023-04-02 19:03:00,110 INFO [train.py:903] (1/4) Epoch 21, batch 2900, loss[loss=0.2272, simple_loss=0.3061, pruned_loss=0.07412, over 19661.00 frames. ], tot_loss[loss=0.2079, simple_loss=0.2877, pruned_loss=0.06404, over 3826976.46 frames. ], batch size: 55, lr: 3.93e-03, grad_scale: 8.0 2023-04-02 19:03:31,310 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.6484, 1.1506, 1.4790, 1.5916, 2.9676, 1.2742, 2.4083, 3.4472], device='cuda:1'), covar=tensor([0.0644, 0.3496, 0.3293, 0.2147, 0.1085, 0.2746, 0.1439, 0.0421], device='cuda:1'), in_proj_covar=tensor([0.0406, 0.0361, 0.0380, 0.0342, 0.0368, 0.0345, 0.0372, 0.0398], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 19:04:04,518 INFO [train.py:903] (1/4) Epoch 21, batch 2950, loss[loss=0.2383, simple_loss=0.3198, pruned_loss=0.07835, over 19665.00 frames. ], tot_loss[loss=0.2089, simple_loss=0.2885, pruned_loss=0.06462, over 3808413.24 frames. ], batch size: 55, lr: 3.93e-03, grad_scale: 8.0 2023-04-02 19:04:28,835 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.677e+02 4.684e+02 5.947e+02 7.442e+02 1.403e+03, threshold=1.189e+03, percent-clipped=5.0 2023-04-02 19:05:01,580 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.0864, 1.2197, 1.5738, 0.9606, 2.4020, 3.0399, 2.7461, 3.1900], device='cuda:1'), covar=tensor([0.1648, 0.3752, 0.3385, 0.2605, 0.0566, 0.0218, 0.0234, 0.0293], device='cuda:1'), in_proj_covar=tensor([0.0270, 0.0320, 0.0350, 0.0264, 0.0242, 0.0186, 0.0216, 0.0253], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-02 19:05:06,921 INFO [train.py:903] (1/4) Epoch 21, batch 3000, loss[loss=0.2183, simple_loss=0.3002, pruned_loss=0.06825, over 19526.00 frames. ], tot_loss[loss=0.2087, simple_loss=0.2882, pruned_loss=0.06459, over 3830605.97 frames. ], batch size: 54, lr: 3.93e-03, grad_scale: 8.0 2023-04-02 19:05:06,921 INFO [train.py:928] (1/4) Computing validation loss 2023-04-02 19:05:13,987 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.8351, 3.5730, 2.7360, 3.2371, 0.8405, 3.5354, 3.2838, 3.6026], device='cuda:1'), covar=tensor([0.0788, 0.0728, 0.1893, 0.0819, 0.4429, 0.0699, 0.0791, 0.0903], device='cuda:1'), in_proj_covar=tensor([0.0495, 0.0407, 0.0488, 0.0342, 0.0398, 0.0424, 0.0419, 0.0455], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 19:05:20,614 INFO [train.py:937] (1/4) Epoch 21, validation: loss=0.1693, simple_loss=0.2693, pruned_loss=0.03465, over 944034.00 frames. 2023-04-02 19:05:20,614 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18821MB 2023-04-02 19:05:24,360 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_rvb from training. Duration: 29.735 2023-04-02 19:05:57,545 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=139589.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:06:05,902 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.81 vs. limit=2.0 2023-04-02 19:06:12,384 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=139600.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:06:12,539 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=139600.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:06:23,828 INFO [train.py:903] (1/4) Epoch 21, batch 3050, loss[loss=0.2004, simple_loss=0.2807, pruned_loss=0.06004, over 19615.00 frames. ], tot_loss[loss=0.2102, simple_loss=0.2897, pruned_loss=0.06535, over 3828417.51 frames. ], batch size: 50, lr: 3.93e-03, grad_scale: 8.0 2023-04-02 19:06:35,312 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 2023-04-02 19:06:48,012 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.185e+02 5.171e+02 6.119e+02 7.965e+02 1.649e+03, threshold=1.224e+03, percent-clipped=3.0 2023-04-02 19:06:57,090 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=139638.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:07:03,688 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=139642.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:07:14,487 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=139651.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:07:24,495 INFO [train.py:903] (1/4) Epoch 21, batch 3100, loss[loss=0.2079, simple_loss=0.2948, pruned_loss=0.06054, over 19664.00 frames. ], tot_loss[loss=0.2107, simple_loss=0.2902, pruned_loss=0.06562, over 3839421.17 frames. ], batch size: 58, lr: 3.93e-03, grad_scale: 8.0 2023-04-02 19:07:34,153 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=139667.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:07:43,956 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.8906, 1.5812, 1.8816, 1.6400, 4.3676, 0.9843, 2.6094, 4.7293], device='cuda:1'), covar=tensor([0.0421, 0.2834, 0.2727, 0.2010, 0.0761, 0.2835, 0.1396, 0.0189], device='cuda:1'), in_proj_covar=tensor([0.0407, 0.0363, 0.0381, 0.0342, 0.0369, 0.0346, 0.0373, 0.0399], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 19:08:25,950 INFO [train.py:903] (1/4) Epoch 21, batch 3150, loss[loss=0.1932, simple_loss=0.2832, pruned_loss=0.0516, over 19745.00 frames. ], tot_loss[loss=0.2096, simple_loss=0.2891, pruned_loss=0.06505, over 3838685.59 frames. ], batch size: 63, lr: 3.93e-03, grad_scale: 8.0 2023-04-02 19:08:32,255 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=139715.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:08:51,318 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.442e+02 5.084e+02 6.643e+02 9.166e+02 2.493e+03, threshold=1.329e+03, percent-clipped=12.0 2023-04-02 19:08:52,511 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 2023-04-02 19:09:19,229 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=139753.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:09:28,040 INFO [train.py:903] (1/4) Epoch 21, batch 3200, loss[loss=0.1798, simple_loss=0.2601, pruned_loss=0.04979, over 19849.00 frames. ], tot_loss[loss=0.2097, simple_loss=0.2893, pruned_loss=0.06511, over 3834327.18 frames. ], batch size: 52, lr: 3.93e-03, grad_scale: 8.0 2023-04-02 19:09:36,946 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=139766.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:10:01,980 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=139787.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:10:27,856 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=139807.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:10:30,931 INFO [train.py:903] (1/4) Epoch 21, batch 3250, loss[loss=0.2209, simple_loss=0.3054, pruned_loss=0.06827, over 19667.00 frames. ], tot_loss[loss=0.2103, simple_loss=0.2899, pruned_loss=0.06537, over 3831760.39 frames. ], batch size: 58, lr: 3.92e-03, grad_scale: 8.0 2023-04-02 19:10:46,030 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=139822.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:10:48,411 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=139824.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:10:55,283 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.059e+02 5.030e+02 6.558e+02 8.674e+02 2.471e+03, threshold=1.312e+03, percent-clipped=9.0 2023-04-02 19:11:31,708 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.7480, 1.7015, 1.6234, 1.4407, 1.3610, 1.4114, 0.3062, 0.6767], device='cuda:1'), covar=tensor([0.0658, 0.0639, 0.0433, 0.0614, 0.1292, 0.0751, 0.1294, 0.1125], device='cuda:1'), in_proj_covar=tensor([0.0356, 0.0351, 0.0354, 0.0379, 0.0456, 0.0384, 0.0331, 0.0339], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 19:11:32,327 INFO [train.py:903] (1/4) Epoch 21, batch 3300, loss[loss=0.2214, simple_loss=0.3029, pruned_loss=0.06999, over 19534.00 frames. ], tot_loss[loss=0.2102, simple_loss=0.2899, pruned_loss=0.06528, over 3827835.25 frames. ], batch size: 54, lr: 3.92e-03, grad_scale: 8.0 2023-04-02 19:11:35,819 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 2023-04-02 19:11:43,182 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=139868.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:12:34,964 INFO [train.py:903] (1/4) Epoch 21, batch 3350, loss[loss=0.2267, simple_loss=0.3043, pruned_loss=0.07451, over 19523.00 frames. ], tot_loss[loss=0.2118, simple_loss=0.2914, pruned_loss=0.06616, over 3818836.05 frames. ], batch size: 64, lr: 3.92e-03, grad_scale: 8.0 2023-04-02 19:12:36,483 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([0.9893, 1.1991, 1.5570, 1.1604, 2.5183, 3.5328, 3.2499, 3.6669], device='cuda:1'), covar=tensor([0.1813, 0.3930, 0.3554, 0.2478, 0.0646, 0.0177, 0.0225, 0.0282], device='cuda:1'), in_proj_covar=tensor([0.0270, 0.0321, 0.0350, 0.0263, 0.0242, 0.0185, 0.0216, 0.0253], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-02 19:12:36,719 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.13 vs. limit=5.0 2023-04-02 19:12:59,279 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.69 vs. limit=5.0 2023-04-02 19:13:00,646 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 2023-04-02 19:13:00,886 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.203e+02 4.734e+02 5.660e+02 7.256e+02 1.171e+03, threshold=1.132e+03, percent-clipped=0.0 2023-04-02 19:13:04,626 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=139933.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:13:08,525 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.82 vs. limit=5.0 2023-04-02 19:13:17,284 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=139944.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:13:37,525 INFO [train.py:903] (1/4) Epoch 21, batch 3400, loss[loss=0.1937, simple_loss=0.2674, pruned_loss=0.06002, over 16031.00 frames. ], tot_loss[loss=0.2106, simple_loss=0.29, pruned_loss=0.06554, over 3827885.49 frames. ], batch size: 35, lr: 3.92e-03, grad_scale: 8.0 2023-04-02 19:13:45,843 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9776, 2.8993, 1.8671, 1.9303, 2.6505, 1.6257, 1.5919, 2.2272], device='cuda:1'), covar=tensor([0.1299, 0.0785, 0.1053, 0.0817, 0.0559, 0.1219, 0.0894, 0.0654], device='cuda:1'), in_proj_covar=tensor([0.0299, 0.0310, 0.0333, 0.0258, 0.0243, 0.0332, 0.0286, 0.0270], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 19:13:52,377 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=139971.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:14:22,167 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=139996.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:14:41,357 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140009.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:14:42,108 INFO [train.py:903] (1/4) Epoch 21, batch 3450, loss[loss=0.1913, simple_loss=0.2631, pruned_loss=0.05976, over 19738.00 frames. ], tot_loss[loss=0.2106, simple_loss=0.2898, pruned_loss=0.06574, over 3825302.18 frames. ], batch size: 45, lr: 3.92e-03, grad_scale: 8.0 2023-04-02 19:14:43,274 WARNING [train.py:1073] (1/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9_rvb from training. Duration: 25.2444375 2023-04-02 19:14:56,560 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140022.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:15:06,287 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.793e+02 5.068e+02 6.577e+02 8.644e+02 2.362e+03, threshold=1.315e+03, percent-clipped=9.0 2023-04-02 19:15:11,005 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140034.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:15:27,227 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140047.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:15:28,360 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140048.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:15:41,653 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140059.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:15:42,387 INFO [train.py:903] (1/4) Epoch 21, batch 3500, loss[loss=0.2276, simple_loss=0.307, pruned_loss=0.07415, over 19359.00 frames. ], tot_loss[loss=0.2121, simple_loss=0.2912, pruned_loss=0.06654, over 3826894.79 frames. ], batch size: 70, lr: 3.92e-03, grad_scale: 8.0 2023-04-02 19:16:45,563 INFO [train.py:903] (1/4) Epoch 21, batch 3550, loss[loss=0.1718, simple_loss=0.251, pruned_loss=0.04626, over 19734.00 frames. ], tot_loss[loss=0.2125, simple_loss=0.2914, pruned_loss=0.06682, over 3822659.68 frames. ], batch size: 47, lr: 3.92e-03, grad_scale: 8.0 2023-04-02 19:16:59,729 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=140121.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:17:08,444 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=140128.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:17:10,319 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.090e+02 5.095e+02 6.549e+02 8.089e+02 2.006e+03, threshold=1.310e+03, percent-clipped=2.0 2023-04-02 19:17:11,696 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=140131.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:17:25,658 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.4856, 1.2879, 1.5394, 1.5134, 3.0753, 1.1939, 2.3630, 3.4101], device='cuda:1'), covar=tensor([0.0505, 0.2744, 0.2743, 0.1819, 0.0724, 0.2375, 0.1154, 0.0279], device='cuda:1'), in_proj_covar=tensor([0.0405, 0.0362, 0.0381, 0.0341, 0.0368, 0.0346, 0.0373, 0.0397], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 19:17:30,343 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1167, 1.2541, 1.6736, 0.9872, 2.3584, 2.9729, 2.6940, 3.1287], device='cuda:1'), covar=tensor([0.1642, 0.3866, 0.3341, 0.2712, 0.0653, 0.0278, 0.0268, 0.0307], device='cuda:1'), in_proj_covar=tensor([0.0270, 0.0322, 0.0351, 0.0263, 0.0243, 0.0185, 0.0216, 0.0253], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-02 19:17:36,884 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=140151.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:17:47,035 INFO [train.py:903] (1/4) Epoch 21, batch 3600, loss[loss=0.1726, simple_loss=0.2491, pruned_loss=0.04798, over 19357.00 frames. ], tot_loss[loss=0.2119, simple_loss=0.2909, pruned_loss=0.06643, over 3813126.67 frames. ], batch size: 47, lr: 3.92e-03, grad_scale: 8.0 2023-04-02 19:17:56,425 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=140166.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:17:58,706 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=140168.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:18:51,574 INFO [train.py:903] (1/4) Epoch 21, batch 3650, loss[loss=0.2366, simple_loss=0.2974, pruned_loss=0.08783, over 19865.00 frames. ], tot_loss[loss=0.2112, simple_loss=0.2904, pruned_loss=0.06596, over 3824360.66 frames. ], batch size: 52, lr: 3.92e-03, grad_scale: 8.0 2023-04-02 19:18:54,129 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=140212.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:19:06,103 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.46 vs. limit=2.0 2023-04-02 19:19:15,567 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.814e+02 5.123e+02 6.079e+02 7.474e+02 1.635e+03, threshold=1.216e+03, percent-clipped=1.0 2023-04-02 19:19:37,412 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140246.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:19:38,724 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0722, 2.2037, 2.3983, 2.7540, 2.1009, 2.6278, 2.4397, 2.2332], device='cuda:1'), covar=tensor([0.4160, 0.3819, 0.1829, 0.2281, 0.3982, 0.2057, 0.4269, 0.3128], device='cuda:1'), in_proj_covar=tensor([0.0893, 0.0954, 0.0715, 0.0926, 0.0871, 0.0807, 0.0835, 0.0776], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 19:19:41,124 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6897, 1.6704, 1.5406, 1.3867, 1.3020, 1.4302, 0.2854, 0.7163], device='cuda:1'), covar=tensor([0.0582, 0.0604, 0.0384, 0.0588, 0.1048, 0.0675, 0.1182, 0.0959], device='cuda:1'), in_proj_covar=tensor([0.0356, 0.0353, 0.0357, 0.0382, 0.0458, 0.0386, 0.0334, 0.0340], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 19:19:54,186 INFO [train.py:903] (1/4) Epoch 21, batch 3700, loss[loss=0.1983, simple_loss=0.2836, pruned_loss=0.05648, over 19570.00 frames. ], tot_loss[loss=0.2103, simple_loss=0.2897, pruned_loss=0.06542, over 3830093.69 frames. ], batch size: 61, lr: 3.92e-03, grad_scale: 8.0 2023-04-02 19:20:01,702 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140266.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:20:15,867 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=140277.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:20:20,519 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140281.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:20:22,770 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140283.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:20:49,148 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140304.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:20:55,391 INFO [train.py:903] (1/4) Epoch 21, batch 3750, loss[loss=0.2149, simple_loss=0.2836, pruned_loss=0.07312, over 19621.00 frames. ], tot_loss[loss=0.2098, simple_loss=0.2894, pruned_loss=0.06508, over 3832908.18 frames. ], batch size: 50, lr: 3.92e-03, grad_scale: 4.0 2023-04-02 19:21:02,662 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140315.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:21:17,856 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140327.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:21:20,269 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140329.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:21:22,187 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.949e+02 4.554e+02 6.128e+02 7.118e+02 1.255e+03, threshold=1.226e+03, percent-clipped=1.0 2023-04-02 19:21:33,824 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140340.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:21:57,847 INFO [train.py:903] (1/4) Epoch 21, batch 3800, loss[loss=0.1954, simple_loss=0.2839, pruned_loss=0.05342, over 19628.00 frames. ], tot_loss[loss=0.2093, simple_loss=0.2891, pruned_loss=0.06479, over 3824250.74 frames. ], batch size: 57, lr: 3.92e-03, grad_scale: 4.0 2023-04-02 19:22:19,205 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.1049, 3.5060, 3.7513, 3.8371, 1.5607, 3.5126, 3.0519, 3.2263], device='cuda:1'), covar=tensor([0.2533, 0.1658, 0.1215, 0.1511, 0.7686, 0.2153, 0.1325, 0.2424], device='cuda:1'), in_proj_covar=tensor([0.0775, 0.0730, 0.0932, 0.0819, 0.0822, 0.0695, 0.0567, 0.0869], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-02 19:22:30,440 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9_rvb from training. Duration: 29.1166875 2023-04-02 19:22:54,677 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=140405.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:23:01,181 INFO [train.py:903] (1/4) Epoch 21, batch 3850, loss[loss=0.2322, simple_loss=0.3098, pruned_loss=0.0773, over 19728.00 frames. ], tot_loss[loss=0.2096, simple_loss=0.2895, pruned_loss=0.06485, over 3838249.12 frames. ], batch size: 51, lr: 3.92e-03, grad_scale: 4.0 2023-04-02 19:23:15,788 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.14 vs. limit=2.0 2023-04-02 19:23:25,867 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.131e+02 5.035e+02 6.153e+02 7.922e+02 1.662e+03, threshold=1.231e+03, percent-clipped=3.0 2023-04-02 19:24:03,194 INFO [train.py:903] (1/4) Epoch 21, batch 3900, loss[loss=0.2041, simple_loss=0.2897, pruned_loss=0.0592, over 19681.00 frames. ], tot_loss[loss=0.2104, simple_loss=0.2902, pruned_loss=0.06531, over 3834485.28 frames. ], batch size: 59, lr: 3.92e-03, grad_scale: 4.0 2023-04-02 19:24:09,094 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=140465.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:24:17,262 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=140472.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:24:51,310 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4826, 1.6118, 1.6180, 1.9795, 1.5650, 1.8245, 1.7171, 1.4330], device='cuda:1'), covar=tensor([0.4973, 0.4420, 0.3018, 0.2671, 0.4131, 0.2523, 0.6570, 0.5310], device='cuda:1'), in_proj_covar=tensor([0.0894, 0.0956, 0.0716, 0.0931, 0.0873, 0.0811, 0.0838, 0.0779], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 19:24:55,671 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140502.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:25:01,340 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=140507.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:25:04,352 INFO [train.py:903] (1/4) Epoch 21, batch 3950, loss[loss=0.1747, simple_loss=0.2722, pruned_loss=0.03861, over 19673.00 frames. ], tot_loss[loss=0.2095, simple_loss=0.2895, pruned_loss=0.06476, over 3840047.70 frames. ], batch size: 58, lr: 3.91e-03, grad_scale: 4.0 2023-04-02 19:25:10,227 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 2023-04-02 19:25:20,285 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140522.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:25:26,816 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=140527.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:25:26,940 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140527.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:25:31,859 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.897e+02 4.763e+02 5.824e+02 7.544e+02 1.413e+03, threshold=1.165e+03, percent-clipped=2.0 2023-04-02 19:25:40,212 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140537.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:25:42,508 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140539.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:25:51,820 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140547.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:26:07,098 INFO [train.py:903] (1/4) Epoch 21, batch 4000, loss[loss=0.1789, simple_loss=0.2605, pruned_loss=0.04869, over 19763.00 frames. ], tot_loss[loss=0.2087, simple_loss=0.2884, pruned_loss=0.06448, over 3847460.63 frames. ], batch size: 47, lr: 3.91e-03, grad_scale: 8.0 2023-04-02 19:26:09,710 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140562.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:26:13,052 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140564.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:26:33,149 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140580.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:26:36,735 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140583.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:26:41,300 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140587.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:26:55,176 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1_rvb from training. Duration: 0.7545625 2023-04-02 19:26:58,526 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.48 vs. limit=5.0 2023-04-02 19:27:02,783 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.13 vs. limit=2.0 2023-04-02 19:27:09,311 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140608.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:27:11,219 INFO [train.py:903] (1/4) Epoch 21, batch 4050, loss[loss=0.2133, simple_loss=0.2948, pruned_loss=0.06587, over 19541.00 frames. ], tot_loss[loss=0.2094, simple_loss=0.2893, pruned_loss=0.06478, over 3837974.93 frames. ], batch size: 54, lr: 3.91e-03, grad_scale: 8.0 2023-04-02 19:27:24,120 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0395, 1.9373, 1.8402, 1.6608, 1.3668, 1.5920, 0.6673, 1.0008], device='cuda:1'), covar=tensor([0.0778, 0.0769, 0.0497, 0.0915, 0.1417, 0.1117, 0.1462, 0.1303], device='cuda:1'), in_proj_covar=tensor([0.0354, 0.0351, 0.0356, 0.0380, 0.0456, 0.0384, 0.0332, 0.0338], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 19:27:25,060 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=140621.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:27:36,528 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.169e+02 5.165e+02 6.710e+02 8.332e+02 1.443e+03, threshold=1.342e+03, percent-clipped=3.0 2023-04-02 19:27:41,321 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5833, 1.3130, 1.2216, 1.4355, 1.1246, 1.3317, 1.1953, 1.3948], device='cuda:1'), covar=tensor([0.1162, 0.1152, 0.1630, 0.1077, 0.1364, 0.0695, 0.1660, 0.0894], device='cuda:1'), in_proj_covar=tensor([0.0270, 0.0359, 0.0311, 0.0249, 0.0300, 0.0252, 0.0310, 0.0254], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 19:28:13,830 INFO [train.py:903] (1/4) Epoch 21, batch 4100, loss[loss=0.197, simple_loss=0.2656, pruned_loss=0.06417, over 19304.00 frames. ], tot_loss[loss=0.2093, simple_loss=0.2891, pruned_loss=0.06475, over 3838603.58 frames. ], batch size: 44, lr: 3.91e-03, grad_scale: 8.0 2023-04-02 19:28:28,665 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=5.00 vs. limit=5.0 2023-04-02 19:28:52,828 WARNING [train.py:1073] (1/4) Exclude cut with ID 497-129325-0061-9566_sp1.1_rvb from training. Duration: 0.97725 2023-04-02 19:29:15,740 INFO [train.py:903] (1/4) Epoch 21, batch 4150, loss[loss=0.1741, simple_loss=0.2639, pruned_loss=0.04216, over 19841.00 frames. ], tot_loss[loss=0.2094, simple_loss=0.2891, pruned_loss=0.06482, over 3834396.83 frames. ], batch size: 52, lr: 3.91e-03, grad_scale: 8.0 2023-04-02 19:29:42,681 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.451e+02 5.318e+02 6.390e+02 7.957e+02 1.686e+03, threshold=1.278e+03, percent-clipped=3.0 2023-04-02 19:29:50,018 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140736.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:29:55,900 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=140741.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:30:00,649 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.7716, 4.3755, 2.7102, 3.7388, 0.8830, 4.3270, 4.1534, 4.2862], device='cuda:1'), covar=tensor([0.0621, 0.0973, 0.1982, 0.0950, 0.4101, 0.0620, 0.0893, 0.1153], device='cuda:1'), in_proj_covar=tensor([0.0500, 0.0409, 0.0490, 0.0347, 0.0399, 0.0430, 0.0423, 0.0458], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 19:30:05,104 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=140749.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:30:17,539 INFO [train.py:903] (1/4) Epoch 21, batch 4200, loss[loss=0.2178, simple_loss=0.2837, pruned_loss=0.07589, over 19782.00 frames. ], tot_loss[loss=0.2101, simple_loss=0.2894, pruned_loss=0.06543, over 3823721.69 frames. ], batch size: 48, lr: 3.91e-03, grad_scale: 8.0 2023-04-02 19:30:24,381 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 2023-04-02 19:31:15,136 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.0825, 1.2308, 1.4385, 1.4019, 2.7339, 3.6904, 3.4085, 3.9696], device='cuda:1'), covar=tensor([0.1861, 0.4005, 0.3897, 0.2410, 0.0642, 0.0193, 0.0233, 0.0262], device='cuda:1'), in_proj_covar=tensor([0.0270, 0.0320, 0.0349, 0.0263, 0.0241, 0.0184, 0.0215, 0.0252], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-02 19:31:21,349 INFO [train.py:903] (1/4) Epoch 21, batch 4250, loss[loss=0.2211, simple_loss=0.3075, pruned_loss=0.06734, over 19543.00 frames. ], tot_loss[loss=0.2098, simple_loss=0.2894, pruned_loss=0.0651, over 3817435.38 frames. ], batch size: 56, lr: 3.91e-03, grad_scale: 8.0 2023-04-02 19:31:40,121 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9_rvb from training. Duration: 29.816625 2023-04-02 19:31:42,789 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=140827.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:31:47,368 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.155e+02 5.291e+02 6.401e+02 8.566e+02 1.506e+03, threshold=1.280e+03, percent-clipped=6.0 2023-04-02 19:31:50,843 WARNING [train.py:1073] (1/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 2023-04-02 19:31:51,262 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.7245, 1.7107, 1.6119, 1.3648, 1.3443, 1.4088, 0.2924, 0.6824], device='cuda:1'), covar=tensor([0.0622, 0.0644, 0.0407, 0.0630, 0.1232, 0.0697, 0.1244, 0.1081], device='cuda:1'), in_proj_covar=tensor([0.0357, 0.0353, 0.0358, 0.0383, 0.0459, 0.0386, 0.0335, 0.0341], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 19:31:53,749 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140836.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:32:01,558 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140843.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:32:13,492 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=140851.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:32:24,304 INFO [train.py:903] (1/4) Epoch 21, batch 4300, loss[loss=0.1848, simple_loss=0.2767, pruned_loss=0.04645, over 19792.00 frames. ], tot_loss[loss=0.2089, simple_loss=0.2886, pruned_loss=0.06456, over 3821477.15 frames. ], batch size: 56, lr: 3.91e-03, grad_scale: 8.0 2023-04-02 19:32:25,822 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140861.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:32:29,377 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140864.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:32:33,859 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140868.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:32:37,169 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=140871.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:32:53,903 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.8723, 1.8426, 1.6831, 1.4302, 1.2259, 1.4014, 0.5115, 0.8036], device='cuda:1'), covar=tensor([0.0821, 0.0781, 0.0481, 0.0910, 0.1503, 0.1129, 0.1446, 0.1309], device='cuda:1'), in_proj_covar=tensor([0.0356, 0.0353, 0.0357, 0.0382, 0.0460, 0.0385, 0.0335, 0.0340], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 19:32:56,049 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.5840, 4.6951, 5.2700, 5.2866, 2.1899, 4.9314, 4.3130, 4.9558], device='cuda:1'), covar=tensor([0.1506, 0.1365, 0.0570, 0.0624, 0.5808, 0.0860, 0.0609, 0.1176], device='cuda:1'), in_proj_covar=tensor([0.0785, 0.0736, 0.0950, 0.0831, 0.0831, 0.0704, 0.0573, 0.0878], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-02 19:33:18,586 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 2023-04-02 19:33:24,432 INFO [train.py:903] (1/4) Epoch 21, batch 4350, loss[loss=0.2009, simple_loss=0.2924, pruned_loss=0.05463, over 18795.00 frames. ], tot_loss[loss=0.2106, simple_loss=0.2903, pruned_loss=0.06544, over 3817955.34 frames. ], batch size: 74, lr: 3.91e-03, grad_scale: 8.0 2023-04-02 19:33:51,500 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.395e+02 4.814e+02 5.846e+02 6.905e+02 1.613e+03, threshold=1.169e+03, percent-clipped=2.0 2023-04-02 19:33:51,985 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5027, 1.6722, 2.0055, 1.8092, 3.2492, 2.6628, 3.6720, 1.5958], device='cuda:1'), covar=tensor([0.2499, 0.4267, 0.2727, 0.1863, 0.1423, 0.2048, 0.1395, 0.4182], device='cuda:1'), in_proj_covar=tensor([0.0530, 0.0639, 0.0707, 0.0481, 0.0614, 0.0530, 0.0660, 0.0545], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 19:33:56,421 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=140935.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:34:25,325 INFO [train.py:903] (1/4) Epoch 21, batch 4400, loss[loss=0.1853, simple_loss=0.2611, pruned_loss=0.05475, over 19368.00 frames. ], tot_loss[loss=0.2118, simple_loss=0.2913, pruned_loss=0.06614, over 3812208.06 frames. ], batch size: 47, lr: 3.91e-03, grad_scale: 8.0 2023-04-02 19:34:29,844 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9089, 1.6488, 1.5734, 1.9222, 1.5881, 1.6369, 1.5582, 1.7617], device='cuda:1'), covar=tensor([0.1139, 0.1507, 0.1600, 0.1049, 0.1449, 0.0588, 0.1528, 0.0779], device='cuda:1'), in_proj_covar=tensor([0.0273, 0.0362, 0.0314, 0.0252, 0.0302, 0.0254, 0.0312, 0.0257], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 19:34:32,127 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=140965.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:34:33,377 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140966.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:34:53,843 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_rvb from training. Duration: 25.285 2023-04-02 19:34:58,767 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140986.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:35:02,859 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 2023-04-02 19:35:05,678 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140992.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:35:26,727 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.6865, 2.4780, 2.3288, 2.7887, 2.5426, 2.2560, 2.2334, 2.5588], device='cuda:1'), covar=tensor([0.0920, 0.1473, 0.1322, 0.0935, 0.1262, 0.0486, 0.1249, 0.0623], device='cuda:1'), in_proj_covar=tensor([0.0272, 0.0360, 0.0312, 0.0250, 0.0301, 0.0252, 0.0310, 0.0256], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 19:35:27,476 INFO [train.py:903] (1/4) Epoch 21, batch 4450, loss[loss=0.1924, simple_loss=0.279, pruned_loss=0.05291, over 19691.00 frames. ], tot_loss[loss=0.212, simple_loss=0.2914, pruned_loss=0.06637, over 3820043.88 frames. ], batch size: 59, lr: 3.91e-03, grad_scale: 8.0 2023-04-02 19:35:37,899 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=141017.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:35:54,619 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.268e+02 5.226e+02 6.450e+02 8.222e+02 2.218e+03, threshold=1.290e+03, percent-clipped=9.0 2023-04-02 19:36:06,103 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 2023-04-02 19:36:32,491 INFO [train.py:903] (1/4) Epoch 21, batch 4500, loss[loss=0.2168, simple_loss=0.2982, pruned_loss=0.06774, over 19708.00 frames. ], tot_loss[loss=0.212, simple_loss=0.2915, pruned_loss=0.06629, over 3809570.88 frames. ], batch size: 59, lr: 3.91e-03, grad_scale: 8.0 2023-04-02 19:37:01,382 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=141085.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:37:34,661 INFO [train.py:903] (1/4) Epoch 21, batch 4550, loss[loss=0.2123, simple_loss=0.2834, pruned_loss=0.07061, over 19590.00 frames. ], tot_loss[loss=0.213, simple_loss=0.2923, pruned_loss=0.06688, over 3806712.23 frames. ], batch size: 52, lr: 3.91e-03, grad_scale: 8.0 2023-04-02 19:37:46,079 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9_rvb from training. Duration: 28.72225 2023-04-02 19:37:46,498 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=141120.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:37:59,732 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.292e+02 5.246e+02 6.195e+02 7.478e+02 1.454e+03, threshold=1.239e+03, percent-clipped=4.0 2023-04-02 19:38:10,839 WARNING [train.py:1073] (1/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 2023-04-02 19:38:18,657 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=141145.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:38:28,874 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.3790, 3.9592, 2.4410, 3.5508, 0.9344, 3.8955, 3.8246, 3.9252], device='cuda:1'), covar=tensor([0.0705, 0.1160, 0.2217, 0.0897, 0.4024, 0.0777, 0.0888, 0.1157], device='cuda:1'), in_proj_covar=tensor([0.0499, 0.0408, 0.0489, 0.0345, 0.0397, 0.0428, 0.0420, 0.0455], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 19:38:35,234 INFO [train.py:903] (1/4) Epoch 21, batch 4600, loss[loss=0.2445, simple_loss=0.3146, pruned_loss=0.08717, over 13439.00 frames. ], tot_loss[loss=0.2134, simple_loss=0.2929, pruned_loss=0.06697, over 3794228.71 frames. ], batch size: 136, lr: 3.91e-03, grad_scale: 8.0 2023-04-02 19:38:36,700 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=141161.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:38:49,776 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=141171.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:39:25,216 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=141200.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:39:36,159 INFO [train.py:903] (1/4) Epoch 21, batch 4650, loss[loss=0.2028, simple_loss=0.2808, pruned_loss=0.06244, over 17469.00 frames. ], tot_loss[loss=0.2121, simple_loss=0.2919, pruned_loss=0.06619, over 3799264.48 frames. ], batch size: 101, lr: 3.91e-03, grad_scale: 8.0 2023-04-02 19:39:53,044 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=141222.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:39:56,082 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_rvb from training. Duration: 0.92 2023-04-02 19:40:02,331 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.173e+02 4.513e+02 5.853e+02 7.712e+02 1.984e+03, threshold=1.171e+03, percent-clipped=6.0 2023-04-02 19:40:05,984 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_rvb from training. Duration: 0.83 2023-04-02 19:40:12,917 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([6.1249, 5.5323, 2.7719, 4.9282, 1.0665, 5.7271, 5.5590, 5.7462], device='cuda:1'), covar=tensor([0.0359, 0.0735, 0.2004, 0.0633, 0.3859, 0.0409, 0.0703, 0.0841], device='cuda:1'), in_proj_covar=tensor([0.0495, 0.0406, 0.0486, 0.0343, 0.0395, 0.0424, 0.0419, 0.0451], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 19:40:15,513 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=141242.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:40:18,356 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0 2023-04-02 19:40:21,266 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=141247.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:40:37,901 INFO [train.py:903] (1/4) Epoch 21, batch 4700, loss[loss=0.1845, simple_loss=0.2652, pruned_loss=0.05185, over 19765.00 frames. ], tot_loss[loss=0.2123, simple_loss=0.2918, pruned_loss=0.06644, over 3806280.82 frames. ], batch size: 48, lr: 3.90e-03, grad_scale: 8.0 2023-04-02 19:40:45,381 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 2023-04-02 19:40:47,274 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=141267.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:41:00,533 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9_rvb from training. Duration: 0.92225 2023-04-02 19:41:00,651 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=141279.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:41:08,829 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=141286.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:41:38,494 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=141309.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:41:39,464 INFO [train.py:903] (1/4) Epoch 21, batch 4750, loss[loss=0.2141, simple_loss=0.2964, pruned_loss=0.06589, over 19865.00 frames. ], tot_loss[loss=0.2124, simple_loss=0.2918, pruned_loss=0.06654, over 3800583.02 frames. ], batch size: 52, lr: 3.90e-03, grad_scale: 8.0 2023-04-02 19:41:59,019 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.1492, 1.8651, 1.7956, 2.0772, 1.8884, 1.8275, 1.7492, 1.9695], device='cuda:1'), covar=tensor([0.0979, 0.1371, 0.1371, 0.0987, 0.1238, 0.0532, 0.1277, 0.0689], device='cuda:1'), in_proj_covar=tensor([0.0271, 0.0360, 0.0312, 0.0251, 0.0301, 0.0253, 0.0309, 0.0256], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 19:42:03,068 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.546e+02 5.235e+02 6.308e+02 8.195e+02 2.468e+03, threshold=1.262e+03, percent-clipped=8.0 2023-04-02 19:42:25,005 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=141347.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:42:30,013 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 2023-04-02 19:42:39,827 INFO [train.py:903] (1/4) Epoch 21, batch 4800, loss[loss=0.2495, simple_loss=0.3247, pruned_loss=0.08713, over 18233.00 frames. ], tot_loss[loss=0.2133, simple_loss=0.2925, pruned_loss=0.06703, over 3813519.89 frames. ], batch size: 83, lr: 3.90e-03, grad_scale: 8.0 2023-04-02 19:43:23,121 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=141394.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:43:41,803 INFO [train.py:903] (1/4) Epoch 21, batch 4850, loss[loss=0.2302, simple_loss=0.3071, pruned_loss=0.07663, over 19505.00 frames. ], tot_loss[loss=0.2126, simple_loss=0.2917, pruned_loss=0.06677, over 3818219.94 frames. ], batch size: 64, lr: 3.90e-03, grad_scale: 8.0 2023-04-02 19:44:01,449 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=141424.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:44:07,726 WARNING [train.py:1073] (1/4) Exclude cut with ID 774-127930-0014-48411_sp1.1_rvb from training. Duration: 0.95 2023-04-02 19:44:08,841 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.291e+02 5.192e+02 6.534e+02 8.985e+02 2.500e+03, threshold=1.307e+03, percent-clipped=12.0 2023-04-02 19:44:13,596 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=141435.0, num_to_drop=1, layers_to_drop={1} 2023-04-02 19:44:28,294 WARNING [train.py:1073] (1/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1_rvb from training. Duration: 0.9409375 2023-04-02 19:44:33,280 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 2023-04-02 19:44:34,433 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 2023-04-02 19:44:39,592 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=141456.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:44:44,245 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=141459.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:44:45,062 INFO [train.py:903] (1/4) Epoch 21, batch 4900, loss[loss=0.2238, simple_loss=0.303, pruned_loss=0.07226, over 19643.00 frames. ], tot_loss[loss=0.2117, simple_loss=0.2907, pruned_loss=0.06631, over 3808593.44 frames. ], batch size: 55, lr: 3.90e-03, grad_scale: 8.0 2023-04-02 19:44:47,084 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1_rvb from training. Duration: 25.3818125 2023-04-02 19:45:05,756 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 2023-04-02 19:45:10,369 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=141481.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:45:18,369 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([0.8370, 0.8917, 0.8596, 0.7453, 0.7431, 0.7719, 0.1057, 0.2921], device='cuda:1'), covar=tensor([0.0468, 0.0472, 0.0311, 0.0444, 0.0800, 0.0479, 0.1113, 0.0791], device='cuda:1'), in_proj_covar=tensor([0.0354, 0.0350, 0.0353, 0.0378, 0.0457, 0.0383, 0.0332, 0.0338], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 19:45:40,780 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=141505.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:45:46,104 INFO [train.py:903] (1/4) Epoch 21, batch 4950, loss[loss=0.2063, simple_loss=0.2756, pruned_loss=0.06849, over 19722.00 frames. ], tot_loss[loss=0.2126, simple_loss=0.2912, pruned_loss=0.06696, over 3799029.96 frames. ], batch size: 51, lr: 3.90e-03, grad_scale: 8.0 2023-04-02 19:46:03,680 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 2023-04-02 19:46:10,482 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.127e+02 5.116e+02 6.031e+02 7.721e+02 1.403e+03, threshold=1.206e+03, percent-clipped=1.0 2023-04-02 19:46:26,541 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=141542.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:46:28,472 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 2023-04-02 19:46:46,921 INFO [train.py:903] (1/4) Epoch 21, batch 5000, loss[loss=0.1865, simple_loss=0.265, pruned_loss=0.05404, over 19376.00 frames. ], tot_loss[loss=0.2133, simple_loss=0.292, pruned_loss=0.06733, over 3784245.50 frames. ], batch size: 47, lr: 3.90e-03, grad_scale: 8.0 2023-04-02 19:46:53,574 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_rvb from training. Duration: 27.14 2023-04-02 19:46:55,131 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=141567.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:47:08,253 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 2023-04-02 19:47:48,210 INFO [train.py:903] (1/4) Epoch 21, batch 5050, loss[loss=0.2143, simple_loss=0.3015, pruned_loss=0.06352, over 19744.00 frames. ], tot_loss[loss=0.2137, simple_loss=0.2925, pruned_loss=0.06751, over 3789495.74 frames. ], batch size: 63, lr: 3.90e-03, grad_scale: 8.0 2023-04-02 19:48:03,201 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=141620.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:48:16,452 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.325e+02 4.754e+02 5.832e+02 6.826e+02 1.423e+03, threshold=1.166e+03, percent-clipped=2.0 2023-04-02 19:48:25,900 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9_rvb from training. Duration: 26.62775 2023-04-02 19:48:39,227 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=141650.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:48:52,475 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([6.1976, 5.5753, 2.9975, 4.8510, 1.3107, 5.7205, 5.5800, 5.8033], device='cuda:1'), covar=tensor([0.0381, 0.0806, 0.1978, 0.0788, 0.3791, 0.0489, 0.0708, 0.0865], device='cuda:1'), in_proj_covar=tensor([0.0499, 0.0408, 0.0490, 0.0346, 0.0400, 0.0429, 0.0422, 0.0455], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 19:48:53,397 INFO [train.py:903] (1/4) Epoch 21, batch 5100, loss[loss=0.1698, simple_loss=0.2515, pruned_loss=0.04408, over 19626.00 frames. ], tot_loss[loss=0.2117, simple_loss=0.2908, pruned_loss=0.06627, over 3797243.41 frames. ], batch size: 50, lr: 3.90e-03, grad_scale: 8.0 2023-04-02 19:49:04,590 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 2023-04-02 19:49:06,994 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_rvb from training. Duration: 25.035 2023-04-02 19:49:11,629 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_rvb from training. Duration: 27.92 2023-04-02 19:49:12,005 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=141675.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:49:17,687 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=141680.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:49:30,080 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=141691.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:49:50,715 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=141705.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:49:56,348 INFO [train.py:903] (1/4) Epoch 21, batch 5150, loss[loss=0.2123, simple_loss=0.2807, pruned_loss=0.072, over 19413.00 frames. ], tot_loss[loss=0.2123, simple_loss=0.2913, pruned_loss=0.06664, over 3806696.27 frames. ], batch size: 48, lr: 3.90e-03, grad_scale: 8.0 2023-04-02 19:50:09,422 WARNING [train.py:1073] (1/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 2023-04-02 19:50:16,621 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.4977, 0.9535, 1.2885, 1.1548, 1.9604, 1.0741, 2.1418, 2.3047], device='cuda:1'), covar=tensor([0.1000, 0.3857, 0.3402, 0.2213, 0.1406, 0.2515, 0.1111, 0.0687], device='cuda:1'), in_proj_covar=tensor([0.0404, 0.0361, 0.0381, 0.0343, 0.0369, 0.0345, 0.0373, 0.0395], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 19:50:20,920 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.369e+02 4.996e+02 6.424e+02 8.131e+02 1.633e+03, threshold=1.285e+03, percent-clipped=6.0 2023-04-02 19:50:46,532 WARNING [train.py:1073] (1/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1_rvb from training. Duration: 0.9681875 2023-04-02 19:50:57,359 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0623, 2.1320, 2.3381, 2.6488, 2.0662, 2.5459, 2.3534, 2.1295], device='cuda:1'), covar=tensor([0.4011, 0.3808, 0.1795, 0.2404, 0.4111, 0.2113, 0.4350, 0.3247], device='cuda:1'), in_proj_covar=tensor([0.0888, 0.0952, 0.0710, 0.0925, 0.0869, 0.0807, 0.0830, 0.0774], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 19:50:58,082 INFO [train.py:903] (1/4) Epoch 21, batch 5200, loss[loss=0.2174, simple_loss=0.3031, pruned_loss=0.06582, over 18084.00 frames. ], tot_loss[loss=0.2129, simple_loss=0.2922, pruned_loss=0.06683, over 3824298.85 frames. ], batch size: 83, lr: 3.90e-03, grad_scale: 8.0 2023-04-02 19:51:14,042 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 2023-04-02 19:51:22,159 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=141779.0, num_to_drop=1, layers_to_drop={1} 2023-04-02 19:51:51,427 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=141803.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:51:54,947 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=141806.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:51:57,322 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([0.9174, 1.2720, 1.5963, 0.6609, 1.9611, 2.3951, 2.0997, 2.5957], device='cuda:1'), covar=tensor([0.1670, 0.3777, 0.3387, 0.2731, 0.0647, 0.0271, 0.0339, 0.0339], device='cuda:1'), in_proj_covar=tensor([0.0270, 0.0320, 0.0349, 0.0263, 0.0241, 0.0184, 0.0215, 0.0253], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-02 19:51:58,266 WARNING [train.py:1073] (1/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 2023-04-02 19:51:59,437 INFO [train.py:903] (1/4) Epoch 21, batch 5250, loss[loss=0.2286, simple_loss=0.3123, pruned_loss=0.07241, over 19656.00 frames. ], tot_loss[loss=0.2133, simple_loss=0.2924, pruned_loss=0.06706, over 3824084.74 frames. ], batch size: 60, lr: 3.90e-03, grad_scale: 8.0 2023-04-02 19:52:28,862 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.823e+02 4.884e+02 6.185e+02 7.574e+02 1.457e+03, threshold=1.237e+03, percent-clipped=1.0 2023-04-02 19:52:56,687 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.52 vs. limit=2.0 2023-04-02 19:53:02,568 INFO [train.py:903] (1/4) Epoch 21, batch 5300, loss[loss=0.1886, simple_loss=0.2654, pruned_loss=0.05591, over 19844.00 frames. ], tot_loss[loss=0.2137, simple_loss=0.293, pruned_loss=0.06716, over 3823331.75 frames. ], batch size: 52, lr: 3.90e-03, grad_scale: 8.0 2023-04-02 19:53:23,419 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=141876.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:53:24,173 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 2023-04-02 19:53:45,676 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=141894.0, num_to_drop=1, layers_to_drop={1} 2023-04-02 19:53:52,726 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4102, 1.4338, 1.7944, 1.6538, 2.6641, 2.2839, 2.7695, 1.2638], device='cuda:1'), covar=tensor([0.2687, 0.4627, 0.2871, 0.2167, 0.1668, 0.2329, 0.1743, 0.4732], device='cuda:1'), in_proj_covar=tensor([0.0532, 0.0641, 0.0710, 0.0485, 0.0617, 0.0531, 0.0661, 0.0549], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 19:53:55,963 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=141901.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:54:06,886 INFO [train.py:903] (1/4) Epoch 21, batch 5350, loss[loss=0.2543, simple_loss=0.341, pruned_loss=0.0838, over 19347.00 frames. ], tot_loss[loss=0.2129, simple_loss=0.2924, pruned_loss=0.06666, over 3833754.23 frames. ], batch size: 66, lr: 3.90e-03, grad_scale: 8.0 2023-04-02 19:54:11,013 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.55 vs. limit=2.0 2023-04-02 19:54:16,545 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=141918.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:54:32,491 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.261e+02 4.997e+02 5.942e+02 7.628e+02 1.099e+03, threshold=1.188e+03, percent-clipped=0.0 2023-04-02 19:54:35,105 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=141934.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:54:43,723 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9_rvb from training. Duration: 26.438875 2023-04-02 19:54:51,469 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=141946.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:54:58,493 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5870, 1.7149, 2.1053, 1.8263, 3.0522, 2.6329, 3.3553, 1.6828], device='cuda:1'), covar=tensor([0.2397, 0.4123, 0.2525, 0.1883, 0.1527, 0.2028, 0.1529, 0.4126], device='cuda:1'), in_proj_covar=tensor([0.0532, 0.0640, 0.0710, 0.0484, 0.0617, 0.0531, 0.0661, 0.0548], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 19:55:07,460 INFO [train.py:903] (1/4) Epoch 21, batch 5400, loss[loss=0.2091, simple_loss=0.2944, pruned_loss=0.06186, over 17366.00 frames. ], tot_loss[loss=0.212, simple_loss=0.2915, pruned_loss=0.06628, over 3836115.25 frames. ], batch size: 101, lr: 3.89e-03, grad_scale: 8.0 2023-04-02 19:55:22,664 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5904, 1.6884, 1.9476, 1.9151, 1.4212, 1.8932, 1.9611, 1.8214], device='cuda:1'), covar=tensor([0.4094, 0.3466, 0.1837, 0.2330, 0.3776, 0.2106, 0.4872, 0.3270], device='cuda:1'), in_proj_covar=tensor([0.0889, 0.0953, 0.0712, 0.0926, 0.0869, 0.0808, 0.0833, 0.0775], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 19:56:10,559 INFO [train.py:903] (1/4) Epoch 21, batch 5450, loss[loss=0.1704, simple_loss=0.2463, pruned_loss=0.04729, over 19758.00 frames. ], tot_loss[loss=0.2111, simple_loss=0.2908, pruned_loss=0.06569, over 3833399.00 frames. ], batch size: 45, lr: 3.89e-03, grad_scale: 8.0 2023-04-02 19:56:39,852 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.702e+02 5.354e+02 6.392e+02 8.231e+02 1.329e+03, threshold=1.278e+03, percent-clipped=1.0 2023-04-02 19:57:14,050 INFO [train.py:903] (1/4) Epoch 21, batch 5500, loss[loss=0.2251, simple_loss=0.303, pruned_loss=0.07361, over 19778.00 frames. ], tot_loss[loss=0.2112, simple_loss=0.2907, pruned_loss=0.06584, over 3824207.46 frames. ], batch size: 56, lr: 3.89e-03, grad_scale: 8.0 2023-04-02 19:57:17,989 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=142062.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:57:42,007 WARNING [train.py:1073] (1/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9_rvb from training. Duration: 27.47775 2023-04-02 19:57:48,335 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=142087.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:58:17,753 INFO [train.py:903] (1/4) Epoch 21, batch 5550, loss[loss=0.1942, simple_loss=0.2842, pruned_loss=0.05213, over 19601.00 frames. ], tot_loss[loss=0.2109, simple_loss=0.2903, pruned_loss=0.06572, over 3822525.48 frames. ], batch size: 52, lr: 3.89e-03, grad_scale: 8.0 2023-04-02 19:58:26,219 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 2023-04-02 19:58:43,634 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.159e+02 4.904e+02 5.778e+02 7.569e+02 2.193e+03, threshold=1.156e+03, percent-clipped=4.0 2023-04-02 19:59:03,236 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([5.6817, 5.1480, 3.0237, 4.4948, 1.4615, 5.1959, 5.0774, 5.2169], device='cuda:1'), covar=tensor([0.0366, 0.0788, 0.1903, 0.0743, 0.3631, 0.0561, 0.0691, 0.1053], device='cuda:1'), in_proj_covar=tensor([0.0498, 0.0409, 0.0490, 0.0345, 0.0401, 0.0428, 0.0422, 0.0456], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 19:59:07,842 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=142150.0, num_to_drop=1, layers_to_drop={0} 2023-04-02 19:59:15,742 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 2023-04-02 19:59:19,356 INFO [train.py:903] (1/4) Epoch 21, batch 5600, loss[loss=0.1772, simple_loss=0.2615, pruned_loss=0.04648, over 19725.00 frames. ], tot_loss[loss=0.2104, simple_loss=0.2896, pruned_loss=0.0656, over 3815990.56 frames. ], batch size: 51, lr: 3.89e-03, grad_scale: 8.0 2023-04-02 19:59:36,334 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=142174.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 19:59:37,406 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=142175.0, num_to_drop=1, layers_to_drop={1} 2023-04-02 19:59:58,293 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1784, 1.2675, 1.7759, 1.0696, 2.5009, 3.3464, 3.0338, 3.5968], device='cuda:1'), covar=tensor([0.1618, 0.3884, 0.3227, 0.2603, 0.0613, 0.0215, 0.0243, 0.0275], device='cuda:1'), in_proj_covar=tensor([0.0270, 0.0319, 0.0349, 0.0262, 0.0241, 0.0183, 0.0215, 0.0252], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-02 20:00:08,640 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=142199.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 20:00:20,409 INFO [train.py:903] (1/4) Epoch 21, batch 5650, loss[loss=0.1948, simple_loss=0.2752, pruned_loss=0.05716, over 19352.00 frames. ], tot_loss[loss=0.211, simple_loss=0.2901, pruned_loss=0.06594, over 3823613.83 frames. ], batch size: 47, lr: 3.89e-03, grad_scale: 8.0 2023-04-02 20:00:49,291 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.717e+02 4.896e+02 5.835e+02 7.915e+02 1.772e+03, threshold=1.167e+03, percent-clipped=3.0 2023-04-02 20:01:10,302 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9_rvb from training. Duration: 28.0944375 2023-04-02 20:01:22,637 INFO [train.py:903] (1/4) Epoch 21, batch 5700, loss[loss=0.2094, simple_loss=0.2846, pruned_loss=0.06715, over 19733.00 frames. ], tot_loss[loss=0.2118, simple_loss=0.2911, pruned_loss=0.06624, over 3823299.06 frames. ], batch size: 51, lr: 3.89e-03, grad_scale: 8.0 2023-04-02 20:01:45,771 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=142278.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 20:01:59,759 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=142290.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 20:02:09,013 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5250, 1.2700, 1.3028, 2.0848, 1.6157, 1.6207, 1.9327, 1.4464], device='cuda:1'), covar=tensor([0.0986, 0.1174, 0.1191, 0.0762, 0.0889, 0.0927, 0.0853, 0.0869], device='cuda:1'), in_proj_covar=tensor([0.0211, 0.0221, 0.0226, 0.0239, 0.0224, 0.0210, 0.0186, 0.0204], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-02 20:02:26,061 INFO [train.py:903] (1/4) Epoch 21, batch 5750, loss[loss=0.2379, simple_loss=0.2971, pruned_loss=0.08935, over 19484.00 frames. ], tot_loss[loss=0.2106, simple_loss=0.2897, pruned_loss=0.06581, over 3824499.09 frames. ], batch size: 49, lr: 3.89e-03, grad_scale: 8.0 2023-04-02 20:02:28,347 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9_rvb from training. Duration: 33.038875 2023-04-02 20:02:36,341 WARNING [train.py:1073] (1/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 2023-04-02 20:02:41,002 WARNING [train.py:1073] (1/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 2023-04-02 20:02:51,482 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.086e+02 5.170e+02 6.700e+02 8.460e+02 1.665e+03, threshold=1.340e+03, percent-clipped=6.0 2023-04-02 20:03:26,600 INFO [train.py:903] (1/4) Epoch 21, batch 5800, loss[loss=0.2006, simple_loss=0.2729, pruned_loss=0.06421, over 19345.00 frames. ], tot_loss[loss=0.2103, simple_loss=0.2896, pruned_loss=0.0655, over 3824644.57 frames. ], batch size: 47, lr: 3.89e-03, grad_scale: 8.0 2023-04-02 20:04:08,908 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=142393.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 20:04:22,748 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=142405.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 20:04:28,056 INFO [train.py:903] (1/4) Epoch 21, batch 5850, loss[loss=0.2163, simple_loss=0.2991, pruned_loss=0.06681, over 18839.00 frames. ], tot_loss[loss=0.2116, simple_loss=0.2906, pruned_loss=0.06635, over 3811546.75 frames. ], batch size: 74, lr: 3.89e-03, grad_scale: 8.0 2023-04-02 20:04:57,527 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.411e+02 4.867e+02 6.014e+02 7.504e+02 1.855e+03, threshold=1.203e+03, percent-clipped=4.0 2023-04-02 20:05:31,488 INFO [train.py:903] (1/4) Epoch 21, batch 5900, loss[loss=0.2068, simple_loss=0.2935, pruned_loss=0.06011, over 19501.00 frames. ], tot_loss[loss=0.2103, simple_loss=0.2898, pruned_loss=0.06539, over 3820932.80 frames. ], batch size: 64, lr: 3.89e-03, grad_scale: 8.0 2023-04-02 20:05:35,050 WARNING [train.py:1073] (1/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 2023-04-02 20:05:58,749 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9_rvb from training. Duration: 27.511125 2023-04-02 20:06:36,836 INFO [train.py:903] (1/4) Epoch 21, batch 5950, loss[loss=0.2606, simple_loss=0.3279, pruned_loss=0.09664, over 18462.00 frames. ], tot_loss[loss=0.2105, simple_loss=0.2897, pruned_loss=0.06561, over 3818327.02 frames. ], batch size: 84, lr: 3.89e-03, grad_scale: 8.0 2023-04-02 20:07:02,040 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.165e+02 4.752e+02 5.691e+02 7.550e+02 2.003e+03, threshold=1.138e+03, percent-clipped=5.0 2023-04-02 20:07:37,089 INFO [train.py:903] (1/4) Epoch 21, batch 6000, loss[loss=0.2022, simple_loss=0.2952, pruned_loss=0.0546, over 19081.00 frames. ], tot_loss[loss=0.2102, simple_loss=0.2895, pruned_loss=0.06542, over 3823681.36 frames. ], batch size: 69, lr: 3.89e-03, grad_scale: 8.0 2023-04-02 20:07:37,090 INFO [train.py:928] (1/4) Computing validation loss 2023-04-02 20:07:50,388 INFO [train.py:937] (1/4) Epoch 21, validation: loss=0.1692, simple_loss=0.2693, pruned_loss=0.03459, over 944034.00 frames. 2023-04-02 20:07:50,389 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18821MB 2023-04-02 20:08:26,825 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=142591.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 20:08:52,457 INFO [train.py:903] (1/4) Epoch 21, batch 6050, loss[loss=0.2075, simple_loss=0.2922, pruned_loss=0.06138, over 19768.00 frames. ], tot_loss[loss=0.2099, simple_loss=0.2895, pruned_loss=0.06515, over 3813544.24 frames. ], batch size: 54, lr: 3.89e-03, grad_scale: 8.0 2023-04-02 20:09:18,890 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.159e+02 4.812e+02 5.741e+02 7.692e+02 1.541e+03, threshold=1.148e+03, percent-clipped=3.0 2023-04-02 20:09:41,427 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=142649.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 20:09:53,725 INFO [train.py:903] (1/4) Epoch 21, batch 6100, loss[loss=0.2416, simple_loss=0.3157, pruned_loss=0.08377, over 19685.00 frames. ], tot_loss[loss=0.2092, simple_loss=0.2888, pruned_loss=0.06479, over 3820085.87 frames. ], batch size: 59, lr: 3.89e-03, grad_scale: 8.0 2023-04-02 20:09:55,369 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=142661.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 20:10:12,477 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=142674.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 20:10:17,586 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 2023-04-02 20:10:28,113 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=142686.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 20:10:55,903 INFO [train.py:903] (1/4) Epoch 21, batch 6150, loss[loss=0.2064, simple_loss=0.2872, pruned_loss=0.06287, over 19129.00 frames. ], tot_loss[loss=0.2077, simple_loss=0.2874, pruned_loss=0.06404, over 3831278.83 frames. ], batch size: 69, lr: 3.88e-03, grad_scale: 8.0 2023-04-02 20:11:25,006 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.643e+02 5.511e+02 6.918e+02 9.659e+02 2.206e+03, threshold=1.384e+03, percent-clipped=13.0 2023-04-02 20:11:26,176 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9_rvb from training. Duration: 31.02225 2023-04-02 20:11:36,780 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=142742.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 20:11:59,805 INFO [train.py:903] (1/4) Epoch 21, batch 6200, loss[loss=0.2103, simple_loss=0.2932, pruned_loss=0.06365, over 19124.00 frames. ], tot_loss[loss=0.2075, simple_loss=0.2871, pruned_loss=0.06391, over 3837321.94 frames. ], batch size: 69, lr: 3.88e-03, grad_scale: 8.0 2023-04-02 20:12:15,936 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3951, 1.4885, 1.5567, 1.5523, 1.7362, 1.9116, 1.7175, 0.5218], device='cuda:1'), covar=tensor([0.2336, 0.3948, 0.2498, 0.1917, 0.1616, 0.2256, 0.1483, 0.4700], device='cuda:1'), in_proj_covar=tensor([0.0530, 0.0643, 0.0710, 0.0485, 0.0619, 0.0532, 0.0664, 0.0548], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 20:13:02,738 INFO [train.py:903] (1/4) Epoch 21, batch 6250, loss[loss=0.2177, simple_loss=0.3069, pruned_loss=0.06424, over 19606.00 frames. ], tot_loss[loss=0.2086, simple_loss=0.2879, pruned_loss=0.06466, over 3817592.14 frames. ], batch size: 61, lr: 3.88e-03, grad_scale: 8.0 2023-04-02 20:13:28,470 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.899e+02 5.305e+02 6.117e+02 7.859e+02 2.157e+03, threshold=1.223e+03, percent-clipped=2.0 2023-04-02 20:13:30,683 WARNING [train.py:1073] (1/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9_rvb from training. Duration: 25.988875 2023-04-02 20:13:46,816 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3586, 1.2002, 1.7067, 1.3339, 2.8002, 3.6437, 3.3491, 3.8309], device='cuda:1'), covar=tensor([0.1593, 0.4037, 0.3529, 0.2418, 0.0578, 0.0199, 0.0212, 0.0228], device='cuda:1'), in_proj_covar=tensor([0.0270, 0.0319, 0.0349, 0.0263, 0.0240, 0.0184, 0.0214, 0.0252], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-02 20:14:05,180 INFO [train.py:903] (1/4) Epoch 21, batch 6300, loss[loss=0.181, simple_loss=0.2707, pruned_loss=0.04566, over 19647.00 frames. ], tot_loss[loss=0.2072, simple_loss=0.287, pruned_loss=0.06373, over 3821780.91 frames. ], batch size: 53, lr: 3.88e-03, grad_scale: 8.0 2023-04-02 20:14:47,686 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9681, 1.8708, 1.7960, 1.5387, 1.4289, 1.5568, 0.4021, 0.8693], device='cuda:1'), covar=tensor([0.0641, 0.0638, 0.0416, 0.0729, 0.1221, 0.0883, 0.1262, 0.1082], device='cuda:1'), in_proj_covar=tensor([0.0356, 0.0355, 0.0354, 0.0382, 0.0459, 0.0386, 0.0335, 0.0341], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 20:15:07,133 INFO [train.py:903] (1/4) Epoch 21, batch 6350, loss[loss=0.2327, simple_loss=0.3277, pruned_loss=0.06889, over 19670.00 frames. ], tot_loss[loss=0.2081, simple_loss=0.2879, pruned_loss=0.0642, over 3815175.14 frames. ], batch size: 55, lr: 3.88e-03, grad_scale: 8.0 2023-04-02 20:15:36,285 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.232e+02 4.679e+02 5.550e+02 7.220e+02 1.923e+03, threshold=1.110e+03, percent-clipped=1.0 2023-04-02 20:15:39,976 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=142935.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 20:16:11,266 INFO [train.py:903] (1/4) Epoch 21, batch 6400, loss[loss=0.1833, simple_loss=0.2681, pruned_loss=0.04924, over 19683.00 frames. ], tot_loss[loss=0.2081, simple_loss=0.2878, pruned_loss=0.0642, over 3822043.89 frames. ], batch size: 53, lr: 3.88e-03, grad_scale: 8.0 2023-04-02 20:17:14,709 INFO [train.py:903] (1/4) Epoch 21, batch 6450, loss[loss=0.2735, simple_loss=0.332, pruned_loss=0.1075, over 12994.00 frames. ], tot_loss[loss=0.2085, simple_loss=0.2879, pruned_loss=0.06449, over 3819475.95 frames. ], batch size: 135, lr: 3.88e-03, grad_scale: 8.0 2023-04-02 20:17:40,509 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.231e+02 4.995e+02 6.270e+02 8.275e+02 2.312e+03, threshold=1.254e+03, percent-clipped=6.0 2023-04-02 20:18:01,471 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9_rvb from training. Duration: 28.638875 2023-04-02 20:18:04,947 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=143050.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 20:18:16,202 INFO [train.py:903] (1/4) Epoch 21, batch 6500, loss[loss=0.1782, simple_loss=0.2593, pruned_loss=0.04854, over 19733.00 frames. ], tot_loss[loss=0.209, simple_loss=0.2884, pruned_loss=0.06484, over 3821514.48 frames. ], batch size: 51, lr: 3.88e-03, grad_scale: 8.0 2023-04-02 20:18:23,429 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1_rvb from training. Duration: 0.836375 2023-04-02 20:18:24,240 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 2023-04-02 20:18:48,994 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=143086.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 20:19:16,589 INFO [train.py:903] (1/4) Epoch 21, batch 6550, loss[loss=0.1924, simple_loss=0.2762, pruned_loss=0.05432, over 19517.00 frames. ], tot_loss[loss=0.2101, simple_loss=0.2894, pruned_loss=0.06546, over 3810708.10 frames. ], batch size: 54, lr: 3.88e-03, grad_scale: 8.0 2023-04-02 20:19:44,582 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.797e+02 5.073e+02 6.169e+02 7.633e+02 2.146e+03, threshold=1.234e+03, percent-clipped=4.0 2023-04-02 20:20:01,369 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0425, 1.7324, 1.6592, 1.9865, 1.6765, 1.6688, 1.6098, 1.9196], device='cuda:1'), covar=tensor([0.0972, 0.1459, 0.1424, 0.1013, 0.1374, 0.0596, 0.1395, 0.0699], device='cuda:1'), in_proj_covar=tensor([0.0270, 0.0356, 0.0308, 0.0248, 0.0299, 0.0251, 0.0310, 0.0256], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 20:20:19,908 INFO [train.py:903] (1/4) Epoch 21, batch 6600, loss[loss=0.201, simple_loss=0.2684, pruned_loss=0.06687, over 19807.00 frames. ], tot_loss[loss=0.2101, simple_loss=0.2893, pruned_loss=0.0654, over 3805950.12 frames. ], batch size: 49, lr: 3.88e-03, grad_scale: 8.0 2023-04-02 20:21:10,262 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=143201.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 20:21:20,459 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4228, 2.0536, 1.6560, 1.4258, 1.8623, 1.3907, 1.3244, 1.8766], device='cuda:1'), covar=tensor([0.0903, 0.0845, 0.1047, 0.0846, 0.0557, 0.1305, 0.0701, 0.0440], device='cuda:1'), in_proj_covar=tensor([0.0302, 0.0313, 0.0336, 0.0261, 0.0246, 0.0335, 0.0290, 0.0274], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 20:21:22,428 INFO [train.py:903] (1/4) Epoch 21, batch 6650, loss[loss=0.2044, simple_loss=0.2849, pruned_loss=0.06197, over 19751.00 frames. ], tot_loss[loss=0.2101, simple_loss=0.2896, pruned_loss=0.06532, over 3813361.33 frames. ], batch size: 63, lr: 3.88e-03, grad_scale: 8.0 2023-04-02 20:21:47,870 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.522e+02 4.940e+02 5.672e+02 7.065e+02 1.538e+03, threshold=1.134e+03, percent-clipped=2.0 2023-04-02 20:22:23,645 INFO [train.py:903] (1/4) Epoch 21, batch 6700, loss[loss=0.2323, simple_loss=0.3031, pruned_loss=0.08074, over 13060.00 frames. ], tot_loss[loss=0.2106, simple_loss=0.2899, pruned_loss=0.0656, over 3807415.94 frames. ], batch size: 136, lr: 3.88e-03, grad_scale: 8.0 2023-04-02 20:22:52,037 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.81 vs. limit=2.0 2023-04-02 20:23:19,568 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=143306.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 20:23:23,654 INFO [train.py:903] (1/4) Epoch 21, batch 6750, loss[loss=0.1869, simple_loss=0.2712, pruned_loss=0.0513, over 19597.00 frames. ], tot_loss[loss=0.2097, simple_loss=0.2891, pruned_loss=0.06517, over 3803822.71 frames. ], batch size: 52, lr: 3.88e-03, grad_scale: 8.0 2023-04-02 20:23:48,060 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=143331.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 20:23:48,849 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.668e+02 4.995e+02 6.197e+02 7.772e+02 2.067e+03, threshold=1.239e+03, percent-clipped=6.0 2023-04-02 20:24:20,243 INFO [train.py:903] (1/4) Epoch 21, batch 6800, loss[loss=0.2061, simple_loss=0.2925, pruned_loss=0.05987, over 18241.00 frames. ], tot_loss[loss=0.2108, simple_loss=0.2898, pruned_loss=0.06584, over 3805950.98 frames. ], batch size: 84, lr: 3.88e-03, grad_scale: 8.0 2023-04-02 20:25:05,675 WARNING [train.py:1073] (1/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 2023-04-02 20:25:06,810 WARNING [train.py:1073] (1/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 2023-04-02 20:25:09,791 INFO [train.py:903] (1/4) Epoch 22, batch 0, loss[loss=0.2118, simple_loss=0.2951, pruned_loss=0.06429, over 19525.00 frames. ], tot_loss[loss=0.2118, simple_loss=0.2951, pruned_loss=0.06429, over 19525.00 frames. ], batch size: 56, lr: 3.78e-03, grad_scale: 8.0 2023-04-02 20:25:09,792 INFO [train.py:928] (1/4) Computing validation loss 2023-04-02 20:25:18,557 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1943, 1.1069, 1.1619, 1.4035, 1.0347, 1.2562, 1.3172, 1.2270], device='cuda:1'), covar=tensor([0.0653, 0.0771, 0.0813, 0.0515, 0.0744, 0.0684, 0.0713, 0.0577], device='cuda:1'), in_proj_covar=tensor([0.0213, 0.0223, 0.0228, 0.0240, 0.0227, 0.0213, 0.0187, 0.0204], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-04-02 20:25:20,458 INFO [train.py:937] (1/4) Epoch 22, validation: loss=0.1683, simple_loss=0.2691, pruned_loss=0.03373, over 944034.00 frames. 2023-04-02 20:25:20,459 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18821MB 2023-04-02 20:25:31,901 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 2023-04-02 20:25:55,286 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=143418.0, num_to_drop=1, layers_to_drop={0} 2023-04-02 20:26:14,247 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.150e+02 4.908e+02 5.891e+02 8.006e+02 1.582e+03, threshold=1.178e+03, percent-clipped=4.0 2023-04-02 20:26:21,016 INFO [train.py:903] (1/4) Epoch 22, batch 50, loss[loss=0.1966, simple_loss=0.2783, pruned_loss=0.05746, over 19040.00 frames. ], tot_loss[loss=0.2103, simple_loss=0.2895, pruned_loss=0.06554, over 870575.16 frames. ], batch size: 69, lr: 3.78e-03, grad_scale: 8.0 2023-04-02 20:26:24,449 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.6736, 2.5413, 2.3075, 2.6955, 2.5515, 2.3078, 2.1627, 2.5429], device='cuda:1'), covar=tensor([0.0921, 0.1536, 0.1422, 0.0990, 0.1213, 0.0492, 0.1324, 0.0650], device='cuda:1'), in_proj_covar=tensor([0.0267, 0.0353, 0.0308, 0.0246, 0.0296, 0.0248, 0.0307, 0.0254], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 20:26:42,107 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=143457.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 20:26:53,918 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1_rvb from training. Duration: 27.0318125 2023-04-02 20:27:13,735 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=143482.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 20:27:19,994 INFO [train.py:903] (1/4) Epoch 22, batch 100, loss[loss=0.2068, simple_loss=0.2837, pruned_loss=0.06495, over 19757.00 frames. ], tot_loss[loss=0.2132, simple_loss=0.2923, pruned_loss=0.06706, over 1532209.94 frames. ], batch size: 51, lr: 3.78e-03, grad_scale: 8.0 2023-04-02 20:27:23,806 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=143491.0, num_to_drop=1, layers_to_drop={1} 2023-04-02 20:27:31,528 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 2023-04-02 20:27:56,476 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.1550, 1.9969, 2.0701, 2.4968, 2.1473, 2.2906, 2.3535, 2.1385], device='cuda:1'), covar=tensor([0.0640, 0.0721, 0.0763, 0.0619, 0.0760, 0.0646, 0.0726, 0.0584], device='cuda:1'), in_proj_covar=tensor([0.0212, 0.0222, 0.0226, 0.0239, 0.0225, 0.0212, 0.0186, 0.0204], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-02 20:28:12,241 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.507e+02 5.227e+02 6.391e+02 8.671e+02 1.540e+03, threshold=1.278e+03, percent-clipped=3.0 2023-04-02 20:28:19,054 INFO [train.py:903] (1/4) Epoch 22, batch 150, loss[loss=0.2245, simple_loss=0.2927, pruned_loss=0.0781, over 19694.00 frames. ], tot_loss[loss=0.2123, simple_loss=0.2907, pruned_loss=0.06691, over 2041066.55 frames. ], batch size: 53, lr: 3.78e-03, grad_scale: 8.0 2023-04-02 20:28:31,077 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.2866, 3.0102, 2.1811, 2.7290, 0.8692, 2.9812, 2.8971, 2.8972], device='cuda:1'), covar=tensor([0.1097, 0.1477, 0.2032, 0.1087, 0.3676, 0.1039, 0.1193, 0.1481], device='cuda:1'), in_proj_covar=tensor([0.0499, 0.0405, 0.0490, 0.0343, 0.0401, 0.0427, 0.0421, 0.0457], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 20:29:18,917 INFO [train.py:903] (1/4) Epoch 22, batch 200, loss[loss=0.2455, simple_loss=0.3183, pruned_loss=0.08631, over 19684.00 frames. ], tot_loss[loss=0.2122, simple_loss=0.2907, pruned_loss=0.06681, over 2446015.12 frames. ], batch size: 60, lr: 3.78e-03, grad_scale: 8.0 2023-04-02 20:29:18,956 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 2023-04-02 20:30:12,489 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.534e+02 5.144e+02 6.083e+02 7.742e+02 1.350e+03, threshold=1.217e+03, percent-clipped=1.0 2023-04-02 20:30:20,926 INFO [train.py:903] (1/4) Epoch 22, batch 250, loss[loss=0.2417, simple_loss=0.3137, pruned_loss=0.08482, over 19514.00 frames. ], tot_loss[loss=0.2115, simple_loss=0.2907, pruned_loss=0.06616, over 2747651.77 frames. ], batch size: 54, lr: 3.78e-03, grad_scale: 8.0 2023-04-02 20:31:20,917 INFO [train.py:903] (1/4) Epoch 22, batch 300, loss[loss=0.2191, simple_loss=0.2944, pruned_loss=0.07187, over 19541.00 frames. ], tot_loss[loss=0.2114, simple_loss=0.2907, pruned_loss=0.06611, over 2987866.61 frames. ], batch size: 56, lr: 3.78e-03, grad_scale: 8.0 2023-04-02 20:31:56,662 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 2023-04-02 20:32:15,075 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.028e+02 5.065e+02 6.247e+02 8.237e+02 1.383e+03, threshold=1.249e+03, percent-clipped=3.0 2023-04-02 20:32:22,211 INFO [train.py:903] (1/4) Epoch 22, batch 350, loss[loss=0.221, simple_loss=0.2995, pruned_loss=0.07123, over 19458.00 frames. ], tot_loss[loss=0.2095, simple_loss=0.2892, pruned_loss=0.06495, over 3179244.53 frames. ], batch size: 64, lr: 3.78e-03, grad_scale: 8.0 2023-04-02 20:32:29,138 WARNING [train.py:1073] (1/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1_rvb from training. Duration: 0.9681875 2023-04-02 20:32:51,098 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=143762.0, num_to_drop=1, layers_to_drop={1} 2023-04-02 20:32:52,791 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.75 vs. limit=2.0 2023-04-02 20:33:20,944 INFO [train.py:903] (1/4) Epoch 22, batch 400, loss[loss=0.2008, simple_loss=0.271, pruned_loss=0.06534, over 19742.00 frames. ], tot_loss[loss=0.2075, simple_loss=0.287, pruned_loss=0.06399, over 3339227.24 frames. ], batch size: 45, lr: 3.78e-03, grad_scale: 8.0 2023-04-02 20:34:15,303 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.174e+02 5.215e+02 6.557e+02 8.093e+02 2.351e+03, threshold=1.311e+03, percent-clipped=8.0 2023-04-02 20:34:17,781 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=143835.0, num_to_drop=1, layers_to_drop={0} 2023-04-02 20:34:20,915 INFO [train.py:903] (1/4) Epoch 22, batch 450, loss[loss=0.2014, simple_loss=0.2821, pruned_loss=0.06036, over 19541.00 frames. ], tot_loss[loss=0.2073, simple_loss=0.287, pruned_loss=0.06377, over 3453743.16 frames. ], batch size: 56, lr: 3.78e-03, grad_scale: 8.0 2023-04-02 20:34:57,885 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 2023-04-02 20:34:58,983 WARNING [train.py:1073] (1/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 2023-04-02 20:35:08,538 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=143877.0, num_to_drop=1, layers_to_drop={1} 2023-04-02 20:35:22,928 INFO [train.py:903] (1/4) Epoch 22, batch 500, loss[loss=0.1605, simple_loss=0.2394, pruned_loss=0.04077, over 19769.00 frames. ], tot_loss[loss=0.2084, simple_loss=0.2881, pruned_loss=0.06434, over 3531591.17 frames. ], batch size: 47, lr: 3.78e-03, grad_scale: 8.0 2023-04-02 20:36:17,498 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.793e+02 5.123e+02 6.359e+02 8.434e+02 1.804e+03, threshold=1.272e+03, percent-clipped=4.0 2023-04-02 20:36:23,280 INFO [train.py:903] (1/4) Epoch 22, batch 550, loss[loss=0.2165, simple_loss=0.3012, pruned_loss=0.06587, over 18824.00 frames. ], tot_loss[loss=0.2086, simple_loss=0.2883, pruned_loss=0.06441, over 3588101.12 frames. ], batch size: 74, lr: 3.78e-03, grad_scale: 8.0 2023-04-02 20:36:37,254 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=143950.0, num_to_drop=1, layers_to_drop={0} 2023-04-02 20:37:00,091 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4194, 1.3258, 1.3851, 1.7266, 1.3321, 1.6567, 1.7357, 1.5232], device='cuda:1'), covar=tensor([0.0948, 0.1061, 0.1075, 0.0764, 0.0890, 0.0823, 0.0820, 0.0741], device='cuda:1'), in_proj_covar=tensor([0.0214, 0.0223, 0.0227, 0.0240, 0.0226, 0.0213, 0.0187, 0.0204], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-04-02 20:37:23,312 INFO [train.py:903] (1/4) Epoch 22, batch 600, loss[loss=0.2018, simple_loss=0.2924, pruned_loss=0.05564, over 19686.00 frames. ], tot_loss[loss=0.2089, simple_loss=0.2886, pruned_loss=0.06464, over 3636692.66 frames. ], batch size: 59, lr: 3.78e-03, grad_scale: 8.0 2023-04-02 20:38:02,269 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.5989, 4.1938, 2.5957, 3.6252, 0.8824, 4.1446, 3.9366, 4.1314], device='cuda:1'), covar=tensor([0.0646, 0.0916, 0.2029, 0.0919, 0.4077, 0.0699, 0.1046, 0.1146], device='cuda:1'), in_proj_covar=tensor([0.0500, 0.0408, 0.0492, 0.0345, 0.0402, 0.0430, 0.0426, 0.0459], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 20:38:06,644 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 2023-04-02 20:38:17,797 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.931e+02 4.914e+02 6.190e+02 8.004e+02 1.732e+03, threshold=1.238e+03, percent-clipped=3.0 2023-04-02 20:38:23,572 INFO [train.py:903] (1/4) Epoch 22, batch 650, loss[loss=0.2255, simple_loss=0.3024, pruned_loss=0.07427, over 19575.00 frames. ], tot_loss[loss=0.2082, simple_loss=0.2876, pruned_loss=0.06436, over 3686918.02 frames. ], batch size: 61, lr: 3.78e-03, grad_scale: 8.0 2023-04-02 20:39:26,361 INFO [train.py:903] (1/4) Epoch 22, batch 700, loss[loss=0.2328, simple_loss=0.3124, pruned_loss=0.07664, over 18663.00 frames. ], tot_loss[loss=0.2081, simple_loss=0.288, pruned_loss=0.06412, over 3722385.86 frames. ], batch size: 74, lr: 3.78e-03, grad_scale: 8.0 2023-04-02 20:40:19,659 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.965e+02 4.796e+02 6.107e+02 7.975e+02 1.533e+03, threshold=1.221e+03, percent-clipped=5.0 2023-04-02 20:40:20,084 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=144133.0, num_to_drop=1, layers_to_drop={0} 2023-04-02 20:40:21,171 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([0.9805, 1.2274, 1.5963, 0.6737, 2.0337, 2.4731, 2.1460, 2.6339], device='cuda:1'), covar=tensor([0.1516, 0.3796, 0.3229, 0.2640, 0.0619, 0.0266, 0.0345, 0.0364], device='cuda:1'), in_proj_covar=tensor([0.0272, 0.0321, 0.0350, 0.0265, 0.0243, 0.0185, 0.0215, 0.0254], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-02 20:40:26,360 INFO [train.py:903] (1/4) Epoch 22, batch 750, loss[loss=0.1942, simple_loss=0.2845, pruned_loss=0.05193, over 19698.00 frames. ], tot_loss[loss=0.2092, simple_loss=0.2886, pruned_loss=0.06491, over 3741866.27 frames. ], batch size: 59, lr: 3.77e-03, grad_scale: 8.0 2023-04-02 20:40:49,192 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=144158.0, num_to_drop=1, layers_to_drop={0} 2023-04-02 20:41:21,372 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.08 vs. limit=2.0 2023-04-02 20:41:26,351 INFO [train.py:903] (1/4) Epoch 22, batch 800, loss[loss=0.1806, simple_loss=0.2564, pruned_loss=0.05235, over 19776.00 frames. ], tot_loss[loss=0.209, simple_loss=0.2887, pruned_loss=0.06461, over 3761239.47 frames. ], batch size: 47, lr: 3.77e-03, grad_scale: 8.0 2023-04-02 20:41:44,760 WARNING [train.py:1073] (1/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 2023-04-02 20:41:48,043 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=144206.0, num_to_drop=1, layers_to_drop={1} 2023-04-02 20:42:19,071 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=144231.0, num_to_drop=1, layers_to_drop={0} 2023-04-02 20:42:20,961 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.163e+02 5.021e+02 6.351e+02 8.019e+02 1.751e+03, threshold=1.270e+03, percent-clipped=5.0 2023-04-02 20:42:26,677 INFO [train.py:903] (1/4) Epoch 22, batch 850, loss[loss=0.2356, simple_loss=0.3154, pruned_loss=0.07784, over 19301.00 frames. ], tot_loss[loss=0.2102, simple_loss=0.2897, pruned_loss=0.06532, over 3762143.64 frames. ], batch size: 70, lr: 3.77e-03, grad_scale: 8.0 2023-04-02 20:43:19,877 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9_rvb from training. Duration: 25.061125 2023-04-02 20:43:26,415 INFO [train.py:903] (1/4) Epoch 22, batch 900, loss[loss=0.2188, simple_loss=0.3031, pruned_loss=0.06727, over 19773.00 frames. ], tot_loss[loss=0.2105, simple_loss=0.2901, pruned_loss=0.06543, over 3777699.30 frames. ], batch size: 54, lr: 3.77e-03, grad_scale: 8.0 2023-04-02 20:43:42,716 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.9251, 4.4520, 2.8907, 3.9600, 0.8052, 4.4832, 4.3386, 4.4935], device='cuda:1'), covar=tensor([0.0536, 0.0949, 0.1830, 0.0830, 0.4409, 0.0598, 0.0766, 0.1114], device='cuda:1'), in_proj_covar=tensor([0.0497, 0.0408, 0.0488, 0.0343, 0.0400, 0.0426, 0.0421, 0.0457], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 20:44:05,049 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.4700, 2.3030, 2.0823, 2.5879, 2.3211, 2.0526, 2.0883, 2.3170], device='cuda:1'), covar=tensor([0.0990, 0.1616, 0.1521, 0.1112, 0.1390, 0.0549, 0.1355, 0.0756], device='cuda:1'), in_proj_covar=tensor([0.0269, 0.0357, 0.0312, 0.0250, 0.0300, 0.0249, 0.0308, 0.0256], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 20:44:21,526 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.404e+02 5.111e+02 6.361e+02 7.451e+02 1.172e+03, threshold=1.272e+03, percent-clipped=0.0 2023-04-02 20:44:26,104 INFO [train.py:903] (1/4) Epoch 22, batch 950, loss[loss=0.1974, simple_loss=0.2866, pruned_loss=0.05411, over 19577.00 frames. ], tot_loss[loss=0.2096, simple_loss=0.2892, pruned_loss=0.06506, over 3789932.21 frames. ], batch size: 61, lr: 3.77e-03, grad_scale: 4.0 2023-04-02 20:44:30,635 WARNING [train.py:1073] (1/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9_rvb from training. Duration: 26.32775 2023-04-02 20:45:27,352 INFO [train.py:903] (1/4) Epoch 22, batch 1000, loss[loss=0.1526, simple_loss=0.2363, pruned_loss=0.03449, over 19756.00 frames. ], tot_loss[loss=0.2095, simple_loss=0.2891, pruned_loss=0.06494, over 3800058.93 frames. ], batch size: 48, lr: 3.77e-03, grad_scale: 4.0 2023-04-02 20:46:17,113 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=144429.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 20:46:17,963 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 2023-04-02 20:46:22,212 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.871e+02 5.215e+02 6.539e+02 8.059e+02 1.779e+03, threshold=1.308e+03, percent-clipped=4.0 2023-04-02 20:46:26,890 INFO [train.py:903] (1/4) Epoch 22, batch 1050, loss[loss=0.177, simple_loss=0.2611, pruned_loss=0.04643, over 19619.00 frames. ], tot_loss[loss=0.2085, simple_loss=0.2883, pruned_loss=0.06432, over 3820814.09 frames. ], batch size: 50, lr: 3.77e-03, grad_scale: 4.0 2023-04-02 20:46:53,402 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([5.2207, 5.1682, 5.9906, 6.0056, 1.9882, 5.5613, 4.7968, 5.6339], device='cuda:1'), covar=tensor([0.1661, 0.0761, 0.0535, 0.0607, 0.6204, 0.0778, 0.0616, 0.1091], device='cuda:1'), in_proj_covar=tensor([0.0771, 0.0731, 0.0933, 0.0821, 0.0823, 0.0695, 0.0564, 0.0865], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-02 20:47:00,693 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 2023-04-02 20:47:07,858 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.6729, 4.2879, 2.6408, 3.7846, 1.0396, 4.2034, 4.1284, 4.1947], device='cuda:1'), covar=tensor([0.0578, 0.0904, 0.1980, 0.0822, 0.3959, 0.0639, 0.0868, 0.0961], device='cuda:1'), in_proj_covar=tensor([0.0496, 0.0407, 0.0486, 0.0342, 0.0400, 0.0427, 0.0421, 0.0457], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 20:47:26,635 INFO [train.py:903] (1/4) Epoch 22, batch 1100, loss[loss=0.2043, simple_loss=0.2875, pruned_loss=0.06056, over 19674.00 frames. ], tot_loss[loss=0.2083, simple_loss=0.2878, pruned_loss=0.06435, over 3825026.54 frames. ], batch size: 53, lr: 3.77e-03, grad_scale: 4.0 2023-04-02 20:48:21,825 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.380e+02 5.103e+02 6.169e+02 7.547e+02 2.403e+03, threshold=1.234e+03, percent-clipped=2.0 2023-04-02 20:48:27,954 INFO [train.py:903] (1/4) Epoch 22, batch 1150, loss[loss=0.2095, simple_loss=0.2926, pruned_loss=0.06327, over 19653.00 frames. ], tot_loss[loss=0.2097, simple_loss=0.2896, pruned_loss=0.0649, over 3825877.42 frames. ], batch size: 58, lr: 3.77e-03, grad_scale: 4.0 2023-04-02 20:49:28,339 INFO [train.py:903] (1/4) Epoch 22, batch 1200, loss[loss=0.1893, simple_loss=0.2644, pruned_loss=0.05709, over 19471.00 frames. ], tot_loss[loss=0.2084, simple_loss=0.2883, pruned_loss=0.06429, over 3837215.38 frames. ], batch size: 49, lr: 3.77e-03, grad_scale: 8.0 2023-04-02 20:49:59,948 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 2023-04-02 20:50:23,754 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.253e+02 4.877e+02 6.112e+02 7.869e+02 2.071e+03, threshold=1.222e+03, percent-clipped=4.0 2023-04-02 20:50:27,125 INFO [train.py:903] (1/4) Epoch 22, batch 1250, loss[loss=0.2197, simple_loss=0.2928, pruned_loss=0.07328, over 19844.00 frames. ], tot_loss[loss=0.2087, simple_loss=0.2886, pruned_loss=0.06445, over 3824841.49 frames. ], batch size: 52, lr: 3.77e-03, grad_scale: 4.0 2023-04-02 20:50:33,308 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.4403, 2.2844, 2.2132, 2.5344, 2.3233, 2.0433, 2.0947, 2.3597], device='cuda:1'), covar=tensor([0.0745, 0.1230, 0.1065, 0.0765, 0.1004, 0.0458, 0.1125, 0.0568], device='cuda:1'), in_proj_covar=tensor([0.0270, 0.0357, 0.0312, 0.0249, 0.0299, 0.0250, 0.0308, 0.0255], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 20:51:26,175 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4724, 1.5930, 2.0357, 1.8476, 3.1381, 4.0712, 3.9788, 4.4361], device='cuda:1'), covar=tensor([0.1594, 0.3622, 0.3115, 0.2175, 0.0679, 0.0263, 0.0181, 0.0248], device='cuda:1'), in_proj_covar=tensor([0.0270, 0.0320, 0.0349, 0.0263, 0.0241, 0.0184, 0.0214, 0.0253], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-02 20:51:28,163 INFO [train.py:903] (1/4) Epoch 22, batch 1300, loss[loss=0.205, simple_loss=0.2916, pruned_loss=0.05922, over 19640.00 frames. ], tot_loss[loss=0.2087, simple_loss=0.2887, pruned_loss=0.06441, over 3818831.04 frames. ], batch size: 57, lr: 3.77e-03, grad_scale: 4.0 2023-04-02 20:51:33,959 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.7550, 1.2329, 1.4491, 1.5773, 3.3222, 1.0093, 2.2389, 3.7586], device='cuda:1'), covar=tensor([0.0483, 0.2842, 0.2927, 0.1763, 0.0730, 0.2605, 0.1428, 0.0236], device='cuda:1'), in_proj_covar=tensor([0.0404, 0.0361, 0.0381, 0.0343, 0.0370, 0.0347, 0.0374, 0.0398], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 20:51:46,031 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.39 vs. limit=5.0 2023-04-02 20:51:59,483 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0 2023-04-02 20:52:26,777 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.265e+02 4.679e+02 5.951e+02 8.140e+02 2.957e+03, threshold=1.190e+03, percent-clipped=7.0 2023-04-02 20:52:30,268 INFO [train.py:903] (1/4) Epoch 22, batch 1350, loss[loss=0.2281, simple_loss=0.3063, pruned_loss=0.07497, over 13646.00 frames. ], tot_loss[loss=0.2093, simple_loss=0.2891, pruned_loss=0.06469, over 3816948.96 frames. ], batch size: 138, lr: 3.77e-03, grad_scale: 4.0 2023-04-02 20:52:39,356 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6360, 1.6965, 1.5689, 1.3250, 1.3287, 1.3828, 0.3052, 0.6802], device='cuda:1'), covar=tensor([0.0679, 0.0610, 0.0419, 0.0637, 0.1233, 0.0751, 0.1323, 0.1071], device='cuda:1'), in_proj_covar=tensor([0.0354, 0.0355, 0.0357, 0.0381, 0.0457, 0.0385, 0.0334, 0.0339], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 20:53:12,557 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=144773.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 20:53:31,338 INFO [train.py:903] (1/4) Epoch 22, batch 1400, loss[loss=0.2108, simple_loss=0.2858, pruned_loss=0.06785, over 19583.00 frames. ], tot_loss[loss=0.2087, simple_loss=0.2883, pruned_loss=0.06457, over 3820599.14 frames. ], batch size: 52, lr: 3.77e-03, grad_scale: 4.0 2023-04-02 20:53:43,605 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=4.41 vs. limit=5.0 2023-04-02 20:54:08,463 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.9868, 4.5150, 2.7722, 3.9478, 1.2871, 4.4573, 4.3176, 4.4969], device='cuda:1'), covar=tensor([0.0495, 0.0921, 0.1943, 0.0832, 0.3622, 0.0651, 0.0885, 0.0925], device='cuda:1'), in_proj_covar=tensor([0.0498, 0.0411, 0.0494, 0.0344, 0.0404, 0.0431, 0.0424, 0.0460], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 20:54:28,426 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.383e+02 4.808e+02 5.945e+02 7.380e+02 1.517e+03, threshold=1.189e+03, percent-clipped=2.0 2023-04-02 20:54:29,516 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_rvb from training. Duration: 25.85 2023-04-02 20:54:31,678 INFO [train.py:903] (1/4) Epoch 22, batch 1450, loss[loss=0.2607, simple_loss=0.3257, pruned_loss=0.09787, over 19518.00 frames. ], tot_loss[loss=0.2094, simple_loss=0.2888, pruned_loss=0.06496, over 3809331.47 frames. ], batch size: 54, lr: 3.77e-03, grad_scale: 4.0 2023-04-02 20:55:30,804 INFO [train.py:903] (1/4) Epoch 22, batch 1500, loss[loss=0.1986, simple_loss=0.2793, pruned_loss=0.05901, over 19603.00 frames. ], tot_loss[loss=0.2092, simple_loss=0.2885, pruned_loss=0.06496, over 3818399.87 frames. ], batch size: 52, lr: 3.76e-03, grad_scale: 4.0 2023-04-02 20:55:31,148 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=144888.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 20:55:33,125 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1352, 1.3280, 1.6728, 0.9503, 2.3594, 3.0518, 2.7529, 3.2032], device='cuda:1'), covar=tensor([0.1632, 0.3665, 0.3287, 0.2639, 0.0593, 0.0206, 0.0264, 0.0320], device='cuda:1'), in_proj_covar=tensor([0.0271, 0.0320, 0.0350, 0.0265, 0.0242, 0.0185, 0.0215, 0.0254], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-02 20:55:33,218 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3620, 1.3940, 1.6141, 1.5817, 2.2769, 1.9225, 2.3039, 0.8426], device='cuda:1'), covar=tensor([0.2679, 0.4535, 0.2934, 0.2100, 0.1603, 0.2432, 0.1531, 0.4834], device='cuda:1'), in_proj_covar=tensor([0.0537, 0.0649, 0.0718, 0.0488, 0.0625, 0.0533, 0.0666, 0.0554], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 20:56:27,870 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.316e+02 4.957e+02 5.987e+02 8.036e+02 1.770e+03, threshold=1.197e+03, percent-clipped=5.0 2023-04-02 20:56:31,410 INFO [train.py:903] (1/4) Epoch 22, batch 1550, loss[loss=0.1772, simple_loss=0.2616, pruned_loss=0.04637, over 19425.00 frames. ], tot_loss[loss=0.2077, simple_loss=0.2867, pruned_loss=0.0643, over 3819379.91 frames. ], batch size: 48, lr: 3.76e-03, grad_scale: 4.0 2023-04-02 20:57:30,408 INFO [train.py:903] (1/4) Epoch 22, batch 1600, loss[loss=0.2514, simple_loss=0.3217, pruned_loss=0.09048, over 18743.00 frames. ], tot_loss[loss=0.2071, simple_loss=0.2863, pruned_loss=0.06393, over 3826642.48 frames. ], batch size: 74, lr: 3.76e-03, grad_scale: 8.0 2023-04-02 20:57:50,817 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9_rvb from training. Duration: 30.1555625 2023-04-02 20:58:02,387 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=145014.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 20:58:27,680 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.160e+02 4.884e+02 5.870e+02 7.908e+02 1.403e+03, threshold=1.174e+03, percent-clipped=3.0 2023-04-02 20:58:31,152 INFO [train.py:903] (1/4) Epoch 22, batch 1650, loss[loss=0.1978, simple_loss=0.2769, pruned_loss=0.05931, over 19379.00 frames. ], tot_loss[loss=0.2072, simple_loss=0.2864, pruned_loss=0.06395, over 3831597.56 frames. ], batch size: 48, lr: 3.76e-03, grad_scale: 8.0 2023-04-02 20:58:39,542 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=145045.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 20:59:26,986 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=145084.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 20:59:31,372 INFO [train.py:903] (1/4) Epoch 22, batch 1700, loss[loss=0.2146, simple_loss=0.3013, pruned_loss=0.06392, over 19787.00 frames. ], tot_loss[loss=0.2069, simple_loss=0.2861, pruned_loss=0.06388, over 3834290.20 frames. ], batch size: 56, lr: 3.76e-03, grad_scale: 8.0 2023-04-02 20:59:52,393 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.86 vs. limit=5.0 2023-04-02 21:00:08,584 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 2023-04-02 21:00:27,996 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.426e+02 4.845e+02 6.228e+02 7.810e+02 2.223e+03, threshold=1.246e+03, percent-clipped=2.0 2023-04-02 21:00:33,049 INFO [train.py:903] (1/4) Epoch 22, batch 1750, loss[loss=0.2358, simple_loss=0.304, pruned_loss=0.08382, over 19739.00 frames. ], tot_loss[loss=0.2063, simple_loss=0.2857, pruned_loss=0.06349, over 3837104.23 frames. ], batch size: 51, lr: 3.76e-03, grad_scale: 8.0 2023-04-02 21:00:40,273 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=145144.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 21:01:09,176 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=145169.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 21:01:31,689 INFO [train.py:903] (1/4) Epoch 22, batch 1800, loss[loss=0.217, simple_loss=0.2914, pruned_loss=0.0713, over 19469.00 frames. ], tot_loss[loss=0.2062, simple_loss=0.2859, pruned_loss=0.06323, over 3848924.60 frames. ], batch size: 49, lr: 3.76e-03, grad_scale: 8.0 2023-04-02 21:01:55,259 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.0220, 1.2111, 1.5531, 0.6281, 2.0342, 2.4376, 2.1440, 2.6342], device='cuda:1'), covar=tensor([0.1576, 0.3911, 0.3482, 0.2799, 0.0623, 0.0300, 0.0355, 0.0362], device='cuda:1'), in_proj_covar=tensor([0.0272, 0.0321, 0.0352, 0.0265, 0.0243, 0.0186, 0.0216, 0.0255], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-02 21:02:27,946 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.845e+02 5.086e+02 5.993e+02 7.804e+02 1.410e+03, threshold=1.199e+03, percent-clipped=2.0 2023-04-02 21:02:27,979 WARNING [train.py:1073] (1/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9_rvb from training. Duration: 25.0944375 2023-04-02 21:02:31,535 INFO [train.py:903] (1/4) Epoch 22, batch 1850, loss[loss=0.201, simple_loss=0.2765, pruned_loss=0.0627, over 19412.00 frames. ], tot_loss[loss=0.207, simple_loss=0.2867, pruned_loss=0.06368, over 3836767.57 frames. ], batch size: 48, lr: 3.76e-03, grad_scale: 8.0 2023-04-02 21:03:04,110 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9_rvb from training. Duration: 27.8166875 2023-04-02 21:03:30,820 INFO [train.py:903] (1/4) Epoch 22, batch 1900, loss[loss=0.1901, simple_loss=0.2822, pruned_loss=0.04898, over 19783.00 frames. ], tot_loss[loss=0.2071, simple_loss=0.2867, pruned_loss=0.0637, over 3834075.88 frames. ], batch size: 54, lr: 3.76e-03, grad_scale: 8.0 2023-04-02 21:03:45,039 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.9405, 4.2687, 4.6805, 4.6681, 2.0814, 4.3290, 3.7562, 4.3859], device='cuda:1'), covar=tensor([0.1533, 0.1281, 0.0572, 0.0645, 0.5598, 0.1050, 0.0685, 0.1035], device='cuda:1'), in_proj_covar=tensor([0.0775, 0.0737, 0.0934, 0.0825, 0.0825, 0.0698, 0.0561, 0.0872], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-02 21:03:48,272 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 2023-04-02 21:03:52,776 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9_rvb from training. Duration: 27.02225 2023-04-02 21:04:15,286 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 2023-04-02 21:04:26,521 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.830e+02 5.286e+02 6.049e+02 6.874e+02 1.450e+03, threshold=1.210e+03, percent-clipped=2.0 2023-04-02 21:04:30,790 INFO [train.py:903] (1/4) Epoch 22, batch 1950, loss[loss=0.2131, simple_loss=0.2911, pruned_loss=0.06753, over 19563.00 frames. ], tot_loss[loss=0.2067, simple_loss=0.2863, pruned_loss=0.06354, over 3818653.83 frames. ], batch size: 61, lr: 3.76e-03, grad_scale: 8.0 2023-04-02 21:04:44,259 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0786, 1.9927, 1.7405, 2.0849, 1.9369, 1.7852, 1.7033, 2.0209], device='cuda:1'), covar=tensor([0.0990, 0.1324, 0.1377, 0.0989, 0.1203, 0.0553, 0.1346, 0.0687], device='cuda:1'), in_proj_covar=tensor([0.0270, 0.0356, 0.0312, 0.0249, 0.0300, 0.0250, 0.0309, 0.0256], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 21:04:55,831 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=145358.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 21:05:31,552 INFO [train.py:903] (1/4) Epoch 22, batch 2000, loss[loss=0.1993, simple_loss=0.2842, pruned_loss=0.05718, over 19681.00 frames. ], tot_loss[loss=0.2062, simple_loss=0.2859, pruned_loss=0.0633, over 3822688.90 frames. ], batch size: 53, lr: 3.76e-03, grad_scale: 8.0 2023-04-02 21:05:32,826 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=145389.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 21:05:32,963 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=145389.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 21:06:19,675 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=145428.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 21:06:27,596 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.585e+02 4.801e+02 6.060e+02 7.916e+02 1.266e+03, threshold=1.212e+03, percent-clipped=1.0 2023-04-02 21:06:27,634 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_rvb from training. Duration: 26.8349375 2023-04-02 21:06:30,165 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1724, 1.3218, 1.7843, 1.4678, 2.7542, 3.8191, 3.5754, 4.0753], device='cuda:1'), covar=tensor([0.1697, 0.3748, 0.3229, 0.2232, 0.0570, 0.0180, 0.0195, 0.0225], device='cuda:1'), in_proj_covar=tensor([0.0272, 0.0321, 0.0352, 0.0265, 0.0243, 0.0186, 0.0216, 0.0256], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-02 21:06:30,896 INFO [train.py:903] (1/4) Epoch 22, batch 2050, loss[loss=0.2086, simple_loss=0.2886, pruned_loss=0.06432, over 19379.00 frames. ], tot_loss[loss=0.2059, simple_loss=0.2859, pruned_loss=0.06291, over 3831988.49 frames. ], batch size: 47, lr: 3.76e-03, grad_scale: 8.0 2023-04-02 21:06:46,543 WARNING [train.py:1073] (1/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 2023-04-02 21:06:46,573 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 2023-04-02 21:07:06,389 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 2023-04-02 21:07:13,295 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=145473.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 21:07:30,911 INFO [train.py:903] (1/4) Epoch 22, batch 2100, loss[loss=0.2418, simple_loss=0.3175, pruned_loss=0.08302, over 19624.00 frames. ], tot_loss[loss=0.2057, simple_loss=0.2856, pruned_loss=0.06294, over 3831770.10 frames. ], batch size: 61, lr: 3.76e-03, grad_scale: 8.0 2023-04-02 21:07:51,393 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=145504.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 21:08:01,435 WARNING [train.py:1073] (1/4) Exclude cut with ID 453-131332-0000-131866_sp0.9_rvb from training. Duration: 25.3333125 2023-04-02 21:08:22,535 WARNING [train.py:1073] (1/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9_rvb from training. Duration: 26.6166875 2023-04-02 21:08:27,086 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.088e+02 4.926e+02 6.113e+02 7.931e+02 1.598e+03, threshold=1.223e+03, percent-clipped=5.0 2023-04-02 21:08:30,642 INFO [train.py:903] (1/4) Epoch 22, batch 2150, loss[loss=0.1966, simple_loss=0.279, pruned_loss=0.05711, over 19784.00 frames. ], tot_loss[loss=0.2066, simple_loss=0.2863, pruned_loss=0.06344, over 3843045.37 frames. ], batch size: 54, lr: 3.76e-03, grad_scale: 8.0 2023-04-02 21:08:38,969 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=145543.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 21:09:32,294 INFO [train.py:903] (1/4) Epoch 22, batch 2200, loss[loss=0.2287, simple_loss=0.3036, pruned_loss=0.07691, over 18798.00 frames. ], tot_loss[loss=0.206, simple_loss=0.2856, pruned_loss=0.06324, over 3840408.08 frames. ], batch size: 74, lr: 3.76e-03, grad_scale: 4.0 2023-04-02 21:10:00,388 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=145612.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 21:10:08,469 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4264, 1.3502, 1.3402, 1.7739, 1.4596, 1.6293, 1.7410, 1.5118], device='cuda:1'), covar=tensor([0.0912, 0.0944, 0.1083, 0.0701, 0.0794, 0.0808, 0.0825, 0.0733], device='cuda:1'), in_proj_covar=tensor([0.0213, 0.0221, 0.0226, 0.0240, 0.0225, 0.0212, 0.0186, 0.0203], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-02 21:10:29,863 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.244e+02 5.056e+02 6.179e+02 8.064e+02 2.249e+03, threshold=1.236e+03, percent-clipped=3.0 2023-04-02 21:10:32,074 INFO [train.py:903] (1/4) Epoch 22, batch 2250, loss[loss=0.1977, simple_loss=0.2828, pruned_loss=0.05628, over 19769.00 frames. ], tot_loss[loss=0.2063, simple_loss=0.2857, pruned_loss=0.06341, over 3840344.42 frames. ], batch size: 54, lr: 3.76e-03, grad_scale: 4.0 2023-04-02 21:11:01,366 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=145661.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 21:11:33,804 INFO [train.py:903] (1/4) Epoch 22, batch 2300, loss[loss=0.2171, simple_loss=0.2984, pruned_loss=0.06785, over 19733.00 frames. ], tot_loss[loss=0.2066, simple_loss=0.2861, pruned_loss=0.06353, over 3832875.71 frames. ], batch size: 63, lr: 3.75e-03, grad_scale: 4.0 2023-04-02 21:11:45,977 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_rvb from training. Duration: 26.205 2023-04-02 21:12:22,713 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=145729.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 21:12:27,054 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=145733.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 21:12:30,301 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.518e+02 4.899e+02 6.109e+02 7.402e+02 2.135e+03, threshold=1.222e+03, percent-clipped=2.0 2023-04-02 21:12:32,755 INFO [train.py:903] (1/4) Epoch 22, batch 2350, loss[loss=0.1651, simple_loss=0.2474, pruned_loss=0.04141, over 15512.00 frames. ], tot_loss[loss=0.2071, simple_loss=0.2869, pruned_loss=0.06371, over 3829834.64 frames. ], batch size: 34, lr: 3.75e-03, grad_scale: 4.0 2023-04-02 21:12:54,262 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=145754.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 21:13:01,087 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=145760.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 21:13:14,123 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_rvb from training. Duration: 25.775 2023-04-02 21:13:31,586 WARNING [train.py:1073] (1/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9_rvb from training. Duration: 25.45 2023-04-02 21:13:31,995 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=145785.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 21:13:35,682 INFO [train.py:903] (1/4) Epoch 22, batch 2400, loss[loss=0.3188, simple_loss=0.3831, pruned_loss=0.1272, over 19533.00 frames. ], tot_loss[loss=0.2076, simple_loss=0.2873, pruned_loss=0.0639, over 3842519.52 frames. ], batch size: 54, lr: 3.75e-03, grad_scale: 8.0 2023-04-02 21:13:45,322 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2696, 1.2598, 1.2813, 1.3924, 1.0890, 1.3469, 1.3958, 1.3521], device='cuda:1'), covar=tensor([0.0955, 0.0993, 0.1088, 0.0642, 0.0834, 0.0880, 0.0850, 0.0766], device='cuda:1'), in_proj_covar=tensor([0.0212, 0.0220, 0.0225, 0.0239, 0.0225, 0.0211, 0.0185, 0.0202], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-02 21:13:48,841 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=145799.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 21:14:20,063 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=145824.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 21:14:34,015 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.210e+02 4.826e+02 5.747e+02 7.009e+02 1.532e+03, threshold=1.149e+03, percent-clipped=5.0 2023-04-02 21:14:35,788 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4109, 1.4926, 1.6568, 1.7372, 1.3206, 1.6648, 1.6677, 1.5846], device='cuda:1'), covar=tensor([0.3277, 0.2931, 0.1676, 0.1878, 0.3077, 0.1742, 0.4115, 0.2714], device='cuda:1'), in_proj_covar=tensor([0.0893, 0.0957, 0.0714, 0.0929, 0.0873, 0.0811, 0.0840, 0.0780], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 21:14:36,508 INFO [train.py:903] (1/4) Epoch 22, batch 2450, loss[loss=0.1898, simple_loss=0.2634, pruned_loss=0.05814, over 16074.00 frames. ], tot_loss[loss=0.2069, simple_loss=0.2866, pruned_loss=0.06359, over 3828308.16 frames. ], batch size: 35, lr: 3.75e-03, grad_scale: 8.0 2023-04-02 21:14:37,948 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=145839.0, num_to_drop=1, layers_to_drop={0} 2023-04-02 21:14:49,298 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=145848.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 21:14:52,761 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.8268, 1.3009, 1.4346, 1.7311, 3.4061, 1.2596, 2.4192, 3.8802], device='cuda:1'), covar=tensor([0.0515, 0.2909, 0.3078, 0.1825, 0.0765, 0.2439, 0.1388, 0.0232], device='cuda:1'), in_proj_covar=tensor([0.0409, 0.0366, 0.0387, 0.0350, 0.0375, 0.0350, 0.0380, 0.0406], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 21:15:28,951 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3816, 1.4832, 1.7844, 1.5960, 2.5442, 2.1582, 2.7491, 1.1979], device='cuda:1'), covar=tensor([0.2596, 0.4367, 0.2747, 0.2015, 0.1548, 0.2236, 0.1442, 0.4468], device='cuda:1'), in_proj_covar=tensor([0.0535, 0.0643, 0.0712, 0.0481, 0.0618, 0.0529, 0.0663, 0.0549], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 21:15:37,466 INFO [train.py:903] (1/4) Epoch 22, batch 2500, loss[loss=0.2036, simple_loss=0.2848, pruned_loss=0.06126, over 19783.00 frames. ], tot_loss[loss=0.2075, simple_loss=0.287, pruned_loss=0.06398, over 3812621.44 frames. ], batch size: 48, lr: 3.75e-03, grad_scale: 8.0 2023-04-02 21:16:25,695 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.8788, 4.8973, 5.6078, 5.6038, 1.9579, 5.2917, 4.5778, 5.2905], device='cuda:1'), covar=tensor([0.1852, 0.1026, 0.0625, 0.0638, 0.6436, 0.0980, 0.0622, 0.1319], device='cuda:1'), in_proj_covar=tensor([0.0781, 0.0741, 0.0941, 0.0828, 0.0830, 0.0704, 0.0565, 0.0878], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-02 21:16:34,420 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.066e+02 4.836e+02 5.791e+02 7.519e+02 1.267e+03, threshold=1.158e+03, percent-clipped=1.0 2023-04-02 21:16:36,602 INFO [train.py:903] (1/4) Epoch 22, batch 2550, loss[loss=0.2009, simple_loss=0.2888, pruned_loss=0.05654, over 19493.00 frames. ], tot_loss[loss=0.2082, simple_loss=0.2875, pruned_loss=0.06444, over 3814651.30 frames. ], batch size: 64, lr: 3.75e-03, grad_scale: 8.0 2023-04-02 21:16:40,550 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.27 vs. limit=5.0 2023-04-02 21:16:59,317 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=145956.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 21:17:15,010 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.9767, 4.5054, 2.8416, 3.9886, 1.0027, 4.5432, 4.4013, 4.4866], device='cuda:1'), covar=tensor([0.0530, 0.0860, 0.1789, 0.0796, 0.3822, 0.0611, 0.0814, 0.1140], device='cuda:1'), in_proj_covar=tensor([0.0501, 0.0407, 0.0489, 0.0342, 0.0399, 0.0427, 0.0422, 0.0456], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 21:17:33,995 WARNING [train.py:1073] (1/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 2023-04-02 21:17:38,152 INFO [train.py:903] (1/4) Epoch 22, batch 2600, loss[loss=0.1915, simple_loss=0.2802, pruned_loss=0.05137, over 19683.00 frames. ], tot_loss[loss=0.2091, simple_loss=0.2887, pruned_loss=0.06479, over 3813894.62 frames. ], batch size: 59, lr: 3.75e-03, grad_scale: 8.0 2023-04-02 21:17:59,427 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=146005.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 21:18:28,341 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1837, 1.8268, 1.4900, 1.2939, 1.6325, 1.2295, 1.2563, 1.6742], device='cuda:1'), covar=tensor([0.0791, 0.0763, 0.1100, 0.0773, 0.0559, 0.1297, 0.0561, 0.0414], device='cuda:1'), in_proj_covar=tensor([0.0301, 0.0315, 0.0338, 0.0264, 0.0248, 0.0336, 0.0290, 0.0274], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 21:18:38,072 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.365e+02 5.022e+02 6.231e+02 7.783e+02 1.698e+03, threshold=1.246e+03, percent-clipped=5.0 2023-04-02 21:18:40,370 INFO [train.py:903] (1/4) Epoch 22, batch 2650, loss[loss=0.1975, simple_loss=0.2882, pruned_loss=0.05339, over 19523.00 frames. ], tot_loss[loss=0.2081, simple_loss=0.2877, pruned_loss=0.0642, over 3819597.21 frames. ], batch size: 54, lr: 3.75e-03, grad_scale: 8.0 2023-04-02 21:19:00,423 WARNING [train.py:1073] (1/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9_rvb from training. Duration: 27.25 2023-04-02 21:19:21,366 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=146071.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 21:19:35,495 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.53 vs. limit=2.0 2023-04-02 21:19:41,307 INFO [train.py:903] (1/4) Epoch 22, batch 2700, loss[loss=0.2258, simple_loss=0.3074, pruned_loss=0.07211, over 19672.00 frames. ], tot_loss[loss=0.2092, simple_loss=0.289, pruned_loss=0.06474, over 3812464.60 frames. ], batch size: 55, lr: 3.75e-03, grad_scale: 8.0 2023-04-02 21:20:00,917 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=146104.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 21:20:20,723 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=146120.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 21:20:32,037 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=146129.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 21:20:39,333 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.068e+02 4.787e+02 6.288e+02 8.148e+02 2.582e+03, threshold=1.258e+03, percent-clipped=4.0 2023-04-02 21:20:41,747 INFO [train.py:903] (1/4) Epoch 22, batch 2750, loss[loss=0.2207, simple_loss=0.3027, pruned_loss=0.06933, over 18724.00 frames. ], tot_loss[loss=0.2075, simple_loss=0.2875, pruned_loss=0.06381, over 3811786.29 frames. ], batch size: 74, lr: 3.75e-03, grad_scale: 8.0 2023-04-02 21:21:05,843 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.68 vs. limit=5.0 2023-04-02 21:21:08,977 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.0818, 1.0767, 1.6362, 1.0798, 2.2014, 2.9857, 2.8054, 3.3518], device='cuda:1'), covar=tensor([0.1776, 0.5135, 0.4355, 0.2589, 0.0707, 0.0290, 0.0307, 0.0297], device='cuda:1'), in_proj_covar=tensor([0.0272, 0.0320, 0.0352, 0.0265, 0.0242, 0.0186, 0.0215, 0.0256], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-02 21:21:18,147 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=146167.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 21:21:37,433 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=146183.0, num_to_drop=1, layers_to_drop={1} 2023-04-02 21:21:43,820 INFO [train.py:903] (1/4) Epoch 22, batch 2800, loss[loss=0.1858, simple_loss=0.2656, pruned_loss=0.05297, over 19612.00 frames. ], tot_loss[loss=0.2073, simple_loss=0.287, pruned_loss=0.0638, over 3805835.40 frames. ], batch size: 50, lr: 3.75e-03, grad_scale: 8.0 2023-04-02 21:21:47,751 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=4.06 vs. limit=5.0 2023-04-02 21:22:42,905 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.980e+02 4.555e+02 5.863e+02 7.335e+02 1.249e+03, threshold=1.173e+03, percent-clipped=1.0 2023-04-02 21:22:45,127 INFO [train.py:903] (1/4) Epoch 22, batch 2850, loss[loss=0.2474, simple_loss=0.3229, pruned_loss=0.08599, over 19669.00 frames. ], tot_loss[loss=0.208, simple_loss=0.2879, pruned_loss=0.06406, over 3799440.55 frames. ], batch size: 60, lr: 3.75e-03, grad_scale: 8.0 2023-04-02 21:23:42,924 WARNING [train.py:1073] (1/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 2023-04-02 21:23:45,161 INFO [train.py:903] (1/4) Epoch 22, batch 2900, loss[loss=0.2058, simple_loss=0.2718, pruned_loss=0.06989, over 19758.00 frames. ], tot_loss[loss=0.2077, simple_loss=0.2876, pruned_loss=0.06397, over 3803056.07 frames. ], batch size: 47, lr: 3.75e-03, grad_scale: 8.0 2023-04-02 21:23:51,884 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4128, 1.3031, 1.3314, 1.6924, 1.3573, 1.5496, 1.6280, 1.4725], device='cuda:1'), covar=tensor([0.0868, 0.0990, 0.1064, 0.0723, 0.0948, 0.0889, 0.0922, 0.0741], device='cuda:1'), in_proj_covar=tensor([0.0211, 0.0221, 0.0224, 0.0239, 0.0225, 0.0210, 0.0186, 0.0202], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-02 21:23:57,291 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=146298.0, num_to_drop=1, layers_to_drop={1} 2023-04-02 21:24:33,155 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=146327.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 21:24:43,683 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.365e+02 4.789e+02 5.854e+02 7.393e+02 1.532e+03, threshold=1.171e+03, percent-clipped=5.0 2023-04-02 21:24:45,870 INFO [train.py:903] (1/4) Epoch 22, batch 2950, loss[loss=0.1721, simple_loss=0.2464, pruned_loss=0.04885, over 19756.00 frames. ], tot_loss[loss=0.2078, simple_loss=0.2878, pruned_loss=0.06395, over 3806054.92 frames. ], batch size: 48, lr: 3.75e-03, grad_scale: 8.0 2023-04-02 21:25:04,169 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=146352.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 21:25:32,559 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=146376.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 21:25:33,970 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.70 vs. limit=2.0 2023-04-02 21:25:46,774 INFO [train.py:903] (1/4) Epoch 22, batch 3000, loss[loss=0.191, simple_loss=0.2709, pruned_loss=0.05556, over 19719.00 frames. ], tot_loss[loss=0.2074, simple_loss=0.2871, pruned_loss=0.06386, over 3817281.09 frames. ], batch size: 51, lr: 3.75e-03, grad_scale: 4.0 2023-04-02 21:25:46,774 INFO [train.py:928] (1/4) Computing validation loss 2023-04-02 21:25:59,181 INFO [train.py:937] (1/4) Epoch 22, validation: loss=0.1687, simple_loss=0.2687, pruned_loss=0.0344, over 944034.00 frames. 2023-04-02 21:25:59,182 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18821MB 2023-04-02 21:26:02,615 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_rvb from training. Duration: 29.735 2023-04-02 21:26:16,133 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=146401.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 21:26:58,620 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.135e+02 5.066e+02 6.686e+02 8.533e+02 1.871e+03, threshold=1.337e+03, percent-clipped=6.0 2023-04-02 21:26:59,750 INFO [train.py:903] (1/4) Epoch 22, batch 3050, loss[loss=0.2029, simple_loss=0.2862, pruned_loss=0.05974, over 19754.00 frames. ], tot_loss[loss=0.2077, simple_loss=0.2875, pruned_loss=0.06395, over 3820348.66 frames. ], batch size: 54, lr: 3.74e-03, grad_scale: 4.0 2023-04-02 21:28:00,892 INFO [train.py:903] (1/4) Epoch 22, batch 3100, loss[loss=0.2053, simple_loss=0.2868, pruned_loss=0.06187, over 19516.00 frames. ], tot_loss[loss=0.2076, simple_loss=0.2873, pruned_loss=0.06398, over 3809103.37 frames. ], batch size: 54, lr: 3.74e-03, grad_scale: 4.0 2023-04-02 21:28:15,869 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1286, 1.3243, 1.9863, 1.4742, 3.1442, 4.5514, 4.3980, 4.9521], device='cuda:1'), covar=tensor([0.1780, 0.4091, 0.3373, 0.2442, 0.0633, 0.0216, 0.0187, 0.0227], device='cuda:1'), in_proj_covar=tensor([0.0273, 0.0321, 0.0352, 0.0265, 0.0243, 0.0186, 0.0215, 0.0257], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-02 21:28:27,617 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=146511.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 21:28:38,309 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.7434, 1.4177, 1.4459, 2.3017, 1.7396, 2.0542, 2.1989, 1.6987], device='cuda:1'), covar=tensor([0.0841, 0.0989, 0.1083, 0.0728, 0.0869, 0.0725, 0.0791, 0.0737], device='cuda:1'), in_proj_covar=tensor([0.0211, 0.0221, 0.0225, 0.0239, 0.0225, 0.0209, 0.0186, 0.0203], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-02 21:28:59,181 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.441e+02 5.077e+02 6.322e+02 8.082e+02 1.628e+03, threshold=1.264e+03, percent-clipped=2.0 2023-04-02 21:29:00,363 INFO [train.py:903] (1/4) Epoch 22, batch 3150, loss[loss=0.1974, simple_loss=0.2821, pruned_loss=0.05632, over 19687.00 frames. ], tot_loss[loss=0.208, simple_loss=0.2879, pruned_loss=0.06403, over 3824209.95 frames. ], batch size: 59, lr: 3.74e-03, grad_scale: 4.0 2023-04-02 21:29:19,896 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=146554.0, num_to_drop=1, layers_to_drop={1} 2023-04-02 21:29:29,200 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 2023-04-02 21:29:31,673 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2144, 1.2972, 1.2205, 1.0570, 1.0646, 1.0556, 0.0690, 0.3451], device='cuda:1'), covar=tensor([0.0663, 0.0634, 0.0457, 0.0578, 0.1329, 0.0658, 0.1366, 0.1194], device='cuda:1'), in_proj_covar=tensor([0.0358, 0.0357, 0.0359, 0.0384, 0.0460, 0.0388, 0.0338, 0.0343], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 21:29:47,812 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.4551, 2.5091, 2.6635, 3.1068, 2.5732, 3.0229, 2.6600, 2.4915], device='cuda:1'), covar=tensor([0.3574, 0.3041, 0.1535, 0.2021, 0.3253, 0.1577, 0.3662, 0.2629], device='cuda:1'), in_proj_covar=tensor([0.0897, 0.0961, 0.0717, 0.0930, 0.0878, 0.0814, 0.0844, 0.0781], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 21:29:51,122 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=146579.0, num_to_drop=1, layers_to_drop={1} 2023-04-02 21:30:00,733 INFO [train.py:903] (1/4) Epoch 22, batch 3200, loss[loss=0.2133, simple_loss=0.2882, pruned_loss=0.0692, over 19669.00 frames. ], tot_loss[loss=0.2078, simple_loss=0.2877, pruned_loss=0.06396, over 3822971.15 frames. ], batch size: 53, lr: 3.74e-03, grad_scale: 8.0 2023-04-02 21:30:28,084 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=146609.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 21:30:47,436 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=146626.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 21:31:01,670 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.447e+02 4.719e+02 5.878e+02 7.469e+02 1.229e+03, threshold=1.176e+03, percent-clipped=0.0 2023-04-02 21:31:02,532 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.31 vs. limit=2.0 2023-04-02 21:31:02,813 INFO [train.py:903] (1/4) Epoch 22, batch 3250, loss[loss=0.1756, simple_loss=0.2644, pruned_loss=0.0434, over 19777.00 frames. ], tot_loss[loss=0.2082, simple_loss=0.2882, pruned_loss=0.06406, over 3815569.91 frames. ], batch size: 54, lr: 3.74e-03, grad_scale: 8.0 2023-04-02 21:31:10,874 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=146644.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 21:32:03,196 INFO [train.py:903] (1/4) Epoch 22, batch 3300, loss[loss=0.2327, simple_loss=0.3086, pruned_loss=0.07838, over 19690.00 frames. ], tot_loss[loss=0.2079, simple_loss=0.2884, pruned_loss=0.06376, over 3833403.88 frames. ], batch size: 59, lr: 3.74e-03, grad_scale: 8.0 2023-04-02 21:32:08,318 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 2023-04-02 21:32:09,864 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 2023-04-02 21:32:59,756 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.231e+02 5.268e+02 6.638e+02 8.509e+02 1.642e+03, threshold=1.328e+03, percent-clipped=7.0 2023-04-02 21:33:00,762 INFO [train.py:903] (1/4) Epoch 22, batch 3350, loss[loss=0.1785, simple_loss=0.2543, pruned_loss=0.05134, over 19292.00 frames. ], tot_loss[loss=0.2072, simple_loss=0.2872, pruned_loss=0.06364, over 3837585.19 frames. ], batch size: 44, lr: 3.74e-03, grad_scale: 8.0 2023-04-02 21:34:00,058 INFO [train.py:903] (1/4) Epoch 22, batch 3400, loss[loss=0.2081, simple_loss=0.2833, pruned_loss=0.06643, over 19850.00 frames. ], tot_loss[loss=0.2062, simple_loss=0.2866, pruned_loss=0.06291, over 3843775.38 frames. ], batch size: 52, lr: 3.74e-03, grad_scale: 8.0 2023-04-02 21:34:59,708 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.964e+02 5.013e+02 6.159e+02 8.066e+02 2.491e+03, threshold=1.232e+03, percent-clipped=4.0 2023-04-02 21:35:00,914 INFO [train.py:903] (1/4) Epoch 22, batch 3450, loss[loss=0.1978, simple_loss=0.2736, pruned_loss=0.06101, over 19577.00 frames. ], tot_loss[loss=0.2073, simple_loss=0.2875, pruned_loss=0.06359, over 3842949.17 frames. ], batch size: 52, lr: 3.74e-03, grad_scale: 8.0 2023-04-02 21:35:04,226 WARNING [train.py:1073] (1/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9_rvb from training. Duration: 25.2444375 2023-04-02 21:35:28,804 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=146862.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 21:35:54,084 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=146882.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 21:36:01,443 INFO [train.py:903] (1/4) Epoch 22, batch 3500, loss[loss=0.2047, simple_loss=0.2715, pruned_loss=0.06895, over 19743.00 frames. ], tot_loss[loss=0.2079, simple_loss=0.2878, pruned_loss=0.06396, over 3834974.49 frames. ], batch size: 45, lr: 3.74e-03, grad_scale: 8.0 2023-04-02 21:36:23,412 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=146907.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 21:37:00,107 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.850e+02 4.615e+02 6.325e+02 8.235e+02 2.059e+03, threshold=1.265e+03, percent-clipped=6.0 2023-04-02 21:37:01,338 INFO [train.py:903] (1/4) Epoch 22, batch 3550, loss[loss=0.1782, simple_loss=0.2596, pruned_loss=0.04837, over 19371.00 frames. ], tot_loss[loss=0.208, simple_loss=0.2875, pruned_loss=0.06428, over 3821011.81 frames. ], batch size: 47, lr: 3.74e-03, grad_scale: 8.0 2023-04-02 21:37:18,262 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=146953.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 21:37:40,823 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.71 vs. limit=5.0 2023-04-02 21:38:02,337 INFO [train.py:903] (1/4) Epoch 22, batch 3600, loss[loss=0.1942, simple_loss=0.2655, pruned_loss=0.06145, over 19740.00 frames. ], tot_loss[loss=0.2087, simple_loss=0.2884, pruned_loss=0.06447, over 3811509.18 frames. ], batch size: 45, lr: 3.74e-03, grad_scale: 8.0 2023-04-02 21:38:02,532 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=146988.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 21:39:01,661 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.391e+02 4.948e+02 6.297e+02 8.667e+02 2.605e+03, threshold=1.259e+03, percent-clipped=8.0 2023-04-02 21:39:02,706 INFO [train.py:903] (1/4) Epoch 22, batch 3650, loss[loss=0.2564, simple_loss=0.3361, pruned_loss=0.08838, over 19622.00 frames. ], tot_loss[loss=0.2089, simple_loss=0.2885, pruned_loss=0.06466, over 3817285.16 frames. ], batch size: 57, lr: 3.74e-03, grad_scale: 8.0 2023-04-02 21:39:39,077 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=147068.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 21:40:03,868 INFO [train.py:903] (1/4) Epoch 22, batch 3700, loss[loss=0.2531, simple_loss=0.339, pruned_loss=0.08358, over 19577.00 frames. ], tot_loss[loss=0.2092, simple_loss=0.2889, pruned_loss=0.0648, over 3823069.26 frames. ], batch size: 61, lr: 3.74e-03, grad_scale: 8.0 2023-04-02 21:40:21,170 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=147103.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 21:40:58,544 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.5636, 4.1181, 2.6189, 3.6112, 0.7811, 4.1088, 3.9876, 4.0993], device='cuda:1'), covar=tensor([0.0656, 0.1066, 0.2119, 0.0937, 0.4330, 0.0761, 0.0980, 0.1224], device='cuda:1'), in_proj_covar=tensor([0.0503, 0.0410, 0.0492, 0.0343, 0.0401, 0.0431, 0.0425, 0.0459], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 21:41:02,895 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.261e+02 4.816e+02 5.960e+02 7.144e+02 1.653e+03, threshold=1.192e+03, percent-clipped=4.0 2023-04-02 21:41:04,071 INFO [train.py:903] (1/4) Epoch 22, batch 3750, loss[loss=0.1724, simple_loss=0.2535, pruned_loss=0.04566, over 19389.00 frames. ], tot_loss[loss=0.2088, simple_loss=0.2884, pruned_loss=0.06454, over 3829886.35 frames. ], batch size: 48, lr: 3.74e-03, grad_scale: 8.0 2023-04-02 21:41:21,578 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4023, 1.4379, 1.7058, 1.6379, 2.3337, 2.1968, 2.5079, 1.1321], device='cuda:1'), covar=tensor([0.2581, 0.4296, 0.2633, 0.1969, 0.1621, 0.2166, 0.1493, 0.4478], device='cuda:1'), in_proj_covar=tensor([0.0536, 0.0643, 0.0714, 0.0482, 0.0618, 0.0531, 0.0663, 0.0549], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 21:42:04,512 INFO [train.py:903] (1/4) Epoch 22, batch 3800, loss[loss=0.2086, simple_loss=0.2875, pruned_loss=0.06488, over 19521.00 frames. ], tot_loss[loss=0.2098, simple_loss=0.2894, pruned_loss=0.06511, over 3833693.16 frames. ], batch size: 54, lr: 3.74e-03, grad_scale: 8.0 2023-04-02 21:42:26,706 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=147206.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 21:42:38,900 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9_rvb from training. Duration: 29.1166875 2023-04-02 21:43:02,581 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.867e+02 5.077e+02 5.970e+02 7.548e+02 1.289e+03, threshold=1.194e+03, percent-clipped=1.0 2023-04-02 21:43:03,540 INFO [train.py:903] (1/4) Epoch 22, batch 3850, loss[loss=0.2514, simple_loss=0.319, pruned_loss=0.09196, over 18751.00 frames. ], tot_loss[loss=0.2094, simple_loss=0.2893, pruned_loss=0.0648, over 3828487.18 frames. ], batch size: 74, lr: 3.73e-03, grad_scale: 8.0 2023-04-02 21:43:09,388 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.7791, 2.4749, 2.2775, 2.7835, 2.4455, 2.2949, 2.3573, 2.6424], device='cuda:1'), covar=tensor([0.0969, 0.1716, 0.1524, 0.1045, 0.1406, 0.0580, 0.1266, 0.0704], device='cuda:1'), in_proj_covar=tensor([0.0269, 0.0354, 0.0309, 0.0249, 0.0301, 0.0251, 0.0308, 0.0255], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 21:43:31,164 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=147259.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 21:44:05,364 INFO [train.py:903] (1/4) Epoch 22, batch 3900, loss[loss=0.195, simple_loss=0.2804, pruned_loss=0.05477, over 19685.00 frames. ], tot_loss[loss=0.208, simple_loss=0.2881, pruned_loss=0.06394, over 3832700.53 frames. ], batch size: 58, lr: 3.73e-03, grad_scale: 8.0 2023-04-02 21:44:28,433 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.49 vs. limit=5.0 2023-04-02 21:44:41,115 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.5992, 1.2650, 1.5503, 1.3237, 2.2367, 1.0269, 2.0980, 2.5276], device='cuda:1'), covar=tensor([0.0745, 0.2860, 0.2737, 0.1652, 0.0951, 0.2064, 0.1038, 0.0460], device='cuda:1'), in_proj_covar=tensor([0.0406, 0.0364, 0.0385, 0.0348, 0.0372, 0.0347, 0.0379, 0.0402], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 21:44:46,187 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=147321.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 21:44:49,711 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=147324.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 21:44:55,086 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=147328.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 21:45:04,841 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.696e+02 4.890e+02 6.799e+02 8.609e+02 1.784e+03, threshold=1.360e+03, percent-clipped=9.0 2023-04-02 21:45:05,887 INFO [train.py:903] (1/4) Epoch 22, batch 3950, loss[loss=0.2207, simple_loss=0.3052, pruned_loss=0.06814, over 19652.00 frames. ], tot_loss[loss=0.2084, simple_loss=0.2882, pruned_loss=0.06432, over 3821676.50 frames. ], batch size: 58, lr: 3.73e-03, grad_scale: 8.0 2023-04-02 21:45:08,142 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 2023-04-02 21:45:18,239 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=147349.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 21:45:30,140 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=147359.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 21:46:00,818 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=147384.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 21:46:04,831 INFO [train.py:903] (1/4) Epoch 22, batch 4000, loss[loss=0.2135, simple_loss=0.3019, pruned_loss=0.06257, over 19455.00 frames. ], tot_loss[loss=0.2079, simple_loss=0.2881, pruned_loss=0.06383, over 3836546.56 frames. ], batch size: 64, lr: 3.73e-03, grad_scale: 8.0 2023-04-02 21:46:50,142 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=147425.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 21:46:52,244 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1_rvb from training. Duration: 0.7545625 2023-04-02 21:46:54,086 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 2023-04-02 21:47:03,832 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.354e+02 4.920e+02 5.925e+02 7.748e+02 1.160e+03, threshold=1.185e+03, percent-clipped=0.0 2023-04-02 21:47:05,868 INFO [train.py:903] (1/4) Epoch 22, batch 4050, loss[loss=0.1854, simple_loss=0.2636, pruned_loss=0.05363, over 15171.00 frames. ], tot_loss[loss=0.2074, simple_loss=0.2877, pruned_loss=0.06358, over 3822472.96 frames. ], batch size: 33, lr: 3.73e-03, grad_scale: 8.0 2023-04-02 21:47:24,618 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=147452.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 21:47:45,016 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 2023-04-02 21:48:07,527 INFO [train.py:903] (1/4) Epoch 22, batch 4100, loss[loss=0.2298, simple_loss=0.308, pruned_loss=0.07584, over 19740.00 frames. ], tot_loss[loss=0.2079, simple_loss=0.288, pruned_loss=0.06387, over 3839847.84 frames. ], batch size: 63, lr: 3.73e-03, grad_scale: 8.0 2023-04-02 21:48:44,892 WARNING [train.py:1073] (1/4) Exclude cut with ID 497-129325-0061-9566_sp1.1_rvb from training. Duration: 0.97725 2023-04-02 21:49:07,909 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.284e+02 5.254e+02 6.349e+02 7.493e+02 1.711e+03, threshold=1.270e+03, percent-clipped=4.0 2023-04-02 21:49:09,106 INFO [train.py:903] (1/4) Epoch 22, batch 4150, loss[loss=0.1768, simple_loss=0.2591, pruned_loss=0.04722, over 19598.00 frames. ], tot_loss[loss=0.2068, simple_loss=0.287, pruned_loss=0.06332, over 3848078.47 frames. ], batch size: 52, lr: 3.73e-03, grad_scale: 8.0 2023-04-02 21:49:57,000 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=147577.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 21:50:09,107 INFO [train.py:903] (1/4) Epoch 22, batch 4200, loss[loss=0.2392, simple_loss=0.3245, pruned_loss=0.07695, over 19659.00 frames. ], tot_loss[loss=0.2073, simple_loss=0.2875, pruned_loss=0.0635, over 3846698.27 frames. ], batch size: 55, lr: 3.73e-03, grad_scale: 8.0 2023-04-02 21:50:13,787 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 2023-04-02 21:50:26,602 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=147602.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 21:50:27,412 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=147603.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 21:51:09,770 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.286e+02 4.933e+02 5.671e+02 7.351e+02 2.024e+03, threshold=1.134e+03, percent-clipped=5.0 2023-04-02 21:51:10,933 INFO [train.py:903] (1/4) Epoch 22, batch 4250, loss[loss=0.1822, simple_loss=0.2656, pruned_loss=0.04936, over 19580.00 frames. ], tot_loss[loss=0.2062, simple_loss=0.2863, pruned_loss=0.06303, over 3840922.97 frames. ], batch size: 52, lr: 3.73e-03, grad_scale: 8.0 2023-04-02 21:51:25,894 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9_rvb from training. Duration: 29.816625 2023-04-02 21:51:38,186 WARNING [train.py:1073] (1/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 2023-04-02 21:51:51,752 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=147672.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 21:51:57,371 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=147677.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 21:52:11,226 INFO [train.py:903] (1/4) Epoch 22, batch 4300, loss[loss=0.1756, simple_loss=0.2597, pruned_loss=0.04575, over 19614.00 frames. ], tot_loss[loss=0.2064, simple_loss=0.2864, pruned_loss=0.0632, over 3840748.32 frames. ], batch size: 50, lr: 3.73e-03, grad_scale: 4.0 2023-04-02 21:52:47,215 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=147718.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 21:53:02,954 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 2023-04-02 21:53:11,620 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.973e+02 5.055e+02 6.219e+02 8.115e+02 2.735e+03, threshold=1.244e+03, percent-clipped=11.0 2023-04-02 21:53:11,639 INFO [train.py:903] (1/4) Epoch 22, batch 4350, loss[loss=0.2077, simple_loss=0.2997, pruned_loss=0.05784, over 18832.00 frames. ], tot_loss[loss=0.2079, simple_loss=0.2876, pruned_loss=0.06411, over 3828832.11 frames. ], batch size: 74, lr: 3.73e-03, grad_scale: 4.0 2023-04-02 21:53:48,691 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=147769.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 21:54:11,229 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=147787.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 21:54:12,000 INFO [train.py:903] (1/4) Epoch 22, batch 4400, loss[loss=0.2152, simple_loss=0.3025, pruned_loss=0.06396, over 19674.00 frames. ], tot_loss[loss=0.207, simple_loss=0.287, pruned_loss=0.06353, over 3831588.41 frames. ], batch size: 58, lr: 3.73e-03, grad_scale: 8.0 2023-04-02 21:54:20,675 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=147796.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 21:54:37,150 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_rvb from training. Duration: 25.285 2023-04-02 21:54:46,692 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 2023-04-02 21:55:06,301 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=147833.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 21:55:12,515 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.305e+02 5.137e+02 6.374e+02 7.743e+02 1.534e+03, threshold=1.275e+03, percent-clipped=3.0 2023-04-02 21:55:12,533 INFO [train.py:903] (1/4) Epoch 22, batch 4450, loss[loss=0.219, simple_loss=0.2997, pruned_loss=0.06913, over 18269.00 frames. ], tot_loss[loss=0.2076, simple_loss=0.2876, pruned_loss=0.06381, over 3826666.03 frames. ], batch size: 83, lr: 3.73e-03, grad_scale: 8.0 2023-04-02 21:55:57,428 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=147875.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 21:55:57,463 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.9751, 1.7245, 1.9939, 1.7900, 4.5113, 1.2804, 2.5746, 4.9116], device='cuda:1'), covar=tensor([0.0440, 0.2596, 0.2700, 0.2013, 0.0788, 0.2569, 0.1416, 0.0157], device='cuda:1'), in_proj_covar=tensor([0.0408, 0.0366, 0.0387, 0.0348, 0.0373, 0.0349, 0.0382, 0.0404], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 21:56:08,760 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=147884.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 21:56:12,929 INFO [train.py:903] (1/4) Epoch 22, batch 4500, loss[loss=0.1587, simple_loss=0.241, pruned_loss=0.03815, over 19396.00 frames. ], tot_loss[loss=0.2069, simple_loss=0.2869, pruned_loss=0.06345, over 3830977.56 frames. ], batch size: 48, lr: 3.73e-03, grad_scale: 8.0 2023-04-02 21:56:41,845 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=147911.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 21:57:03,781 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=147929.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 21:57:15,409 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.490e+02 4.535e+02 5.624e+02 7.235e+02 1.683e+03, threshold=1.125e+03, percent-clipped=3.0 2023-04-02 21:57:15,427 INFO [train.py:903] (1/4) Epoch 22, batch 4550, loss[loss=0.2421, simple_loss=0.3227, pruned_loss=0.08081, over 18594.00 frames. ], tot_loss[loss=0.2072, simple_loss=0.2868, pruned_loss=0.06375, over 3820877.14 frames. ], batch size: 74, lr: 3.73e-03, grad_scale: 8.0 2023-04-02 21:57:23,455 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9_rvb from training. Duration: 28.72225 2023-04-02 21:57:46,097 WARNING [train.py:1073] (1/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 2023-04-02 21:57:58,605 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=147974.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 21:58:15,826 INFO [train.py:903] (1/4) Epoch 22, batch 4600, loss[loss=0.2101, simple_loss=0.3013, pruned_loss=0.05942, over 19526.00 frames. ], tot_loss[loss=0.2075, simple_loss=0.2872, pruned_loss=0.06389, over 3827398.68 frames. ], batch size: 56, lr: 3.73e-03, grad_scale: 8.0 2023-04-02 21:58:28,677 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=147999.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 21:58:56,679 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=148021.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 21:59:04,092 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 2023-04-02 21:59:12,793 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.1030, 1.2607, 1.4097, 1.5548, 2.7231, 1.0820, 2.1813, 3.1063], device='cuda:1'), covar=tensor([0.0584, 0.2741, 0.3046, 0.1610, 0.0754, 0.2312, 0.1164, 0.0313], device='cuda:1'), in_proj_covar=tensor([0.0408, 0.0366, 0.0387, 0.0347, 0.0374, 0.0348, 0.0381, 0.0403], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 21:59:16,087 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.327e+02 4.857e+02 6.032e+02 8.227e+02 1.754e+03, threshold=1.206e+03, percent-clipped=4.0 2023-04-02 21:59:16,110 INFO [train.py:903] (1/4) Epoch 22, batch 4650, loss[loss=0.1726, simple_loss=0.2488, pruned_loss=0.04822, over 19772.00 frames. ], tot_loss[loss=0.2081, simple_loss=0.2877, pruned_loss=0.0643, over 3814285.73 frames. ], batch size: 47, lr: 3.72e-03, grad_scale: 8.0 2023-04-02 21:59:22,648 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=148043.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 21:59:32,296 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_rvb from training. Duration: 0.92 2023-04-02 21:59:43,962 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_rvb from training. Duration: 0.83 2023-04-02 21:59:53,309 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=148068.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 22:00:16,031 INFO [train.py:903] (1/4) Epoch 22, batch 4700, loss[loss=0.2327, simple_loss=0.3147, pruned_loss=0.07539, over 19678.00 frames. ], tot_loss[loss=0.2081, simple_loss=0.2876, pruned_loss=0.06432, over 3822819.49 frames. ], batch size: 59, lr: 3.72e-03, grad_scale: 8.0 2023-04-02 22:00:39,952 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9_rvb from training. Duration: 0.92225 2023-04-02 22:01:15,707 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=148136.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 22:01:17,615 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.017e+02 5.448e+02 6.322e+02 7.572e+02 1.580e+03, threshold=1.264e+03, percent-clipped=4.0 2023-04-02 22:01:17,634 INFO [train.py:903] (1/4) Epoch 22, batch 4750, loss[loss=0.201, simple_loss=0.2865, pruned_loss=0.05772, over 19594.00 frames. ], tot_loss[loss=0.2073, simple_loss=0.2868, pruned_loss=0.06384, over 3830722.86 frames. ], batch size: 61, lr: 3.72e-03, grad_scale: 8.0 2023-04-02 22:01:21,283 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=148140.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 22:01:34,524 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 2023-04-02 22:01:50,456 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=148165.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 22:01:52,735 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=148167.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 22:02:05,544 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=148177.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 22:02:06,208 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.51 vs. limit=2.0 2023-04-02 22:02:18,752 INFO [train.py:903] (1/4) Epoch 22, batch 4800, loss[loss=0.2177, simple_loss=0.3034, pruned_loss=0.06599, over 19506.00 frames. ], tot_loss[loss=0.2078, simple_loss=0.2875, pruned_loss=0.06408, over 3820936.06 frames. ], batch size: 64, lr: 3.72e-03, grad_scale: 8.0 2023-04-02 22:02:23,657 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=148192.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 22:02:57,694 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=148219.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 22:03:18,987 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.136e+02 4.799e+02 5.785e+02 7.279e+02 1.291e+03, threshold=1.157e+03, percent-clipped=1.0 2023-04-02 22:03:19,007 INFO [train.py:903] (1/4) Epoch 22, batch 4850, loss[loss=0.209, simple_loss=0.296, pruned_loss=0.06099, over 19608.00 frames. ], tot_loss[loss=0.2085, simple_loss=0.2876, pruned_loss=0.06467, over 3802562.29 frames. ], batch size: 57, lr: 3.72e-03, grad_scale: 8.0 2023-04-02 22:03:44,995 WARNING [train.py:1073] (1/4) Exclude cut with ID 774-127930-0014-48411_sp1.1_rvb from training. Duration: 0.95 2023-04-02 22:04:01,912 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=148273.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 22:04:02,921 WARNING [train.py:1073] (1/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1_rvb from training. Duration: 0.9409375 2023-04-02 22:04:08,081 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 2023-04-02 22:04:09,252 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 2023-04-02 22:04:18,711 INFO [train.py:903] (1/4) Epoch 22, batch 4900, loss[loss=0.2404, simple_loss=0.3105, pruned_loss=0.08518, over 17511.00 frames. ], tot_loss[loss=0.2095, simple_loss=0.2886, pruned_loss=0.06519, over 3814466.88 frames. ], batch size: 101, lr: 3.72e-03, grad_scale: 8.0 2023-04-02 22:04:18,721 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1_rvb from training. Duration: 25.3818125 2023-04-02 22:04:24,312 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=148292.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 22:04:39,207 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 2023-04-02 22:04:40,586 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=148305.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 22:05:14,498 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=148334.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 22:05:19,513 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.031e+02 5.396e+02 6.572e+02 8.433e+02 1.736e+03, threshold=1.314e+03, percent-clipped=6.0 2023-04-02 22:05:19,535 INFO [train.py:903] (1/4) Epoch 22, batch 4950, loss[loss=0.1897, simple_loss=0.2679, pruned_loss=0.0557, over 19836.00 frames. ], tot_loss[loss=0.208, simple_loss=0.2871, pruned_loss=0.06443, over 3820343.74 frames. ], batch size: 52, lr: 3.72e-03, grad_scale: 8.0 2023-04-02 22:05:26,426 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.29 vs. limit=5.0 2023-04-02 22:05:35,936 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 2023-04-02 22:06:01,177 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 2023-04-02 22:06:20,897 INFO [train.py:903] (1/4) Epoch 22, batch 5000, loss[loss=0.2055, simple_loss=0.2864, pruned_loss=0.06232, over 19737.00 frames. ], tot_loss[loss=0.208, simple_loss=0.2873, pruned_loss=0.06438, over 3813921.00 frames. ], batch size: 51, lr: 3.72e-03, grad_scale: 8.0 2023-04-02 22:06:21,239 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=148388.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 22:06:25,711 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=148392.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 22:06:29,617 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_rvb from training. Duration: 27.14 2023-04-02 22:06:40,551 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 2023-04-02 22:06:55,410 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=148417.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 22:07:19,245 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.597e+02 4.828e+02 6.337e+02 7.923e+02 1.456e+03, threshold=1.267e+03, percent-clipped=1.0 2023-04-02 22:07:19,264 INFO [train.py:903] (1/4) Epoch 22, batch 5050, loss[loss=0.2108, simple_loss=0.2931, pruned_loss=0.06424, over 19535.00 frames. ], tot_loss[loss=0.2094, simple_loss=0.2887, pruned_loss=0.06499, over 3807704.39 frames. ], batch size: 54, lr: 3.72e-03, grad_scale: 8.0 2023-04-02 22:07:30,067 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=148447.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 22:07:46,920 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.3990, 4.0032, 2.6769, 3.4976, 1.0026, 3.9587, 3.8211, 3.9145], device='cuda:1'), covar=tensor([0.0661, 0.0987, 0.1916, 0.0894, 0.3850, 0.0656, 0.0962, 0.1094], device='cuda:1'), in_proj_covar=tensor([0.0502, 0.0409, 0.0492, 0.0344, 0.0398, 0.0432, 0.0424, 0.0458], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 22:07:54,473 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9_rvb from training. Duration: 26.62775 2023-04-02 22:08:19,394 INFO [train.py:903] (1/4) Epoch 22, batch 5100, loss[loss=0.1788, simple_loss=0.2677, pruned_loss=0.04495, over 19670.00 frames. ], tot_loss[loss=0.2094, simple_loss=0.2886, pruned_loss=0.06504, over 3817801.77 frames. ], batch size: 55, lr: 3.72e-03, grad_scale: 8.0 2023-04-02 22:08:21,060 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 2023-04-02 22:08:30,475 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 2023-04-02 22:08:33,794 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_rvb from training. Duration: 25.035 2023-04-02 22:08:39,204 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_rvb from training. Duration: 27.92 2023-04-02 22:08:53,267 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=148516.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 22:09:19,535 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.928e+02 5.468e+02 6.941e+02 9.893e+02 2.948e+03, threshold=1.388e+03, percent-clipped=12.0 2023-04-02 22:09:19,553 INFO [train.py:903] (1/4) Epoch 22, batch 5150, loss[loss=0.2211, simple_loss=0.3021, pruned_loss=0.07006, over 18797.00 frames. ], tot_loss[loss=0.2092, simple_loss=0.2886, pruned_loss=0.06486, over 3817907.22 frames. ], batch size: 74, lr: 3.72e-03, grad_scale: 8.0 2023-04-02 22:09:31,354 WARNING [train.py:1073] (1/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 2023-04-02 22:09:32,946 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=148548.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 22:09:40,839 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 2023-04-02 22:10:02,829 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=148573.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 22:10:05,908 WARNING [train.py:1073] (1/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1_rvb from training. Duration: 0.9681875 2023-04-02 22:10:20,790 INFO [train.py:903] (1/4) Epoch 22, batch 5200, loss[loss=0.1853, simple_loss=0.2634, pruned_loss=0.05362, over 19490.00 frames. ], tot_loss[loss=0.2088, simple_loss=0.288, pruned_loss=0.06477, over 3808869.84 frames. ], batch size: 49, lr: 3.72e-03, grad_scale: 8.0 2023-04-02 22:10:23,519 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=148590.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 22:10:33,168 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 2023-04-02 22:10:53,683 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=148615.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 22:10:58,738 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 2023-04-02 22:11:17,518 WARNING [train.py:1073] (1/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 2023-04-02 22:11:21,001 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.230e+02 4.558e+02 5.776e+02 7.251e+02 2.001e+03, threshold=1.155e+03, percent-clipped=2.0 2023-04-02 22:11:21,019 INFO [train.py:903] (1/4) Epoch 22, batch 5250, loss[loss=0.1705, simple_loss=0.2444, pruned_loss=0.04833, over 19760.00 frames. ], tot_loss[loss=0.2072, simple_loss=0.287, pruned_loss=0.06375, over 3821597.60 frames. ], batch size: 48, lr: 3.72e-03, grad_scale: 8.0 2023-04-02 22:11:27,728 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.9674, 2.0908, 2.2922, 2.0812, 3.6661, 1.7374, 2.9798, 3.7244], device='cuda:1'), covar=tensor([0.0439, 0.2094, 0.2101, 0.1649, 0.0603, 0.2147, 0.1496, 0.0271], device='cuda:1'), in_proj_covar=tensor([0.0410, 0.0367, 0.0387, 0.0348, 0.0375, 0.0351, 0.0384, 0.0405], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 22:11:27,830 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=148644.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 22:11:33,742 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=148649.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 22:11:35,101 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5712, 1.4853, 2.1561, 1.6603, 3.1236, 4.7985, 4.6905, 5.1383], device='cuda:1'), covar=tensor([0.1500, 0.3643, 0.3011, 0.2253, 0.0594, 0.0159, 0.0153, 0.0204], device='cuda:1'), in_proj_covar=tensor([0.0276, 0.0326, 0.0357, 0.0268, 0.0248, 0.0190, 0.0218, 0.0262], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-02 22:11:58,260 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=148669.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 22:11:58,362 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=148669.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 22:12:20,277 INFO [train.py:903] (1/4) Epoch 22, batch 5300, loss[loss=0.1998, simple_loss=0.2772, pruned_loss=0.06126, over 19423.00 frames. ], tot_loss[loss=0.2087, simple_loss=0.2883, pruned_loss=0.06455, over 3820165.61 frames. ], batch size: 48, lr: 3.72e-03, grad_scale: 8.0 2023-04-02 22:12:39,144 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 2023-04-02 22:13:17,745 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=148734.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 22:13:22,172 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.679e+02 5.309e+02 6.457e+02 8.011e+02 2.116e+03, threshold=1.291e+03, percent-clipped=5.0 2023-04-02 22:13:22,191 INFO [train.py:903] (1/4) Epoch 22, batch 5350, loss[loss=0.2323, simple_loss=0.3091, pruned_loss=0.07772, over 17452.00 frames. ], tot_loss[loss=0.2078, simple_loss=0.2876, pruned_loss=0.06399, over 3819967.31 frames. ], batch size: 101, lr: 3.72e-03, grad_scale: 8.0 2023-04-02 22:13:47,488 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9913, 2.0747, 2.3372, 2.7153, 2.0539, 2.5802, 2.3409, 2.1166], device='cuda:1'), covar=tensor([0.4385, 0.4044, 0.1866, 0.2444, 0.4180, 0.2128, 0.4675, 0.3252], device='cuda:1'), in_proj_covar=tensor([0.0897, 0.0961, 0.0716, 0.0929, 0.0877, 0.0814, 0.0840, 0.0780], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 22:13:50,705 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.8867, 1.2688, 1.5944, 1.6758, 4.1031, 1.3149, 2.7871, 4.5799], device='cuda:1'), covar=tensor([0.0559, 0.3872, 0.3526, 0.2434, 0.1195, 0.3084, 0.1467, 0.0296], device='cuda:1'), in_proj_covar=tensor([0.0411, 0.0368, 0.0387, 0.0347, 0.0375, 0.0350, 0.0383, 0.0405], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 22:13:53,945 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=148764.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 22:13:55,496 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9_rvb from training. Duration: 26.438875 2023-04-02 22:14:03,664 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.1066, 1.2502, 1.4372, 1.5003, 2.6969, 1.1284, 2.1886, 3.1009], device='cuda:1'), covar=tensor([0.0601, 0.2948, 0.3117, 0.1784, 0.0792, 0.2496, 0.1277, 0.0322], device='cuda:1'), in_proj_covar=tensor([0.0412, 0.0369, 0.0388, 0.0348, 0.0376, 0.0351, 0.0384, 0.0406], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 22:14:24,408 INFO [train.py:903] (1/4) Epoch 22, batch 5400, loss[loss=0.2263, simple_loss=0.3054, pruned_loss=0.07358, over 19649.00 frames. ], tot_loss[loss=0.2077, simple_loss=0.2876, pruned_loss=0.06392, over 3817966.00 frames. ], batch size: 58, lr: 3.72e-03, grad_scale: 8.0 2023-04-02 22:14:28,068 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=148791.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 22:15:24,092 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.599e+02 4.882e+02 5.931e+02 8.184e+02 2.288e+03, threshold=1.186e+03, percent-clipped=7.0 2023-04-02 22:15:24,110 INFO [train.py:903] (1/4) Epoch 22, batch 5450, loss[loss=0.2151, simple_loss=0.2954, pruned_loss=0.06742, over 19424.00 frames. ], tot_loss[loss=0.2071, simple_loss=0.2872, pruned_loss=0.06354, over 3816661.47 frames. ], batch size: 70, lr: 3.71e-03, grad_scale: 8.0 2023-04-02 22:15:50,187 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=148860.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 22:16:24,051 INFO [train.py:903] (1/4) Epoch 22, batch 5500, loss[loss=0.1696, simple_loss=0.2458, pruned_loss=0.04664, over 19792.00 frames. ], tot_loss[loss=0.2075, simple_loss=0.2872, pruned_loss=0.06395, over 3817001.84 frames. ], batch size: 47, lr: 3.71e-03, grad_scale: 8.0 2023-04-02 22:16:47,564 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=148906.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 22:16:49,402 WARNING [train.py:1073] (1/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9_rvb from training. Duration: 27.47775 2023-04-02 22:16:50,910 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6289, 1.4201, 1.9335, 1.6337, 3.0242, 4.5042, 4.2885, 4.8607], device='cuda:1'), covar=tensor([0.1458, 0.3750, 0.3346, 0.2316, 0.0645, 0.0197, 0.0183, 0.0208], device='cuda:1'), in_proj_covar=tensor([0.0275, 0.0324, 0.0356, 0.0266, 0.0247, 0.0189, 0.0216, 0.0261], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-02 22:17:25,271 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.265e+02 4.956e+02 6.031e+02 8.410e+02 1.981e+03, threshold=1.206e+03, percent-clipped=11.0 2023-04-02 22:17:25,289 INFO [train.py:903] (1/4) Epoch 22, batch 5550, loss[loss=0.2247, simple_loss=0.3102, pruned_loss=0.06963, over 19614.00 frames. ], tot_loss[loss=0.2085, simple_loss=0.2883, pruned_loss=0.06442, over 3813979.62 frames. ], batch size: 57, lr: 3.71e-03, grad_scale: 8.0 2023-04-02 22:17:33,842 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 2023-04-02 22:17:57,660 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.5801, 1.9796, 2.1562, 2.0640, 3.3021, 1.7606, 2.8617, 3.5319], device='cuda:1'), covar=tensor([0.0434, 0.2278, 0.2284, 0.1598, 0.0569, 0.2080, 0.1697, 0.0285], device='cuda:1'), in_proj_covar=tensor([0.0409, 0.0367, 0.0386, 0.0347, 0.0375, 0.0351, 0.0382, 0.0406], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 22:18:10,799 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=148975.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 22:18:21,568 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 2023-04-02 22:18:27,000 INFO [train.py:903] (1/4) Epoch 22, batch 5600, loss[loss=0.2006, simple_loss=0.2903, pruned_loss=0.05544, over 19784.00 frames. ], tot_loss[loss=0.2085, simple_loss=0.2881, pruned_loss=0.06447, over 3815062.74 frames. ], batch size: 56, lr: 3.71e-03, grad_scale: 8.0 2023-04-02 22:18:56,786 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=149013.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 22:19:06,002 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=149020.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 22:19:10,576 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 2023-04-02 22:19:27,601 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.511e+02 5.096e+02 6.059e+02 8.103e+02 1.757e+03, threshold=1.212e+03, percent-clipped=10.0 2023-04-02 22:19:27,620 INFO [train.py:903] (1/4) Epoch 22, batch 5650, loss[loss=0.2005, simple_loss=0.2829, pruned_loss=0.05905, over 19669.00 frames. ], tot_loss[loss=0.2075, simple_loss=0.2873, pruned_loss=0.06387, over 3816669.56 frames. ], batch size: 60, lr: 3.71e-03, grad_scale: 8.0 2023-04-02 22:19:30,328 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.3751, 2.2364, 2.0271, 1.8841, 1.7501, 1.9535, 0.6165, 1.3172], device='cuda:1'), covar=tensor([0.0585, 0.0552, 0.0485, 0.0853, 0.1178, 0.0882, 0.1357, 0.0988], device='cuda:1'), in_proj_covar=tensor([0.0356, 0.0357, 0.0359, 0.0382, 0.0462, 0.0390, 0.0337, 0.0341], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 22:19:35,829 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=149045.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 22:20:15,160 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9_rvb from training. Duration: 28.0944375 2023-04-02 22:20:16,295 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=149078.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 22:20:28,284 INFO [train.py:903] (1/4) Epoch 22, batch 5700, loss[loss=0.2157, simple_loss=0.2984, pruned_loss=0.06654, over 19730.00 frames. ], tot_loss[loss=0.2069, simple_loss=0.2867, pruned_loss=0.06361, over 3821109.62 frames. ], batch size: 63, lr: 3.71e-03, grad_scale: 8.0 2023-04-02 22:21:17,779 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=149128.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 22:21:29,596 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.928e+02 4.618e+02 6.077e+02 7.635e+02 1.470e+03, threshold=1.215e+03, percent-clipped=6.0 2023-04-02 22:21:29,615 INFO [train.py:903] (1/4) Epoch 22, batch 5750, loss[loss=0.1856, simple_loss=0.2773, pruned_loss=0.04688, over 19685.00 frames. ], tot_loss[loss=0.2066, simple_loss=0.2865, pruned_loss=0.06341, over 3822131.67 frames. ], batch size: 59, lr: 3.71e-03, grad_scale: 8.0 2023-04-02 22:21:30,805 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9_rvb from training. Duration: 33.038875 2023-04-02 22:21:39,627 WARNING [train.py:1073] (1/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 2023-04-02 22:21:46,351 WARNING [train.py:1073] (1/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 2023-04-02 22:21:59,246 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=149162.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 22:22:03,892 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1473, 1.2068, 1.3305, 1.3898, 1.0620, 1.3212, 1.3676, 1.2478], device='cuda:1'), covar=tensor([0.2889, 0.2303, 0.1380, 0.1586, 0.2573, 0.1453, 0.3400, 0.2362], device='cuda:1'), in_proj_covar=tensor([0.0897, 0.0962, 0.0718, 0.0933, 0.0880, 0.0814, 0.0841, 0.0781], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 22:22:29,841 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=149187.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 22:22:30,653 INFO [train.py:903] (1/4) Epoch 22, batch 5800, loss[loss=0.1856, simple_loss=0.2804, pruned_loss=0.04533, over 19534.00 frames. ], tot_loss[loss=0.2069, simple_loss=0.2865, pruned_loss=0.06364, over 3826718.57 frames. ], batch size: 54, lr: 3.71e-03, grad_scale: 4.0 2023-04-02 22:22:37,174 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=149193.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 22:23:23,071 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=149231.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 22:23:30,407 INFO [train.py:903] (1/4) Epoch 22, batch 5850, loss[loss=0.201, simple_loss=0.2856, pruned_loss=0.0582, over 19511.00 frames. ], tot_loss[loss=0.2074, simple_loss=0.2873, pruned_loss=0.06376, over 3832307.62 frames. ], batch size: 56, lr: 3.71e-03, grad_scale: 4.0 2023-04-02 22:23:31,585 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.414e+02 5.174e+02 6.346e+02 7.936e+02 1.645e+03, threshold=1.269e+03, percent-clipped=7.0 2023-04-02 22:23:39,957 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.6645, 2.3897, 2.1529, 2.6756, 2.2682, 2.2289, 2.0925, 2.5521], device='cuda:1'), covar=tensor([0.0974, 0.1631, 0.1473, 0.1159, 0.1473, 0.0553, 0.1401, 0.0718], device='cuda:1'), in_proj_covar=tensor([0.0268, 0.0354, 0.0312, 0.0250, 0.0300, 0.0250, 0.0309, 0.0255], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 22:23:52,928 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=149256.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 22:24:19,399 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.7538, 1.5729, 1.5922, 2.1792, 1.6259, 2.0784, 2.1321, 1.8179], device='cuda:1'), covar=tensor([0.0861, 0.0991, 0.1034, 0.0804, 0.0938, 0.0735, 0.0826, 0.0706], device='cuda:1'), in_proj_covar=tensor([0.0214, 0.0223, 0.0226, 0.0241, 0.0229, 0.0213, 0.0187, 0.0205], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-04-02 22:24:30,286 INFO [train.py:903] (1/4) Epoch 22, batch 5900, loss[loss=0.1984, simple_loss=0.2802, pruned_loss=0.05836, over 19846.00 frames. ], tot_loss[loss=0.2077, simple_loss=0.2874, pruned_loss=0.06405, over 3830493.56 frames. ], batch size: 52, lr: 3.71e-03, grad_scale: 4.0 2023-04-02 22:24:35,580 WARNING [train.py:1073] (1/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 2023-04-02 22:24:56,377 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9_rvb from training. Duration: 27.511125 2023-04-02 22:25:31,743 INFO [train.py:903] (1/4) Epoch 22, batch 5950, loss[loss=0.2853, simple_loss=0.3481, pruned_loss=0.1112, over 19787.00 frames. ], tot_loss[loss=0.2071, simple_loss=0.2865, pruned_loss=0.06382, over 3835156.17 frames. ], batch size: 56, lr: 3.71e-03, grad_scale: 4.0 2023-04-02 22:25:32,889 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.792e+02 4.957e+02 5.985e+02 7.132e+02 1.534e+03, threshold=1.197e+03, percent-clipped=1.0 2023-04-02 22:26:26,172 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.13 vs. limit=2.0 2023-04-02 22:26:28,504 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=149384.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 22:26:33,220 INFO [train.py:903] (1/4) Epoch 22, batch 6000, loss[loss=0.2604, simple_loss=0.3259, pruned_loss=0.09746, over 13451.00 frames. ], tot_loss[loss=0.2071, simple_loss=0.2867, pruned_loss=0.06369, over 3831150.10 frames. ], batch size: 136, lr: 3.71e-03, grad_scale: 8.0 2023-04-02 22:26:33,220 INFO [train.py:928] (1/4) Computing validation loss 2023-04-02 22:26:40,115 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4093, 1.4842, 1.4390, 1.8287, 1.4789, 1.6622, 1.6477, 1.5987], device='cuda:1'), covar=tensor([0.1012, 0.0960, 0.1066, 0.0686, 0.0960, 0.0863, 0.0993, 0.0732], device='cuda:1'), in_proj_covar=tensor([0.0214, 0.0223, 0.0226, 0.0241, 0.0229, 0.0214, 0.0188, 0.0205], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-04-02 22:26:43,948 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.0791, 3.5143, 3.5901, 3.6154, 1.9995, 3.2660, 3.1611, 3.3976], device='cuda:1'), covar=tensor([0.1512, 0.0705, 0.0618, 0.0625, 0.4855, 0.1120, 0.0622, 0.0934], device='cuda:1'), in_proj_covar=tensor([0.0783, 0.0743, 0.0949, 0.0833, 0.0835, 0.0714, 0.0566, 0.0883], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-02 22:26:46,899 INFO [train.py:937] (1/4) Epoch 22, validation: loss=0.1681, simple_loss=0.2682, pruned_loss=0.03398, over 944034.00 frames. 2023-04-02 22:26:46,900 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18821MB 2023-04-02 22:27:13,578 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=149409.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 22:27:48,666 INFO [train.py:903] (1/4) Epoch 22, batch 6050, loss[loss=0.1637, simple_loss=0.2482, pruned_loss=0.03959, over 19401.00 frames. ], tot_loss[loss=0.2053, simple_loss=0.2854, pruned_loss=0.06258, over 3838344.81 frames. ], batch size: 48, lr: 3.71e-03, grad_scale: 8.0 2023-04-02 22:27:49,809 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.518e+02 4.880e+02 5.766e+02 7.280e+02 1.810e+03, threshold=1.153e+03, percent-clipped=3.0 2023-04-02 22:27:52,759 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.65 vs. limit=5.0 2023-04-02 22:28:02,643 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=149449.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 22:28:07,324 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5495, 2.2309, 1.6113, 1.5975, 2.0967, 1.3646, 1.4690, 1.9527], device='cuda:1'), covar=tensor([0.1090, 0.0933, 0.1177, 0.0816, 0.0524, 0.1317, 0.0777, 0.0563], device='cuda:1'), in_proj_covar=tensor([0.0304, 0.0318, 0.0340, 0.0266, 0.0248, 0.0338, 0.0291, 0.0276], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 22:28:32,585 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=149474.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 22:28:50,329 INFO [train.py:903] (1/4) Epoch 22, batch 6100, loss[loss=0.2116, simple_loss=0.2995, pruned_loss=0.06191, over 19775.00 frames. ], tot_loss[loss=0.2056, simple_loss=0.286, pruned_loss=0.06258, over 3840365.36 frames. ], batch size: 56, lr: 3.71e-03, grad_scale: 8.0 2023-04-02 22:28:59,850 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9820, 1.8919, 1.7321, 1.5636, 1.4914, 1.5596, 0.3658, 0.8752], device='cuda:1'), covar=tensor([0.0659, 0.0674, 0.0473, 0.0737, 0.1299, 0.0893, 0.1399, 0.1179], device='cuda:1'), in_proj_covar=tensor([0.0356, 0.0355, 0.0358, 0.0381, 0.0462, 0.0389, 0.0336, 0.0341], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 22:29:02,457 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.56 vs. limit=5.0 2023-04-02 22:29:42,763 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=149532.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 22:29:48,971 INFO [train.py:903] (1/4) Epoch 22, batch 6150, loss[loss=0.182, simple_loss=0.2565, pruned_loss=0.05376, over 19756.00 frames. ], tot_loss[loss=0.2062, simple_loss=0.2863, pruned_loss=0.06308, over 3825131.97 frames. ], batch size: 46, lr: 3.71e-03, grad_scale: 8.0 2023-04-02 22:29:49,331 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5040, 1.4542, 1.4543, 1.7888, 1.2180, 1.6675, 1.6835, 1.6102], device='cuda:1'), covar=tensor([0.0848, 0.0925, 0.0968, 0.0678, 0.0917, 0.0798, 0.0863, 0.0702], device='cuda:1'), in_proj_covar=tensor([0.0212, 0.0222, 0.0224, 0.0240, 0.0227, 0.0212, 0.0187, 0.0204], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-02 22:29:50,024 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.191e+02 4.761e+02 6.063e+02 7.648e+02 1.908e+03, threshold=1.213e+03, percent-clipped=8.0 2023-04-02 22:29:50,467 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.7446, 2.4478, 2.2548, 2.6804, 2.3302, 2.2897, 2.1898, 2.6562], device='cuda:1'), covar=tensor([0.0890, 0.1557, 0.1398, 0.1064, 0.1385, 0.0512, 0.1314, 0.0657], device='cuda:1'), in_proj_covar=tensor([0.0269, 0.0354, 0.0312, 0.0250, 0.0301, 0.0250, 0.0309, 0.0255], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 22:30:15,360 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.6466, 2.8775, 3.0728, 3.0651, 1.7322, 2.8727, 2.6259, 2.8943], device='cuda:1'), covar=tensor([0.1362, 0.3088, 0.0657, 0.0790, 0.4176, 0.1534, 0.0630, 0.1003], device='cuda:1'), in_proj_covar=tensor([0.0779, 0.0740, 0.0943, 0.0827, 0.0831, 0.0708, 0.0563, 0.0877], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-02 22:30:19,811 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9_rvb from training. Duration: 31.02225 2023-04-02 22:30:29,460 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=4.93 vs. limit=5.0 2023-04-02 22:30:43,281 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=149583.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 22:30:49,419 INFO [train.py:903] (1/4) Epoch 22, batch 6200, loss[loss=0.1942, simple_loss=0.2845, pruned_loss=0.052, over 17959.00 frames. ], tot_loss[loss=0.2074, simple_loss=0.2874, pruned_loss=0.06365, over 3819611.58 frames. ], batch size: 83, lr: 3.71e-03, grad_scale: 8.0 2023-04-02 22:31:51,294 INFO [train.py:903] (1/4) Epoch 22, batch 6250, loss[loss=0.2085, simple_loss=0.2967, pruned_loss=0.06011, over 19544.00 frames. ], tot_loss[loss=0.2074, simple_loss=0.2876, pruned_loss=0.06361, over 3796598.80 frames. ], batch size: 54, lr: 3.70e-03, grad_scale: 8.0 2023-04-02 22:31:52,383 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.896e+02 5.068e+02 6.178e+02 8.268e+02 1.694e+03, threshold=1.236e+03, percent-clipped=5.0 2023-04-02 22:32:21,864 WARNING [train.py:1073] (1/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9_rvb from training. Duration: 25.988875 2023-04-02 22:32:52,464 INFO [train.py:903] (1/4) Epoch 22, batch 6300, loss[loss=0.1999, simple_loss=0.2875, pruned_loss=0.05615, over 19566.00 frames. ], tot_loss[loss=0.2087, simple_loss=0.2887, pruned_loss=0.06437, over 3783776.07 frames. ], batch size: 61, lr: 3.70e-03, grad_scale: 8.0 2023-04-02 22:32:59,448 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.7285, 1.6538, 1.5907, 2.1449, 1.5104, 2.0486, 2.0808, 1.8577], device='cuda:1'), covar=tensor([0.0842, 0.0941, 0.0980, 0.0796, 0.0935, 0.0734, 0.0870, 0.0694], device='cuda:1'), in_proj_covar=tensor([0.0211, 0.0220, 0.0222, 0.0238, 0.0226, 0.0211, 0.0186, 0.0203], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-02 22:33:51,829 INFO [train.py:903] (1/4) Epoch 22, batch 6350, loss[loss=0.1771, simple_loss=0.249, pruned_loss=0.05265, over 19736.00 frames. ], tot_loss[loss=0.2082, simple_loss=0.288, pruned_loss=0.06418, over 3795373.76 frames. ], batch size: 46, lr: 3.70e-03, grad_scale: 8.0 2023-04-02 22:33:52,935 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.101e+02 5.551e+02 6.531e+02 8.044e+02 1.579e+03, threshold=1.306e+03, percent-clipped=6.0 2023-04-02 22:34:23,920 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=149764.0, num_to_drop=1, layers_to_drop={1} 2023-04-02 22:34:52,496 INFO [train.py:903] (1/4) Epoch 22, batch 6400, loss[loss=0.2181, simple_loss=0.2997, pruned_loss=0.06827, over 17251.00 frames. ], tot_loss[loss=0.2087, simple_loss=0.2884, pruned_loss=0.06448, over 3785350.83 frames. ], batch size: 101, lr: 3.70e-03, grad_scale: 8.0 2023-04-02 22:35:54,292 INFO [train.py:903] (1/4) Epoch 22, batch 6450, loss[loss=0.2151, simple_loss=0.2952, pruned_loss=0.06755, over 19753.00 frames. ], tot_loss[loss=0.2093, simple_loss=0.2892, pruned_loss=0.06473, over 3783558.55 frames. ], batch size: 63, lr: 3.70e-03, grad_scale: 8.0 2023-04-02 22:35:55,271 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.826e+02 4.788e+02 5.699e+02 7.070e+02 1.580e+03, threshold=1.140e+03, percent-clipped=2.0 2023-04-02 22:36:39,133 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9_rvb from training. Duration: 28.638875 2023-04-02 22:36:40,524 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=149876.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 22:36:54,745 INFO [train.py:903] (1/4) Epoch 22, batch 6500, loss[loss=0.2511, simple_loss=0.3315, pruned_loss=0.0853, over 19664.00 frames. ], tot_loss[loss=0.2096, simple_loss=0.289, pruned_loss=0.06506, over 3787583.70 frames. ], batch size: 55, lr: 3.70e-03, grad_scale: 8.0 2023-04-02 22:37:00,223 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1_rvb from training. Duration: 0.836375 2023-04-02 22:37:25,307 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.1568, 4.3140, 4.7686, 4.7901, 2.8512, 4.4191, 4.0499, 4.5034], device='cuda:1'), covar=tensor([0.1408, 0.2473, 0.0596, 0.0619, 0.4160, 0.1078, 0.0616, 0.0990], device='cuda:1'), in_proj_covar=tensor([0.0784, 0.0743, 0.0950, 0.0827, 0.0835, 0.0710, 0.0565, 0.0879], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-02 22:37:41,914 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=149927.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 22:37:55,100 INFO [train.py:903] (1/4) Epoch 22, batch 6550, loss[loss=0.1717, simple_loss=0.2546, pruned_loss=0.0444, over 19596.00 frames. ], tot_loss[loss=0.2094, simple_loss=0.2892, pruned_loss=0.06474, over 3794266.44 frames. ], batch size: 52, lr: 3.70e-03, grad_scale: 8.0 2023-04-02 22:37:56,255 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.115e+02 4.654e+02 5.933e+02 7.304e+02 1.667e+03, threshold=1.187e+03, percent-clipped=4.0 2023-04-02 22:38:45,123 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.1389, 1.9680, 1.7645, 2.1136, 1.8778, 1.7806, 1.7678, 1.9940], device='cuda:1'), covar=tensor([0.0963, 0.1431, 0.1406, 0.0972, 0.1320, 0.0552, 0.1302, 0.0726], device='cuda:1'), in_proj_covar=tensor([0.0270, 0.0354, 0.0312, 0.0250, 0.0302, 0.0251, 0.0308, 0.0256], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 22:38:55,839 INFO [train.py:903] (1/4) Epoch 22, batch 6600, loss[loss=0.2087, simple_loss=0.2929, pruned_loss=0.06221, over 17548.00 frames. ], tot_loss[loss=0.2069, simple_loss=0.2869, pruned_loss=0.06342, over 3798973.22 frames. ], batch size: 101, lr: 3.70e-03, grad_scale: 8.0 2023-04-02 22:38:59,623 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=149991.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 22:39:59,884 INFO [train.py:903] (1/4) Epoch 22, batch 6650, loss[loss=0.2122, simple_loss=0.2987, pruned_loss=0.06286, over 19544.00 frames. ], tot_loss[loss=0.2067, simple_loss=0.287, pruned_loss=0.06326, over 3811233.09 frames. ], batch size: 56, lr: 3.70e-03, grad_scale: 8.0 2023-04-02 22:40:01,054 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.517e+02 4.673e+02 5.867e+02 7.414e+02 1.313e+03, threshold=1.173e+03, percent-clipped=2.0 2023-04-02 22:40:04,739 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=150042.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 22:40:59,358 INFO [train.py:903] (1/4) Epoch 22, batch 6700, loss[loss=0.2154, simple_loss=0.3055, pruned_loss=0.06261, over 19778.00 frames. ], tot_loss[loss=0.2077, simple_loss=0.2878, pruned_loss=0.06383, over 3810735.37 frames. ], batch size: 56, lr: 3.70e-03, grad_scale: 8.0 2023-04-02 22:41:10,389 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=150097.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 22:41:10,939 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.82 vs. limit=2.0 2023-04-02 22:41:23,647 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=150108.0, num_to_drop=1, layers_to_drop={0} 2023-04-02 22:41:57,264 INFO [train.py:903] (1/4) Epoch 22, batch 6750, loss[loss=0.1642, simple_loss=0.2484, pruned_loss=0.04003, over 19828.00 frames. ], tot_loss[loss=0.2075, simple_loss=0.2874, pruned_loss=0.06375, over 3811375.19 frames. ], batch size: 52, lr: 3.70e-03, grad_scale: 8.0 2023-04-02 22:41:58,374 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.145e+02 4.879e+02 6.504e+02 7.654e+02 1.720e+03, threshold=1.301e+03, percent-clipped=5.0 2023-04-02 22:42:17,387 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.7468, 2.5236, 2.2551, 2.7125, 2.3468, 1.9777, 2.1510, 2.7049], device='cuda:1'), covar=tensor([0.0915, 0.1562, 0.1434, 0.1018, 0.1423, 0.0706, 0.1473, 0.0650], device='cuda:1'), in_proj_covar=tensor([0.0269, 0.0352, 0.0312, 0.0249, 0.0301, 0.0250, 0.0308, 0.0255], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 22:42:53,133 INFO [train.py:903] (1/4) Epoch 22, batch 6800, loss[loss=0.2197, simple_loss=0.2979, pruned_loss=0.07077, over 19443.00 frames. ], tot_loss[loss=0.208, simple_loss=0.2877, pruned_loss=0.06414, over 3812725.19 frames. ], batch size: 70, lr: 3.70e-03, grad_scale: 8.0 2023-04-02 22:43:38,752 WARNING [train.py:1073] (1/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 2023-04-02 22:43:39,212 WARNING [train.py:1073] (1/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 2023-04-02 22:43:42,629 INFO [train.py:903] (1/4) Epoch 23, batch 0, loss[loss=0.2153, simple_loss=0.2952, pruned_loss=0.06771, over 19039.00 frames. ], tot_loss[loss=0.2153, simple_loss=0.2952, pruned_loss=0.06771, over 19039.00 frames. ], batch size: 69, lr: 3.61e-03, grad_scale: 8.0 2023-04-02 22:43:42,630 INFO [train.py:928] (1/4) Computing validation loss 2023-04-02 22:43:54,256 INFO [train.py:937] (1/4) Epoch 23, validation: loss=0.1688, simple_loss=0.2693, pruned_loss=0.03418, over 944034.00 frames. 2023-04-02 22:43:54,257 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18821MB 2023-04-02 22:43:54,666 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4582, 1.6005, 2.0530, 1.7057, 3.3077, 2.6561, 3.5596, 1.6921], device='cuda:1'), covar=tensor([0.2489, 0.4191, 0.2533, 0.1879, 0.1352, 0.2004, 0.1414, 0.3989], device='cuda:1'), in_proj_covar=tensor([0.0536, 0.0643, 0.0714, 0.0483, 0.0620, 0.0531, 0.0662, 0.0548], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 22:44:03,498 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=150223.0, num_to_drop=1, layers_to_drop={0} 2023-04-02 22:44:06,626 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 2023-04-02 22:44:21,436 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.019e+02 4.848e+02 5.561e+02 7.527e+02 1.735e+03, threshold=1.112e+03, percent-clipped=5.0 2023-04-02 22:44:31,849 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=150247.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 22:44:55,828 INFO [train.py:903] (1/4) Epoch 23, batch 50, loss[loss=0.2111, simple_loss=0.2905, pruned_loss=0.06591, over 19674.00 frames. ], tot_loss[loss=0.2056, simple_loss=0.2865, pruned_loss=0.06235, over 858439.33 frames. ], batch size: 58, lr: 3.61e-03, grad_scale: 8.0 2023-04-02 22:45:03,070 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=150272.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 22:45:27,176 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=150292.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 22:45:30,282 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1_rvb from training. Duration: 27.0318125 2023-04-02 22:45:36,450 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=150298.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 22:45:57,949 INFO [train.py:903] (1/4) Epoch 23, batch 100, loss[loss=0.2134, simple_loss=0.2974, pruned_loss=0.06474, over 19726.00 frames. ], tot_loss[loss=0.2042, simple_loss=0.2856, pruned_loss=0.06142, over 1524057.76 frames. ], batch size: 59, lr: 3.61e-03, grad_scale: 8.0 2023-04-02 22:46:06,459 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=150323.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 22:46:07,253 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 2023-04-02 22:46:26,516 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.179e+02 4.910e+02 5.630e+02 7.676e+02 1.557e+03, threshold=1.126e+03, percent-clipped=7.0 2023-04-02 22:46:59,548 INFO [train.py:903] (1/4) Epoch 23, batch 150, loss[loss=0.2, simple_loss=0.282, pruned_loss=0.05904, over 19613.00 frames. ], tot_loss[loss=0.2041, simple_loss=0.285, pruned_loss=0.06159, over 2048709.22 frames. ], batch size: 57, lr: 3.61e-03, grad_scale: 8.0 2023-04-02 22:47:22,284 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.8060, 4.3865, 2.7556, 3.8423, 0.9441, 4.3052, 4.2113, 4.3422], device='cuda:1'), covar=tensor([0.0569, 0.0872, 0.1918, 0.0798, 0.3969, 0.0621, 0.0851, 0.1000], device='cuda:1'), in_proj_covar=tensor([0.0503, 0.0411, 0.0496, 0.0346, 0.0399, 0.0432, 0.0425, 0.0459], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 22:47:23,512 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.8490, 1.7981, 1.8148, 2.2254, 1.8026, 2.0418, 2.1080, 1.9804], device='cuda:1'), covar=tensor([0.0761, 0.0820, 0.0855, 0.0723, 0.0859, 0.0797, 0.0866, 0.0608], device='cuda:1'), in_proj_covar=tensor([0.0212, 0.0222, 0.0222, 0.0240, 0.0227, 0.0213, 0.0187, 0.0204], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-02 22:47:59,882 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 2023-04-02 22:48:00,288 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6365, 1.5468, 1.5834, 2.0532, 1.5519, 1.8694, 1.9324, 1.7492], device='cuda:1'), covar=tensor([0.0839, 0.0962, 0.0972, 0.0764, 0.0852, 0.0823, 0.0917, 0.0698], device='cuda:1'), in_proj_covar=tensor([0.0211, 0.0221, 0.0222, 0.0239, 0.0226, 0.0212, 0.0187, 0.0203], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-02 22:48:01,003 INFO [train.py:903] (1/4) Epoch 23, batch 200, loss[loss=0.1975, simple_loss=0.2782, pruned_loss=0.05846, over 19758.00 frames. ], tot_loss[loss=0.2062, simple_loss=0.2867, pruned_loss=0.06281, over 2453725.38 frames. ], batch size: 46, lr: 3.61e-03, grad_scale: 8.0 2023-04-02 22:48:01,414 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6136, 1.5548, 1.5556, 1.9976, 1.5392, 1.8872, 1.8693, 1.7551], device='cuda:1'), covar=tensor([0.0870, 0.0950, 0.0998, 0.0801, 0.0873, 0.0804, 0.0942, 0.0709], device='cuda:1'), in_proj_covar=tensor([0.0211, 0.0221, 0.0222, 0.0239, 0.0226, 0.0212, 0.0187, 0.0203], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-02 22:48:26,087 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.41 vs. limit=2.0 2023-04-02 22:48:30,849 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.764e+02 5.350e+02 6.653e+02 9.712e+02 2.771e+03, threshold=1.331e+03, percent-clipped=16.0 2023-04-02 22:48:33,192 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=150441.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 22:48:41,449 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5062, 2.2226, 1.6677, 1.5334, 2.1116, 1.4018, 1.3898, 1.8680], device='cuda:1'), covar=tensor([0.1141, 0.0844, 0.1089, 0.0796, 0.0567, 0.1188, 0.0810, 0.0585], device='cuda:1'), in_proj_covar=tensor([0.0304, 0.0317, 0.0339, 0.0267, 0.0247, 0.0336, 0.0291, 0.0274], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 22:49:02,669 INFO [train.py:903] (1/4) Epoch 23, batch 250, loss[loss=0.1981, simple_loss=0.2631, pruned_loss=0.06657, over 19752.00 frames. ], tot_loss[loss=0.2079, simple_loss=0.2878, pruned_loss=0.064, over 2760206.11 frames. ], batch size: 46, lr: 3.61e-03, grad_scale: 8.0 2023-04-02 22:49:20,004 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=150479.0, num_to_drop=1, layers_to_drop={1} 2023-04-02 22:49:36,997 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.9191, 4.3877, 4.6664, 4.6571, 1.8415, 4.3881, 3.8672, 4.3962], device='cuda:1'), covar=tensor([0.1673, 0.0792, 0.0613, 0.0647, 0.6020, 0.0839, 0.0630, 0.1142], device='cuda:1'), in_proj_covar=tensor([0.0784, 0.0744, 0.0951, 0.0832, 0.0836, 0.0712, 0.0566, 0.0878], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-02 22:49:49,524 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.3061, 3.0140, 2.1894, 2.6761, 0.7278, 2.9867, 2.8731, 3.0049], device='cuda:1'), covar=tensor([0.1172, 0.1393, 0.2134, 0.1156, 0.3926, 0.0964, 0.1158, 0.1353], device='cuda:1'), in_proj_covar=tensor([0.0504, 0.0412, 0.0497, 0.0345, 0.0400, 0.0433, 0.0425, 0.0460], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 22:49:49,695 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=150504.0, num_to_drop=1, layers_to_drop={0} 2023-04-02 22:50:05,858 INFO [train.py:903] (1/4) Epoch 23, batch 300, loss[loss=0.241, simple_loss=0.3225, pruned_loss=0.07969, over 19506.00 frames. ], tot_loss[loss=0.2072, simple_loss=0.2872, pruned_loss=0.06363, over 3007829.08 frames. ], batch size: 64, lr: 3.61e-03, grad_scale: 4.0 2023-04-02 22:50:34,494 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.361e+02 5.024e+02 5.928e+02 7.198e+02 2.066e+03, threshold=1.186e+03, percent-clipped=3.0 2023-04-02 22:50:49,307 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.0821, 4.5050, 4.8154, 4.8263, 1.9234, 4.4904, 3.9552, 4.5132], device='cuda:1'), covar=tensor([0.1616, 0.0731, 0.0572, 0.0633, 0.5686, 0.0733, 0.0643, 0.1130], device='cuda:1'), in_proj_covar=tensor([0.0788, 0.0745, 0.0955, 0.0833, 0.0838, 0.0713, 0.0568, 0.0880], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-02 22:50:54,923 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=150555.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 22:50:56,024 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=150556.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 22:51:07,068 INFO [train.py:903] (1/4) Epoch 23, batch 350, loss[loss=0.1849, simple_loss=0.2764, pruned_loss=0.04665, over 19670.00 frames. ], tot_loss[loss=0.2068, simple_loss=0.2865, pruned_loss=0.06354, over 3186207.46 frames. ], batch size: 58, lr: 3.61e-03, grad_scale: 4.0 2023-04-02 22:51:11,927 WARNING [train.py:1073] (1/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1_rvb from training. Duration: 0.9681875 2023-04-02 22:52:09,968 INFO [train.py:903] (1/4) Epoch 23, batch 400, loss[loss=0.2083, simple_loss=0.272, pruned_loss=0.07233, over 19778.00 frames. ], tot_loss[loss=0.2075, simple_loss=0.2867, pruned_loss=0.06412, over 3325213.49 frames. ], batch size: 46, lr: 3.61e-03, grad_scale: 8.0 2023-04-02 22:52:17,553 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3179, 1.9104, 1.5066, 1.3525, 1.7944, 1.3055, 1.3505, 1.7563], device='cuda:1'), covar=tensor([0.0929, 0.0894, 0.0960, 0.0801, 0.0532, 0.1163, 0.0656, 0.0458], device='cuda:1'), in_proj_covar=tensor([0.0304, 0.0316, 0.0339, 0.0266, 0.0246, 0.0336, 0.0290, 0.0274], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 22:52:36,591 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=150636.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 22:52:40,893 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.962e+02 5.093e+02 6.486e+02 8.008e+02 1.724e+03, threshold=1.297e+03, percent-clipped=3.0 2023-04-02 22:53:11,935 INFO [train.py:903] (1/4) Epoch 23, batch 450, loss[loss=0.2166, simple_loss=0.287, pruned_loss=0.07312, over 19464.00 frames. ], tot_loss[loss=0.2062, simple_loss=0.2855, pruned_loss=0.06346, over 3447235.69 frames. ], batch size: 49, lr: 3.61e-03, grad_scale: 8.0 2023-04-02 22:53:46,053 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 2023-04-02 22:53:46,076 WARNING [train.py:1073] (1/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 2023-04-02 22:54:13,701 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.7856, 2.1711, 1.6990, 1.6663, 2.0596, 1.5575, 1.6857, 1.9501], device='cuda:1'), covar=tensor([0.0862, 0.0698, 0.0778, 0.0727, 0.0493, 0.1066, 0.0582, 0.0476], device='cuda:1'), in_proj_covar=tensor([0.0302, 0.0315, 0.0339, 0.0265, 0.0246, 0.0336, 0.0289, 0.0273], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 22:54:15,720 INFO [train.py:903] (1/4) Epoch 23, batch 500, loss[loss=0.2313, simple_loss=0.3088, pruned_loss=0.0769, over 19118.00 frames. ], tot_loss[loss=0.2064, simple_loss=0.2855, pruned_loss=0.06366, over 3497281.78 frames. ], batch size: 69, lr: 3.61e-03, grad_scale: 8.0 2023-04-02 22:54:45,187 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.118e+02 5.191e+02 6.635e+02 8.528e+02 2.142e+03, threshold=1.327e+03, percent-clipped=5.0 2023-04-02 22:54:57,353 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=150750.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 22:54:58,656 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=150751.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 22:55:17,063 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.62 vs. limit=5.0 2023-04-02 22:55:17,408 INFO [train.py:903] (1/4) Epoch 23, batch 550, loss[loss=0.2072, simple_loss=0.2881, pruned_loss=0.06313, over 19608.00 frames. ], tot_loss[loss=0.2069, simple_loss=0.2861, pruned_loss=0.06383, over 3573527.71 frames. ], batch size: 50, lr: 3.61e-03, grad_scale: 8.0 2023-04-02 22:55:30,133 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0081, 1.8856, 1.6760, 2.0738, 1.8415, 1.7374, 1.6942, 1.9774], device='cuda:1'), covar=tensor([0.0980, 0.1330, 0.1407, 0.0932, 0.1203, 0.0541, 0.1397, 0.0670], device='cuda:1'), in_proj_covar=tensor([0.0270, 0.0353, 0.0313, 0.0250, 0.0302, 0.0251, 0.0309, 0.0257], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 22:56:09,314 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 2023-04-02 22:56:14,500 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=150812.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 22:56:18,589 INFO [train.py:903] (1/4) Epoch 23, batch 600, loss[loss=0.1765, simple_loss=0.2517, pruned_loss=0.0507, over 19742.00 frames. ], tot_loss[loss=0.2067, simple_loss=0.2859, pruned_loss=0.06374, over 3624508.43 frames. ], batch size: 46, lr: 3.61e-03, grad_scale: 8.0 2023-04-02 22:56:38,343 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9837, 2.0701, 2.2431, 2.7310, 2.0512, 2.5802, 2.3051, 2.1001], device='cuda:1'), covar=tensor([0.4324, 0.3922, 0.1908, 0.2370, 0.4199, 0.2198, 0.4771, 0.3331], device='cuda:1'), in_proj_covar=tensor([0.0905, 0.0971, 0.0720, 0.0934, 0.0886, 0.0822, 0.0846, 0.0788], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 22:56:45,511 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=150837.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 22:56:48,646 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.714e+02 5.000e+02 5.859e+02 6.998e+02 1.831e+03, threshold=1.172e+03, percent-clipped=2.0 2023-04-02 22:56:59,190 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 2023-04-02 22:57:21,218 INFO [train.py:903] (1/4) Epoch 23, batch 650, loss[loss=0.2209, simple_loss=0.3004, pruned_loss=0.07076, over 19506.00 frames. ], tot_loss[loss=0.2081, simple_loss=0.2873, pruned_loss=0.06441, over 3666923.47 frames. ], batch size: 64, lr: 3.61e-03, grad_scale: 8.0 2023-04-02 22:58:02,429 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=150899.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 22:58:23,644 INFO [train.py:903] (1/4) Epoch 23, batch 700, loss[loss=0.2035, simple_loss=0.2956, pruned_loss=0.0557, over 18165.00 frames. ], tot_loss[loss=0.2079, simple_loss=0.2872, pruned_loss=0.06423, over 3711399.18 frames. ], batch size: 84, lr: 3.61e-03, grad_scale: 8.0 2023-04-02 22:58:27,264 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=150918.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 22:58:52,730 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.226e+02 5.019e+02 5.794e+02 7.164e+02 1.349e+03, threshold=1.159e+03, percent-clipped=1.0 2023-04-02 22:59:09,193 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.4405, 1.1977, 1.3721, 1.3778, 3.0010, 1.0433, 2.2872, 3.3556], device='cuda:1'), covar=tensor([0.0534, 0.2927, 0.3115, 0.1907, 0.0735, 0.2555, 0.1268, 0.0301], device='cuda:1'), in_proj_covar=tensor([0.0416, 0.0370, 0.0389, 0.0349, 0.0379, 0.0353, 0.0384, 0.0408], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 22:59:23,021 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.2727, 3.5792, 2.0785, 2.2445, 3.3300, 1.8672, 1.6874, 2.4191], device='cuda:1'), covar=tensor([0.1380, 0.0613, 0.1148, 0.0905, 0.0489, 0.1313, 0.1050, 0.0712], device='cuda:1'), in_proj_covar=tensor([0.0304, 0.0318, 0.0340, 0.0266, 0.0248, 0.0339, 0.0292, 0.0275], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 22:59:26,031 INFO [train.py:903] (1/4) Epoch 23, batch 750, loss[loss=0.1793, simple_loss=0.2487, pruned_loss=0.05492, over 19793.00 frames. ], tot_loss[loss=0.2073, simple_loss=0.2867, pruned_loss=0.06395, over 3729035.64 frames. ], batch size: 47, lr: 3.61e-03, grad_scale: 8.0 2023-04-02 22:59:51,792 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.7970, 1.8877, 2.1740, 2.3514, 1.7005, 2.2361, 2.1533, 2.0126], device='cuda:1'), covar=tensor([0.4536, 0.4005, 0.2014, 0.2388, 0.4189, 0.2176, 0.5299, 0.3474], device='cuda:1'), in_proj_covar=tensor([0.0903, 0.0969, 0.0718, 0.0933, 0.0885, 0.0820, 0.0846, 0.0785], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 23:00:17,626 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=151007.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 23:00:26,915 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=151014.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 23:00:28,660 INFO [train.py:903] (1/4) Epoch 23, batch 800, loss[loss=0.2253, simple_loss=0.294, pruned_loss=0.07832, over 19580.00 frames. ], tot_loss[loss=0.2074, simple_loss=0.2869, pruned_loss=0.06398, over 3746935.69 frames. ], batch size: 52, lr: 3.61e-03, grad_scale: 8.0 2023-04-02 23:00:46,712 WARNING [train.py:1073] (1/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 2023-04-02 23:00:48,233 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=151032.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 23:00:57,133 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.089e+02 5.089e+02 6.161e+02 7.478e+02 1.780e+03, threshold=1.232e+03, percent-clipped=6.0 2023-04-02 23:01:29,764 INFO [train.py:903] (1/4) Epoch 23, batch 850, loss[loss=0.214, simple_loss=0.2955, pruned_loss=0.06631, over 19664.00 frames. ], tot_loss[loss=0.2083, simple_loss=0.2876, pruned_loss=0.06445, over 3757580.89 frames. ], batch size: 55, lr: 3.60e-03, grad_scale: 8.0 2023-04-02 23:02:04,866 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=151094.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 23:02:25,412 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9_rvb from training. Duration: 25.061125 2023-04-02 23:02:31,809 INFO [train.py:903] (1/4) Epoch 23, batch 900, loss[loss=0.2188, simple_loss=0.2883, pruned_loss=0.07462, over 19391.00 frames. ], tot_loss[loss=0.2073, simple_loss=0.2869, pruned_loss=0.06383, over 3776418.42 frames. ], batch size: 48, lr: 3.60e-03, grad_scale: 8.0 2023-04-02 23:03:02,018 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.084e+02 5.070e+02 6.543e+02 8.443e+02 1.332e+03, threshold=1.309e+03, percent-clipped=3.0 2023-04-02 23:03:31,177 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 2023-04-02 23:03:32,679 INFO [train.py:903] (1/4) Epoch 23, batch 950, loss[loss=0.2008, simple_loss=0.2776, pruned_loss=0.06202, over 19421.00 frames. ], tot_loss[loss=0.2091, simple_loss=0.2884, pruned_loss=0.06493, over 3787335.86 frames. ], batch size: 48, lr: 3.60e-03, grad_scale: 8.0 2023-04-02 23:03:39,545 WARNING [train.py:1073] (1/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9_rvb from training. Duration: 26.32775 2023-04-02 23:04:17,085 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 2023-04-02 23:04:17,819 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=151202.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 23:04:27,683 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=151209.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 23:04:35,170 INFO [train.py:903] (1/4) Epoch 23, batch 1000, loss[loss=0.2481, simple_loss=0.3271, pruned_loss=0.08455, over 18305.00 frames. ], tot_loss[loss=0.2098, simple_loss=0.2892, pruned_loss=0.06524, over 3785275.40 frames. ], batch size: 83, lr: 3.60e-03, grad_scale: 8.0 2023-04-02 23:05:04,889 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.333e+02 5.146e+02 6.401e+02 7.951e+02 1.702e+03, threshold=1.280e+03, percent-clipped=4.0 2023-04-02 23:05:32,307 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 2023-04-02 23:05:33,592 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=151262.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 23:05:37,845 INFO [train.py:903] (1/4) Epoch 23, batch 1050, loss[loss=0.2153, simple_loss=0.2991, pruned_loss=0.06571, over 19702.00 frames. ], tot_loss[loss=0.2098, simple_loss=0.2894, pruned_loss=0.06514, over 3801193.81 frames. ], batch size: 59, lr: 3.60e-03, grad_scale: 8.0 2023-04-02 23:05:42,821 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=151270.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 23:06:12,810 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 2023-04-02 23:06:14,315 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=151295.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 23:06:21,334 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=151300.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 23:06:40,101 INFO [train.py:903] (1/4) Epoch 23, batch 1100, loss[loss=0.2044, simple_loss=0.2954, pruned_loss=0.05671, over 19697.00 frames. ], tot_loss[loss=0.2092, simple_loss=0.2888, pruned_loss=0.06473, over 3819186.78 frames. ], batch size: 58, lr: 3.60e-03, grad_scale: 8.0 2023-04-02 23:06:40,413 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=151316.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 23:07:09,143 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.937e+02 5.070e+02 6.152e+02 7.617e+02 1.362e+03, threshold=1.230e+03, percent-clipped=2.0 2023-04-02 23:07:40,866 INFO [train.py:903] (1/4) Epoch 23, batch 1150, loss[loss=0.2367, simple_loss=0.3248, pruned_loss=0.07432, over 19610.00 frames. ], tot_loss[loss=0.2087, simple_loss=0.2884, pruned_loss=0.06452, over 3829385.08 frames. ], batch size: 61, lr: 3.60e-03, grad_scale: 4.0 2023-04-02 23:07:55,722 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=151377.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 23:08:43,930 INFO [train.py:903] (1/4) Epoch 23, batch 1200, loss[loss=0.2331, simple_loss=0.312, pruned_loss=0.07705, over 19701.00 frames. ], tot_loss[loss=0.2098, simple_loss=0.2897, pruned_loss=0.06494, over 3816772.05 frames. ], batch size: 59, lr: 3.60e-03, grad_scale: 8.0 2023-04-02 23:09:10,476 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.8778, 2.6287, 2.5271, 3.0130, 2.7340, 2.4210, 2.3868, 2.9571], device='cuda:1'), covar=tensor([0.0848, 0.1528, 0.1318, 0.0961, 0.1227, 0.0484, 0.1251, 0.0551], device='cuda:1'), in_proj_covar=tensor([0.0272, 0.0355, 0.0314, 0.0251, 0.0303, 0.0252, 0.0309, 0.0256], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 23:09:14,769 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.106e+02 4.896e+02 6.001e+02 7.643e+02 1.247e+03, threshold=1.200e+03, percent-clipped=2.0 2023-04-02 23:09:18,073 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 2023-04-02 23:09:45,306 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=151465.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 23:09:46,022 INFO [train.py:903] (1/4) Epoch 23, batch 1250, loss[loss=0.1926, simple_loss=0.2731, pruned_loss=0.05603, over 19585.00 frames. ], tot_loss[loss=0.209, simple_loss=0.2883, pruned_loss=0.06483, over 3820874.72 frames. ], batch size: 52, lr: 3.60e-03, grad_scale: 8.0 2023-04-02 23:10:16,512 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=151490.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 23:10:27,794 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 2023-04-02 23:10:46,707 INFO [train.py:903] (1/4) Epoch 23, batch 1300, loss[loss=0.2524, simple_loss=0.3341, pruned_loss=0.0853, over 19554.00 frames. ], tot_loss[loss=0.2101, simple_loss=0.2893, pruned_loss=0.06539, over 3816754.21 frames. ], batch size: 61, lr: 3.60e-03, grad_scale: 8.0 2023-04-02 23:10:55,136 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4184, 1.5499, 1.5750, 1.9179, 1.5031, 1.7931, 1.7008, 1.3619], device='cuda:1'), covar=tensor([0.4805, 0.4208, 0.2808, 0.2871, 0.4131, 0.2526, 0.6279, 0.5089], device='cuda:1'), in_proj_covar=tensor([0.0906, 0.0973, 0.0719, 0.0936, 0.0884, 0.0817, 0.0844, 0.0787], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 23:10:55,327 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 2023-04-02 23:11:16,029 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.826e+02 5.118e+02 6.046e+02 8.032e+02 1.744e+03, threshold=1.209e+03, percent-clipped=5.0 2023-04-02 23:11:22,847 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=151546.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 23:11:25,506 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.7821, 1.8346, 2.0014, 1.8990, 2.6440, 2.3226, 2.6609, 1.6912], device='cuda:1'), covar=tensor([0.1972, 0.3350, 0.2225, 0.1630, 0.1228, 0.1762, 0.1249, 0.3722], device='cuda:1'), in_proj_covar=tensor([0.0540, 0.0650, 0.0719, 0.0490, 0.0622, 0.0534, 0.0668, 0.0555], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 23:11:46,631 INFO [train.py:903] (1/4) Epoch 23, batch 1350, loss[loss=0.1746, simple_loss=0.2471, pruned_loss=0.05107, over 19768.00 frames. ], tot_loss[loss=0.2102, simple_loss=0.2893, pruned_loss=0.06552, over 3804780.64 frames. ], batch size: 47, lr: 3.60e-03, grad_scale: 8.0 2023-04-02 23:12:48,223 INFO [train.py:903] (1/4) Epoch 23, batch 1400, loss[loss=0.1768, simple_loss=0.2621, pruned_loss=0.04579, over 19396.00 frames. ], tot_loss[loss=0.2092, simple_loss=0.2885, pruned_loss=0.06493, over 3804109.82 frames. ], batch size: 48, lr: 3.60e-03, grad_scale: 8.0 2023-04-02 23:13:08,447 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=151633.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 23:13:17,804 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.750e+02 5.043e+02 6.240e+02 8.237e+02 1.280e+03, threshold=1.248e+03, percent-clipped=3.0 2023-04-02 23:13:21,398 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=151644.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 23:13:38,321 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=151658.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 23:13:40,483 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=151660.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 23:13:41,856 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=151661.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 23:13:48,344 INFO [train.py:903] (1/4) Epoch 23, batch 1450, loss[loss=0.2022, simple_loss=0.2845, pruned_loss=0.06, over 19671.00 frames. ], tot_loss[loss=0.2083, simple_loss=0.2878, pruned_loss=0.06439, over 3812182.74 frames. ], batch size: 53, lr: 3.60e-03, grad_scale: 8.0 2023-04-02 23:13:48,375 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_rvb from training. Duration: 25.85 2023-04-02 23:14:36,980 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=151706.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 23:14:49,056 INFO [train.py:903] (1/4) Epoch 23, batch 1500, loss[loss=0.1839, simple_loss=0.2707, pruned_loss=0.04858, over 19671.00 frames. ], tot_loss[loss=0.209, simple_loss=0.2886, pruned_loss=0.06469, over 3802191.18 frames. ], batch size: 58, lr: 3.60e-03, grad_scale: 8.0 2023-04-02 23:15:18,488 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.327e+02 4.905e+02 6.054e+02 7.299e+02 2.065e+03, threshold=1.211e+03, percent-clipped=4.0 2023-04-02 23:15:40,398 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=151759.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 23:15:47,608 INFO [train.py:903] (1/4) Epoch 23, batch 1550, loss[loss=0.2025, simple_loss=0.2836, pruned_loss=0.06069, over 19530.00 frames. ], tot_loss[loss=0.2092, simple_loss=0.2888, pruned_loss=0.06478, over 3809072.23 frames. ], batch size: 54, lr: 3.60e-03, grad_scale: 8.0 2023-04-02 23:15:59,927 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=151775.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 23:16:50,042 INFO [train.py:903] (1/4) Epoch 23, batch 1600, loss[loss=0.2046, simple_loss=0.2823, pruned_loss=0.06349, over 19613.00 frames. ], tot_loss[loss=0.2085, simple_loss=0.2881, pruned_loss=0.06443, over 3789491.42 frames. ], batch size: 50, lr: 3.60e-03, grad_scale: 8.0 2023-04-02 23:16:53,591 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=151819.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 23:17:10,285 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9_rvb from training. Duration: 30.1555625 2023-04-02 23:17:20,214 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.366e+02 4.863e+02 5.887e+02 6.951e+02 2.426e+03, threshold=1.177e+03, percent-clipped=3.0 2023-04-02 23:17:50,185 INFO [train.py:903] (1/4) Epoch 23, batch 1650, loss[loss=0.1794, simple_loss=0.2694, pruned_loss=0.04472, over 19790.00 frames. ], tot_loss[loss=0.2092, simple_loss=0.2888, pruned_loss=0.06477, over 3793398.36 frames. ], batch size: 56, lr: 3.60e-03, grad_scale: 8.0 2023-04-02 23:18:39,685 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1123, 1.2647, 1.6793, 1.2563, 2.7544, 3.6869, 3.4290, 3.9614], device='cuda:1'), covar=tensor([0.1842, 0.3973, 0.3570, 0.2645, 0.0646, 0.0219, 0.0235, 0.0272], device='cuda:1'), in_proj_covar=tensor([0.0275, 0.0323, 0.0355, 0.0265, 0.0244, 0.0189, 0.0217, 0.0260], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-02 23:18:51,776 INFO [train.py:903] (1/4) Epoch 23, batch 1700, loss[loss=0.1917, simple_loss=0.2811, pruned_loss=0.05115, over 19733.00 frames. ], tot_loss[loss=0.2089, simple_loss=0.2883, pruned_loss=0.06475, over 3784139.88 frames. ], batch size: 63, lr: 3.59e-03, grad_scale: 8.0 2023-04-02 23:18:53,362 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=151917.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 23:19:21,471 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.136e+02 5.018e+02 6.073e+02 7.613e+02 1.748e+03, threshold=1.215e+03, percent-clipped=5.0 2023-04-02 23:19:23,163 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=151942.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 23:19:26,196 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 2023-04-02 23:19:52,379 INFO [train.py:903] (1/4) Epoch 23, batch 1750, loss[loss=0.254, simple_loss=0.3194, pruned_loss=0.09432, over 13248.00 frames. ], tot_loss[loss=0.2083, simple_loss=0.2879, pruned_loss=0.06438, over 3781077.05 frames. ], batch size: 135, lr: 3.59e-03, grad_scale: 8.0 2023-04-02 23:20:53,697 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=152015.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 23:20:54,416 INFO [train.py:903] (1/4) Epoch 23, batch 1800, loss[loss=0.2133, simple_loss=0.2969, pruned_loss=0.06485, over 17428.00 frames. ], tot_loss[loss=0.2072, simple_loss=0.2865, pruned_loss=0.06391, over 3783832.86 frames. ], batch size: 101, lr: 3.59e-03, grad_scale: 4.0 2023-04-02 23:21:13,138 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=152031.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 23:21:24,577 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=152040.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 23:21:26,671 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.151e+02 5.200e+02 6.601e+02 8.823e+02 1.720e+03, threshold=1.320e+03, percent-clipped=12.0 2023-04-02 23:21:36,337 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=152050.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 23:21:40,757 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2114, 1.4122, 1.7495, 1.4241, 2.8442, 3.7018, 3.4236, 3.8427], device='cuda:1'), covar=tensor([0.1746, 0.3730, 0.3426, 0.2454, 0.0622, 0.0215, 0.0218, 0.0282], device='cuda:1'), in_proj_covar=tensor([0.0275, 0.0324, 0.0355, 0.0265, 0.0245, 0.0188, 0.0217, 0.0259], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-02 23:21:44,253 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=152056.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 23:21:48,437 WARNING [train.py:1073] (1/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9_rvb from training. Duration: 25.0944375 2023-04-02 23:21:55,071 INFO [train.py:903] (1/4) Epoch 23, batch 1850, loss[loss=0.2063, simple_loss=0.2891, pruned_loss=0.06179, over 19669.00 frames. ], tot_loss[loss=0.208, simple_loss=0.2877, pruned_loss=0.0641, over 3781869.49 frames. ], batch size: 60, lr: 3.59e-03, grad_scale: 4.0 2023-04-02 23:22:26,618 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9_rvb from training. Duration: 27.8166875 2023-04-02 23:22:54,485 INFO [train.py:903] (1/4) Epoch 23, batch 1900, loss[loss=0.2494, simple_loss=0.3229, pruned_loss=0.08793, over 18797.00 frames. ], tot_loss[loss=0.2089, simple_loss=0.2886, pruned_loss=0.06457, over 3789334.74 frames. ], batch size: 74, lr: 3.59e-03, grad_scale: 4.0 2023-04-02 23:23:09,879 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 2023-04-02 23:23:16,372 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9_rvb from training. Duration: 27.02225 2023-04-02 23:23:26,803 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.505e+02 4.994e+02 5.968e+02 7.712e+02 2.482e+03, threshold=1.194e+03, percent-clipped=3.0 2023-04-02 23:23:36,535 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4957, 1.5774, 1.8764, 1.7079, 2.9797, 2.4654, 3.2515, 1.6393], device='cuda:1'), covar=tensor([0.2563, 0.4457, 0.2854, 0.2028, 0.1719, 0.2221, 0.1678, 0.4330], device='cuda:1'), in_proj_covar=tensor([0.0534, 0.0643, 0.0715, 0.0486, 0.0618, 0.0530, 0.0663, 0.0549], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 23:23:41,413 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 2023-04-02 23:23:52,764 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=152163.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 23:23:55,364 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=152165.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 23:23:56,154 INFO [train.py:903] (1/4) Epoch 23, batch 1950, loss[loss=0.1964, simple_loss=0.2872, pruned_loss=0.05276, over 19603.00 frames. ], tot_loss[loss=0.2067, simple_loss=0.2866, pruned_loss=0.06343, over 3795660.98 frames. ], batch size: 57, lr: 3.59e-03, grad_scale: 4.0 2023-04-02 23:24:58,530 INFO [train.py:903] (1/4) Epoch 23, batch 2000, loss[loss=0.2289, simple_loss=0.3132, pruned_loss=0.07237, over 19610.00 frames. ], tot_loss[loss=0.2062, simple_loss=0.2864, pruned_loss=0.06302, over 3794320.36 frames. ], batch size: 57, lr: 3.59e-03, grad_scale: 8.0 2023-04-02 23:25:06,465 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.47 vs. limit=5.0 2023-04-02 23:25:28,732 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.131e+02 4.661e+02 5.613e+02 7.041e+02 1.127e+03, threshold=1.123e+03, percent-clipped=0.0 2023-04-02 23:25:30,260 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=152243.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 23:25:53,680 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_rvb from training. Duration: 26.8349375 2023-04-02 23:25:58,367 INFO [train.py:903] (1/4) Epoch 23, batch 2050, loss[loss=0.1864, simple_loss=0.2765, pruned_loss=0.04813, over 19772.00 frames. ], tot_loss[loss=0.2056, simple_loss=0.2858, pruned_loss=0.06265, over 3806169.45 frames. ], batch size: 56, lr: 3.59e-03, grad_scale: 8.0 2023-04-02 23:26:13,113 WARNING [train.py:1073] (1/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 2023-04-02 23:26:13,448 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=152278.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 23:26:14,192 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 2023-04-02 23:26:29,838 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=152291.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 23:26:36,227 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 2023-04-02 23:26:58,782 INFO [train.py:903] (1/4) Epoch 23, batch 2100, loss[loss=0.2171, simple_loss=0.2934, pruned_loss=0.07042, over 19764.00 frames. ], tot_loss[loss=0.2061, simple_loss=0.2865, pruned_loss=0.06284, over 3803912.57 frames. ], batch size: 54, lr: 3.59e-03, grad_scale: 8.0 2023-04-02 23:27:27,815 WARNING [train.py:1073] (1/4) Exclude cut with ID 453-131332-0000-131866_sp0.9_rvb from training. Duration: 25.3333125 2023-04-02 23:27:31,217 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.686e+02 4.808e+02 5.705e+02 7.050e+02 1.568e+03, threshold=1.141e+03, percent-clipped=6.0 2023-04-02 23:27:48,004 WARNING [train.py:1073] (1/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9_rvb from training. Duration: 26.6166875 2023-04-02 23:27:59,782 INFO [train.py:903] (1/4) Epoch 23, batch 2150, loss[loss=0.2099, simple_loss=0.2844, pruned_loss=0.06775, over 19772.00 frames. ], tot_loss[loss=0.2065, simple_loss=0.2866, pruned_loss=0.06318, over 3817332.58 frames. ], batch size: 54, lr: 3.59e-03, grad_scale: 8.0 2023-04-02 23:28:24,359 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.8901, 1.7657, 1.5622, 1.9770, 1.7107, 1.6487, 1.6002, 1.8115], device='cuda:1'), covar=tensor([0.1109, 0.1561, 0.1618, 0.1108, 0.1406, 0.0624, 0.1447, 0.0848], device='cuda:1'), in_proj_covar=tensor([0.0273, 0.0355, 0.0314, 0.0253, 0.0304, 0.0252, 0.0310, 0.0257], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 23:28:35,511 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0 2023-04-02 23:29:00,653 INFO [train.py:903] (1/4) Epoch 23, batch 2200, loss[loss=0.2506, simple_loss=0.3296, pruned_loss=0.0858, over 19662.00 frames. ], tot_loss[loss=0.2073, simple_loss=0.2871, pruned_loss=0.06372, over 3815394.35 frames. ], batch size: 60, lr: 3.59e-03, grad_scale: 8.0 2023-04-02 23:29:07,494 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=152421.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 23:29:31,741 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.170e+02 5.084e+02 6.029e+02 8.181e+02 1.825e+03, threshold=1.206e+03, percent-clipped=10.0 2023-04-02 23:29:37,606 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=152446.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 23:29:58,447 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=152463.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 23:30:01,449 INFO [train.py:903] (1/4) Epoch 23, batch 2250, loss[loss=0.2549, simple_loss=0.336, pruned_loss=0.08687, over 19379.00 frames. ], tot_loss[loss=0.2066, simple_loss=0.2868, pruned_loss=0.06317, over 3825369.81 frames. ], batch size: 70, lr: 3.59e-03, grad_scale: 8.0 2023-04-02 23:31:01,771 INFO [train.py:903] (1/4) Epoch 23, batch 2300, loss[loss=0.1973, simple_loss=0.2701, pruned_loss=0.06222, over 19472.00 frames. ], tot_loss[loss=0.2066, simple_loss=0.2865, pruned_loss=0.0634, over 3818113.32 frames. ], batch size: 49, lr: 3.59e-03, grad_scale: 4.0 2023-04-02 23:31:17,227 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_rvb from training. Duration: 26.205 2023-04-02 23:31:25,349 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=152534.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 23:31:36,142 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.429e+02 4.927e+02 5.902e+02 7.617e+02 2.113e+03, threshold=1.180e+03, percent-clipped=5.0 2023-04-02 23:31:55,694 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=152559.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 23:32:04,343 INFO [train.py:903] (1/4) Epoch 23, batch 2350, loss[loss=0.2014, simple_loss=0.2891, pruned_loss=0.05689, over 19769.00 frames. ], tot_loss[loss=0.2057, simple_loss=0.2858, pruned_loss=0.06283, over 3820894.65 frames. ], batch size: 56, lr: 3.59e-03, grad_scale: 4.0 2023-04-02 23:32:30,248 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=152587.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 23:32:39,675 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 2023-04-02 23:32:43,252 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_rvb from training. Duration: 25.775 2023-04-02 23:32:45,481 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=152600.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 23:32:54,447 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=152607.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 23:33:03,103 WARNING [train.py:1073] (1/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9_rvb from training. Duration: 25.45 2023-04-02 23:33:05,166 INFO [train.py:903] (1/4) Epoch 23, batch 2400, loss[loss=0.243, simple_loss=0.3213, pruned_loss=0.08231, over 19553.00 frames. ], tot_loss[loss=0.2072, simple_loss=0.2872, pruned_loss=0.06362, over 3812289.31 frames. ], batch size: 56, lr: 3.59e-03, grad_scale: 8.0 2023-04-02 23:33:28,192 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=152635.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 23:33:38,290 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.633e+02 5.347e+02 6.485e+02 7.646e+02 1.871e+03, threshold=1.297e+03, percent-clipped=3.0 2023-04-02 23:34:06,659 INFO [train.py:903] (1/4) Epoch 23, batch 2450, loss[loss=0.2067, simple_loss=0.2951, pruned_loss=0.05914, over 19665.00 frames. ], tot_loss[loss=0.2071, simple_loss=0.2873, pruned_loss=0.0635, over 3800060.17 frames. ], batch size: 58, lr: 3.59e-03, grad_scale: 8.0 2023-04-02 23:34:37,185 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.5364, 1.5407, 1.5898, 1.8264, 3.1204, 1.3047, 2.3845, 3.6101], device='cuda:1'), covar=tensor([0.0523, 0.2686, 0.2846, 0.1651, 0.0690, 0.2425, 0.1377, 0.0250], device='cuda:1'), in_proj_covar=tensor([0.0413, 0.0369, 0.0390, 0.0350, 0.0377, 0.0355, 0.0385, 0.0408], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 23:34:51,448 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=152702.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 23:35:06,561 INFO [train.py:903] (1/4) Epoch 23, batch 2500, loss[loss=0.1807, simple_loss=0.275, pruned_loss=0.04322, over 19780.00 frames. ], tot_loss[loss=0.2069, simple_loss=0.2873, pruned_loss=0.06328, over 3810255.41 frames. ], batch size: 56, lr: 3.59e-03, grad_scale: 8.0 2023-04-02 23:35:12,381 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0307, 1.8023, 1.5995, 1.8961, 1.6940, 1.7439, 1.6258, 1.8884], device='cuda:1'), covar=tensor([0.1072, 0.1340, 0.1605, 0.1197, 0.1369, 0.0561, 0.1483, 0.0764], device='cuda:1'), in_proj_covar=tensor([0.0271, 0.0355, 0.0313, 0.0252, 0.0302, 0.0250, 0.0309, 0.0257], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 23:35:29,554 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([6.1468, 5.6014, 3.0218, 4.8595, 0.9041, 5.6793, 5.4496, 5.7395], device='cuda:1'), covar=tensor([0.0346, 0.0743, 0.1954, 0.0739, 0.4339, 0.0513, 0.0750, 0.1022], device='cuda:1'), in_proj_covar=tensor([0.0515, 0.0416, 0.0502, 0.0351, 0.0404, 0.0440, 0.0431, 0.0464], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 23:35:40,606 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.829e+02 4.968e+02 5.949e+02 7.714e+02 2.745e+03, threshold=1.190e+03, percent-clipped=5.0 2023-04-02 23:35:42,075 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=152744.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 23:35:48,943 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=152750.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 23:36:08,241 INFO [train.py:903] (1/4) Epoch 23, batch 2550, loss[loss=0.203, simple_loss=0.2807, pruned_loss=0.06268, over 19625.00 frames. ], tot_loss[loss=0.2059, simple_loss=0.2866, pruned_loss=0.06263, over 3815924.90 frames. ], batch size: 50, lr: 3.58e-03, grad_scale: 8.0 2023-04-02 23:36:28,485 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.1606, 3.4492, 2.0116, 2.0719, 3.0956, 1.6510, 1.6404, 2.3340], device='cuda:1'), covar=tensor([0.1276, 0.0596, 0.1152, 0.0835, 0.0610, 0.1385, 0.0958, 0.0659], device='cuda:1'), in_proj_covar=tensor([0.0301, 0.0314, 0.0334, 0.0265, 0.0246, 0.0336, 0.0288, 0.0275], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 23:36:57,880 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=152807.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 23:37:01,064 WARNING [train.py:1073] (1/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 2023-04-02 23:37:08,439 INFO [train.py:903] (1/4) Epoch 23, batch 2600, loss[loss=0.1939, simple_loss=0.2737, pruned_loss=0.0571, over 19736.00 frames. ], tot_loss[loss=0.2052, simple_loss=0.2854, pruned_loss=0.06253, over 3821603.47 frames. ], batch size: 63, lr: 3.58e-03, grad_scale: 8.0 2023-04-02 23:37:37,470 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([0.9590, 0.9871, 1.0954, 1.0814, 1.3098, 1.2874, 1.2468, 0.5561], device='cuda:1'), covar=tensor([0.1806, 0.3170, 0.1870, 0.1495, 0.1222, 0.1759, 0.1114, 0.4019], device='cuda:1'), in_proj_covar=tensor([0.0536, 0.0647, 0.0718, 0.0487, 0.0620, 0.0533, 0.0664, 0.0552], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 23:37:40,420 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.218e+02 4.724e+02 5.510e+02 7.301e+02 1.657e+03, threshold=1.102e+03, percent-clipped=4.0 2023-04-02 23:38:08,552 INFO [train.py:903] (1/4) Epoch 23, batch 2650, loss[loss=0.1807, simple_loss=0.26, pruned_loss=0.05066, over 19801.00 frames. ], tot_loss[loss=0.2068, simple_loss=0.2867, pruned_loss=0.06345, over 3809121.19 frames. ], batch size: 49, lr: 3.58e-03, grad_scale: 8.0 2023-04-02 23:38:27,766 WARNING [train.py:1073] (1/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9_rvb from training. Duration: 27.25 2023-04-02 23:39:08,680 INFO [train.py:903] (1/4) Epoch 23, batch 2700, loss[loss=0.2227, simple_loss=0.3021, pruned_loss=0.07167, over 19305.00 frames. ], tot_loss[loss=0.206, simple_loss=0.2859, pruned_loss=0.06309, over 3809971.37 frames. ], batch size: 66, lr: 3.58e-03, grad_scale: 8.0 2023-04-02 23:39:16,550 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=152922.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 23:39:42,349 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.879e+02 4.895e+02 5.730e+02 7.672e+02 1.465e+03, threshold=1.146e+03, percent-clipped=5.0 2023-04-02 23:39:43,656 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=152944.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 23:39:51,725 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=152951.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 23:40:00,073 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=152958.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 23:40:09,445 INFO [train.py:903] (1/4) Epoch 23, batch 2750, loss[loss=0.2449, simple_loss=0.3248, pruned_loss=0.08247, over 19671.00 frames. ], tot_loss[loss=0.2062, simple_loss=0.2867, pruned_loss=0.06288, over 3828799.69 frames. ], batch size: 53, lr: 3.58e-03, grad_scale: 8.0 2023-04-02 23:40:23,475 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.4368, 1.4132, 1.6437, 1.4243, 3.0616, 1.1128, 2.2750, 3.4954], device='cuda:1'), covar=tensor([0.0530, 0.2700, 0.2619, 0.1832, 0.0697, 0.2457, 0.1214, 0.0235], device='cuda:1'), in_proj_covar=tensor([0.0414, 0.0370, 0.0390, 0.0350, 0.0375, 0.0354, 0.0383, 0.0407], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 23:40:29,119 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([0.9355, 0.8812, 0.9079, 1.0319, 0.8475, 0.9933, 0.9344, 0.9646], device='cuda:1'), covar=tensor([0.0691, 0.0764, 0.0816, 0.0549, 0.0814, 0.0682, 0.0748, 0.0623], device='cuda:1'), in_proj_covar=tensor([0.0210, 0.0220, 0.0224, 0.0238, 0.0227, 0.0211, 0.0187, 0.0204], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-02 23:40:31,191 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=152983.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 23:40:57,898 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=153006.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 23:41:10,184 INFO [train.py:903] (1/4) Epoch 23, batch 2800, loss[loss=0.2099, simple_loss=0.2832, pruned_loss=0.06826, over 19607.00 frames. ], tot_loss[loss=0.2078, simple_loss=0.2879, pruned_loss=0.06382, over 3809337.71 frames. ], batch size: 52, lr: 3.58e-03, grad_scale: 8.0 2023-04-02 23:41:27,938 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=153031.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 23:41:42,352 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.106e+02 4.940e+02 6.140e+02 7.865e+02 1.529e+03, threshold=1.228e+03, percent-clipped=3.0 2023-04-02 23:41:54,199 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 2023-04-02 23:42:02,472 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=153059.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 23:42:10,891 INFO [train.py:903] (1/4) Epoch 23, batch 2850, loss[loss=0.1883, simple_loss=0.2631, pruned_loss=0.05679, over 19749.00 frames. ], tot_loss[loss=0.2095, simple_loss=0.2897, pruned_loss=0.06463, over 3809520.76 frames. ], batch size: 45, lr: 3.58e-03, grad_scale: 8.0 2023-04-02 23:42:11,236 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=153066.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 23:42:33,234 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.8009, 3.2534, 3.3262, 3.3299, 1.3022, 3.2173, 2.7946, 3.0932], device='cuda:1'), covar=tensor([0.1873, 0.1198, 0.0845, 0.0989, 0.6009, 0.1119, 0.0870, 0.1417], device='cuda:1'), in_proj_covar=tensor([0.0786, 0.0750, 0.0957, 0.0840, 0.0844, 0.0718, 0.0570, 0.0892], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-02 23:42:36,390 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=153088.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 23:42:51,798 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6477, 1.5091, 1.6194, 2.1826, 1.6222, 1.9309, 1.8566, 1.7753], device='cuda:1'), covar=tensor([0.0819, 0.0891, 0.0918, 0.0654, 0.0841, 0.0750, 0.0857, 0.0675], device='cuda:1'), in_proj_covar=tensor([0.0211, 0.0221, 0.0224, 0.0239, 0.0228, 0.0212, 0.0188, 0.0205], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-04-02 23:43:09,810 WARNING [train.py:1073] (1/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 2023-04-02 23:43:11,000 INFO [train.py:903] (1/4) Epoch 23, batch 2900, loss[loss=0.1563, simple_loss=0.2369, pruned_loss=0.03785, over 19777.00 frames. ], tot_loss[loss=0.2081, simple_loss=0.2883, pruned_loss=0.06393, over 3807867.58 frames. ], batch size: 48, lr: 3.58e-03, grad_scale: 8.0 2023-04-02 23:43:12,415 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.6332, 1.3026, 1.4682, 1.5467, 3.2451, 1.0695, 2.2896, 3.6486], device='cuda:1'), covar=tensor([0.0503, 0.2856, 0.3081, 0.1876, 0.0708, 0.2656, 0.1396, 0.0246], device='cuda:1'), in_proj_covar=tensor([0.0413, 0.0368, 0.0390, 0.0350, 0.0374, 0.0354, 0.0383, 0.0407], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 23:43:34,541 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=153135.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 23:43:45,170 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.545e+02 5.074e+02 6.117e+02 7.609e+02 1.538e+03, threshold=1.223e+03, percent-clipped=2.0 2023-04-02 23:44:10,298 INFO [train.py:903] (1/4) Epoch 23, batch 2950, loss[loss=0.2172, simple_loss=0.287, pruned_loss=0.07366, over 19573.00 frames. ], tot_loss[loss=0.2084, simple_loss=0.2884, pruned_loss=0.06426, over 3812890.12 frames. ], batch size: 52, lr: 3.58e-03, grad_scale: 4.0 2023-04-02 23:44:23,187 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([5.0840, 5.0846, 5.8666, 5.8567, 1.9619, 5.5747, 4.6232, 5.5118], device='cuda:1'), covar=tensor([0.1536, 0.0854, 0.0518, 0.0548, 0.6278, 0.0771, 0.0630, 0.1087], device='cuda:1'), in_proj_covar=tensor([0.0780, 0.0743, 0.0950, 0.0833, 0.0838, 0.0711, 0.0566, 0.0885], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-02 23:44:25,472 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=153178.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 23:44:54,243 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=153203.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 23:44:54,283 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=153203.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 23:45:09,888 INFO [train.py:903] (1/4) Epoch 23, batch 3000, loss[loss=0.2857, simple_loss=0.3446, pruned_loss=0.1134, over 13079.00 frames. ], tot_loss[loss=0.2087, simple_loss=0.2882, pruned_loss=0.06459, over 3818882.34 frames. ], batch size: 135, lr: 3.58e-03, grad_scale: 4.0 2023-04-02 23:45:09,889 INFO [train.py:928] (1/4) Computing validation loss 2023-04-02 23:45:23,388 INFO [train.py:937] (1/4) Epoch 23, validation: loss=0.1686, simple_loss=0.2685, pruned_loss=0.03441, over 944034.00 frames. 2023-04-02 23:45:23,388 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18821MB 2023-04-02 23:45:26,695 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_rvb from training. Duration: 29.735 2023-04-02 23:45:27,892 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.9138, 4.4513, 2.8304, 3.8722, 0.9899, 4.4562, 4.2634, 4.4544], device='cuda:1'), covar=tensor([0.0571, 0.0944, 0.1936, 0.0861, 0.4170, 0.0651, 0.0929, 0.1036], device='cuda:1'), in_proj_covar=tensor([0.0511, 0.0414, 0.0498, 0.0349, 0.0402, 0.0437, 0.0429, 0.0463], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 23:45:40,636 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 2023-04-02 23:45:55,421 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9794, 2.0667, 2.3231, 2.6056, 1.9204, 2.4815, 2.3571, 2.1083], device='cuda:1'), covar=tensor([0.4470, 0.4310, 0.2112, 0.2460, 0.4359, 0.2304, 0.5158, 0.3618], device='cuda:1'), in_proj_covar=tensor([0.0908, 0.0975, 0.0721, 0.0934, 0.0885, 0.0822, 0.0848, 0.0787], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 23:45:57,208 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.963e+02 5.132e+02 6.544e+02 7.997e+02 1.730e+03, threshold=1.309e+03, percent-clipped=4.0 2023-04-02 23:46:23,490 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.57 vs. limit=2.0 2023-04-02 23:46:24,009 INFO [train.py:903] (1/4) Epoch 23, batch 3050, loss[loss=0.2225, simple_loss=0.2992, pruned_loss=0.07293, over 17326.00 frames. ], tot_loss[loss=0.208, simple_loss=0.2875, pruned_loss=0.06424, over 3839338.59 frames. ], batch size: 101, lr: 3.58e-03, grad_scale: 4.0 2023-04-02 23:46:27,330 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.7282, 4.3138, 2.8926, 3.8142, 0.8923, 4.2672, 4.1579, 4.2350], device='cuda:1'), covar=tensor([0.0616, 0.0839, 0.1671, 0.0771, 0.4103, 0.0590, 0.0889, 0.0982], device='cuda:1'), in_proj_covar=tensor([0.0511, 0.0414, 0.0497, 0.0349, 0.0401, 0.0437, 0.0430, 0.0463], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 23:47:00,337 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=153296.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 23:47:05,913 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4398, 1.5286, 1.7692, 1.7246, 2.6352, 2.3465, 2.8248, 1.2000], device='cuda:1'), covar=tensor([0.2478, 0.4315, 0.2690, 0.1904, 0.1473, 0.2088, 0.1343, 0.4453], device='cuda:1'), in_proj_covar=tensor([0.0535, 0.0644, 0.0715, 0.0486, 0.0617, 0.0531, 0.0662, 0.0551], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 23:47:25,750 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=153315.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 23:47:26,471 INFO [train.py:903] (1/4) Epoch 23, batch 3100, loss[loss=0.209, simple_loss=0.2881, pruned_loss=0.06493, over 19723.00 frames. ], tot_loss[loss=0.207, simple_loss=0.2863, pruned_loss=0.0638, over 3836689.33 frames. ], batch size: 63, lr: 3.58e-03, grad_scale: 4.0 2023-04-02 23:47:33,582 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=153322.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 23:47:54,373 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=153340.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 23:47:59,279 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.375e+02 4.897e+02 6.414e+02 9.491e+02 6.432e+03, threshold=1.283e+03, percent-clipped=11.0 2023-04-02 23:48:01,106 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.80 vs. limit=2.0 2023-04-02 23:48:03,113 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=153347.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 23:48:13,743 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6782, 1.4813, 1.2484, 1.5446, 1.4463, 1.3329, 1.2456, 1.4874], device='cuda:1'), covar=tensor([0.1239, 0.1527, 0.1961, 0.1251, 0.1414, 0.0987, 0.1930, 0.1058], device='cuda:1'), in_proj_covar=tensor([0.0273, 0.0357, 0.0314, 0.0254, 0.0305, 0.0252, 0.0310, 0.0258], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 23:48:25,986 INFO [train.py:903] (1/4) Epoch 23, batch 3150, loss[loss=0.1701, simple_loss=0.2535, pruned_loss=0.04336, over 19467.00 frames. ], tot_loss[loss=0.2078, simple_loss=0.2871, pruned_loss=0.06421, over 3815914.67 frames. ], batch size: 49, lr: 3.58e-03, grad_scale: 4.0 2023-04-02 23:48:54,111 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 2023-04-02 23:48:57,823 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.4309, 1.4959, 1.5744, 1.4716, 3.0333, 1.1457, 2.4877, 3.3963], device='cuda:1'), covar=tensor([0.0560, 0.2705, 0.2839, 0.1925, 0.0736, 0.2487, 0.1086, 0.0303], device='cuda:1'), in_proj_covar=tensor([0.0415, 0.0370, 0.0392, 0.0352, 0.0376, 0.0354, 0.0385, 0.0408], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 23:49:03,432 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3270, 1.3845, 1.4938, 1.4915, 1.7695, 1.8299, 1.8186, 0.7156], device='cuda:1'), covar=tensor([0.2393, 0.4157, 0.2635, 0.1898, 0.1617, 0.2275, 0.1409, 0.4505], device='cuda:1'), in_proj_covar=tensor([0.0535, 0.0645, 0.0716, 0.0486, 0.0617, 0.0532, 0.0662, 0.0552], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-02 23:49:03,693 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.59 vs. limit=5.0 2023-04-02 23:49:26,056 INFO [train.py:903] (1/4) Epoch 23, batch 3200, loss[loss=0.1941, simple_loss=0.2788, pruned_loss=0.05463, over 19476.00 frames. ], tot_loss[loss=0.2059, simple_loss=0.2853, pruned_loss=0.06327, over 3818076.98 frames. ], batch size: 64, lr: 3.58e-03, grad_scale: 8.0 2023-04-02 23:49:54,888 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=153439.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 23:50:00,125 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.031e+02 5.043e+02 6.093e+02 8.078e+02 1.420e+03, threshold=1.219e+03, percent-clipped=2.0 2023-04-02 23:50:17,468 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=153459.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 23:50:26,643 INFO [train.py:903] (1/4) Epoch 23, batch 3250, loss[loss=0.232, simple_loss=0.3117, pruned_loss=0.07612, over 19608.00 frames. ], tot_loss[loss=0.2068, simple_loss=0.2861, pruned_loss=0.06377, over 3812304.00 frames. ], batch size: 57, lr: 3.58e-03, grad_scale: 8.0 2023-04-02 23:50:43,099 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=153479.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 23:50:48,913 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=153484.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 23:51:05,697 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 2023-04-02 23:51:27,758 INFO [train.py:903] (1/4) Epoch 23, batch 3300, loss[loss=0.2059, simple_loss=0.2902, pruned_loss=0.06077, over 19667.00 frames. ], tot_loss[loss=0.2062, simple_loss=0.2858, pruned_loss=0.06333, over 3830810.13 frames. ], batch size: 55, lr: 3.58e-03, grad_scale: 8.0 2023-04-02 23:51:34,823 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 2023-04-02 23:52:00,745 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.814e+02 5.081e+02 6.216e+02 8.009e+02 2.047e+03, threshold=1.243e+03, percent-clipped=5.0 2023-04-02 23:52:26,361 INFO [train.py:903] (1/4) Epoch 23, batch 3350, loss[loss=0.2015, simple_loss=0.2817, pruned_loss=0.06063, over 19693.00 frames. ], tot_loss[loss=0.2075, simple_loss=0.2871, pruned_loss=0.06389, over 3824776.19 frames. ], batch size: 53, lr: 3.58e-03, grad_scale: 8.0 2023-04-02 23:52:56,913 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5387, 2.2339, 1.6479, 1.5296, 2.0110, 1.3343, 1.4182, 1.9451], device='cuda:1'), covar=tensor([0.0993, 0.0736, 0.1207, 0.0858, 0.0621, 0.1354, 0.0755, 0.0483], device='cuda:1'), in_proj_covar=tensor([0.0307, 0.0319, 0.0344, 0.0269, 0.0252, 0.0343, 0.0295, 0.0280], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 23:53:00,283 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=153594.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 23:53:26,223 INFO [train.py:903] (1/4) Epoch 23, batch 3400, loss[loss=0.2236, simple_loss=0.3033, pruned_loss=0.07193, over 17530.00 frames. ], tot_loss[loss=0.2087, simple_loss=0.2883, pruned_loss=0.06458, over 3824975.96 frames. ], batch size: 101, lr: 3.57e-03, grad_scale: 8.0 2023-04-02 23:53:37,515 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.45 vs. limit=2.0 2023-04-02 23:53:47,127 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.2720, 1.8731, 1.8926, 2.7753, 1.9178, 2.3982, 2.3614, 2.2673], device='cuda:1'), covar=tensor([0.0745, 0.0911, 0.0943, 0.0778, 0.0875, 0.0746, 0.0922, 0.0622], device='cuda:1'), in_proj_covar=tensor([0.0210, 0.0219, 0.0223, 0.0237, 0.0225, 0.0211, 0.0186, 0.0202], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-02 23:53:56,984 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=153640.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 23:53:58,257 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3262, 1.3195, 1.7276, 1.1594, 2.5044, 3.4379, 3.1328, 3.6252], device='cuda:1'), covar=tensor([0.1461, 0.3653, 0.3149, 0.2507, 0.0604, 0.0175, 0.0210, 0.0245], device='cuda:1'), in_proj_covar=tensor([0.0274, 0.0324, 0.0353, 0.0266, 0.0245, 0.0189, 0.0218, 0.0260], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-02 23:54:01,365 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.785e+02 5.295e+02 6.743e+02 8.549e+02 2.424e+03, threshold=1.349e+03, percent-clipped=5.0 2023-04-02 23:54:17,488 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.4614, 2.2472, 2.3386, 2.6240, 2.3448, 2.1244, 2.1887, 2.4472], device='cuda:1'), covar=tensor([0.0777, 0.1265, 0.1037, 0.0743, 0.1061, 0.0453, 0.1045, 0.0532], device='cuda:1'), in_proj_covar=tensor([0.0276, 0.0360, 0.0317, 0.0256, 0.0308, 0.0254, 0.0313, 0.0260], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 23:54:28,038 INFO [train.py:903] (1/4) Epoch 23, batch 3450, loss[loss=0.2103, simple_loss=0.2978, pruned_loss=0.06143, over 19791.00 frames. ], tot_loss[loss=0.2077, simple_loss=0.2875, pruned_loss=0.064, over 3821758.50 frames. ], batch size: 56, lr: 3.57e-03, grad_scale: 8.0 2023-04-02 23:54:31,533 WARNING [train.py:1073] (1/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9_rvb from training. Duration: 25.2444375 2023-04-02 23:54:52,895 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0040, 2.0848, 2.3317, 2.7603, 2.0495, 2.5985, 2.4036, 2.1848], device='cuda:1'), covar=tensor([0.4108, 0.3889, 0.1871, 0.2269, 0.3960, 0.2099, 0.4595, 0.3275], device='cuda:1'), in_proj_covar=tensor([0.0907, 0.0972, 0.0719, 0.0934, 0.0882, 0.0819, 0.0847, 0.0788], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-02 23:55:30,145 INFO [train.py:903] (1/4) Epoch 23, batch 3500, loss[loss=0.2131, simple_loss=0.2961, pruned_loss=0.06507, over 19430.00 frames. ], tot_loss[loss=0.2086, simple_loss=0.2886, pruned_loss=0.06432, over 3821952.66 frames. ], batch size: 70, lr: 3.57e-03, grad_scale: 8.0 2023-04-02 23:56:02,440 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.496e+02 4.938e+02 5.821e+02 7.521e+02 2.332e+03, threshold=1.164e+03, percent-clipped=1.0 2023-04-02 23:56:17,986 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=153755.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 23:56:30,177 INFO [train.py:903] (1/4) Epoch 23, batch 3550, loss[loss=0.184, simple_loss=0.2705, pruned_loss=0.0488, over 19682.00 frames. ], tot_loss[loss=0.2084, simple_loss=0.2883, pruned_loss=0.06431, over 3816141.73 frames. ], batch size: 58, lr: 3.57e-03, grad_scale: 8.0 2023-04-02 23:56:50,064 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=153783.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 23:57:30,146 INFO [train.py:903] (1/4) Epoch 23, batch 3600, loss[loss=0.2021, simple_loss=0.2881, pruned_loss=0.0581, over 19541.00 frames. ], tot_loss[loss=0.2079, simple_loss=0.2879, pruned_loss=0.06398, over 3827467.60 frames. ], batch size: 54, lr: 3.57e-03, grad_scale: 8.0 2023-04-02 23:57:47,202 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 2023-04-02 23:58:05,069 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.445e+02 5.155e+02 6.351e+02 8.015e+02 2.586e+03, threshold=1.270e+03, percent-clipped=6.0 2023-04-02 23:58:12,392 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=153850.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 23:58:30,878 INFO [train.py:903] (1/4) Epoch 23, batch 3650, loss[loss=0.2385, simple_loss=0.3102, pruned_loss=0.08343, over 18131.00 frames. ], tot_loss[loss=0.2083, simple_loss=0.2878, pruned_loss=0.06437, over 3819588.94 frames. ], batch size: 83, lr: 3.57e-03, grad_scale: 8.0 2023-04-02 23:58:42,967 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=153875.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 23:59:08,879 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=153898.0, num_to_drop=0, layers_to_drop=set() 2023-04-02 23:59:31,665 INFO [train.py:903] (1/4) Epoch 23, batch 3700, loss[loss=0.1933, simple_loss=0.2785, pruned_loss=0.05404, over 19490.00 frames. ], tot_loss[loss=0.2087, simple_loss=0.2881, pruned_loss=0.06464, over 3818250.13 frames. ], batch size: 49, lr: 3.57e-03, grad_scale: 8.0 2023-04-02 23:59:57,424 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.5583, 1.1860, 1.4909, 1.3613, 3.0931, 1.0581, 2.2951, 3.4781], device='cuda:1'), covar=tensor([0.0625, 0.3177, 0.3072, 0.2198, 0.0838, 0.2778, 0.1453, 0.0326], device='cuda:1'), in_proj_covar=tensor([0.0413, 0.0369, 0.0391, 0.0352, 0.0375, 0.0352, 0.0385, 0.0406], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-02 23:59:57,956 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.74 vs. limit=2.0 2023-04-03 00:00:00,754 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=153941.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:00:04,690 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.006e+02 4.617e+02 5.510e+02 6.874e+02 2.344e+03, threshold=1.102e+03, percent-clipped=3.0 2023-04-03 00:00:31,958 INFO [train.py:903] (1/4) Epoch 23, batch 3750, loss[loss=0.1913, simple_loss=0.2789, pruned_loss=0.05184, over 19529.00 frames. ], tot_loss[loss=0.2093, simple_loss=0.2889, pruned_loss=0.06486, over 3821262.88 frames. ], batch size: 56, lr: 3.57e-03, grad_scale: 8.0 2023-04-03 00:01:24,683 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 2023-04-03 00:01:27,853 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=154011.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:01:33,254 INFO [train.py:903] (1/4) Epoch 23, batch 3800, loss[loss=0.2552, simple_loss=0.3185, pruned_loss=0.09598, over 13371.00 frames. ], tot_loss[loss=0.2086, simple_loss=0.2885, pruned_loss=0.0644, over 3826137.30 frames. ], batch size: 136, lr: 3.57e-03, grad_scale: 8.0 2023-04-03 00:01:59,812 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=154036.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:02:05,115 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9_rvb from training. Duration: 29.1166875 2023-04-03 00:02:08,274 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.228e+02 4.923e+02 6.103e+02 7.526e+02 2.694e+03, threshold=1.221e+03, percent-clipped=9.0 2023-04-03 00:02:33,013 INFO [train.py:903] (1/4) Epoch 23, batch 3850, loss[loss=0.2183, simple_loss=0.3014, pruned_loss=0.06762, over 19393.00 frames. ], tot_loss[loss=0.2076, simple_loss=0.2876, pruned_loss=0.06383, over 3834218.90 frames. ], batch size: 70, lr: 3.57e-03, grad_scale: 8.0 2023-04-03 00:02:35,318 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=154067.0, num_to_drop=1, layers_to_drop={1} 2023-04-03 00:03:01,627 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.82 vs. limit=2.0 2023-04-03 00:03:09,351 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=154094.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:03:18,588 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=154102.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:03:35,946 INFO [train.py:903] (1/4) Epoch 23, batch 3900, loss[loss=0.164, simple_loss=0.2457, pruned_loss=0.04116, over 16880.00 frames. ], tot_loss[loss=0.2071, simple_loss=0.2873, pruned_loss=0.06347, over 3837586.00 frames. ], batch size: 37, lr: 3.57e-03, grad_scale: 8.0 2023-04-03 00:04:09,480 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.867e+02 4.608e+02 5.656e+02 7.392e+02 1.919e+03, threshold=1.131e+03, percent-clipped=3.0 2023-04-03 00:04:22,819 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=154154.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:04:37,502 INFO [train.py:903] (1/4) Epoch 23, batch 3950, loss[loss=0.2633, simple_loss=0.3186, pruned_loss=0.104, over 12655.00 frames. ], tot_loss[loss=0.2083, simple_loss=0.2881, pruned_loss=0.06419, over 3824307.54 frames. ], batch size: 136, lr: 3.57e-03, grad_scale: 8.0 2023-04-03 00:04:44,234 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 2023-04-03 00:04:52,286 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=154179.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:05:37,018 INFO [train.py:903] (1/4) Epoch 23, batch 4000, loss[loss=0.2124, simple_loss=0.2873, pruned_loss=0.06873, over 19666.00 frames. ], tot_loss[loss=0.2079, simple_loss=0.2875, pruned_loss=0.06414, over 3832389.04 frames. ], batch size: 55, lr: 3.57e-03, grad_scale: 8.0 2023-04-03 00:06:06,068 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=154239.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:06:12,360 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.243e+02 5.130e+02 6.145e+02 8.525e+02 2.203e+03, threshold=1.229e+03, percent-clipped=9.0 2023-04-03 00:06:27,019 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1_rvb from training. Duration: 0.7545625 2023-04-03 00:06:37,055 INFO [train.py:903] (1/4) Epoch 23, batch 4050, loss[loss=0.2126, simple_loss=0.2912, pruned_loss=0.06703, over 19771.00 frames. ], tot_loss[loss=0.2081, simple_loss=0.2877, pruned_loss=0.06422, over 3828362.85 frames. ], batch size: 63, lr: 3.57e-03, grad_scale: 8.0 2023-04-03 00:07:01,635 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=154285.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:07:17,531 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.8317, 1.1666, 1.4885, 1.5317, 3.3443, 1.2466, 2.4959, 3.8940], device='cuda:1'), covar=tensor([0.0548, 0.3372, 0.3251, 0.2229, 0.0813, 0.2648, 0.1389, 0.0251], device='cuda:1'), in_proj_covar=tensor([0.0416, 0.0371, 0.0394, 0.0355, 0.0377, 0.0354, 0.0387, 0.0409], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 00:07:20,309 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.41 vs. limit=2.0 2023-04-03 00:07:23,123 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=154304.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:07:29,631 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=154309.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:07:37,599 INFO [train.py:903] (1/4) Epoch 23, batch 4100, loss[loss=0.2174, simple_loss=0.2783, pruned_loss=0.07826, over 19718.00 frames. ], tot_loss[loss=0.2074, simple_loss=0.2871, pruned_loss=0.0639, over 3821072.94 frames. ], batch size: 46, lr: 3.57e-03, grad_scale: 8.0 2023-04-03 00:08:11,157 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.231e+02 5.105e+02 5.940e+02 7.682e+02 1.555e+03, threshold=1.188e+03, percent-clipped=4.0 2023-04-03 00:08:13,557 WARNING [train.py:1073] (1/4) Exclude cut with ID 497-129325-0061-9566_sp1.1_rvb from training. Duration: 0.97725 2023-04-03 00:08:16,176 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 2023-04-03 00:08:25,599 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=154355.0, num_to_drop=1, layers_to_drop={0} 2023-04-03 00:08:39,413 INFO [train.py:903] (1/4) Epoch 23, batch 4150, loss[loss=0.1788, simple_loss=0.2612, pruned_loss=0.04823, over 19798.00 frames. ], tot_loss[loss=0.2071, simple_loss=0.2871, pruned_loss=0.06357, over 3798473.47 frames. ], batch size: 48, lr: 3.57e-03, grad_scale: 8.0 2023-04-03 00:09:22,103 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=154400.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:09:33,978 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=154411.0, num_to_drop=1, layers_to_drop={1} 2023-04-03 00:09:39,111 INFO [train.py:903] (1/4) Epoch 23, batch 4200, loss[loss=0.2018, simple_loss=0.2751, pruned_loss=0.0642, over 19473.00 frames. ], tot_loss[loss=0.2071, simple_loss=0.2872, pruned_loss=0.06351, over 3797668.01 frames. ], batch size: 49, lr: 3.57e-03, grad_scale: 8.0 2023-04-03 00:09:41,419 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 2023-04-03 00:10:07,065 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=154438.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:10:14,757 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.103e+02 4.744e+02 5.863e+02 7.378e+02 1.705e+03, threshold=1.173e+03, percent-clipped=3.0 2023-04-03 00:10:17,212 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=154446.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:10:40,150 INFO [train.py:903] (1/4) Epoch 23, batch 4250, loss[loss=0.2466, simple_loss=0.3161, pruned_loss=0.08852, over 12949.00 frames. ], tot_loss[loss=0.2074, simple_loss=0.2875, pruned_loss=0.06369, over 3790844.94 frames. ], batch size: 136, lr: 3.56e-03, grad_scale: 8.0 2023-04-03 00:10:54,221 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9_rvb from training. Duration: 29.816625 2023-04-03 00:11:05,304 WARNING [train.py:1073] (1/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 2023-04-03 00:11:12,072 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=154492.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:11:40,259 INFO [train.py:903] (1/4) Epoch 23, batch 4300, loss[loss=0.2035, simple_loss=0.2843, pruned_loss=0.06135, over 17278.00 frames. ], tot_loss[loss=0.2075, simple_loss=0.2874, pruned_loss=0.0638, over 3792930.58 frames. ], batch size: 102, lr: 3.56e-03, grad_scale: 8.0 2023-04-03 00:11:53,655 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=154526.0, num_to_drop=1, layers_to_drop={0} 2023-04-03 00:12:02,702 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.2879, 3.7732, 3.8908, 3.8949, 1.6773, 3.7285, 3.2532, 3.6613], device='cuda:1'), covar=tensor([0.1691, 0.1267, 0.0687, 0.0810, 0.5685, 0.1167, 0.0752, 0.1113], device='cuda:1'), in_proj_covar=tensor([0.0784, 0.0745, 0.0950, 0.0835, 0.0835, 0.0712, 0.0568, 0.0883], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-03 00:12:13,327 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.268e+02 4.652e+02 5.888e+02 7.584e+02 1.931e+03, threshold=1.178e+03, percent-clipped=3.0 2023-04-03 00:12:24,494 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=154553.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:12:33,748 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 2023-04-03 00:12:35,167 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=154561.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:12:41,454 INFO [train.py:903] (1/4) Epoch 23, batch 4350, loss[loss=0.2378, simple_loss=0.3243, pruned_loss=0.07565, over 18824.00 frames. ], tot_loss[loss=0.2081, simple_loss=0.2881, pruned_loss=0.064, over 3807840.35 frames. ], batch size: 74, lr: 3.56e-03, grad_scale: 8.0 2023-04-03 00:13:01,188 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=154583.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:13:40,250 INFO [train.py:903] (1/4) Epoch 23, batch 4400, loss[loss=0.2193, simple_loss=0.2986, pruned_loss=0.06996, over 19772.00 frames. ], tot_loss[loss=0.2073, simple_loss=0.2875, pruned_loss=0.06355, over 3813599.65 frames. ], batch size: 56, lr: 3.56e-03, grad_scale: 8.0 2023-04-03 00:13:46,123 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([0.9602, 1.2420, 1.6013, 0.9621, 2.2864, 3.0106, 2.6935, 3.2366], device='cuda:1'), covar=tensor([0.1829, 0.4041, 0.3523, 0.2677, 0.0669, 0.0230, 0.0300, 0.0330], device='cuda:1'), in_proj_covar=tensor([0.0272, 0.0322, 0.0351, 0.0263, 0.0243, 0.0187, 0.0215, 0.0259], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-03 00:14:04,354 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_rvb from training. Duration: 25.285 2023-04-03 00:14:14,044 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.271e+02 5.197e+02 6.555e+02 7.915e+02 1.480e+03, threshold=1.311e+03, percent-clipped=6.0 2023-04-03 00:14:15,157 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 2023-04-03 00:14:16,530 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=154646.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:14:18,492 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=154648.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:14:23,881 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=154653.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:14:27,557 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=154656.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:14:38,570 INFO [train.py:903] (1/4) Epoch 23, batch 4450, loss[loss=0.2431, simple_loss=0.3147, pruned_loss=0.08578, over 18848.00 frames. ], tot_loss[loss=0.2078, simple_loss=0.2879, pruned_loss=0.06387, over 3819920.88 frames. ], batch size: 74, lr: 3.56e-03, grad_scale: 8.0 2023-04-03 00:14:58,456 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=154681.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:15:18,465 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=154698.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:15:19,380 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=154699.0, num_to_drop=1, layers_to_drop={1} 2023-04-03 00:15:38,748 INFO [train.py:903] (1/4) Epoch 23, batch 4500, loss[loss=0.233, simple_loss=0.3136, pruned_loss=0.07623, over 19574.00 frames. ], tot_loss[loss=0.2092, simple_loss=0.2888, pruned_loss=0.06486, over 3795496.61 frames. ], batch size: 61, lr: 3.56e-03, grad_scale: 4.0 2023-04-03 00:16:06,371 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=154738.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:16:13,919 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.080e+02 4.922e+02 6.448e+02 7.735e+02 1.395e+03, threshold=1.290e+03, percent-clipped=1.0 2023-04-03 00:16:31,820 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.5669, 4.7448, 5.2815, 5.3072, 2.3088, 5.0043, 4.3537, 5.0107], device='cuda:1'), covar=tensor([0.1618, 0.1519, 0.0512, 0.0638, 0.5545, 0.0857, 0.0595, 0.1010], device='cuda:1'), in_proj_covar=tensor([0.0783, 0.0748, 0.0953, 0.0839, 0.0837, 0.0715, 0.0570, 0.0886], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-03 00:16:37,165 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=154763.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:16:41,137 INFO [train.py:903] (1/4) Epoch 23, batch 4550, loss[loss=0.1736, simple_loss=0.2512, pruned_loss=0.04805, over 19780.00 frames. ], tot_loss[loss=0.2089, simple_loss=0.2887, pruned_loss=0.06459, over 3811578.84 frames. ], batch size: 47, lr: 3.56e-03, grad_scale: 4.0 2023-04-03 00:16:43,821 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=154768.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:16:48,240 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9_rvb from training. Duration: 28.72225 2023-04-03 00:17:00,059 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=154782.0, num_to_drop=1, layers_to_drop={1} 2023-04-03 00:17:11,893 WARNING [train.py:1073] (1/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 2023-04-03 00:17:31,869 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=154807.0, num_to_drop=1, layers_to_drop={1} 2023-04-03 00:17:34,136 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=154809.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:17:39,882 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=154814.0, num_to_drop=1, layers_to_drop={1} 2023-04-03 00:17:41,843 INFO [train.py:903] (1/4) Epoch 23, batch 4600, loss[loss=0.1703, simple_loss=0.2525, pruned_loss=0.04408, over 19748.00 frames. ], tot_loss[loss=0.2083, simple_loss=0.2881, pruned_loss=0.06418, over 3815528.64 frames. ], batch size: 47, lr: 3.56e-03, grad_scale: 4.0 2023-04-03 00:17:43,482 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=154817.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:18:02,715 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=154834.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:18:04,539 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=154836.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:18:13,493 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=154842.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:18:17,273 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.766e+02 4.757e+02 5.456e+02 7.137e+02 2.039e+03, threshold=1.091e+03, percent-clipped=4.0 2023-04-03 00:18:41,884 INFO [train.py:903] (1/4) Epoch 23, batch 4650, loss[loss=0.2289, simple_loss=0.3018, pruned_loss=0.078, over 19497.00 frames. ], tot_loss[loss=0.2071, simple_loss=0.2874, pruned_loss=0.06336, over 3820494.03 frames. ], batch size: 64, lr: 3.56e-03, grad_scale: 4.0 2023-04-03 00:18:57,532 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_rvb from training. Duration: 0.92 2023-04-03 00:19:09,930 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_rvb from training. Duration: 0.83 2023-04-03 00:19:37,575 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.7578, 1.7735, 1.6866, 1.4555, 1.4158, 1.5279, 0.3558, 0.7174], device='cuda:1'), covar=tensor([0.0618, 0.0578, 0.0379, 0.0534, 0.1173, 0.0687, 0.1220, 0.1031], device='cuda:1'), in_proj_covar=tensor([0.0358, 0.0354, 0.0360, 0.0383, 0.0462, 0.0389, 0.0337, 0.0340], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 00:19:42,545 INFO [train.py:903] (1/4) Epoch 23, batch 4700, loss[loss=0.2285, simple_loss=0.3028, pruned_loss=0.07709, over 18070.00 frames. ], tot_loss[loss=0.207, simple_loss=0.287, pruned_loss=0.06353, over 3811285.20 frames. ], batch size: 83, lr: 3.56e-03, grad_scale: 4.0 2023-04-03 00:20:04,429 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9_rvb from training. Duration: 0.92225 2023-04-03 00:20:17,976 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.052e+02 4.550e+02 5.511e+02 7.065e+02 1.410e+03, threshold=1.102e+03, percent-clipped=2.0 2023-04-03 00:20:25,052 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=154951.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:20:28,231 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=154954.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:20:39,150 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.1171, 1.7735, 1.9069, 2.6549, 1.9575, 2.4025, 2.3953, 2.1569], device='cuda:1'), covar=tensor([0.0818, 0.0966, 0.0972, 0.0826, 0.0893, 0.0748, 0.0873, 0.0698], device='cuda:1'), in_proj_covar=tensor([0.0211, 0.0221, 0.0225, 0.0240, 0.0226, 0.0212, 0.0187, 0.0203], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-03 00:20:44,141 INFO [train.py:903] (1/4) Epoch 23, batch 4750, loss[loss=0.199, simple_loss=0.2932, pruned_loss=0.0524, over 19666.00 frames. ], tot_loss[loss=0.2064, simple_loss=0.2861, pruned_loss=0.06335, over 3830047.50 frames. ], batch size: 55, lr: 3.56e-03, grad_scale: 4.0 2023-04-03 00:21:00,301 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=154979.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:21:12,337 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=154990.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:21:17,077 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=154994.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:21:45,304 INFO [train.py:903] (1/4) Epoch 23, batch 4800, loss[loss=0.2063, simple_loss=0.2953, pruned_loss=0.05863, over 19676.00 frames. ], tot_loss[loss=0.2073, simple_loss=0.2871, pruned_loss=0.06371, over 3818614.79 frames. ], batch size: 55, lr: 3.56e-03, grad_scale: 8.0 2023-04-03 00:21:49,048 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=155019.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:21:54,274 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=155024.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:21:56,087 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.76 vs. limit=5.0 2023-04-03 00:22:18,887 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=155044.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:22:19,573 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.470e+02 5.324e+02 6.216e+02 7.674e+02 2.163e+03, threshold=1.243e+03, percent-clipped=8.0 2023-04-03 00:22:26,085 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=155049.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:22:44,536 INFO [train.py:903] (1/4) Epoch 23, batch 4850, loss[loss=0.234, simple_loss=0.2989, pruned_loss=0.08457, over 19694.00 frames. ], tot_loss[loss=0.2082, simple_loss=0.2876, pruned_loss=0.06435, over 3819118.15 frames. ], batch size: 53, lr: 3.56e-03, grad_scale: 8.0 2023-04-03 00:22:49,279 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=155070.0, num_to_drop=1, layers_to_drop={0} 2023-04-03 00:23:03,361 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=155082.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:23:09,614 WARNING [train.py:1073] (1/4) Exclude cut with ID 774-127930-0014-48411_sp1.1_rvb from training. Duration: 0.95 2023-04-03 00:23:11,851 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=155087.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:23:21,468 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=155095.0, num_to_drop=1, layers_to_drop={1} 2023-04-03 00:23:29,055 WARNING [train.py:1073] (1/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1_rvb from training. Duration: 0.9409375 2023-04-03 00:23:32,629 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=155105.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:23:34,405 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 2023-04-03 00:23:34,434 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 2023-04-03 00:23:44,624 INFO [train.py:903] (1/4) Epoch 23, batch 4900, loss[loss=0.2023, simple_loss=0.289, pruned_loss=0.05783, over 19670.00 frames. ], tot_loss[loss=0.2086, simple_loss=0.2879, pruned_loss=0.06461, over 3821805.34 frames. ], batch size: 58, lr: 3.56e-03, grad_scale: 8.0 2023-04-03 00:23:44,639 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1_rvb from training. Duration: 25.3818125 2023-04-03 00:24:04,404 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 2023-04-03 00:24:20,243 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.924e+02 5.163e+02 5.938e+02 7.647e+02 1.407e+03, threshold=1.188e+03, percent-clipped=5.0 2023-04-03 00:24:46,211 INFO [train.py:903] (1/4) Epoch 23, batch 4950, loss[loss=0.2119, simple_loss=0.2962, pruned_loss=0.06383, over 19680.00 frames. ], tot_loss[loss=0.2081, simple_loss=0.2878, pruned_loss=0.06419, over 3824034.26 frames. ], batch size: 59, lr: 3.56e-03, grad_scale: 8.0 2023-04-03 00:25:01,061 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 2023-04-03 00:25:19,885 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.73 vs. limit=5.0 2023-04-03 00:25:21,558 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=155197.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:25:22,302 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 2023-04-03 00:25:30,762 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.3587, 2.0026, 2.0450, 2.8190, 2.0169, 2.5534, 2.5570, 2.2883], device='cuda:1'), covar=tensor([0.0725, 0.0878, 0.0923, 0.0796, 0.0849, 0.0774, 0.0883, 0.0649], device='cuda:1'), in_proj_covar=tensor([0.0210, 0.0222, 0.0225, 0.0240, 0.0226, 0.0212, 0.0187, 0.0204], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-03 00:25:34,798 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=155207.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:25:44,533 INFO [train.py:903] (1/4) Epoch 23, batch 5000, loss[loss=0.2027, simple_loss=0.2921, pruned_loss=0.05665, over 19320.00 frames. ], tot_loss[loss=0.2086, simple_loss=0.2885, pruned_loss=0.06436, over 3822511.55 frames. ], batch size: 66, lr: 3.56e-03, grad_scale: 8.0 2023-04-03 00:25:52,524 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_rvb from training. Duration: 27.14 2023-04-03 00:26:02,939 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=155232.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:26:03,632 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 2023-04-03 00:26:19,060 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.096e+02 4.751e+02 5.889e+02 7.363e+02 1.722e+03, threshold=1.178e+03, percent-clipped=5.0 2023-04-03 00:26:28,348 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2484, 1.4702, 1.8396, 1.4743, 2.9546, 4.5622, 4.4487, 4.9589], device='cuda:1'), covar=tensor([0.1695, 0.3819, 0.3498, 0.2409, 0.0650, 0.0194, 0.0185, 0.0201], device='cuda:1'), in_proj_covar=tensor([0.0274, 0.0324, 0.0353, 0.0265, 0.0246, 0.0188, 0.0217, 0.0261], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-03 00:26:30,619 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([0.8234, 1.1089, 1.4425, 0.5854, 1.9234, 2.1862, 1.9580, 2.3532], device='cuda:1'), covar=tensor([0.1661, 0.3668, 0.3221, 0.2850, 0.0787, 0.0354, 0.0381, 0.0438], device='cuda:1'), in_proj_covar=tensor([0.0274, 0.0324, 0.0353, 0.0265, 0.0246, 0.0188, 0.0217, 0.0261], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-03 00:26:43,566 INFO [train.py:903] (1/4) Epoch 23, batch 5050, loss[loss=0.265, simple_loss=0.3426, pruned_loss=0.09368, over 17492.00 frames. ], tot_loss[loss=0.2079, simple_loss=0.2881, pruned_loss=0.06386, over 3821031.50 frames. ], batch size: 101, lr: 3.56e-03, grad_scale: 8.0 2023-04-03 00:27:17,602 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9_rvb from training. Duration: 26.62775 2023-04-03 00:27:42,608 INFO [train.py:903] (1/4) Epoch 23, batch 5100, loss[loss=0.2429, simple_loss=0.3129, pruned_loss=0.08647, over 13155.00 frames. ], tot_loss[loss=0.2084, simple_loss=0.2884, pruned_loss=0.06422, over 3820649.90 frames. ], batch size: 136, lr: 3.56e-03, grad_scale: 8.0 2023-04-03 00:27:53,102 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 2023-04-03 00:27:56,487 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_rvb from training. Duration: 25.035 2023-04-03 00:28:01,507 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_rvb from training. Duration: 27.92 2023-04-03 00:28:10,591 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=155338.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:28:18,269 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.450e+02 5.084e+02 6.467e+02 7.878e+02 1.414e+03, threshold=1.293e+03, percent-clipped=6.0 2023-04-03 00:28:36,971 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=155361.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:28:43,673 INFO [train.py:903] (1/4) Epoch 23, batch 5150, loss[loss=0.223, simple_loss=0.3007, pruned_loss=0.07263, over 19693.00 frames. ], tot_loss[loss=0.2074, simple_loss=0.2876, pruned_loss=0.06366, over 3823659.95 frames. ], batch size: 59, lr: 3.55e-03, grad_scale: 8.0 2023-04-03 00:28:56,722 WARNING [train.py:1073] (1/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 2023-04-03 00:29:03,276 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 2023-04-03 00:29:08,592 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=155386.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:29:31,259 WARNING [train.py:1073] (1/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1_rvb from training. Duration: 0.9681875 2023-04-03 00:29:45,063 INFO [train.py:903] (1/4) Epoch 23, batch 5200, loss[loss=0.1965, simple_loss=0.2813, pruned_loss=0.05585, over 19537.00 frames. ], tot_loss[loss=0.2088, simple_loss=0.2887, pruned_loss=0.06448, over 3830764.64 frames. ], batch size: 56, lr: 3.55e-03, grad_scale: 8.0 2023-04-03 00:29:58,658 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 2023-04-03 00:30:02,272 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=155431.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:30:19,766 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.141e+02 5.305e+02 6.432e+02 7.969e+02 2.733e+03, threshold=1.286e+03, percent-clipped=6.0 2023-04-03 00:30:30,727 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=155453.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:30:30,794 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=155453.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:30:41,483 WARNING [train.py:1073] (1/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 2023-04-03 00:30:44,913 INFO [train.py:903] (1/4) Epoch 23, batch 5250, loss[loss=0.2118, simple_loss=0.2934, pruned_loss=0.06514, over 19590.00 frames. ], tot_loss[loss=0.2092, simple_loss=0.289, pruned_loss=0.06466, over 3815931.12 frames. ], batch size: 61, lr: 3.55e-03, grad_scale: 8.0 2023-04-03 00:30:55,549 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=155475.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:30:59,146 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=155478.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:31:35,362 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.7400, 1.6321, 1.4963, 2.1575, 1.6351, 2.0820, 2.0334, 1.8829], device='cuda:1'), covar=tensor([0.0811, 0.0893, 0.1002, 0.0743, 0.0879, 0.0705, 0.0832, 0.0649], device='cuda:1'), in_proj_covar=tensor([0.0210, 0.0220, 0.0225, 0.0239, 0.0225, 0.0212, 0.0186, 0.0203], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-03 00:31:45,243 INFO [train.py:903] (1/4) Epoch 23, batch 5300, loss[loss=0.2243, simple_loss=0.2983, pruned_loss=0.0751, over 19287.00 frames. ], tot_loss[loss=0.2078, simple_loss=0.2875, pruned_loss=0.06404, over 3825268.06 frames. ], batch size: 66, lr: 3.55e-03, grad_scale: 8.0 2023-04-03 00:32:03,706 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 2023-04-03 00:32:06,104 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4711, 1.5089, 1.7901, 1.6622, 2.4480, 2.1457, 2.4919, 1.2176], device='cuda:1'), covar=tensor([0.2555, 0.4466, 0.2815, 0.2119, 0.1613, 0.2314, 0.1574, 0.4668], device='cuda:1'), in_proj_covar=tensor([0.0537, 0.0647, 0.0718, 0.0489, 0.0621, 0.0535, 0.0659, 0.0552], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 00:32:21,414 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.105e+02 4.703e+02 5.856e+02 7.687e+02 1.612e+03, threshold=1.171e+03, percent-clipped=4.0 2023-04-03 00:32:22,915 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=155546.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:32:35,456 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4588, 1.5435, 1.8098, 1.6605, 2.8365, 2.2753, 2.9389, 1.4234], device='cuda:1'), covar=tensor([0.2553, 0.4341, 0.2725, 0.1996, 0.1468, 0.2224, 0.1456, 0.4205], device='cuda:1'), in_proj_covar=tensor([0.0538, 0.0649, 0.0720, 0.0490, 0.0623, 0.0536, 0.0661, 0.0554], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 00:32:46,437 INFO [train.py:903] (1/4) Epoch 23, batch 5350, loss[loss=0.2348, simple_loss=0.3056, pruned_loss=0.08199, over 13248.00 frames. ], tot_loss[loss=0.208, simple_loss=0.2881, pruned_loss=0.06401, over 3823252.64 frames. ], batch size: 136, lr: 3.55e-03, grad_scale: 8.0 2023-04-03 00:33:18,091 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9_rvb from training. Duration: 26.438875 2023-04-03 00:33:46,934 INFO [train.py:903] (1/4) Epoch 23, batch 5400, loss[loss=0.2095, simple_loss=0.2792, pruned_loss=0.06988, over 19488.00 frames. ], tot_loss[loss=0.2069, simple_loss=0.287, pruned_loss=0.06338, over 3817315.79 frames. ], batch size: 49, lr: 3.55e-03, grad_scale: 8.0 2023-04-03 00:33:56,241 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=155623.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:34:21,898 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.892e+02 4.747e+02 5.806e+02 7.220e+02 1.360e+03, threshold=1.161e+03, percent-clipped=1.0 2023-04-03 00:34:48,075 INFO [train.py:903] (1/4) Epoch 23, batch 5450, loss[loss=0.223, simple_loss=0.3029, pruned_loss=0.07151, over 19670.00 frames. ], tot_loss[loss=0.2056, simple_loss=0.2858, pruned_loss=0.06265, over 3823613.15 frames. ], batch size: 55, lr: 3.55e-03, grad_scale: 8.0 2023-04-03 00:34:58,361 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=155675.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:35:08,414 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.2690, 2.2806, 2.4596, 2.9657, 2.2289, 2.8781, 2.6891, 2.3566], device='cuda:1'), covar=tensor([0.4371, 0.4042, 0.1955, 0.2635, 0.4449, 0.2193, 0.4513, 0.3316], device='cuda:1'), in_proj_covar=tensor([0.0908, 0.0977, 0.0726, 0.0938, 0.0888, 0.0824, 0.0846, 0.0789], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-03 00:35:39,952 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=155709.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:35:47,565 INFO [train.py:903] (1/4) Epoch 23, batch 5500, loss[loss=0.2033, simple_loss=0.2834, pruned_loss=0.06161, over 18138.00 frames. ], tot_loss[loss=0.2058, simple_loss=0.2863, pruned_loss=0.06268, over 3832570.59 frames. ], batch size: 83, lr: 3.55e-03, grad_scale: 4.0 2023-04-03 00:36:10,021 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=155734.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:36:13,480 WARNING [train.py:1073] (1/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9_rvb from training. Duration: 27.47775 2023-04-03 00:36:24,171 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.224e+02 5.057e+02 6.298e+02 8.158e+02 1.659e+03, threshold=1.260e+03, percent-clipped=6.0 2023-04-03 00:36:46,706 INFO [train.py:903] (1/4) Epoch 23, batch 5550, loss[loss=0.2075, simple_loss=0.292, pruned_loss=0.06146, over 19580.00 frames. ], tot_loss[loss=0.2063, simple_loss=0.2865, pruned_loss=0.06302, over 3831652.76 frames. ], batch size: 52, lr: 3.55e-03, grad_scale: 4.0 2023-04-03 00:36:50,863 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5067, 1.5032, 1.4453, 1.9792, 1.6124, 1.8077, 1.8590, 1.7106], device='cuda:1'), covar=tensor([0.0898, 0.0959, 0.1053, 0.0700, 0.0788, 0.0757, 0.0767, 0.0695], device='cuda:1'), in_proj_covar=tensor([0.0211, 0.0221, 0.0224, 0.0240, 0.0225, 0.0212, 0.0186, 0.0204], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-03 00:36:56,205 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 2023-04-03 00:37:30,479 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=155802.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:37:30,612 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=155802.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:37:42,232 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 2023-04-03 00:37:48,755 INFO [train.py:903] (1/4) Epoch 23, batch 5600, loss[loss=0.2221, simple_loss=0.3009, pruned_loss=0.07164, over 18736.00 frames. ], tot_loss[loss=0.2072, simple_loss=0.2872, pruned_loss=0.06361, over 3819588.74 frames. ], batch size: 74, lr: 3.55e-03, grad_scale: 8.0 2023-04-03 00:37:52,340 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=155819.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:38:02,587 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=155827.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:38:23,384 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.767e+02 5.365e+02 7.033e+02 8.601e+02 1.530e+03, threshold=1.407e+03, percent-clipped=6.0 2023-04-03 00:38:48,661 INFO [train.py:903] (1/4) Epoch 23, batch 5650, loss[loss=0.2611, simple_loss=0.3248, pruned_loss=0.09873, over 13961.00 frames. ], tot_loss[loss=0.2084, simple_loss=0.2883, pruned_loss=0.06431, over 3817851.16 frames. ], batch size: 137, lr: 3.55e-03, grad_scale: 8.0 2023-04-03 00:39:33,338 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9_rvb from training. Duration: 28.0944375 2023-04-03 00:39:47,802 INFO [train.py:903] (1/4) Epoch 23, batch 5700, loss[loss=0.2007, simple_loss=0.292, pruned_loss=0.05471, over 19545.00 frames. ], tot_loss[loss=0.209, simple_loss=0.2886, pruned_loss=0.06468, over 3804959.98 frames. ], batch size: 56, lr: 3.55e-03, grad_scale: 8.0 2023-04-03 00:39:49,800 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.31 vs. limit=2.0 2023-04-03 00:40:10,841 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=155934.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:40:24,771 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.238e+02 4.906e+02 6.184e+02 7.924e+02 2.131e+03, threshold=1.237e+03, percent-clipped=2.0 2023-04-03 00:40:35,575 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4532, 2.1408, 1.6874, 1.5452, 2.0207, 1.3673, 1.3483, 1.8487], device='cuda:1'), covar=tensor([0.1069, 0.0865, 0.1095, 0.0838, 0.0588, 0.1293, 0.0759, 0.0512], device='cuda:1'), in_proj_covar=tensor([0.0301, 0.0314, 0.0337, 0.0265, 0.0247, 0.0338, 0.0289, 0.0272], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 00:40:47,842 INFO [train.py:903] (1/4) Epoch 23, batch 5750, loss[loss=0.1753, simple_loss=0.267, pruned_loss=0.04184, over 19667.00 frames. ], tot_loss[loss=0.2084, simple_loss=0.2879, pruned_loss=0.06443, over 3816914.09 frames. ], batch size: 58, lr: 3.55e-03, grad_scale: 8.0 2023-04-03 00:40:49,184 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=155967.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:40:51,095 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9_rvb from training. Duration: 33.038875 2023-04-03 00:40:58,763 WARNING [train.py:1073] (1/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 2023-04-03 00:41:04,146 WARNING [train.py:1073] (1/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 2023-04-03 00:41:50,977 INFO [train.py:903] (1/4) Epoch 23, batch 5800, loss[loss=0.1859, simple_loss=0.2767, pruned_loss=0.04754, over 19528.00 frames. ], tot_loss[loss=0.2082, simple_loss=0.2877, pruned_loss=0.06433, over 3813011.26 frames. ], batch size: 54, lr: 3.55e-03, grad_scale: 8.0 2023-04-03 00:41:54,496 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=156019.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:42:25,072 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.381e+02 4.982e+02 6.301e+02 7.857e+02 1.493e+03, threshold=1.260e+03, percent-clipped=3.0 2023-04-03 00:42:50,181 INFO [train.py:903] (1/4) Epoch 23, batch 5850, loss[loss=0.1973, simple_loss=0.2802, pruned_loss=0.05722, over 19599.00 frames. ], tot_loss[loss=0.2074, simple_loss=0.2873, pruned_loss=0.06375, over 3820658.87 frames. ], batch size: 52, lr: 3.55e-03, grad_scale: 8.0 2023-04-03 00:43:08,224 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=156082.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:43:14,881 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=156088.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:43:48,537 INFO [train.py:903] (1/4) Epoch 23, batch 5900, loss[loss=0.1801, simple_loss=0.268, pruned_loss=0.04613, over 19710.00 frames. ], tot_loss[loss=0.2071, simple_loss=0.2873, pruned_loss=0.06345, over 3824934.42 frames. ], batch size: 45, lr: 3.55e-03, grad_scale: 8.0 2023-04-03 00:43:52,964 WARNING [train.py:1073] (1/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 2023-04-03 00:44:10,850 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=156134.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:44:14,060 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9_rvb from training. Duration: 27.511125 2023-04-03 00:44:24,575 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.271e+02 5.043e+02 6.365e+02 8.179e+02 2.050e+03, threshold=1.273e+03, percent-clipped=8.0 2023-04-03 00:44:25,628 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=156146.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:44:48,118 INFO [train.py:903] (1/4) Epoch 23, batch 5950, loss[loss=0.214, simple_loss=0.2927, pruned_loss=0.06766, over 19770.00 frames. ], tot_loss[loss=0.2087, simple_loss=0.2885, pruned_loss=0.06439, over 3819082.76 frames. ], batch size: 54, lr: 3.55e-03, grad_scale: 8.0 2023-04-03 00:45:19,174 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=156190.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:45:47,965 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=156215.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:45:48,677 INFO [train.py:903] (1/4) Epoch 23, batch 6000, loss[loss=0.2009, simple_loss=0.2791, pruned_loss=0.06131, over 19502.00 frames. ], tot_loss[loss=0.2076, simple_loss=0.2874, pruned_loss=0.0639, over 3833407.48 frames. ], batch size: 49, lr: 3.54e-03, grad_scale: 8.0 2023-04-03 00:45:48,677 INFO [train.py:928] (1/4) Computing validation loss 2023-04-03 00:46:01,142 INFO [train.py:937] (1/4) Epoch 23, validation: loss=0.1686, simple_loss=0.2684, pruned_loss=0.03439, over 944034.00 frames. 2023-04-03 00:46:01,143 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18821MB 2023-04-03 00:46:23,274 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=156234.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:46:37,247 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.373e+02 4.873e+02 6.527e+02 8.069e+02 1.468e+03, threshold=1.305e+03, percent-clipped=4.0 2023-04-03 00:46:47,420 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6408, 1.7628, 1.9549, 2.0219, 1.5614, 1.9694, 1.9623, 1.8441], device='cuda:1'), covar=tensor([0.4006, 0.3357, 0.1973, 0.2155, 0.3662, 0.2032, 0.4949, 0.3163], device='cuda:1'), in_proj_covar=tensor([0.0903, 0.0973, 0.0721, 0.0933, 0.0885, 0.0821, 0.0843, 0.0786], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-03 00:46:55,716 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=156261.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:47:01,915 INFO [train.py:903] (1/4) Epoch 23, batch 6050, loss[loss=0.2008, simple_loss=0.2872, pruned_loss=0.05717, over 19516.00 frames. ], tot_loss[loss=0.2083, simple_loss=0.2881, pruned_loss=0.06427, over 3828966.55 frames. ], batch size: 54, lr: 3.54e-03, grad_scale: 8.0 2023-04-03 00:47:02,325 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.0484, 2.8577, 2.3167, 2.1480, 2.0674, 2.4542, 1.1944, 2.0004], device='cuda:1'), covar=tensor([0.0688, 0.0682, 0.0692, 0.1166, 0.1124, 0.1172, 0.1427, 0.1172], device='cuda:1'), in_proj_covar=tensor([0.0359, 0.0354, 0.0360, 0.0385, 0.0464, 0.0391, 0.0338, 0.0342], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 00:47:53,128 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=156308.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:47:57,936 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.1054, 1.7965, 1.7215, 2.0078, 1.7649, 1.7996, 1.6502, 2.0074], device='cuda:1'), covar=tensor([0.1019, 0.1441, 0.1526, 0.1111, 0.1442, 0.0566, 0.1502, 0.0701], device='cuda:1'), in_proj_covar=tensor([0.0274, 0.0355, 0.0314, 0.0252, 0.0304, 0.0253, 0.0311, 0.0257], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 00:48:02,113 INFO [train.py:903] (1/4) Epoch 23, batch 6100, loss[loss=0.1822, simple_loss=0.2605, pruned_loss=0.05202, over 19604.00 frames. ], tot_loss[loss=0.2069, simple_loss=0.2866, pruned_loss=0.06357, over 3837953.43 frames. ], batch size: 52, lr: 3.54e-03, grad_scale: 8.0 2023-04-03 00:48:27,995 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=156338.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:48:37,367 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.018e+02 4.768e+02 6.249e+02 8.138e+02 1.749e+03, threshold=1.250e+03, percent-clipped=2.0 2023-04-03 00:48:58,835 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=156363.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:49:01,836 INFO [train.py:903] (1/4) Epoch 23, batch 6150, loss[loss=0.2799, simple_loss=0.3586, pruned_loss=0.1007, over 19681.00 frames. ], tot_loss[loss=0.2071, simple_loss=0.287, pruned_loss=0.06364, over 3832134.99 frames. ], batch size: 58, lr: 3.54e-03, grad_scale: 8.0 2023-04-03 00:49:31,045 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=156390.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:49:31,821 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9_rvb from training. Duration: 31.02225 2023-04-03 00:50:00,579 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=156415.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:50:01,337 INFO [train.py:903] (1/4) Epoch 23, batch 6200, loss[loss=0.1977, simple_loss=0.2845, pruned_loss=0.05545, over 19652.00 frames. ], tot_loss[loss=0.2053, simple_loss=0.2853, pruned_loss=0.06267, over 3841261.66 frames. ], batch size: 60, lr: 3.54e-03, grad_scale: 8.0 2023-04-03 00:50:20,444 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4470, 1.8020, 1.4451, 1.2956, 1.7837, 1.3116, 1.3620, 1.7637], device='cuda:1'), covar=tensor([0.0850, 0.0766, 0.0817, 0.0858, 0.0495, 0.1053, 0.0601, 0.0425], device='cuda:1'), in_proj_covar=tensor([0.0302, 0.0314, 0.0336, 0.0266, 0.0246, 0.0338, 0.0289, 0.0273], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 00:50:22,385 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=156432.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:50:38,862 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.999e+02 4.815e+02 5.704e+02 6.895e+02 2.552e+03, threshold=1.141e+03, percent-clipped=3.0 2023-04-03 00:51:02,800 INFO [train.py:903] (1/4) Epoch 23, batch 6250, loss[loss=0.299, simple_loss=0.3528, pruned_loss=0.1226, over 13787.00 frames. ], tot_loss[loss=0.2059, simple_loss=0.2858, pruned_loss=0.06305, over 3821549.53 frames. ], batch size: 136, lr: 3.54e-03, grad_scale: 8.0 2023-04-03 00:51:32,693 WARNING [train.py:1073] (1/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9_rvb from training. Duration: 25.988875 2023-04-03 00:52:02,820 INFO [train.py:903] (1/4) Epoch 23, batch 6300, loss[loss=0.2434, simple_loss=0.3231, pruned_loss=0.08189, over 19677.00 frames. ], tot_loss[loss=0.2066, simple_loss=0.2864, pruned_loss=0.06341, over 3822539.55 frames. ], batch size: 60, lr: 3.54e-03, grad_scale: 8.0 2023-04-03 00:52:04,461 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=156517.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:52:34,305 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=156542.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:52:39,263 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.268e+02 5.353e+02 6.743e+02 8.019e+02 1.408e+03, threshold=1.349e+03, percent-clipped=4.0 2023-04-03 00:52:40,672 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=156547.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:53:03,413 INFO [train.py:903] (1/4) Epoch 23, batch 6350, loss[loss=0.2394, simple_loss=0.3188, pruned_loss=0.07994, over 19653.00 frames. ], tot_loss[loss=0.206, simple_loss=0.2861, pruned_loss=0.06297, over 3827628.57 frames. ], batch size: 60, lr: 3.54e-03, grad_scale: 8.0 2023-04-03 00:53:17,257 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=156578.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:53:32,812 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2443, 1.1838, 1.8282, 1.8041, 3.0806, 4.8403, 4.6010, 5.1321], device='cuda:1'), covar=tensor([0.1628, 0.4827, 0.3957, 0.2035, 0.0601, 0.0147, 0.0219, 0.0216], device='cuda:1'), in_proj_covar=tensor([0.0275, 0.0325, 0.0354, 0.0265, 0.0246, 0.0190, 0.0217, 0.0261], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-03 00:54:02,673 INFO [train.py:903] (1/4) Epoch 23, batch 6400, loss[loss=0.1677, simple_loss=0.2448, pruned_loss=0.0453, over 19782.00 frames. ], tot_loss[loss=0.2056, simple_loss=0.2855, pruned_loss=0.06283, over 3841998.03 frames. ], batch size: 47, lr: 3.54e-03, grad_scale: 8.0 2023-04-03 00:54:39,391 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.510e+02 4.834e+02 5.927e+02 7.987e+02 2.615e+03, threshold=1.185e+03, percent-clipped=4.0 2023-04-03 00:54:46,361 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=156652.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:55:04,059 INFO [train.py:903] (1/4) Epoch 23, batch 6450, loss[loss=0.197, simple_loss=0.2865, pruned_loss=0.05378, over 19322.00 frames. ], tot_loss[loss=0.2056, simple_loss=0.2858, pruned_loss=0.06266, over 3832577.34 frames. ], batch size: 70, lr: 3.54e-03, grad_scale: 8.0 2023-04-03 00:55:35,929 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=156693.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:55:47,932 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9_rvb from training. Duration: 28.638875 2023-04-03 00:56:04,449 INFO [train.py:903] (1/4) Epoch 23, batch 6500, loss[loss=0.2338, simple_loss=0.309, pruned_loss=0.07927, over 17451.00 frames. ], tot_loss[loss=0.2054, simple_loss=0.2857, pruned_loss=0.06254, over 3848511.85 frames. ], batch size: 101, lr: 3.54e-03, grad_scale: 8.0 2023-04-03 00:56:10,080 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1_rvb from training. Duration: 0.836375 2023-04-03 00:56:11,235 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4360, 1.5303, 1.8590, 1.4680, 2.3641, 2.6889, 2.5814, 2.8560], device='cuda:1'), covar=tensor([0.1369, 0.3046, 0.2714, 0.2478, 0.1100, 0.0311, 0.0266, 0.0376], device='cuda:1'), in_proj_covar=tensor([0.0274, 0.0324, 0.0353, 0.0264, 0.0245, 0.0189, 0.0216, 0.0260], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-03 00:56:39,933 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.419e+02 5.077e+02 6.090e+02 8.057e+02 1.603e+03, threshold=1.218e+03, percent-clipped=6.0 2023-04-03 00:57:04,758 INFO [train.py:903] (1/4) Epoch 23, batch 6550, loss[loss=0.2082, simple_loss=0.2854, pruned_loss=0.06549, over 19693.00 frames. ], tot_loss[loss=0.2056, simple_loss=0.2857, pruned_loss=0.06277, over 3850175.99 frames. ], batch size: 59, lr: 3.54e-03, grad_scale: 8.0 2023-04-03 00:57:06,345 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=156767.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:57:06,415 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=156767.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:57:08,551 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([5.0634, 1.7653, 1.9672, 1.6589, 4.5653, 1.3298, 2.6237, 4.8947], device='cuda:1'), covar=tensor([0.0412, 0.2683, 0.2636, 0.2084, 0.0700, 0.2509, 0.1355, 0.0183], device='cuda:1'), in_proj_covar=tensor([0.0413, 0.0367, 0.0388, 0.0349, 0.0373, 0.0349, 0.0382, 0.0405], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 00:57:50,663 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=156803.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:58:04,880 INFO [train.py:903] (1/4) Epoch 23, batch 6600, loss[loss=0.2396, simple_loss=0.3171, pruned_loss=0.08101, over 18846.00 frames. ], tot_loss[loss=0.2062, simple_loss=0.2863, pruned_loss=0.06306, over 3834346.08 frames. ], batch size: 74, lr: 3.54e-03, grad_scale: 8.0 2023-04-03 00:58:20,332 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=156828.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 00:58:41,921 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.453e+02 5.163e+02 6.336e+02 8.010e+02 1.885e+03, threshold=1.267e+03, percent-clipped=5.0 2023-04-03 00:58:44,406 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3186, 1.4893, 1.9163, 1.4474, 2.8740, 3.8821, 3.5521, 4.0871], device='cuda:1'), covar=tensor([0.1533, 0.3623, 0.3059, 0.2309, 0.0545, 0.0174, 0.0207, 0.0245], device='cuda:1'), in_proj_covar=tensor([0.0274, 0.0324, 0.0353, 0.0264, 0.0245, 0.0190, 0.0216, 0.0260], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-03 00:59:05,156 INFO [train.py:903] (1/4) Epoch 23, batch 6650, loss[loss=0.2055, simple_loss=0.2912, pruned_loss=0.05991, over 19618.00 frames. ], tot_loss[loss=0.2064, simple_loss=0.2865, pruned_loss=0.06314, over 3840347.68 frames. ], batch size: 57, lr: 3.54e-03, grad_scale: 8.0 2023-04-03 00:59:51,159 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=156903.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 01:00:07,303 INFO [train.py:903] (1/4) Epoch 23, batch 6700, loss[loss=0.2493, simple_loss=0.3235, pruned_loss=0.08756, over 19575.00 frames. ], tot_loss[loss=0.2067, simple_loss=0.2865, pruned_loss=0.06342, over 3833458.53 frames. ], batch size: 61, lr: 3.54e-03, grad_scale: 8.0 2023-04-03 01:00:41,738 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.603e+02 5.252e+02 6.535e+02 7.903e+02 1.565e+03, threshold=1.307e+03, percent-clipped=2.0 2023-04-03 01:00:45,400 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=156949.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 01:01:04,456 INFO [train.py:903] (1/4) Epoch 23, batch 6750, loss[loss=0.2376, simple_loss=0.3222, pruned_loss=0.07655, over 17344.00 frames. ], tot_loss[loss=0.2072, simple_loss=0.2872, pruned_loss=0.06363, over 3835951.88 frames. ], batch size: 101, lr: 3.54e-03, grad_scale: 8.0 2023-04-03 01:01:13,626 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=156974.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 01:01:14,656 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.8678, 1.3164, 1.5272, 1.5293, 3.4675, 1.2009, 2.4739, 3.8974], device='cuda:1'), covar=tensor([0.0466, 0.2955, 0.2859, 0.1903, 0.0649, 0.2562, 0.1335, 0.0218], device='cuda:1'), in_proj_covar=tensor([0.0415, 0.0369, 0.0391, 0.0352, 0.0374, 0.0352, 0.0384, 0.0407], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 01:02:00,906 INFO [train.py:903] (1/4) Epoch 23, batch 6800, loss[loss=0.1673, simple_loss=0.2513, pruned_loss=0.04165, over 19638.00 frames. ], tot_loss[loss=0.2072, simple_loss=0.2869, pruned_loss=0.06375, over 3841611.19 frames. ], batch size: 50, lr: 3.54e-03, grad_scale: 8.0 2023-04-03 01:02:09,504 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=157023.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 01:02:14,936 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=157028.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 01:02:45,018 WARNING [train.py:1073] (1/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 2023-04-03 01:02:45,463 WARNING [train.py:1073] (1/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 2023-04-03 01:02:48,303 INFO [train.py:903] (1/4) Epoch 24, batch 0, loss[loss=0.1963, simple_loss=0.2738, pruned_loss=0.05935, over 19772.00 frames. ], tot_loss[loss=0.1963, simple_loss=0.2738, pruned_loss=0.05935, over 19772.00 frames. ], batch size: 49, lr: 3.46e-03, grad_scale: 8.0 2023-04-03 01:02:48,304 INFO [train.py:928] (1/4) Computing validation loss 2023-04-03 01:02:59,925 INFO [train.py:937] (1/4) Epoch 24, validation: loss=0.1683, simple_loss=0.2685, pruned_loss=0.03408, over 944034.00 frames. 2023-04-03 01:02:59,926 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18821MB 2023-04-03 01:03:03,176 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.276e+02 5.212e+02 6.445e+02 8.399e+02 3.393e+03, threshold=1.289e+03, percent-clipped=7.0 2023-04-03 01:03:05,662 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=157048.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 01:03:12,278 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 2023-04-03 01:03:57,953 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5363, 1.6194, 2.0411, 1.8125, 3.3470, 2.6778, 3.5732, 1.6201], device='cuda:1'), covar=tensor([0.2570, 0.4427, 0.2754, 0.1960, 0.1445, 0.2115, 0.1475, 0.4230], device='cuda:1'), in_proj_covar=tensor([0.0537, 0.0646, 0.0718, 0.0489, 0.0618, 0.0533, 0.0658, 0.0551], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 01:04:00,821 INFO [train.py:903] (1/4) Epoch 24, batch 50, loss[loss=0.1702, simple_loss=0.2511, pruned_loss=0.04465, over 19369.00 frames. ], tot_loss[loss=0.2107, simple_loss=0.2911, pruned_loss=0.06515, over 860406.62 frames. ], batch size: 47, lr: 3.46e-03, grad_scale: 8.0 2023-04-03 01:04:20,660 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=157111.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 01:04:32,470 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1_rvb from training. Duration: 27.0318125 2023-04-03 01:05:01,217 INFO [train.py:903] (1/4) Epoch 24, batch 100, loss[loss=0.1805, simple_loss=0.2663, pruned_loss=0.04733, over 19621.00 frames. ], tot_loss[loss=0.2053, simple_loss=0.2863, pruned_loss=0.06214, over 1521588.15 frames. ], batch size: 50, lr: 3.46e-03, grad_scale: 8.0 2023-04-03 01:05:03,494 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.245e+02 5.500e+02 6.534e+02 8.918e+02 1.825e+03, threshold=1.307e+03, percent-clipped=7.0 2023-04-03 01:05:11,339 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 2023-04-03 01:05:19,846 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=157160.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 01:06:02,079 INFO [train.py:903] (1/4) Epoch 24, batch 150, loss[loss=0.187, simple_loss=0.2604, pruned_loss=0.0568, over 19395.00 frames. ], tot_loss[loss=0.2062, simple_loss=0.2872, pruned_loss=0.06259, over 2023126.98 frames. ], batch size: 47, lr: 3.46e-03, grad_scale: 8.0 2023-04-03 01:06:42,430 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=157226.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 01:07:01,358 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 2023-04-03 01:07:02,474 INFO [train.py:903] (1/4) Epoch 24, batch 200, loss[loss=0.1682, simple_loss=0.2539, pruned_loss=0.0412, over 19397.00 frames. ], tot_loss[loss=0.2057, simple_loss=0.2868, pruned_loss=0.06231, over 2419828.79 frames. ], batch size: 48, lr: 3.46e-03, grad_scale: 8.0 2023-04-03 01:07:04,623 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.193e+02 4.992e+02 5.973e+02 7.088e+02 2.080e+03, threshold=1.195e+03, percent-clipped=2.0 2023-04-03 01:07:05,920 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=157247.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 01:08:03,056 INFO [train.py:903] (1/4) Epoch 24, batch 250, loss[loss=0.2119, simple_loss=0.2912, pruned_loss=0.06633, over 19529.00 frames. ], tot_loss[loss=0.2081, simple_loss=0.2886, pruned_loss=0.06385, over 2737903.09 frames. ], batch size: 54, lr: 3.46e-03, grad_scale: 8.0 2023-04-03 01:08:10,681 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.55 vs. limit=5.0 2023-04-03 01:09:03,288 INFO [train.py:903] (1/4) Epoch 24, batch 300, loss[loss=0.2112, simple_loss=0.3001, pruned_loss=0.06115, over 18719.00 frames. ], tot_loss[loss=0.2082, simple_loss=0.2884, pruned_loss=0.06402, over 2981663.54 frames. ], batch size: 74, lr: 3.46e-03, grad_scale: 8.0 2023-04-03 01:09:06,233 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.399e+02 5.405e+02 6.557e+02 9.024e+02 1.464e+03, threshold=1.311e+03, percent-clipped=9.0 2023-04-03 01:09:25,572 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=157362.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 01:09:36,397 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=157372.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 01:09:39,481 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=157374.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 01:10:05,066 INFO [train.py:903] (1/4) Epoch 24, batch 350, loss[loss=0.1842, simple_loss=0.2707, pruned_loss=0.04887, over 19469.00 frames. ], tot_loss[loss=0.2082, simple_loss=0.2881, pruned_loss=0.06411, over 3160552.07 frames. ], batch size: 49, lr: 3.46e-03, grad_scale: 8.0 2023-04-03 01:10:10,682 WARNING [train.py:1073] (1/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1_rvb from training. Duration: 0.9681875 2023-04-03 01:10:12,239 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=157400.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 01:10:27,236 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0374, 2.1203, 2.4073, 2.7422, 2.0821, 2.6012, 2.3936, 2.2095], device='cuda:1'), covar=tensor([0.4248, 0.4025, 0.1869, 0.2492, 0.4373, 0.2173, 0.4706, 0.3210], device='cuda:1'), in_proj_covar=tensor([0.0906, 0.0975, 0.0722, 0.0932, 0.0887, 0.0823, 0.0846, 0.0787], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-03 01:10:39,437 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=157422.0, num_to_drop=1, layers_to_drop={1} 2023-04-03 01:11:05,171 INFO [train.py:903] (1/4) Epoch 24, batch 400, loss[loss=0.226, simple_loss=0.3067, pruned_loss=0.07266, over 19338.00 frames. ], tot_loss[loss=0.2094, simple_loss=0.2893, pruned_loss=0.06474, over 3304038.07 frames. ], batch size: 70, lr: 3.46e-03, grad_scale: 8.0 2023-04-03 01:11:07,649 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.933e+02 4.825e+02 6.674e+02 8.153e+02 1.427e+03, threshold=1.335e+03, percent-clipped=2.0 2023-04-03 01:11:52,596 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=157482.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 01:11:58,245 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=157487.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 01:12:05,697 INFO [train.py:903] (1/4) Epoch 24, batch 450, loss[loss=0.2235, simple_loss=0.3051, pruned_loss=0.07092, over 19528.00 frames. ], tot_loss[loss=0.2091, simple_loss=0.2891, pruned_loss=0.06453, over 3425809.72 frames. ], batch size: 54, lr: 3.45e-03, grad_scale: 8.0 2023-04-03 01:12:11,725 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.5977, 1.2308, 1.5772, 1.2192, 2.2599, 1.0667, 2.1653, 2.5042], device='cuda:1'), covar=tensor([0.0717, 0.2751, 0.2626, 0.1742, 0.0839, 0.2106, 0.0966, 0.0453], device='cuda:1'), in_proj_covar=tensor([0.0416, 0.0371, 0.0391, 0.0352, 0.0375, 0.0352, 0.0386, 0.0408], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 01:12:20,132 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=157504.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 01:12:22,677 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.3486, 1.4410, 1.6183, 1.5807, 3.0062, 1.3038, 2.3428, 3.3514], device='cuda:1'), covar=tensor([0.0593, 0.2719, 0.2843, 0.1915, 0.0735, 0.2326, 0.1451, 0.0320], device='cuda:1'), in_proj_covar=tensor([0.0416, 0.0371, 0.0391, 0.0352, 0.0375, 0.0352, 0.0385, 0.0407], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 01:12:24,879 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=157507.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 01:12:40,671 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 2023-04-03 01:12:40,702 WARNING [train.py:1073] (1/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 2023-04-03 01:12:44,551 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=157524.0, num_to_drop=1, layers_to_drop={0} 2023-04-03 01:13:08,941 INFO [train.py:903] (1/4) Epoch 24, batch 500, loss[loss=0.1975, simple_loss=0.2641, pruned_loss=0.06543, over 19737.00 frames. ], tot_loss[loss=0.2064, simple_loss=0.2865, pruned_loss=0.06317, over 3521969.98 frames. ], batch size: 45, lr: 3.45e-03, grad_scale: 8.0 2023-04-03 01:13:12,135 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.617e+02 5.203e+02 6.096e+02 8.720e+02 1.456e+03, threshold=1.219e+03, percent-clipped=3.0 2023-04-03 01:14:02,828 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=157586.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 01:14:12,208 INFO [train.py:903] (1/4) Epoch 24, batch 550, loss[loss=0.2, simple_loss=0.2725, pruned_loss=0.0638, over 16426.00 frames. ], tot_loss[loss=0.2047, simple_loss=0.2852, pruned_loss=0.06212, over 3590771.64 frames. ], batch size: 36, lr: 3.45e-03, grad_scale: 8.0 2023-04-03 01:14:20,997 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.6423, 4.2369, 2.6749, 3.6852, 1.0071, 4.1691, 4.0555, 4.1171], device='cuda:1'), covar=tensor([0.0608, 0.0947, 0.1995, 0.0899, 0.3797, 0.0639, 0.0894, 0.1114], device='cuda:1'), in_proj_covar=tensor([0.0515, 0.0415, 0.0499, 0.0349, 0.0403, 0.0439, 0.0433, 0.0469], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 01:14:41,067 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=157618.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 01:14:42,112 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=157619.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 01:15:13,057 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=157643.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 01:15:13,770 INFO [train.py:903] (1/4) Epoch 24, batch 600, loss[loss=0.1804, simple_loss=0.276, pruned_loss=0.04242, over 19532.00 frames. ], tot_loss[loss=0.2054, simple_loss=0.2858, pruned_loss=0.06248, over 3622147.27 frames. ], batch size: 54, lr: 3.45e-03, grad_scale: 8.0 2023-04-03 01:15:15,912 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.272e+02 4.646e+02 5.574e+02 6.767e+02 1.170e+03, threshold=1.115e+03, percent-clipped=0.0 2023-04-03 01:15:53,052 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 2023-04-03 01:16:14,760 INFO [train.py:903] (1/4) Epoch 24, batch 650, loss[loss=0.1778, simple_loss=0.2531, pruned_loss=0.0513, over 19746.00 frames. ], tot_loss[loss=0.2066, simple_loss=0.2867, pruned_loss=0.06327, over 3663844.46 frames. ], batch size: 46, lr: 3.45e-03, grad_scale: 8.0 2023-04-03 01:16:45,957 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=157718.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 01:17:07,775 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3129, 1.3235, 1.5162, 1.4934, 2.3322, 2.0504, 2.5168, 1.0011], device='cuda:1'), covar=tensor([0.2706, 0.4561, 0.2810, 0.2193, 0.1690, 0.2334, 0.1513, 0.4651], device='cuda:1'), in_proj_covar=tensor([0.0542, 0.0651, 0.0725, 0.0494, 0.0623, 0.0537, 0.0664, 0.0555], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 01:17:14,713 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=157743.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 01:17:15,390 INFO [train.py:903] (1/4) Epoch 24, batch 700, loss[loss=0.2415, simple_loss=0.3086, pruned_loss=0.08725, over 19642.00 frames. ], tot_loss[loss=0.2076, simple_loss=0.2879, pruned_loss=0.06361, over 3704982.85 frames. ], batch size: 60, lr: 3.45e-03, grad_scale: 8.0 2023-04-03 01:17:15,547 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=157744.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 01:17:20,753 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.032e+02 5.232e+02 6.869e+02 8.195e+02 1.483e+03, threshold=1.374e+03, percent-clipped=7.0 2023-04-03 01:17:44,054 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=157766.0, num_to_drop=1, layers_to_drop={1} 2023-04-03 01:17:46,361 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=157768.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 01:18:19,076 INFO [train.py:903] (1/4) Epoch 24, batch 750, loss[loss=0.2221, simple_loss=0.305, pruned_loss=0.06962, over 19688.00 frames. ], tot_loss[loss=0.2078, simple_loss=0.288, pruned_loss=0.06383, over 3736291.67 frames. ], batch size: 59, lr: 3.45e-03, grad_scale: 8.0 2023-04-03 01:19:06,793 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=157833.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 01:19:19,804 INFO [train.py:903] (1/4) Epoch 24, batch 800, loss[loss=0.2026, simple_loss=0.2919, pruned_loss=0.05668, over 19537.00 frames. ], tot_loss[loss=0.2073, simple_loss=0.2877, pruned_loss=0.06347, over 3766410.09 frames. ], batch size: 56, lr: 3.45e-03, grad_scale: 8.0 2023-04-03 01:19:22,489 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.5408, 2.5062, 2.1738, 2.7260, 2.4610, 2.1196, 1.9207, 2.5415], device='cuda:1'), covar=tensor([0.0949, 0.1542, 0.1444, 0.1019, 0.1321, 0.0565, 0.1545, 0.0691], device='cuda:1'), in_proj_covar=tensor([0.0274, 0.0356, 0.0315, 0.0254, 0.0305, 0.0255, 0.0314, 0.0259], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 01:19:23,265 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.261e+02 5.074e+02 6.358e+02 8.526e+02 1.766e+03, threshold=1.272e+03, percent-clipped=4.0 2023-04-03 01:19:30,235 WARNING [train.py:1073] (1/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 2023-04-03 01:19:31,738 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.6516, 2.5067, 2.2444, 2.6906, 2.4362, 2.2649, 2.0485, 2.6199], device='cuda:1'), covar=tensor([0.1008, 0.1581, 0.1510, 0.1088, 0.1449, 0.0573, 0.1548, 0.0715], device='cuda:1'), in_proj_covar=tensor([0.0274, 0.0357, 0.0315, 0.0254, 0.0305, 0.0255, 0.0314, 0.0259], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 01:19:37,396 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=157859.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 01:19:48,487 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=157868.0, num_to_drop=1, layers_to_drop={1} 2023-04-03 01:19:52,852 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=157871.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 01:19:59,171 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=157875.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 01:20:05,652 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=157881.0, num_to_drop=1, layers_to_drop={1} 2023-04-03 01:20:20,136 INFO [train.py:903] (1/4) Epoch 24, batch 850, loss[loss=0.1983, simple_loss=0.2764, pruned_loss=0.06004, over 19484.00 frames. ], tot_loss[loss=0.2069, simple_loss=0.2872, pruned_loss=0.06323, over 3777392.73 frames. ], batch size: 49, lr: 3.45e-03, grad_scale: 8.0 2023-04-03 01:20:27,432 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=157900.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 01:21:05,408 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=157930.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 01:21:08,623 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9_rvb from training. Duration: 25.061125 2023-04-03 01:21:21,348 INFO [train.py:903] (1/4) Epoch 24, batch 900, loss[loss=0.1883, simple_loss=0.2616, pruned_loss=0.0575, over 19401.00 frames. ], tot_loss[loss=0.2061, simple_loss=0.2865, pruned_loss=0.06288, over 3785823.87 frames. ], batch size: 47, lr: 3.45e-03, grad_scale: 8.0 2023-04-03 01:21:25,792 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.893e+02 4.800e+02 5.808e+02 7.910e+02 1.683e+03, threshold=1.162e+03, percent-clipped=5.0 2023-04-03 01:21:31,665 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.6374, 4.0373, 4.5156, 4.5690, 1.9113, 4.3155, 3.5424, 3.9065], device='cuda:1'), covar=tensor([0.2422, 0.1544, 0.1014, 0.1236, 0.7230, 0.1713, 0.1281, 0.2191], device='cuda:1'), in_proj_covar=tensor([0.0789, 0.0756, 0.0965, 0.0841, 0.0844, 0.0727, 0.0579, 0.0887], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-03 01:22:10,074 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=157983.0, num_to_drop=1, layers_to_drop={1} 2023-04-03 01:22:22,960 INFO [train.py:903] (1/4) Epoch 24, batch 950, loss[loss=0.2133, simple_loss=0.2922, pruned_loss=0.06717, over 19678.00 frames. ], tot_loss[loss=0.2063, simple_loss=0.2867, pruned_loss=0.06297, over 3800406.20 frames. ], batch size: 55, lr: 3.45e-03, grad_scale: 4.0 2023-04-03 01:22:23,014 WARNING [train.py:1073] (1/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9_rvb from training. Duration: 26.32775 2023-04-03 01:22:38,872 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([6.1982, 5.7082, 3.2546, 4.8251, 0.8535, 5.8381, 5.6448, 5.8058], device='cuda:1'), covar=tensor([0.0334, 0.0692, 0.1591, 0.0693, 0.4148, 0.0434, 0.0690, 0.0905], device='cuda:1'), in_proj_covar=tensor([0.0512, 0.0413, 0.0496, 0.0347, 0.0401, 0.0438, 0.0432, 0.0466], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 01:23:26,860 INFO [train.py:903] (1/4) Epoch 24, batch 1000, loss[loss=0.1907, simple_loss=0.2812, pruned_loss=0.05011, over 19623.00 frames. ], tot_loss[loss=0.2071, simple_loss=0.2874, pruned_loss=0.06338, over 3801823.74 frames. ], batch size: 57, lr: 3.45e-03, grad_scale: 4.0 2023-04-03 01:23:28,194 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=158045.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 01:23:31,292 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.770e+02 5.075e+02 5.979e+02 8.043e+02 1.884e+03, threshold=1.196e+03, percent-clipped=5.0 2023-04-03 01:23:34,160 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.7184, 1.8108, 1.9089, 2.4535, 1.8466, 2.2900, 1.9747, 1.6449], device='cuda:1'), covar=tensor([0.4865, 0.4726, 0.2687, 0.2906, 0.4567, 0.2588, 0.6374, 0.5110], device='cuda:1'), in_proj_covar=tensor([0.0909, 0.0980, 0.0723, 0.0935, 0.0888, 0.0825, 0.0845, 0.0787], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-03 01:24:17,475 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 2023-04-03 01:24:22,235 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=158089.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 01:24:27,542 INFO [train.py:903] (1/4) Epoch 24, batch 1050, loss[loss=0.2117, simple_loss=0.2932, pruned_loss=0.06514, over 19672.00 frames. ], tot_loss[loss=0.2071, simple_loss=0.2875, pruned_loss=0.06332, over 3808671.39 frames. ], batch size: 58, lr: 3.45e-03, grad_scale: 4.0 2023-04-03 01:24:50,225 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=158114.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 01:24:52,312 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=158115.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 01:24:56,178 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 2023-04-03 01:25:18,971 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4387, 1.3521, 1.5605, 1.5656, 1.6894, 1.9562, 1.7120, 0.5707], device='cuda:1'), covar=tensor([0.2326, 0.4177, 0.2595, 0.1866, 0.1683, 0.2191, 0.1454, 0.4687], device='cuda:1'), in_proj_covar=tensor([0.0538, 0.0648, 0.0720, 0.0492, 0.0619, 0.0534, 0.0660, 0.0553], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 01:25:20,064 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=158137.0, num_to_drop=1, layers_to_drop={1} 2023-04-03 01:25:23,266 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=158140.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 01:25:23,354 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=158140.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 01:25:27,550 INFO [train.py:903] (1/4) Epoch 24, batch 1100, loss[loss=0.2314, simple_loss=0.3094, pruned_loss=0.07673, over 19356.00 frames. ], tot_loss[loss=0.2079, simple_loss=0.288, pruned_loss=0.06391, over 3815360.47 frames. ], batch size: 66, lr: 3.45e-03, grad_scale: 4.0 2023-04-03 01:25:31,913 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.699e+02 5.122e+02 6.777e+02 7.992e+02 2.032e+03, threshold=1.355e+03, percent-clipped=5.0 2023-04-03 01:25:51,776 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=158162.0, num_to_drop=1, layers_to_drop={1} 2023-04-03 01:26:28,898 INFO [train.py:903] (1/4) Epoch 24, batch 1150, loss[loss=0.2008, simple_loss=0.2866, pruned_loss=0.05743, over 19595.00 frames. ], tot_loss[loss=0.2082, simple_loss=0.2885, pruned_loss=0.06393, over 3827723.74 frames. ], batch size: 57, lr: 3.45e-03, grad_scale: 4.0 2023-04-03 01:26:56,849 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=158215.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 01:27:25,301 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=158239.0, num_to_drop=1, layers_to_drop={1} 2023-04-03 01:27:31,330 INFO [train.py:903] (1/4) Epoch 24, batch 1200, loss[loss=0.1891, simple_loss=0.2691, pruned_loss=0.05449, over 19618.00 frames. ], tot_loss[loss=0.2084, simple_loss=0.289, pruned_loss=0.06387, over 3835880.02 frames. ], batch size: 50, lr: 3.45e-03, grad_scale: 8.0 2023-04-03 01:27:37,768 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.973e+02 4.926e+02 5.852e+02 7.782e+02 1.430e+03, threshold=1.170e+03, percent-clipped=2.0 2023-04-03 01:27:56,377 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=158264.0, num_to_drop=1, layers_to_drop={0} 2023-04-03 01:28:02,802 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 2023-04-03 01:28:21,997 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6440, 1.7195, 2.0328, 2.0205, 1.5278, 1.9923, 2.0148, 1.8650], device='cuda:1'), covar=tensor([0.4176, 0.3857, 0.1897, 0.2382, 0.3921, 0.2085, 0.5044, 0.3294], device='cuda:1'), in_proj_covar=tensor([0.0908, 0.0981, 0.0725, 0.0939, 0.0890, 0.0826, 0.0846, 0.0788], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-03 01:28:34,707 INFO [train.py:903] (1/4) Epoch 24, batch 1250, loss[loss=0.205, simple_loss=0.2922, pruned_loss=0.05889, over 19324.00 frames. ], tot_loss[loss=0.2085, simple_loss=0.2889, pruned_loss=0.06406, over 3828631.44 frames. ], batch size: 70, lr: 3.45e-03, grad_scale: 8.0 2023-04-03 01:28:43,307 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=158301.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 01:29:15,062 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=158326.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 01:29:20,344 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=158330.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 01:29:35,684 INFO [train.py:903] (1/4) Epoch 24, batch 1300, loss[loss=0.2015, simple_loss=0.2783, pruned_loss=0.06239, over 19488.00 frames. ], tot_loss[loss=0.208, simple_loss=0.2884, pruned_loss=0.06378, over 3824213.74 frames. ], batch size: 49, lr: 3.45e-03, grad_scale: 8.0 2023-04-03 01:29:40,388 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.859e+02 5.283e+02 7.241e+02 8.945e+02 2.355e+03, threshold=1.448e+03, percent-clipped=9.0 2023-04-03 01:30:36,942 INFO [train.py:903] (1/4) Epoch 24, batch 1350, loss[loss=0.168, simple_loss=0.242, pruned_loss=0.047, over 19735.00 frames. ], tot_loss[loss=0.2077, simple_loss=0.2882, pruned_loss=0.06356, over 3831191.63 frames. ], batch size: 46, lr: 3.44e-03, grad_scale: 8.0 2023-04-03 01:30:47,162 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=158402.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 01:31:38,834 INFO [train.py:903] (1/4) Epoch 24, batch 1400, loss[loss=0.2488, simple_loss=0.3224, pruned_loss=0.08762, over 18732.00 frames. ], tot_loss[loss=0.208, simple_loss=0.2884, pruned_loss=0.06385, over 3834322.65 frames. ], batch size: 74, lr: 3.44e-03, grad_scale: 8.0 2023-04-03 01:31:43,422 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.414e+02 5.106e+02 6.656e+02 8.188e+02 2.197e+03, threshold=1.331e+03, percent-clipped=4.0 2023-04-03 01:32:03,726 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.0187, 1.0479, 1.4122, 1.3958, 2.4088, 1.1641, 2.2830, 2.9019], device='cuda:1'), covar=tensor([0.0788, 0.3769, 0.3291, 0.2202, 0.1255, 0.2704, 0.1304, 0.0484], device='cuda:1'), in_proj_covar=tensor([0.0417, 0.0371, 0.0391, 0.0352, 0.0375, 0.0354, 0.0386, 0.0408], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 01:32:16,295 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2998, 1.9621, 1.5880, 1.3525, 1.8397, 1.2582, 1.1426, 1.8079], device='cuda:1'), covar=tensor([0.0968, 0.0898, 0.1033, 0.0884, 0.0539, 0.1306, 0.0781, 0.0462], device='cuda:1'), in_proj_covar=tensor([0.0304, 0.0318, 0.0340, 0.0270, 0.0249, 0.0340, 0.0292, 0.0274], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 01:32:27,056 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=158484.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 01:32:39,522 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4539, 2.1345, 1.6585, 1.5113, 1.9763, 1.3461, 1.3682, 1.8424], device='cuda:1'), covar=tensor([0.0964, 0.0814, 0.1048, 0.0867, 0.0578, 0.1258, 0.0700, 0.0495], device='cuda:1'), in_proj_covar=tensor([0.0303, 0.0317, 0.0339, 0.0270, 0.0248, 0.0340, 0.0291, 0.0273], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 01:32:40,271 INFO [train.py:903] (1/4) Epoch 24, batch 1450, loss[loss=0.1962, simple_loss=0.2734, pruned_loss=0.05955, over 19700.00 frames. ], tot_loss[loss=0.2091, simple_loss=0.2894, pruned_loss=0.06442, over 3826900.51 frames. ], batch size: 53, lr: 3.44e-03, grad_scale: 8.0 2023-04-03 01:32:42,616 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_rvb from training. Duration: 25.85 2023-04-03 01:33:27,467 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.9512, 4.4124, 4.7445, 4.7398, 1.7436, 4.3999, 3.8530, 4.4694], device='cuda:1'), covar=tensor([0.1841, 0.0922, 0.0603, 0.0716, 0.6378, 0.0974, 0.0686, 0.1128], device='cuda:1'), in_proj_covar=tensor([0.0790, 0.0756, 0.0961, 0.0839, 0.0844, 0.0727, 0.0578, 0.0888], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-03 01:33:41,524 INFO [train.py:903] (1/4) Epoch 24, batch 1500, loss[loss=0.1898, simple_loss=0.2831, pruned_loss=0.04831, over 19525.00 frames. ], tot_loss[loss=0.2086, simple_loss=0.289, pruned_loss=0.06412, over 3835093.96 frames. ], batch size: 56, lr: 3.44e-03, grad_scale: 8.0 2023-04-03 01:33:46,123 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.633e+02 4.870e+02 5.968e+02 7.477e+02 1.869e+03, threshold=1.194e+03, percent-clipped=2.0 2023-04-03 01:34:33,952 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=158586.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 01:34:42,445 INFO [train.py:903] (1/4) Epoch 24, batch 1550, loss[loss=0.1649, simple_loss=0.2444, pruned_loss=0.04274, over 19730.00 frames. ], tot_loss[loss=0.2074, simple_loss=0.288, pruned_loss=0.06341, over 3843943.49 frames. ], batch size: 46, lr: 3.44e-03, grad_scale: 8.0 2023-04-03 01:34:48,606 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=158599.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 01:35:04,336 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=158611.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 01:35:46,052 INFO [train.py:903] (1/4) Epoch 24, batch 1600, loss[loss=0.2027, simple_loss=0.2884, pruned_loss=0.05845, over 19656.00 frames. ], tot_loss[loss=0.2066, simple_loss=0.2869, pruned_loss=0.06317, over 3815889.72 frames. ], batch size: 58, lr: 3.44e-03, grad_scale: 8.0 2023-04-03 01:35:51,812 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.433e+02 4.721e+02 6.000e+02 7.264e+02 1.836e+03, threshold=1.200e+03, percent-clipped=4.0 2023-04-03 01:36:12,024 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4657, 1.6326, 1.9973, 1.7314, 2.9434, 2.4609, 3.2824, 1.5059], device='cuda:1'), covar=tensor([0.2638, 0.4424, 0.2730, 0.2017, 0.1705, 0.2230, 0.1660, 0.4491], device='cuda:1'), in_proj_covar=tensor([0.0540, 0.0650, 0.0721, 0.0492, 0.0621, 0.0537, 0.0663, 0.0557], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 01:36:12,728 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9_rvb from training. Duration: 30.1555625 2023-04-03 01:36:28,083 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5495, 1.6675, 1.9193, 1.9406, 1.4617, 1.8627, 1.9446, 1.7767], device='cuda:1'), covar=tensor([0.4294, 0.3739, 0.1942, 0.2367, 0.3879, 0.2223, 0.5035, 0.3427], device='cuda:1'), in_proj_covar=tensor([0.0911, 0.0981, 0.0724, 0.0937, 0.0890, 0.0827, 0.0845, 0.0789], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-03 01:36:50,311 INFO [train.py:903] (1/4) Epoch 24, batch 1650, loss[loss=0.181, simple_loss=0.2656, pruned_loss=0.04821, over 19474.00 frames. ], tot_loss[loss=0.2068, simple_loss=0.2872, pruned_loss=0.06318, over 3820141.69 frames. ], batch size: 49, lr: 3.44e-03, grad_scale: 8.0 2023-04-03 01:37:52,881 INFO [train.py:903] (1/4) Epoch 24, batch 1700, loss[loss=0.1893, simple_loss=0.272, pruned_loss=0.05326, over 19776.00 frames. ], tot_loss[loss=0.2051, simple_loss=0.2859, pruned_loss=0.06219, over 3832437.27 frames. ], batch size: 47, lr: 3.44e-03, grad_scale: 8.0 2023-04-03 01:37:55,338 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=158746.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 01:37:57,376 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.031e+02 4.745e+02 5.793e+02 7.168e+02 1.456e+03, threshold=1.159e+03, percent-clipped=2.0 2023-04-03 01:38:00,925 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.3741, 3.9866, 2.8339, 3.5364, 0.9420, 3.9588, 3.7911, 3.9003], device='cuda:1'), covar=tensor([0.0674, 0.0972, 0.1722, 0.0868, 0.3883, 0.0720, 0.0905, 0.1091], device='cuda:1'), in_proj_covar=tensor([0.0512, 0.0414, 0.0497, 0.0346, 0.0403, 0.0438, 0.0430, 0.0463], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 01:38:35,610 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 2023-04-03 01:38:54,538 INFO [train.py:903] (1/4) Epoch 24, batch 1750, loss[loss=0.1727, simple_loss=0.2549, pruned_loss=0.04519, over 19755.00 frames. ], tot_loss[loss=0.2059, simple_loss=0.2863, pruned_loss=0.06277, over 3824879.49 frames. ], batch size: 47, lr: 3.44e-03, grad_scale: 8.0 2023-04-03 01:38:56,134 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=158795.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 01:39:03,092 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=158801.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 01:39:57,565 INFO [train.py:903] (1/4) Epoch 24, batch 1800, loss[loss=0.2027, simple_loss=0.2774, pruned_loss=0.06397, over 19652.00 frames. ], tot_loss[loss=0.206, simple_loss=0.2863, pruned_loss=0.06281, over 3823949.92 frames. ], batch size: 53, lr: 3.44e-03, grad_scale: 8.0 2023-04-03 01:40:02,408 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.024e+02 4.690e+02 5.979e+02 7.282e+02 2.087e+03, threshold=1.196e+03, percent-clipped=3.0 2023-04-03 01:40:08,473 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=158853.0, num_to_drop=1, layers_to_drop={1} 2023-04-03 01:40:11,872 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=158855.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 01:40:19,331 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=158861.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 01:40:40,540 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=158880.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 01:40:56,885 WARNING [train.py:1073] (1/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9_rvb from training. Duration: 25.0944375 2023-04-03 01:40:58,009 INFO [train.py:903] (1/4) Epoch 24, batch 1850, loss[loss=0.1999, simple_loss=0.2894, pruned_loss=0.05522, over 19598.00 frames. ], tot_loss[loss=0.2064, simple_loss=0.2867, pruned_loss=0.06306, over 3832757.80 frames. ], batch size: 57, lr: 3.44e-03, grad_scale: 8.0 2023-04-03 01:41:32,428 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9_rvb from training. Duration: 27.8166875 2023-04-03 01:41:49,535 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([6.2373, 5.6480, 3.2179, 4.9512, 1.2094, 5.7262, 5.5770, 5.7519], device='cuda:1'), covar=tensor([0.0329, 0.0781, 0.1744, 0.0651, 0.4039, 0.0489, 0.0788, 0.0824], device='cuda:1'), in_proj_covar=tensor([0.0515, 0.0416, 0.0499, 0.0347, 0.0404, 0.0440, 0.0432, 0.0465], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 01:42:00,397 INFO [train.py:903] (1/4) Epoch 24, batch 1900, loss[loss=0.1866, simple_loss=0.2668, pruned_loss=0.05324, over 19492.00 frames. ], tot_loss[loss=0.2056, simple_loss=0.2862, pruned_loss=0.06251, over 3821853.53 frames. ], batch size: 49, lr: 3.44e-03, grad_scale: 8.0 2023-04-03 01:42:04,927 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.445e+02 4.936e+02 5.873e+02 7.509e+02 2.125e+03, threshold=1.175e+03, percent-clipped=8.0 2023-04-03 01:42:18,827 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 2023-04-03 01:42:24,127 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9_rvb from training. Duration: 27.02225 2023-04-03 01:42:46,701 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 2023-04-03 01:43:00,391 INFO [train.py:903] (1/4) Epoch 24, batch 1950, loss[loss=0.1833, simple_loss=0.275, pruned_loss=0.04587, over 19779.00 frames. ], tot_loss[loss=0.2056, simple_loss=0.2861, pruned_loss=0.06256, over 3817231.54 frames. ], batch size: 56, lr: 3.44e-03, grad_scale: 8.0 2023-04-03 01:44:03,845 INFO [train.py:903] (1/4) Epoch 24, batch 2000, loss[loss=0.2212, simple_loss=0.3062, pruned_loss=0.06813, over 19373.00 frames. ], tot_loss[loss=0.2058, simple_loss=0.2862, pruned_loss=0.06266, over 3810583.11 frames. ], batch size: 70, lr: 3.44e-03, grad_scale: 8.0 2023-04-03 01:44:08,593 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.425e+02 5.106e+02 6.605e+02 9.481e+02 1.726e+03, threshold=1.321e+03, percent-clipped=5.0 2023-04-03 01:44:57,683 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=159087.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 01:44:59,858 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_rvb from training. Duration: 26.8349375 2023-04-03 01:45:06,688 INFO [train.py:903] (1/4) Epoch 24, batch 2050, loss[loss=0.1765, simple_loss=0.2594, pruned_loss=0.04678, over 19413.00 frames. ], tot_loss[loss=0.2064, simple_loss=0.287, pruned_loss=0.06291, over 3809245.75 frames. ], batch size: 48, lr: 3.44e-03, grad_scale: 8.0 2023-04-03 01:45:19,174 WARNING [train.py:1073] (1/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 2023-04-03 01:45:20,305 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 2023-04-03 01:45:35,038 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=159117.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 01:45:41,053 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 2023-04-03 01:46:02,787 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=159139.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 01:46:06,499 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=159142.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 01:46:08,201 INFO [train.py:903] (1/4) Epoch 24, batch 2100, loss[loss=0.2405, simple_loss=0.3084, pruned_loss=0.08625, over 19631.00 frames. ], tot_loss[loss=0.205, simple_loss=0.2858, pruned_loss=0.06211, over 3821401.27 frames. ], batch size: 50, lr: 3.44e-03, grad_scale: 8.0 2023-04-03 01:46:10,470 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=159145.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 01:46:13,635 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.527e+02 4.786e+02 5.922e+02 8.131e+02 1.881e+03, threshold=1.184e+03, percent-clipped=4.0 2023-04-03 01:46:37,350 WARNING [train.py:1073] (1/4) Exclude cut with ID 453-131332-0000-131866_sp0.9_rvb from training. Duration: 25.3333125 2023-04-03 01:46:49,814 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=159177.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 01:46:58,711 WARNING [train.py:1073] (1/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9_rvb from training. Duration: 26.6166875 2023-04-03 01:47:00,992 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.8446, 1.3144, 1.4877, 1.4205, 3.4266, 1.1410, 2.5424, 3.8942], device='cuda:1'), covar=tensor([0.0535, 0.2921, 0.3141, 0.2094, 0.0710, 0.2729, 0.1412, 0.0224], device='cuda:1'), in_proj_covar=tensor([0.0417, 0.0373, 0.0394, 0.0354, 0.0378, 0.0355, 0.0389, 0.0409], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 01:47:10,236 INFO [train.py:903] (1/4) Epoch 24, batch 2150, loss[loss=0.1822, simple_loss=0.2675, pruned_loss=0.04845, over 19655.00 frames. ], tot_loss[loss=0.2059, simple_loss=0.2865, pruned_loss=0.06264, over 3823187.81 frames. ], batch size: 53, lr: 3.44e-03, grad_scale: 8.0 2023-04-03 01:47:13,876 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=159197.0, num_to_drop=1, layers_to_drop={1} 2023-04-03 01:48:12,839 INFO [train.py:903] (1/4) Epoch 24, batch 2200, loss[loss=0.2233, simple_loss=0.3016, pruned_loss=0.07251, over 19849.00 frames. ], tot_loss[loss=0.2061, simple_loss=0.2862, pruned_loss=0.06305, over 3814141.37 frames. ], batch size: 52, lr: 3.44e-03, grad_scale: 8.0 2023-04-03 01:48:18,019 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.316e+02 4.881e+02 6.157e+02 7.813e+02 2.191e+03, threshold=1.231e+03, percent-clipped=6.0 2023-04-03 01:48:25,306 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=159254.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 01:48:33,265 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=159260.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 01:48:36,912 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.7489, 2.7543, 2.1231, 2.1211, 1.6552, 2.1925, 1.0651, 1.9720], device='cuda:1'), covar=tensor([0.1159, 0.0839, 0.0749, 0.1350, 0.1647, 0.1656, 0.1676, 0.1280], device='cuda:1'), in_proj_covar=tensor([0.0357, 0.0357, 0.0360, 0.0385, 0.0462, 0.0390, 0.0338, 0.0342], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 01:49:14,294 INFO [train.py:903] (1/4) Epoch 24, batch 2250, loss[loss=0.2464, simple_loss=0.3216, pruned_loss=0.08556, over 19777.00 frames. ], tot_loss[loss=0.2067, simple_loss=0.2866, pruned_loss=0.06338, over 3824190.83 frames. ], batch size: 54, lr: 3.44e-03, grad_scale: 8.0 2023-04-03 01:49:31,900 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=159308.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 01:49:37,433 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=159312.0, num_to_drop=1, layers_to_drop={0} 2023-04-03 01:50:16,681 INFO [train.py:903] (1/4) Epoch 24, batch 2300, loss[loss=0.1803, simple_loss=0.2568, pruned_loss=0.05189, over 19796.00 frames. ], tot_loss[loss=0.2067, simple_loss=0.2867, pruned_loss=0.06332, over 3821694.13 frames. ], batch size: 49, lr: 3.43e-03, grad_scale: 8.0 2023-04-03 01:50:21,055 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.893e+02 4.864e+02 6.208e+02 8.672e+02 1.812e+03, threshold=1.242e+03, percent-clipped=10.0 2023-04-03 01:50:31,289 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_rvb from training. Duration: 26.205 2023-04-03 01:51:19,168 INFO [train.py:903] (1/4) Epoch 24, batch 2350, loss[loss=0.2352, simple_loss=0.3164, pruned_loss=0.07698, over 19631.00 frames. ], tot_loss[loss=0.2057, simple_loss=0.2858, pruned_loss=0.06284, over 3830079.28 frames. ], batch size: 57, lr: 3.43e-03, grad_scale: 8.0 2023-04-03 01:52:00,180 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_rvb from training. Duration: 25.775 2023-04-03 01:52:04,788 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=159431.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 01:52:17,050 WARNING [train.py:1073] (1/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9_rvb from training. Duration: 25.45 2023-04-03 01:52:19,432 INFO [train.py:903] (1/4) Epoch 24, batch 2400, loss[loss=0.1956, simple_loss=0.2826, pruned_loss=0.05428, over 19745.00 frames. ], tot_loss[loss=0.2068, simple_loss=0.2872, pruned_loss=0.06324, over 3806780.36 frames. ], batch size: 63, lr: 3.43e-03, grad_scale: 8.0 2023-04-03 01:52:21,069 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.3578, 1.9277, 1.9326, 2.1451, 1.8222, 1.8753, 1.6989, 2.1448], device='cuda:1'), covar=tensor([0.0900, 0.1479, 0.1374, 0.1040, 0.1444, 0.0548, 0.1507, 0.0700], device='cuda:1'), in_proj_covar=tensor([0.0276, 0.0358, 0.0313, 0.0255, 0.0304, 0.0253, 0.0315, 0.0259], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 01:52:25,061 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.677e+02 4.946e+02 5.943e+02 8.368e+02 2.189e+03, threshold=1.189e+03, percent-clipped=6.0 2023-04-03 01:53:21,799 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1392, 1.2482, 1.6698, 1.0116, 2.2667, 2.9711, 2.6746, 3.1864], device='cuda:1'), covar=tensor([0.1686, 0.3960, 0.3411, 0.2868, 0.0717, 0.0270, 0.0285, 0.0336], device='cuda:1'), in_proj_covar=tensor([0.0274, 0.0325, 0.0355, 0.0265, 0.0246, 0.0190, 0.0216, 0.0262], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-03 01:53:22,620 INFO [train.py:903] (1/4) Epoch 24, batch 2450, loss[loss=0.2342, simple_loss=0.3203, pruned_loss=0.07405, over 19339.00 frames. ], tot_loss[loss=0.2056, simple_loss=0.2862, pruned_loss=0.06251, over 3821742.36 frames. ], batch size: 70, lr: 3.43e-03, grad_scale: 8.0 2023-04-03 01:53:42,140 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=159510.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 01:53:47,916 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 2023-04-03 01:53:50,000 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=159516.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 01:53:55,513 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=159521.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 01:54:14,824 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=159535.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 01:54:21,730 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=159541.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 01:54:24,657 INFO [train.py:903] (1/4) Epoch 24, batch 2500, loss[loss=0.1695, simple_loss=0.2464, pruned_loss=0.04629, over 19789.00 frames. ], tot_loss[loss=0.2065, simple_loss=0.2867, pruned_loss=0.06315, over 3801755.26 frames. ], batch size: 48, lr: 3.43e-03, grad_scale: 8.0 2023-04-03 01:54:27,451 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=159546.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 01:54:29,312 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.080e+02 4.941e+02 6.110e+02 7.649e+02 1.406e+03, threshold=1.222e+03, percent-clipped=1.0 2023-04-03 01:54:55,472 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=159568.0, num_to_drop=1, layers_to_drop={1} 2023-04-03 01:54:56,911 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.48 vs. limit=2.0 2023-04-03 01:55:25,723 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=159593.0, num_to_drop=1, layers_to_drop={1} 2023-04-03 01:55:26,480 INFO [train.py:903] (1/4) Epoch 24, batch 2550, loss[loss=0.1838, simple_loss=0.259, pruned_loss=0.05433, over 19763.00 frames. ], tot_loss[loss=0.2069, simple_loss=0.2871, pruned_loss=0.0633, over 3814639.99 frames. ], batch size: 46, lr: 3.43e-03, grad_scale: 8.0 2023-04-03 01:56:19,845 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=159636.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 01:56:20,621 WARNING [train.py:1073] (1/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 2023-04-03 01:56:29,402 INFO [train.py:903] (1/4) Epoch 24, batch 2600, loss[loss=0.2521, simple_loss=0.314, pruned_loss=0.09516, over 13853.00 frames. ], tot_loss[loss=0.208, simple_loss=0.2885, pruned_loss=0.06373, over 3809073.78 frames. ], batch size: 138, lr: 3.43e-03, grad_scale: 8.0 2023-04-03 01:56:34,954 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.356e+02 4.779e+02 5.928e+02 8.262e+02 1.528e+03, threshold=1.186e+03, percent-clipped=6.0 2023-04-03 01:56:36,559 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=159649.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 01:56:39,984 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=159652.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 01:57:31,603 INFO [train.py:903] (1/4) Epoch 24, batch 2650, loss[loss=0.2034, simple_loss=0.2892, pruned_loss=0.05882, over 19590.00 frames. ], tot_loss[loss=0.2081, simple_loss=0.2883, pruned_loss=0.06393, over 3801523.65 frames. ], batch size: 61, lr: 3.43e-03, grad_scale: 8.0 2023-04-03 01:57:34,998 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2413, 1.3513, 1.7357, 1.2523, 2.6845, 3.5400, 3.2428, 3.7611], device='cuda:1'), covar=tensor([0.1603, 0.3709, 0.3362, 0.2547, 0.0604, 0.0205, 0.0231, 0.0282], device='cuda:1'), in_proj_covar=tensor([0.0273, 0.0324, 0.0356, 0.0264, 0.0246, 0.0190, 0.0216, 0.0263], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-03 01:57:43,903 WARNING [train.py:1073] (1/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9_rvb from training. Duration: 27.25 2023-04-03 01:58:21,020 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3542, 1.4017, 1.5445, 1.5410, 1.7529, 1.8472, 1.7659, 0.5573], device='cuda:1'), covar=tensor([0.2726, 0.4457, 0.2809, 0.2119, 0.1712, 0.2489, 0.1506, 0.5160], device='cuda:1'), in_proj_covar=tensor([0.0544, 0.0656, 0.0727, 0.0495, 0.0625, 0.0537, 0.0667, 0.0560], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-03 01:58:34,705 INFO [train.py:903] (1/4) Epoch 24, batch 2700, loss[loss=0.1921, simple_loss=0.261, pruned_loss=0.06162, over 19730.00 frames. ], tot_loss[loss=0.2084, simple_loss=0.2887, pruned_loss=0.06406, over 3802885.86 frames. ], batch size: 45, lr: 3.43e-03, grad_scale: 8.0 2023-04-03 01:58:39,053 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.760e+02 5.237e+02 6.508e+02 8.466e+02 2.382e+03, threshold=1.302e+03, percent-clipped=8.0 2023-04-03 01:59:03,878 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=159767.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 01:59:36,001 INFO [train.py:903] (1/4) Epoch 24, batch 2750, loss[loss=0.1878, simple_loss=0.2619, pruned_loss=0.05684, over 19763.00 frames. ], tot_loss[loss=0.2081, simple_loss=0.2884, pruned_loss=0.06391, over 3824420.42 frames. ], batch size: 46, lr: 3.43e-03, grad_scale: 8.0 2023-04-03 01:59:46,871 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=159802.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 02:00:19,530 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=159827.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 02:00:40,377 INFO [train.py:903] (1/4) Epoch 24, batch 2800, loss[loss=0.2273, simple_loss=0.3115, pruned_loss=0.07155, over 19672.00 frames. ], tot_loss[loss=0.2069, simple_loss=0.2872, pruned_loss=0.0633, over 3832490.88 frames. ], batch size: 60, lr: 3.43e-03, grad_scale: 8.0 2023-04-03 02:00:45,935 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.444e+02 4.661e+02 5.641e+02 7.181e+02 2.352e+03, threshold=1.128e+03, percent-clipped=2.0 2023-04-03 02:01:10,957 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.51 vs. limit=2.0 2023-04-03 02:01:16,194 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=159873.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 02:01:40,836 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=159892.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 02:01:42,762 INFO [train.py:903] (1/4) Epoch 24, batch 2850, loss[loss=0.2076, simple_loss=0.2947, pruned_loss=0.06023, over 19431.00 frames. ], tot_loss[loss=0.2066, simple_loss=0.287, pruned_loss=0.06309, over 3838300.99 frames. ], batch size: 70, lr: 3.43e-03, grad_scale: 8.0 2023-04-03 02:02:10,656 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=159917.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 02:02:25,913 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=159928.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 02:02:39,636 WARNING [train.py:1073] (1/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 2023-04-03 02:02:45,315 INFO [train.py:903] (1/4) Epoch 24, batch 2900, loss[loss=0.1938, simple_loss=0.271, pruned_loss=0.05825, over 19608.00 frames. ], tot_loss[loss=0.2063, simple_loss=0.2865, pruned_loss=0.06298, over 3840046.51 frames. ], batch size: 50, lr: 3.43e-03, grad_scale: 8.0 2023-04-03 02:02:51,070 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.210e+02 4.874e+02 6.501e+02 8.672e+02 1.518e+03, threshold=1.300e+03, percent-clipped=5.0 2023-04-03 02:03:45,530 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=159993.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 02:03:46,480 INFO [train.py:903] (1/4) Epoch 24, batch 2950, loss[loss=0.1876, simple_loss=0.2671, pruned_loss=0.05404, over 19760.00 frames. ], tot_loss[loss=0.2057, simple_loss=0.2858, pruned_loss=0.06284, over 3837316.64 frames. ], batch size: 48, lr: 3.43e-03, grad_scale: 8.0 2023-04-03 02:04:24,717 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=160023.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 02:04:45,836 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_rvb from training. Duration: 29.735 2023-04-03 02:04:49,458 INFO [train.py:903] (1/4) Epoch 24, batch 3000, loss[loss=0.2088, simple_loss=0.2947, pruned_loss=0.0615, over 19787.00 frames. ], tot_loss[loss=0.2057, simple_loss=0.2861, pruned_loss=0.06264, over 3835060.87 frames. ], batch size: 54, lr: 3.43e-03, grad_scale: 8.0 2023-04-03 02:04:49,458 INFO [train.py:928] (1/4) Computing validation loss 2023-04-03 02:05:01,999 INFO [train.py:937] (1/4) Epoch 24, validation: loss=0.1679, simple_loss=0.268, pruned_loss=0.03397, over 944034.00 frames. 2023-04-03 02:05:02,000 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18821MB 2023-04-03 02:05:08,064 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=160048.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 02:05:08,848 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.319e+02 4.966e+02 6.275e+02 7.790e+02 1.988e+03, threshold=1.255e+03, percent-clipped=5.0 2023-04-03 02:06:04,591 INFO [train.py:903] (1/4) Epoch 24, batch 3050, loss[loss=0.2855, simple_loss=0.3476, pruned_loss=0.1117, over 13059.00 frames. ], tot_loss[loss=0.2056, simple_loss=0.2861, pruned_loss=0.06258, over 3831124.88 frames. ], batch size: 135, lr: 3.43e-03, grad_scale: 8.0 2023-04-03 02:06:23,643 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=160108.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 02:06:37,819 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.80 vs. limit=2.0 2023-04-03 02:07:06,772 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.3109, 2.1765, 2.0877, 1.9658, 1.6888, 1.8763, 0.5518, 1.3396], device='cuda:1'), covar=tensor([0.0634, 0.0631, 0.0477, 0.0792, 0.1215, 0.0897, 0.1402, 0.1036], device='cuda:1'), in_proj_covar=tensor([0.0358, 0.0356, 0.0361, 0.0383, 0.0461, 0.0391, 0.0338, 0.0344], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 02:07:08,738 INFO [train.py:903] (1/4) Epoch 24, batch 3100, loss[loss=0.2261, simple_loss=0.3036, pruned_loss=0.07433, over 19543.00 frames. ], tot_loss[loss=0.2069, simple_loss=0.2876, pruned_loss=0.06315, over 3818760.79 frames. ], batch size: 56, lr: 3.43e-03, grad_scale: 8.0 2023-04-03 02:07:14,571 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.956e+02 4.859e+02 5.894e+02 7.109e+02 1.682e+03, threshold=1.179e+03, percent-clipped=4.0 2023-04-03 02:07:40,561 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.6802, 4.2243, 4.4708, 4.4620, 1.6139, 4.1988, 3.6171, 4.1830], device='cuda:1'), covar=tensor([0.1856, 0.0972, 0.0652, 0.0722, 0.6701, 0.1036, 0.0725, 0.1201], device='cuda:1'), in_proj_covar=tensor([0.0793, 0.0757, 0.0963, 0.0844, 0.0847, 0.0733, 0.0576, 0.0895], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-03 02:08:10,532 INFO [train.py:903] (1/4) Epoch 24, batch 3150, loss[loss=0.2345, simple_loss=0.3031, pruned_loss=0.08299, over 19523.00 frames. ], tot_loss[loss=0.2057, simple_loss=0.2865, pruned_loss=0.06243, over 3821372.82 frames. ], batch size: 54, lr: 3.43e-03, grad_scale: 8.0 2023-04-03 02:08:36,030 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 2023-04-03 02:08:39,725 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=160217.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 02:09:14,346 INFO [train.py:903] (1/4) Epoch 24, batch 3200, loss[loss=0.1881, simple_loss=0.2768, pruned_loss=0.04971, over 19757.00 frames. ], tot_loss[loss=0.2061, simple_loss=0.2867, pruned_loss=0.0628, over 3818733.01 frames. ], batch size: 54, lr: 3.42e-03, grad_scale: 8.0 2023-04-03 02:09:20,051 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.297e+02 5.134e+02 6.599e+02 8.700e+02 2.161e+03, threshold=1.320e+03, percent-clipped=4.0 2023-04-03 02:09:33,567 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([0.9451, 1.2572, 1.5968, 0.8857, 2.2827, 3.0254, 2.7447, 3.2101], device='cuda:1'), covar=tensor([0.1789, 0.3955, 0.3607, 0.2784, 0.0673, 0.0226, 0.0264, 0.0342], device='cuda:1'), in_proj_covar=tensor([0.0273, 0.0326, 0.0356, 0.0265, 0.0246, 0.0190, 0.0217, 0.0263], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-03 02:09:49,730 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=160272.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 02:10:17,065 INFO [train.py:903] (1/4) Epoch 24, batch 3250, loss[loss=0.1685, simple_loss=0.2472, pruned_loss=0.0449, over 19760.00 frames. ], tot_loss[loss=0.2047, simple_loss=0.2853, pruned_loss=0.06202, over 3830364.70 frames. ], batch size: 45, lr: 3.42e-03, grad_scale: 8.0 2023-04-03 02:10:56,761 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=160325.0, num_to_drop=1, layers_to_drop={0} 2023-04-03 02:11:04,807 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=160332.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 02:11:21,829 INFO [train.py:903] (1/4) Epoch 24, batch 3300, loss[loss=0.1729, simple_loss=0.2519, pruned_loss=0.047, over 19372.00 frames. ], tot_loss[loss=0.2057, simple_loss=0.2859, pruned_loss=0.06271, over 3822277.31 frames. ], batch size: 47, lr: 3.42e-03, grad_scale: 8.0 2023-04-03 02:11:24,399 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 2023-04-03 02:11:27,809 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.182e+02 4.973e+02 5.834e+02 7.675e+02 1.997e+03, threshold=1.167e+03, percent-clipped=3.0 2023-04-03 02:11:46,547 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=160364.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 02:12:15,652 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=160387.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 02:12:18,219 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=160389.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 02:12:23,808 INFO [train.py:903] (1/4) Epoch 24, batch 3350, loss[loss=0.1881, simple_loss=0.264, pruned_loss=0.05605, over 19383.00 frames. ], tot_loss[loss=0.2056, simple_loss=0.2858, pruned_loss=0.06265, over 3833357.85 frames. ], batch size: 47, lr: 3.42e-03, grad_scale: 8.0 2023-04-03 02:12:27,296 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.41 vs. limit=2.0 2023-04-03 02:12:39,927 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.54 vs. limit=2.0 2023-04-03 02:13:26,805 INFO [train.py:903] (1/4) Epoch 24, batch 3400, loss[loss=0.2054, simple_loss=0.2763, pruned_loss=0.06729, over 19726.00 frames. ], tot_loss[loss=0.205, simple_loss=0.2852, pruned_loss=0.06241, over 3835657.75 frames. ], batch size: 51, lr: 3.42e-03, grad_scale: 8.0 2023-04-03 02:13:32,540 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.093e+02 5.133e+02 6.647e+02 9.203e+02 1.938e+03, threshold=1.329e+03, percent-clipped=8.0 2023-04-03 02:13:48,729 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4780, 1.4847, 1.7674, 1.7154, 2.5150, 2.1982, 2.5543, 1.2860], device='cuda:1'), covar=tensor([0.2485, 0.4283, 0.2742, 0.1922, 0.1461, 0.2131, 0.1485, 0.4394], device='cuda:1'), in_proj_covar=tensor([0.0543, 0.0654, 0.0728, 0.0494, 0.0622, 0.0537, 0.0665, 0.0558], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 02:14:28,021 INFO [train.py:903] (1/4) Epoch 24, batch 3450, loss[loss=0.219, simple_loss=0.3068, pruned_loss=0.06565, over 19836.00 frames. ], tot_loss[loss=0.2051, simple_loss=0.2856, pruned_loss=0.06231, over 3839430.51 frames. ], batch size: 52, lr: 3.42e-03, grad_scale: 8.0 2023-04-03 02:14:31,643 WARNING [train.py:1073] (1/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9_rvb from training. Duration: 25.2444375 2023-04-03 02:15:29,542 INFO [train.py:903] (1/4) Epoch 24, batch 3500, loss[loss=0.1765, simple_loss=0.2598, pruned_loss=0.04661, over 19588.00 frames. ], tot_loss[loss=0.2037, simple_loss=0.284, pruned_loss=0.06169, over 3836836.12 frames. ], batch size: 52, lr: 3.42e-03, grad_scale: 8.0 2023-04-03 02:15:38,055 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.068e+02 4.685e+02 5.887e+02 7.904e+02 2.662e+03, threshold=1.177e+03, percent-clipped=4.0 2023-04-03 02:15:42,029 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([0.9902, 1.3340, 1.0691, 0.9724, 1.2313, 0.8774, 1.0046, 1.2394], device='cuda:1'), covar=tensor([0.0664, 0.0682, 0.0774, 0.0642, 0.0443, 0.1039, 0.0486, 0.0365], device='cuda:1'), in_proj_covar=tensor([0.0302, 0.0314, 0.0336, 0.0266, 0.0246, 0.0339, 0.0289, 0.0273], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 02:15:50,496 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 2023-04-03 02:16:26,898 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=160588.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 02:16:33,578 INFO [train.py:903] (1/4) Epoch 24, batch 3550, loss[loss=0.1784, simple_loss=0.2619, pruned_loss=0.04741, over 19848.00 frames. ], tot_loss[loss=0.2051, simple_loss=0.2853, pruned_loss=0.06248, over 3816239.89 frames. ], batch size: 52, lr: 3.42e-03, grad_scale: 8.0 2023-04-03 02:16:50,465 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.0262, 4.4401, 4.7849, 4.7970, 1.7690, 4.5007, 3.8783, 4.4883], device='cuda:1'), covar=tensor([0.1623, 0.0939, 0.0555, 0.0659, 0.5784, 0.0918, 0.0687, 0.1078], device='cuda:1'), in_proj_covar=tensor([0.0793, 0.0757, 0.0963, 0.0842, 0.0846, 0.0733, 0.0576, 0.0895], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-03 02:16:57,371 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=160613.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 02:16:58,911 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 2023-04-03 02:17:20,640 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.81 vs. limit=2.0 2023-04-03 02:17:36,600 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=160643.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 02:17:37,270 INFO [train.py:903] (1/4) Epoch 24, batch 3600, loss[loss=0.1945, simple_loss=0.2645, pruned_loss=0.06225, over 19749.00 frames. ], tot_loss[loss=0.2056, simple_loss=0.2856, pruned_loss=0.06283, over 3800671.85 frames. ], batch size: 45, lr: 3.42e-03, grad_scale: 8.0 2023-04-03 02:17:44,412 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.188e+02 4.811e+02 5.669e+02 7.209e+02 1.690e+03, threshold=1.134e+03, percent-clipped=3.0 2023-04-03 02:18:07,971 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=160668.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 02:18:08,811 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=160669.0, num_to_drop=1, layers_to_drop={1} 2023-04-03 02:18:30,058 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5356, 1.5794, 1.9021, 1.7828, 2.7353, 2.3681, 2.8800, 1.2126], device='cuda:1'), covar=tensor([0.2600, 0.4455, 0.2757, 0.2050, 0.1589, 0.2223, 0.1447, 0.4667], device='cuda:1'), in_proj_covar=tensor([0.0541, 0.0652, 0.0725, 0.0493, 0.0621, 0.0535, 0.0663, 0.0556], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 02:18:40,045 INFO [train.py:903] (1/4) Epoch 24, batch 3650, loss[loss=0.1816, simple_loss=0.2637, pruned_loss=0.04968, over 19368.00 frames. ], tot_loss[loss=0.2048, simple_loss=0.2851, pruned_loss=0.06229, over 3817006.27 frames. ], batch size: 47, lr: 3.42e-03, grad_scale: 8.0 2023-04-03 02:18:48,164 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.7845, 4.2285, 4.4707, 4.4993, 1.7887, 4.2070, 3.6624, 4.2078], device='cuda:1'), covar=tensor([0.1668, 0.0913, 0.0640, 0.0682, 0.5664, 0.0941, 0.0700, 0.1128], device='cuda:1'), in_proj_covar=tensor([0.0794, 0.0757, 0.0962, 0.0842, 0.0847, 0.0733, 0.0577, 0.0894], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-03 02:19:40,904 INFO [train.py:903] (1/4) Epoch 24, batch 3700, loss[loss=0.2577, simple_loss=0.3367, pruned_loss=0.08937, over 13433.00 frames. ], tot_loss[loss=0.2069, simple_loss=0.2871, pruned_loss=0.06335, over 3817067.28 frames. ], batch size: 136, lr: 3.42e-03, grad_scale: 8.0 2023-04-03 02:19:49,418 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.079e+02 4.616e+02 6.347e+02 7.690e+02 1.972e+03, threshold=1.269e+03, percent-clipped=6.0 2023-04-03 02:20:30,326 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=160784.0, num_to_drop=1, layers_to_drop={0} 2023-04-03 02:20:44,502 INFO [train.py:903] (1/4) Epoch 24, batch 3750, loss[loss=0.1993, simple_loss=0.2777, pruned_loss=0.06044, over 19849.00 frames. ], tot_loss[loss=0.2073, simple_loss=0.287, pruned_loss=0.06379, over 3813936.68 frames. ], batch size: 52, lr: 3.42e-03, grad_scale: 8.0 2023-04-03 02:21:45,656 INFO [train.py:903] (1/4) Epoch 24, batch 3800, loss[loss=0.1807, simple_loss=0.274, pruned_loss=0.0437, over 19681.00 frames. ], tot_loss[loss=0.208, simple_loss=0.2878, pruned_loss=0.0641, over 3818105.95 frames. ], batch size: 53, lr: 3.42e-03, grad_scale: 8.0 2023-04-03 02:21:53,460 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.186e+02 4.635e+02 5.250e+02 7.046e+02 1.734e+03, threshold=1.050e+03, percent-clipped=4.0 2023-04-03 02:22:18,588 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9_rvb from training. Duration: 29.1166875 2023-04-03 02:22:22,443 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4736, 1.1591, 1.1297, 1.3818, 1.0815, 1.2587, 1.1384, 1.3413], device='cuda:1'), covar=tensor([0.1112, 0.1244, 0.1530, 0.0989, 0.1267, 0.0633, 0.1531, 0.0810], device='cuda:1'), in_proj_covar=tensor([0.0276, 0.0356, 0.0313, 0.0255, 0.0305, 0.0254, 0.0313, 0.0259], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 02:22:41,689 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([0.7679, 0.9227, 1.1865, 0.5863, 1.5153, 1.7398, 1.5785, 1.8042], device='cuda:1'), covar=tensor([0.1320, 0.2983, 0.2618, 0.2454, 0.0838, 0.0392, 0.0320, 0.0414], device='cuda:1'), in_proj_covar=tensor([0.0273, 0.0325, 0.0355, 0.0264, 0.0245, 0.0190, 0.0216, 0.0262], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-03 02:22:47,096 INFO [train.py:903] (1/4) Epoch 24, batch 3850, loss[loss=0.1985, simple_loss=0.2935, pruned_loss=0.05177, over 19525.00 frames. ], tot_loss[loss=0.2079, simple_loss=0.2875, pruned_loss=0.06408, over 3818144.93 frames. ], batch size: 56, lr: 3.42e-03, grad_scale: 8.0 2023-04-03 02:23:18,520 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=160919.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 02:23:48,689 INFO [train.py:903] (1/4) Epoch 24, batch 3900, loss[loss=0.1792, simple_loss=0.2575, pruned_loss=0.05042, over 16871.00 frames. ], tot_loss[loss=0.2078, simple_loss=0.2876, pruned_loss=0.064, over 3817211.77 frames. ], batch size: 37, lr: 3.42e-03, grad_scale: 8.0 2023-04-03 02:23:58,399 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.922e+02 5.134e+02 6.381e+02 7.692e+02 1.884e+03, threshold=1.276e+03, percent-clipped=12.0 2023-04-03 02:24:38,042 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=160983.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 02:24:53,495 INFO [train.py:903] (1/4) Epoch 24, batch 3950, loss[loss=0.1854, simple_loss=0.2614, pruned_loss=0.05467, over 19802.00 frames. ], tot_loss[loss=0.2067, simple_loss=0.2867, pruned_loss=0.06334, over 3816561.48 frames. ], batch size: 49, lr: 3.42e-03, grad_scale: 8.0 2023-04-03 02:24:57,072 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 2023-04-03 02:25:51,223 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=161040.0, num_to_drop=1, layers_to_drop={0} 2023-04-03 02:25:55,366 INFO [train.py:903] (1/4) Epoch 24, batch 4000, loss[loss=0.1853, simple_loss=0.2658, pruned_loss=0.05239, over 19369.00 frames. ], tot_loss[loss=0.206, simple_loss=0.286, pruned_loss=0.06298, over 3812984.39 frames. ], batch size: 47, lr: 3.42e-03, grad_scale: 8.0 2023-04-03 02:26:00,841 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=4.25 vs. limit=5.0 2023-04-03 02:26:03,423 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.651e+02 5.074e+02 6.327e+02 7.723e+02 1.762e+03, threshold=1.265e+03, percent-clipped=6.0 2023-04-03 02:26:08,671 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6466, 1.2619, 1.2527, 1.4458, 1.0756, 1.2880, 1.1779, 1.4254], device='cuda:1'), covar=tensor([0.1209, 0.1196, 0.1771, 0.1069, 0.1427, 0.0806, 0.1908, 0.0975], device='cuda:1'), in_proj_covar=tensor([0.0277, 0.0357, 0.0313, 0.0255, 0.0305, 0.0254, 0.0314, 0.0259], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 02:26:21,981 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=161065.0, num_to_drop=1, layers_to_drop={0} 2023-04-03 02:26:41,800 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1_rvb from training. Duration: 0.7545625 2023-04-03 02:26:57,071 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.13 vs. limit=2.0 2023-04-03 02:26:57,378 INFO [train.py:903] (1/4) Epoch 24, batch 4050, loss[loss=0.2137, simple_loss=0.2935, pruned_loss=0.0669, over 19649.00 frames. ], tot_loss[loss=0.2072, simple_loss=0.2868, pruned_loss=0.06375, over 3800588.54 frames. ], batch size: 58, lr: 3.42e-03, grad_scale: 8.0 2023-04-03 02:27:50,008 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9840, 1.8581, 1.8332, 2.0621, 1.9161, 1.7351, 1.8215, 1.9968], device='cuda:1'), covar=tensor([0.0859, 0.1257, 0.1147, 0.0800, 0.1015, 0.0481, 0.1124, 0.0562], device='cuda:1'), in_proj_covar=tensor([0.0276, 0.0357, 0.0313, 0.0254, 0.0305, 0.0254, 0.0314, 0.0259], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 02:27:57,727 INFO [train.py:903] (1/4) Epoch 24, batch 4100, loss[loss=0.1784, simple_loss=0.262, pruned_loss=0.0474, over 19661.00 frames. ], tot_loss[loss=0.207, simple_loss=0.2865, pruned_loss=0.06371, over 3794832.40 frames. ], batch size: 53, lr: 3.42e-03, grad_scale: 8.0 2023-04-03 02:27:58,488 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 2023-04-03 02:28:06,052 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.204e+02 4.915e+02 6.129e+02 7.795e+02 1.333e+03, threshold=1.226e+03, percent-clipped=1.0 2023-04-03 02:28:31,036 WARNING [train.py:1073] (1/4) Exclude cut with ID 497-129325-0061-9566_sp1.1_rvb from training. Duration: 0.97725 2023-04-03 02:28:41,668 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.8027, 1.8942, 2.2192, 2.2886, 1.7452, 2.2271, 2.2588, 2.0745], device='cuda:1'), covar=tensor([0.4389, 0.4174, 0.1971, 0.2590, 0.4397, 0.2419, 0.5097, 0.3494], device='cuda:1'), in_proj_covar=tensor([0.0919, 0.0989, 0.0729, 0.0940, 0.0894, 0.0830, 0.0852, 0.0790], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-03 02:29:00,662 INFO [train.py:903] (1/4) Epoch 24, batch 4150, loss[loss=0.2304, simple_loss=0.3078, pruned_loss=0.07648, over 13221.00 frames. ], tot_loss[loss=0.2067, simple_loss=0.2865, pruned_loss=0.0634, over 3791393.53 frames. ], batch size: 135, lr: 3.41e-03, grad_scale: 8.0 2023-04-03 02:29:02,586 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.66 vs. limit=2.0 2023-04-03 02:29:43,135 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.0592, 2.8116, 2.1509, 2.2127, 1.9525, 2.4071, 1.0849, 2.0348], device='cuda:1'), covar=tensor([0.0640, 0.0651, 0.0755, 0.1102, 0.1118, 0.1104, 0.1467, 0.1078], device='cuda:1'), in_proj_covar=tensor([0.0361, 0.0361, 0.0365, 0.0388, 0.0466, 0.0395, 0.0342, 0.0347], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 02:29:47,421 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6097, 1.5467, 1.6063, 2.0798, 1.6751, 1.9425, 1.9721, 1.8077], device='cuda:1'), covar=tensor([0.0869, 0.0958, 0.1009, 0.0806, 0.0902, 0.0760, 0.0886, 0.0680], device='cuda:1'), in_proj_covar=tensor([0.0213, 0.0222, 0.0226, 0.0239, 0.0224, 0.0212, 0.0188, 0.0203], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-03 02:29:50,574 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=161234.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 02:30:01,377 INFO [train.py:903] (1/4) Epoch 24, batch 4200, loss[loss=0.1864, simple_loss=0.2792, pruned_loss=0.04686, over 19654.00 frames. ], tot_loss[loss=0.2077, simple_loss=0.2876, pruned_loss=0.06388, over 3803627.24 frames. ], batch size: 58, lr: 3.41e-03, grad_scale: 8.0 2023-04-03 02:30:02,579 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 2023-04-03 02:30:08,532 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=161249.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 02:30:09,257 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.885e+02 4.684e+02 6.110e+02 7.845e+02 2.290e+03, threshold=1.222e+03, percent-clipped=7.0 2023-04-03 02:30:24,258 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=161263.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 02:31:03,259 INFO [train.py:903] (1/4) Epoch 24, batch 4250, loss[loss=0.1633, simple_loss=0.2401, pruned_loss=0.0433, over 19326.00 frames. ], tot_loss[loss=0.2065, simple_loss=0.2866, pruned_loss=0.0632, over 3810392.87 frames. ], batch size: 44, lr: 3.41e-03, grad_scale: 8.0 2023-04-03 02:31:17,080 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9_rvb from training. Duration: 29.816625 2023-04-03 02:31:28,262 WARNING [train.py:1073] (1/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 2023-04-03 02:31:44,255 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=161327.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 02:31:47,947 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.51 vs. limit=2.0 2023-04-03 02:32:04,741 INFO [train.py:903] (1/4) Epoch 24, batch 4300, loss[loss=0.2024, simple_loss=0.2838, pruned_loss=0.06049, over 19679.00 frames. ], tot_loss[loss=0.2052, simple_loss=0.2855, pruned_loss=0.06249, over 3814824.58 frames. ], batch size: 53, lr: 3.41e-03, grad_scale: 8.0 2023-04-03 02:32:09,679 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([5.1735, 5.1823, 6.0197, 6.0166, 1.9588, 5.6687, 4.7819, 5.6135], device='cuda:1'), covar=tensor([0.1805, 0.0908, 0.0558, 0.0619, 0.6233, 0.0727, 0.0646, 0.1202], device='cuda:1'), in_proj_covar=tensor([0.0794, 0.0760, 0.0961, 0.0842, 0.0844, 0.0728, 0.0575, 0.0893], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-03 02:32:12,548 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.040e+02 4.585e+02 5.768e+02 7.257e+02 2.214e+03, threshold=1.154e+03, percent-clipped=5.0 2023-04-03 02:32:38,074 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0185, 2.0142, 1.7947, 2.1241, 2.0236, 1.7882, 1.7833, 2.0098], device='cuda:1'), covar=tensor([0.1066, 0.1380, 0.1415, 0.0957, 0.1218, 0.0564, 0.1380, 0.0763], device='cuda:1'), in_proj_covar=tensor([0.0277, 0.0358, 0.0314, 0.0256, 0.0306, 0.0254, 0.0315, 0.0260], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 02:32:47,609 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=161378.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 02:32:47,677 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9904, 1.9230, 1.8736, 1.6715, 1.4593, 1.6843, 0.4436, 0.9482], device='cuda:1'), covar=tensor([0.0679, 0.0662, 0.0415, 0.0681, 0.1279, 0.0785, 0.1299, 0.1072], device='cuda:1'), in_proj_covar=tensor([0.0360, 0.0360, 0.0364, 0.0387, 0.0465, 0.0393, 0.0341, 0.0347], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 02:32:57,525 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 2023-04-03 02:33:06,262 INFO [train.py:903] (1/4) Epoch 24, batch 4350, loss[loss=0.18, simple_loss=0.2659, pruned_loss=0.04703, over 19654.00 frames. ], tot_loss[loss=0.2052, simple_loss=0.2857, pruned_loss=0.06238, over 3813228.87 frames. ], batch size: 53, lr: 3.41e-03, grad_scale: 8.0 2023-04-03 02:33:19,515 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5433, 1.4739, 1.4774, 1.9042, 1.5004, 1.6222, 1.7314, 1.5826], device='cuda:1'), covar=tensor([0.0864, 0.0965, 0.1048, 0.0613, 0.0793, 0.0828, 0.0857, 0.0731], device='cuda:1'), in_proj_covar=tensor([0.0213, 0.0222, 0.0226, 0.0238, 0.0224, 0.0212, 0.0188, 0.0203], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-03 02:33:37,843 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=161419.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 02:33:38,148 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.25 vs. limit=5.0 2023-04-03 02:34:07,079 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=161442.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 02:34:08,989 INFO [train.py:903] (1/4) Epoch 24, batch 4400, loss[loss=0.2162, simple_loss=0.2955, pruned_loss=0.06847, over 19667.00 frames. ], tot_loss[loss=0.2053, simple_loss=0.2855, pruned_loss=0.06254, over 3816245.88 frames. ], batch size: 55, lr: 3.41e-03, grad_scale: 8.0 2023-04-03 02:34:15,588 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.985e+02 5.009e+02 6.093e+02 7.233e+02 1.222e+03, threshold=1.219e+03, percent-clipped=3.0 2023-04-03 02:34:31,975 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_rvb from training. Duration: 25.285 2023-04-03 02:34:41,593 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 2023-04-03 02:34:50,948 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=161478.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 02:35:10,542 INFO [train.py:903] (1/4) Epoch 24, batch 4450, loss[loss=0.2512, simple_loss=0.329, pruned_loss=0.08671, over 19493.00 frames. ], tot_loss[loss=0.207, simple_loss=0.2871, pruned_loss=0.06348, over 3819185.19 frames. ], batch size: 64, lr: 3.41e-03, grad_scale: 8.0 2023-04-03 02:35:14,646 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 2023-04-03 02:35:49,135 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.1709, 2.0616, 1.9920, 1.7848, 1.7234, 1.7667, 0.5040, 1.1008], device='cuda:1'), covar=tensor([0.0719, 0.0675, 0.0470, 0.0784, 0.1167, 0.0899, 0.1424, 0.1127], device='cuda:1'), in_proj_covar=tensor([0.0359, 0.0359, 0.0361, 0.0385, 0.0463, 0.0391, 0.0339, 0.0345], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 02:36:10,890 INFO [train.py:903] (1/4) Epoch 24, batch 4500, loss[loss=0.1919, simple_loss=0.2795, pruned_loss=0.05217, over 19673.00 frames. ], tot_loss[loss=0.2075, simple_loss=0.2875, pruned_loss=0.06373, over 3819141.39 frames. ], batch size: 58, lr: 3.41e-03, grad_scale: 8.0 2023-04-03 02:36:17,771 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.067e+02 5.206e+02 6.185e+02 8.214e+02 2.130e+03, threshold=1.237e+03, percent-clipped=6.0 2023-04-03 02:36:48,198 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5782, 1.3248, 1.1848, 1.4392, 1.0832, 1.3140, 1.1662, 1.4249], device='cuda:1'), covar=tensor([0.1250, 0.1174, 0.1781, 0.1166, 0.1400, 0.0729, 0.1759, 0.0945], device='cuda:1'), in_proj_covar=tensor([0.0279, 0.0358, 0.0315, 0.0258, 0.0307, 0.0255, 0.0317, 0.0260], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 02:36:53,395 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=161578.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 02:36:57,974 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5697, 2.2851, 1.6498, 1.5691, 2.0682, 1.3205, 1.4452, 1.8897], device='cuda:1'), covar=tensor([0.1279, 0.0834, 0.1180, 0.0872, 0.0622, 0.1434, 0.0833, 0.0620], device='cuda:1'), in_proj_covar=tensor([0.0304, 0.0317, 0.0338, 0.0267, 0.0248, 0.0341, 0.0292, 0.0275], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 02:37:10,278 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=161593.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 02:37:12,186 INFO [train.py:903] (1/4) Epoch 24, batch 4550, loss[loss=0.1921, simple_loss=0.2824, pruned_loss=0.05088, over 19661.00 frames. ], tot_loss[loss=0.2071, simple_loss=0.287, pruned_loss=0.06358, over 3829987.82 frames. ], batch size: 55, lr: 3.41e-03, grad_scale: 8.0 2023-04-03 02:37:20,034 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9_rvb from training. Duration: 28.72225 2023-04-03 02:37:35,014 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4172, 1.5463, 1.8048, 1.6863, 2.6398, 2.2778, 2.8110, 1.2760], device='cuda:1'), covar=tensor([0.2668, 0.4595, 0.2859, 0.2057, 0.1729, 0.2257, 0.1638, 0.4628], device='cuda:1'), in_proj_covar=tensor([0.0546, 0.0656, 0.0732, 0.0497, 0.0629, 0.0540, 0.0667, 0.0560], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-03 02:37:44,745 WARNING [train.py:1073] (1/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 2023-04-03 02:38:01,553 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=161634.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 02:38:15,062 INFO [train.py:903] (1/4) Epoch 24, batch 4600, loss[loss=0.1866, simple_loss=0.2688, pruned_loss=0.05225, over 19617.00 frames. ], tot_loss[loss=0.2074, simple_loss=0.2873, pruned_loss=0.06381, over 3830456.38 frames. ], batch size: 50, lr: 3.41e-03, grad_scale: 8.0 2023-04-03 02:38:21,974 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.192e+02 4.894e+02 5.872e+02 7.852e+02 1.807e+03, threshold=1.174e+03, percent-clipped=3.0 2023-04-03 02:38:33,828 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=161659.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 02:39:15,627 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=161693.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 02:39:16,441 INFO [train.py:903] (1/4) Epoch 24, batch 4650, loss[loss=0.2302, simple_loss=0.3076, pruned_loss=0.07641, over 19691.00 frames. ], tot_loss[loss=0.2066, simple_loss=0.2866, pruned_loss=0.06328, over 3839098.77 frames. ], batch size: 59, lr: 3.41e-03, grad_scale: 8.0 2023-04-03 02:39:22,565 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=161698.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 02:39:23,689 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2695, 1.3490, 1.8560, 1.5927, 3.0517, 4.5175, 4.3961, 4.9621], device='cuda:1'), covar=tensor([0.1697, 0.3829, 0.3311, 0.2303, 0.0597, 0.0203, 0.0171, 0.0198], device='cuda:1'), in_proj_covar=tensor([0.0274, 0.0326, 0.0356, 0.0266, 0.0246, 0.0191, 0.0217, 0.0264], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-03 02:39:33,574 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_rvb from training. Duration: 0.92 2023-04-03 02:39:33,869 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=161708.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 02:39:44,940 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_rvb from training. Duration: 0.83 2023-04-03 02:39:53,568 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=161723.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 02:39:57,797 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3851, 2.0577, 1.6269, 1.4313, 1.8831, 1.3473, 1.3219, 1.8251], device='cuda:1'), covar=tensor([0.0947, 0.0799, 0.1282, 0.0854, 0.0568, 0.1401, 0.0719, 0.0473], device='cuda:1'), in_proj_covar=tensor([0.0302, 0.0315, 0.0338, 0.0267, 0.0247, 0.0340, 0.0291, 0.0274], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 02:40:19,304 INFO [train.py:903] (1/4) Epoch 24, batch 4700, loss[loss=0.2306, simple_loss=0.3104, pruned_loss=0.07543, over 19687.00 frames. ], tot_loss[loss=0.205, simple_loss=0.2851, pruned_loss=0.06244, over 3832137.13 frames. ], batch size: 60, lr: 3.41e-03, grad_scale: 8.0 2023-04-03 02:40:26,428 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.995e+02 5.147e+02 6.134e+02 7.606e+02 1.537e+03, threshold=1.227e+03, percent-clipped=3.0 2023-04-03 02:40:39,906 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9_rvb from training. Duration: 0.92225 2023-04-03 02:40:43,342 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=161763.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 02:41:20,592 INFO [train.py:903] (1/4) Epoch 24, batch 4750, loss[loss=0.201, simple_loss=0.2589, pruned_loss=0.07159, over 19031.00 frames. ], tot_loss[loss=0.2042, simple_loss=0.2844, pruned_loss=0.06202, over 3830558.37 frames. ], batch size: 42, lr: 3.41e-03, grad_scale: 8.0 2023-04-03 02:41:57,353 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=161822.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 02:42:09,923 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=161833.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 02:42:23,974 INFO [train.py:903] (1/4) Epoch 24, batch 4800, loss[loss=0.195, simple_loss=0.2728, pruned_loss=0.05862, over 19606.00 frames. ], tot_loss[loss=0.2038, simple_loss=0.2841, pruned_loss=0.06181, over 3826727.97 frames. ], batch size: 50, lr: 3.41e-03, grad_scale: 8.0 2023-04-03 02:42:31,546 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.081e+02 5.225e+02 6.101e+02 7.305e+02 1.695e+03, threshold=1.220e+03, percent-clipped=2.0 2023-04-03 02:43:04,858 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=161878.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 02:43:21,623 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.0847, 1.4096, 1.5695, 1.3402, 2.7153, 1.1909, 2.2210, 3.0833], device='cuda:1'), covar=tensor([0.0571, 0.2529, 0.2761, 0.1913, 0.0695, 0.2252, 0.1244, 0.0326], device='cuda:1'), in_proj_covar=tensor([0.0414, 0.0368, 0.0390, 0.0350, 0.0373, 0.0352, 0.0387, 0.0407], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 02:43:25,940 INFO [train.py:903] (1/4) Epoch 24, batch 4850, loss[loss=0.2028, simple_loss=0.2856, pruned_loss=0.06002, over 18177.00 frames. ], tot_loss[loss=0.2029, simple_loss=0.2832, pruned_loss=0.06133, over 3833165.22 frames. ], batch size: 83, lr: 3.41e-03, grad_scale: 8.0 2023-04-03 02:43:50,712 WARNING [train.py:1073] (1/4) Exclude cut with ID 774-127930-0014-48411_sp1.1_rvb from training. Duration: 0.95 2023-04-03 02:44:11,986 WARNING [train.py:1073] (1/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1_rvb from training. Duration: 0.9409375 2023-04-03 02:44:16,567 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 2023-04-03 02:44:17,729 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 2023-04-03 02:44:19,215 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=161937.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 02:44:27,634 INFO [train.py:903] (1/4) Epoch 24, batch 4900, loss[loss=0.213, simple_loss=0.2971, pruned_loss=0.06444, over 18901.00 frames. ], tot_loss[loss=0.2038, simple_loss=0.2842, pruned_loss=0.06173, over 3816879.02 frames. ], batch size: 74, lr: 3.41e-03, grad_scale: 8.0 2023-04-03 02:44:27,676 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1_rvb from training. Duration: 25.3818125 2023-04-03 02:44:33,921 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=161949.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 02:44:34,693 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.672e+02 5.435e+02 6.496e+02 8.047e+02 2.666e+03, threshold=1.299e+03, percent-clipped=6.0 2023-04-03 02:44:47,021 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 2023-04-03 02:44:53,304 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=161964.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 02:44:54,311 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.2164, 3.4965, 3.7389, 3.7530, 2.0839, 3.5021, 3.1817, 3.5248], device='cuda:1'), covar=tensor([0.1550, 0.2936, 0.0700, 0.0758, 0.4740, 0.1403, 0.0639, 0.1062], device='cuda:1'), in_proj_covar=tensor([0.0794, 0.0759, 0.0966, 0.0848, 0.0847, 0.0731, 0.0578, 0.0894], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-03 02:45:05,672 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=161974.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 02:45:23,802 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=161989.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 02:45:28,147 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.7910, 1.5326, 1.4774, 1.7990, 1.4811, 1.5434, 1.4514, 1.6792], device='cuda:1'), covar=tensor([0.1075, 0.1324, 0.1466, 0.0951, 0.1245, 0.0596, 0.1455, 0.0779], device='cuda:1'), in_proj_covar=tensor([0.0276, 0.0357, 0.0311, 0.0255, 0.0305, 0.0253, 0.0314, 0.0259], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 02:45:28,862 INFO [train.py:903] (1/4) Epoch 24, batch 4950, loss[loss=0.2255, simple_loss=0.3096, pruned_loss=0.07065, over 18010.00 frames. ], tot_loss[loss=0.2034, simple_loss=0.2838, pruned_loss=0.06152, over 3812811.27 frames. ], batch size: 83, lr: 3.41e-03, grad_scale: 8.0 2023-04-03 02:45:47,818 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 2023-04-03 02:46:13,013 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 2023-04-03 02:46:33,474 INFO [train.py:903] (1/4) Epoch 24, batch 5000, loss[loss=0.2371, simple_loss=0.3146, pruned_loss=0.07976, over 19666.00 frames. ], tot_loss[loss=0.2033, simple_loss=0.2837, pruned_loss=0.06148, over 3814233.26 frames. ], batch size: 55, lr: 3.41e-03, grad_scale: 8.0 2023-04-03 02:46:41,312 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.890e+02 4.589e+02 5.769e+02 7.322e+02 1.477e+03, threshold=1.154e+03, percent-clipped=3.0 2023-04-03 02:46:44,820 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_rvb from training. Duration: 27.14 2023-04-03 02:46:56,726 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 2023-04-03 02:47:35,230 INFO [train.py:903] (1/4) Epoch 24, batch 5050, loss[loss=0.1991, simple_loss=0.2843, pruned_loss=0.05696, over 19847.00 frames. ], tot_loss[loss=0.2035, simple_loss=0.2839, pruned_loss=0.06151, over 3818497.87 frames. ], batch size: 52, lr: 3.41e-03, grad_scale: 8.0 2023-04-03 02:47:36,675 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=162095.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 02:47:40,202 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.4919, 1.4334, 1.6942, 1.5039, 3.1003, 1.0217, 2.4226, 3.4815], device='cuda:1'), covar=tensor([0.0504, 0.2639, 0.2539, 0.1735, 0.0613, 0.2434, 0.1138, 0.0239], device='cuda:1'), in_proj_covar=tensor([0.0414, 0.0368, 0.0388, 0.0350, 0.0372, 0.0352, 0.0386, 0.0406], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 02:48:14,994 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9_rvb from training. Duration: 26.62775 2023-04-03 02:48:25,521 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=162134.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 02:48:37,740 INFO [train.py:903] (1/4) Epoch 24, batch 5100, loss[loss=0.1945, simple_loss=0.2694, pruned_loss=0.05983, over 19755.00 frames. ], tot_loss[loss=0.2025, simple_loss=0.283, pruned_loss=0.06102, over 3818409.29 frames. ], batch size: 45, lr: 3.40e-03, grad_scale: 8.0 2023-04-03 02:48:44,638 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.860e+02 4.540e+02 5.777e+02 7.463e+02 1.637e+03, threshold=1.155e+03, percent-clipped=6.0 2023-04-03 02:48:51,578 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 2023-04-03 02:48:55,185 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_rvb from training. Duration: 25.035 2023-04-03 02:48:55,619 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=162159.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 02:48:58,702 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_rvb from training. Duration: 27.92 2023-04-03 02:49:18,961 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=162177.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 02:49:39,609 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=162193.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 02:49:40,374 INFO [train.py:903] (1/4) Epoch 24, batch 5150, loss[loss=0.1984, simple_loss=0.285, pruned_loss=0.05596, over 19301.00 frames. ], tot_loss[loss=0.2025, simple_loss=0.2832, pruned_loss=0.06096, over 3813238.19 frames. ], batch size: 66, lr: 3.40e-03, grad_scale: 8.0 2023-04-03 02:49:57,201 WARNING [train.py:1073] (1/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 2023-04-03 02:50:11,328 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=162218.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 02:50:31,577 WARNING [train.py:1073] (1/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1_rvb from training. Duration: 0.9681875 2023-04-03 02:50:43,109 INFO [train.py:903] (1/4) Epoch 24, batch 5200, loss[loss=0.2137, simple_loss=0.295, pruned_loss=0.0662, over 18646.00 frames. ], tot_loss[loss=0.2037, simple_loss=0.2845, pruned_loss=0.06141, over 3805688.50 frames. ], batch size: 74, lr: 3.40e-03, grad_scale: 8.0 2023-04-03 02:50:50,150 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=162249.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 02:50:51,040 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.192e+02 4.774e+02 5.644e+02 7.681e+02 1.514e+03, threshold=1.129e+03, percent-clipped=4.0 2023-04-03 02:50:59,732 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 2023-04-03 02:51:44,107 WARNING [train.py:1073] (1/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 2023-04-03 02:51:44,406 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=162292.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 02:51:46,277 INFO [train.py:903] (1/4) Epoch 24, batch 5250, loss[loss=0.224, simple_loss=0.3057, pruned_loss=0.07113, over 19341.00 frames. ], tot_loss[loss=0.2058, simple_loss=0.2862, pruned_loss=0.06267, over 3784119.35 frames. ], batch size: 66, lr: 3.40e-03, grad_scale: 8.0 2023-04-03 02:52:14,404 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.8797, 1.3970, 1.5814, 1.4984, 3.4222, 1.1668, 2.4025, 3.9184], device='cuda:1'), covar=tensor([0.0455, 0.2692, 0.2867, 0.1889, 0.0692, 0.2507, 0.1399, 0.0206], device='cuda:1'), in_proj_covar=tensor([0.0414, 0.0368, 0.0389, 0.0350, 0.0373, 0.0353, 0.0386, 0.0406], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 02:52:50,073 INFO [train.py:903] (1/4) Epoch 24, batch 5300, loss[loss=0.2205, simple_loss=0.3019, pruned_loss=0.06955, over 17413.00 frames. ], tot_loss[loss=0.2049, simple_loss=0.2854, pruned_loss=0.06219, over 3787356.44 frames. ], batch size: 101, lr: 3.40e-03, grad_scale: 8.0 2023-04-03 02:52:57,101 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.926e+02 4.862e+02 5.825e+02 7.901e+02 2.284e+03, threshold=1.165e+03, percent-clipped=8.0 2023-04-03 02:53:08,449 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 2023-04-03 02:53:33,674 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6420, 1.2153, 1.2827, 1.5078, 1.0790, 1.4071, 1.2272, 1.4904], device='cuda:1'), covar=tensor([0.1137, 0.1231, 0.1523, 0.1024, 0.1298, 0.0627, 0.1536, 0.0807], device='cuda:1'), in_proj_covar=tensor([0.0278, 0.0360, 0.0314, 0.0257, 0.0306, 0.0255, 0.0316, 0.0261], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 02:53:43,024 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=4.07 vs. limit=5.0 2023-04-03 02:53:51,568 INFO [train.py:903] (1/4) Epoch 24, batch 5350, loss[loss=0.2123, simple_loss=0.2915, pruned_loss=0.06658, over 19626.00 frames. ], tot_loss[loss=0.2052, simple_loss=0.2857, pruned_loss=0.06238, over 3792173.29 frames. ], batch size: 57, lr: 3.40e-03, grad_scale: 8.0 2023-04-03 02:54:28,285 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9_rvb from training. Duration: 26.438875 2023-04-03 02:54:47,990 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=162439.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 02:54:54,712 INFO [train.py:903] (1/4) Epoch 24, batch 5400, loss[loss=0.1842, simple_loss=0.2583, pruned_loss=0.05507, over 19748.00 frames. ], tot_loss[loss=0.2054, simple_loss=0.2858, pruned_loss=0.06249, over 3796775.56 frames. ], batch size: 46, lr: 3.40e-03, grad_scale: 8.0 2023-04-03 02:55:00,083 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 2023-04-03 02:55:02,631 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.752e+02 4.703e+02 6.237e+02 7.666e+02 1.372e+03, threshold=1.247e+03, percent-clipped=3.0 2023-04-03 02:55:56,871 INFO [train.py:903] (1/4) Epoch 24, batch 5450, loss[loss=0.2059, simple_loss=0.2921, pruned_loss=0.05986, over 19656.00 frames. ], tot_loss[loss=0.2062, simple_loss=0.2867, pruned_loss=0.06284, over 3796603.55 frames. ], batch size: 58, lr: 3.40e-03, grad_scale: 8.0 2023-04-03 02:56:37,127 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.2734, 1.9378, 2.0794, 2.9623, 1.9864, 2.3408, 2.4954, 2.0976], device='cuda:1'), covar=tensor([0.0759, 0.0908, 0.0939, 0.0751, 0.0908, 0.0774, 0.0882, 0.0679], device='cuda:1'), in_proj_covar=tensor([0.0213, 0.0223, 0.0227, 0.0240, 0.0226, 0.0213, 0.0189, 0.0205], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-04-03 02:56:38,283 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.7710, 1.7344, 1.5695, 1.3722, 1.3532, 1.3654, 0.2462, 0.6521], device='cuda:1'), covar=tensor([0.0598, 0.0581, 0.0414, 0.0649, 0.1145, 0.0736, 0.1243, 0.1054], device='cuda:1'), in_proj_covar=tensor([0.0359, 0.0359, 0.0362, 0.0386, 0.0465, 0.0394, 0.0340, 0.0348], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 02:56:40,433 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=162529.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 02:57:00,250 INFO [train.py:903] (1/4) Epoch 24, batch 5500, loss[loss=0.2303, simple_loss=0.3025, pruned_loss=0.07903, over 18785.00 frames. ], tot_loss[loss=0.2057, simple_loss=0.2863, pruned_loss=0.06253, over 3793091.83 frames. ], batch size: 74, lr: 3.40e-03, grad_scale: 8.0 2023-04-03 02:57:00,687 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.4312, 2.4919, 2.6687, 3.1600, 2.6309, 3.0357, 2.7012, 2.5365], device='cuda:1'), covar=tensor([0.3701, 0.3316, 0.1649, 0.2135, 0.3605, 0.1862, 0.3845, 0.2722], device='cuda:1'), in_proj_covar=tensor([0.0916, 0.0989, 0.0729, 0.0940, 0.0895, 0.0830, 0.0850, 0.0793], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-03 02:57:05,452 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=162548.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 02:57:07,335 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.996e+02 5.077e+02 6.464e+02 7.861e+02 1.317e+03, threshold=1.293e+03, percent-clipped=1.0 2023-04-03 02:57:12,152 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=162554.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 02:57:25,028 WARNING [train.py:1073] (1/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9_rvb from training. Duration: 27.47775 2023-04-03 02:57:36,167 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=162573.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 02:57:49,665 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9532, 1.8745, 1.8232, 1.6119, 1.4603, 1.5854, 0.4639, 0.9417], device='cuda:1'), covar=tensor([0.0656, 0.0641, 0.0427, 0.0725, 0.1219, 0.0860, 0.1329, 0.1127], device='cuda:1'), in_proj_covar=tensor([0.0361, 0.0360, 0.0364, 0.0388, 0.0467, 0.0396, 0.0341, 0.0349], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 02:58:01,101 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=162593.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 02:58:01,999 INFO [train.py:903] (1/4) Epoch 24, batch 5550, loss[loss=0.195, simple_loss=0.273, pruned_loss=0.05848, over 19414.00 frames. ], tot_loss[loss=0.2058, simple_loss=0.2862, pruned_loss=0.06272, over 3769140.87 frames. ], batch size: 48, lr: 3.40e-03, grad_scale: 8.0 2023-04-03 02:58:08,816 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 2023-04-03 02:58:18,362 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.4146, 1.2816, 1.4407, 1.3740, 2.9808, 1.0678, 2.2484, 3.4319], device='cuda:1'), covar=tensor([0.0561, 0.2934, 0.3041, 0.1992, 0.0772, 0.2625, 0.1427, 0.0267], device='cuda:1'), in_proj_covar=tensor([0.0414, 0.0368, 0.0389, 0.0349, 0.0372, 0.0352, 0.0386, 0.0405], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 02:58:50,763 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.3057, 3.6652, 2.2890, 2.1699, 3.3659, 1.9903, 1.7478, 2.5902], device='cuda:1'), covar=tensor([0.1250, 0.0483, 0.1001, 0.0892, 0.0444, 0.1186, 0.0944, 0.0574], device='cuda:1'), in_proj_covar=tensor([0.0303, 0.0316, 0.0340, 0.0268, 0.0248, 0.0342, 0.0292, 0.0274], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 02:58:59,599 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 2023-04-03 02:59:02,952 INFO [train.py:903] (1/4) Epoch 24, batch 5600, loss[loss=0.2376, simple_loss=0.3106, pruned_loss=0.0823, over 18782.00 frames. ], tot_loss[loss=0.2054, simple_loss=0.2859, pruned_loss=0.06245, over 3792935.23 frames. ], batch size: 74, lr: 3.40e-03, grad_scale: 16.0 2023-04-03 02:59:12,271 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.260e+02 5.034e+02 5.949e+02 8.933e+02 2.100e+03, threshold=1.190e+03, percent-clipped=10.0 2023-04-03 02:59:19,259 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0 2023-04-03 03:00:07,504 INFO [train.py:903] (1/4) Epoch 24, batch 5650, loss[loss=0.2072, simple_loss=0.2918, pruned_loss=0.06123, over 19699.00 frames. ], tot_loss[loss=0.2049, simple_loss=0.2854, pruned_loss=0.0622, over 3804418.69 frames. ], batch size: 59, lr: 3.40e-03, grad_scale: 8.0 2023-04-03 03:00:24,969 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=162708.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 03:00:55,023 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9_rvb from training. Duration: 28.0944375 2023-04-03 03:01:09,550 INFO [train.py:903] (1/4) Epoch 24, batch 5700, loss[loss=0.2079, simple_loss=0.2917, pruned_loss=0.06203, over 19768.00 frames. ], tot_loss[loss=0.2056, simple_loss=0.2862, pruned_loss=0.06247, over 3812148.02 frames. ], batch size: 56, lr: 3.40e-03, grad_scale: 8.0 2023-04-03 03:01:17,480 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.412e+02 4.704e+02 5.740e+02 7.100e+02 1.656e+03, threshold=1.148e+03, percent-clipped=4.0 2023-04-03 03:01:23,595 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2452, 1.1813, 1.1810, 1.3462, 1.0462, 1.3418, 1.3186, 1.2435], device='cuda:1'), covar=tensor([0.0907, 0.1010, 0.1093, 0.0672, 0.0874, 0.0846, 0.0874, 0.0788], device='cuda:1'), in_proj_covar=tensor([0.0212, 0.0223, 0.0226, 0.0239, 0.0225, 0.0213, 0.0189, 0.0203], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-04-03 03:01:56,784 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.1735, 2.8693, 2.3156, 2.2253, 1.9268, 2.4744, 0.9242, 2.1441], device='cuda:1'), covar=tensor([0.0628, 0.0588, 0.0691, 0.1172, 0.1306, 0.1128, 0.1538, 0.1037], device='cuda:1'), in_proj_covar=tensor([0.0357, 0.0357, 0.0360, 0.0385, 0.0464, 0.0393, 0.0338, 0.0345], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 03:02:02,759 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.28 vs. limit=5.0 2023-04-03 03:02:11,498 INFO [train.py:903] (1/4) Epoch 24, batch 5750, loss[loss=0.2088, simple_loss=0.2961, pruned_loss=0.0607, over 19374.00 frames. ], tot_loss[loss=0.2055, simple_loss=0.2863, pruned_loss=0.06235, over 3820801.71 frames. ], batch size: 70, lr: 3.40e-03, grad_scale: 8.0 2023-04-03 03:02:13,881 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9_rvb from training. Duration: 33.038875 2023-04-03 03:02:22,207 WARNING [train.py:1073] (1/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 2023-04-03 03:02:28,829 WARNING [train.py:1073] (1/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 2023-04-03 03:02:31,419 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=162810.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 03:03:01,710 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.72 vs. limit=2.0 2023-04-03 03:03:03,489 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=162835.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 03:03:13,925 INFO [train.py:903] (1/4) Epoch 24, batch 5800, loss[loss=0.2355, simple_loss=0.3159, pruned_loss=0.07754, over 19654.00 frames. ], tot_loss[loss=0.2039, simple_loss=0.2849, pruned_loss=0.06142, over 3826337.82 frames. ], batch size: 60, lr: 3.40e-03, grad_scale: 8.0 2023-04-03 03:03:15,414 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=162845.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 03:03:22,913 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.123e+02 4.857e+02 6.549e+02 8.249e+02 1.553e+03, threshold=1.310e+03, percent-clipped=3.0 2023-04-03 03:03:51,947 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=162873.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 03:04:16,919 INFO [train.py:903] (1/4) Epoch 24, batch 5850, loss[loss=0.245, simple_loss=0.3203, pruned_loss=0.08486, over 19700.00 frames. ], tot_loss[loss=0.2041, simple_loss=0.285, pruned_loss=0.0616, over 3825114.87 frames. ], batch size: 59, lr: 3.40e-03, grad_scale: 8.0 2023-04-03 03:05:20,445 INFO [train.py:903] (1/4) Epoch 24, batch 5900, loss[loss=0.1872, simple_loss=0.2725, pruned_loss=0.0509, over 19402.00 frames. ], tot_loss[loss=0.2051, simple_loss=0.2859, pruned_loss=0.06217, over 3819122.96 frames. ], batch size: 66, lr: 3.40e-03, grad_scale: 8.0 2023-04-03 03:05:26,356 WARNING [train.py:1073] (1/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 2023-04-03 03:05:28,676 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.343e+02 4.550e+02 5.410e+02 6.827e+02 1.573e+03, threshold=1.082e+03, percent-clipped=1.0 2023-04-03 03:05:38,168 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9849, 1.9357, 1.8305, 1.6622, 1.5928, 1.5964, 0.5049, 0.9038], device='cuda:1'), covar=tensor([0.0703, 0.0710, 0.0454, 0.0693, 0.1241, 0.0879, 0.1321, 0.1145], device='cuda:1'), in_proj_covar=tensor([0.0360, 0.0359, 0.0362, 0.0387, 0.0465, 0.0394, 0.0341, 0.0347], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 03:05:43,775 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=162964.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 03:05:44,428 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9_rvb from training. Duration: 27.511125 2023-04-03 03:06:15,201 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=162988.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 03:06:16,452 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=162989.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 03:06:21,816 INFO [train.py:903] (1/4) Epoch 24, batch 5950, loss[loss=0.1918, simple_loss=0.277, pruned_loss=0.05327, over 19594.00 frames. ], tot_loss[loss=0.2055, simple_loss=0.286, pruned_loss=0.06252, over 3825538.18 frames. ], batch size: 52, lr: 3.40e-03, grad_scale: 8.0 2023-04-03 03:07:22,796 INFO [train.py:903] (1/4) Epoch 24, batch 6000, loss[loss=0.2076, simple_loss=0.2806, pruned_loss=0.06731, over 19763.00 frames. ], tot_loss[loss=0.2054, simple_loss=0.2862, pruned_loss=0.06232, over 3828468.09 frames. ], batch size: 45, lr: 3.40e-03, grad_scale: 8.0 2023-04-03 03:07:22,797 INFO [train.py:928] (1/4) Computing validation loss 2023-04-03 03:07:35,178 INFO [train.py:937] (1/4) Epoch 24, validation: loss=0.1683, simple_loss=0.2679, pruned_loss=0.03436, over 944034.00 frames. 2023-04-03 03:07:35,180 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18821MB 2023-04-03 03:07:43,473 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.958e+02 5.276e+02 6.488e+02 8.005e+02 1.643e+03, threshold=1.298e+03, percent-clipped=7.0 2023-04-03 03:07:59,723 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9750, 1.9235, 1.7860, 1.5966, 1.4657, 1.5609, 0.4205, 0.8423], device='cuda:1'), covar=tensor([0.0639, 0.0618, 0.0440, 0.0741, 0.1268, 0.0899, 0.1342, 0.1155], device='cuda:1'), in_proj_covar=tensor([0.0361, 0.0361, 0.0364, 0.0388, 0.0467, 0.0395, 0.0342, 0.0349], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 03:08:18,813 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.91 vs. limit=2.0 2023-04-03 03:08:20,490 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5399, 1.5281, 1.8452, 1.7580, 2.5280, 2.3341, 2.7165, 1.0846], device='cuda:1'), covar=tensor([0.2451, 0.4275, 0.2703, 0.1933, 0.1603, 0.2153, 0.1459, 0.4648], device='cuda:1'), in_proj_covar=tensor([0.0542, 0.0653, 0.0730, 0.0494, 0.0624, 0.0540, 0.0664, 0.0559], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 03:08:29,503 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.55 vs. limit=2.0 2023-04-03 03:08:35,923 INFO [train.py:903] (1/4) Epoch 24, batch 6050, loss[loss=0.2206, simple_loss=0.2969, pruned_loss=0.07213, over 19667.00 frames. ], tot_loss[loss=0.2059, simple_loss=0.2866, pruned_loss=0.06264, over 3835875.86 frames. ], batch size: 55, lr: 3.39e-03, grad_scale: 8.0 2023-04-03 03:08:53,234 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=163108.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 03:09:37,884 INFO [train.py:903] (1/4) Epoch 24, batch 6100, loss[loss=0.1818, simple_loss=0.2579, pruned_loss=0.0529, over 19372.00 frames. ], tot_loss[loss=0.2049, simple_loss=0.2855, pruned_loss=0.06218, over 3840563.83 frames. ], batch size: 47, lr: 3.39e-03, grad_scale: 8.0 2023-04-03 03:09:45,776 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.855e+02 5.048e+02 5.892e+02 7.607e+02 1.565e+03, threshold=1.178e+03, percent-clipped=5.0 2023-04-03 03:09:56,032 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([5.1012, 5.1015, 5.8813, 5.9137, 2.1954, 5.5681, 4.7296, 5.5192], device='cuda:1'), covar=tensor([0.1609, 0.0853, 0.0555, 0.0561, 0.5789, 0.0797, 0.0620, 0.1129], device='cuda:1'), in_proj_covar=tensor([0.0797, 0.0755, 0.0965, 0.0844, 0.0844, 0.0731, 0.0577, 0.0896], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-03 03:10:33,906 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=163189.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 03:10:40,167 INFO [train.py:903] (1/4) Epoch 24, batch 6150, loss[loss=0.2341, simple_loss=0.3075, pruned_loss=0.08035, over 18720.00 frames. ], tot_loss[loss=0.2046, simple_loss=0.2852, pruned_loss=0.06194, over 3835559.96 frames. ], batch size: 74, lr: 3.39e-03, grad_scale: 8.0 2023-04-03 03:11:10,602 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9_rvb from training. Duration: 31.02225 2023-04-03 03:11:43,668 INFO [train.py:903] (1/4) Epoch 24, batch 6200, loss[loss=0.1632, simple_loss=0.243, pruned_loss=0.04165, over 16525.00 frames. ], tot_loss[loss=0.204, simple_loss=0.2849, pruned_loss=0.06154, over 3819367.89 frames. ], batch size: 36, lr: 3.39e-03, grad_scale: 8.0 2023-04-03 03:11:44,112 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=163244.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 03:11:51,414 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.768e+02 4.538e+02 5.825e+02 7.342e+02 1.276e+03, threshold=1.165e+03, percent-clipped=3.0 2023-04-03 03:12:00,554 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=163258.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 03:12:14,129 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=163269.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 03:12:43,994 INFO [train.py:903] (1/4) Epoch 24, batch 6250, loss[loss=0.2246, simple_loss=0.3087, pruned_loss=0.07031, over 19339.00 frames. ], tot_loss[loss=0.2051, simple_loss=0.2857, pruned_loss=0.06228, over 3812207.68 frames. ], batch size: 66, lr: 3.39e-03, grad_scale: 8.0 2023-04-03 03:12:49,742 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.3551, 3.0458, 2.1606, 2.7615, 0.7978, 3.0744, 2.9603, 3.0137], device='cuda:1'), covar=tensor([0.1102, 0.1331, 0.2228, 0.1108, 0.3846, 0.0983, 0.1119, 0.1400], device='cuda:1'), in_proj_covar=tensor([0.0511, 0.0418, 0.0502, 0.0351, 0.0401, 0.0442, 0.0436, 0.0466], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 03:12:56,470 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=163304.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 03:13:12,646 WARNING [train.py:1073] (1/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9_rvb from training. Duration: 25.988875 2023-04-03 03:13:45,519 INFO [train.py:903] (1/4) Epoch 24, batch 6300, loss[loss=0.205, simple_loss=0.2897, pruned_loss=0.0601, over 19428.00 frames. ], tot_loss[loss=0.2053, simple_loss=0.2858, pruned_loss=0.0624, over 3826000.36 frames. ], batch size: 70, lr: 3.39e-03, grad_scale: 8.0 2023-04-03 03:13:54,810 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.313e+02 5.008e+02 6.887e+02 8.761e+02 2.377e+03, threshold=1.377e+03, percent-clipped=3.0 2023-04-03 03:14:48,468 INFO [train.py:903] (1/4) Epoch 24, batch 6350, loss[loss=0.2387, simple_loss=0.3262, pruned_loss=0.07565, over 19655.00 frames. ], tot_loss[loss=0.2044, simple_loss=0.2851, pruned_loss=0.06179, over 3830554.75 frames. ], batch size: 58, lr: 3.39e-03, grad_scale: 8.0 2023-04-03 03:15:50,632 INFO [train.py:903] (1/4) Epoch 24, batch 6400, loss[loss=0.1972, simple_loss=0.2823, pruned_loss=0.05606, over 19406.00 frames. ], tot_loss[loss=0.205, simple_loss=0.2859, pruned_loss=0.06208, over 3832409.11 frames. ], batch size: 70, lr: 3.39e-03, grad_scale: 8.0 2023-04-03 03:15:59,002 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.312e+02 4.869e+02 5.854e+02 7.378e+02 1.563e+03, threshold=1.171e+03, percent-clipped=2.0 2023-04-03 03:16:00,333 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=163452.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 03:16:52,294 INFO [train.py:903] (1/4) Epoch 24, batch 6450, loss[loss=0.179, simple_loss=0.2579, pruned_loss=0.05011, over 19743.00 frames. ], tot_loss[loss=0.2036, simple_loss=0.2847, pruned_loss=0.06122, over 3838520.28 frames. ], batch size: 51, lr: 3.39e-03, grad_scale: 8.0 2023-04-03 03:17:29,958 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.9382, 3.6082, 2.6412, 3.2400, 0.9165, 3.5448, 3.4428, 3.5208], device='cuda:1'), covar=tensor([0.0927, 0.1184, 0.1844, 0.0926, 0.3869, 0.0891, 0.1025, 0.1317], device='cuda:1'), in_proj_covar=tensor([0.0512, 0.0418, 0.0502, 0.0352, 0.0402, 0.0443, 0.0436, 0.0467], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 03:17:34,401 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9_rvb from training. Duration: 28.638875 2023-04-03 03:17:53,149 INFO [train.py:903] (1/4) Epoch 24, batch 6500, loss[loss=0.2144, simple_loss=0.2973, pruned_loss=0.06577, over 18076.00 frames. ], tot_loss[loss=0.2047, simple_loss=0.2855, pruned_loss=0.062, over 3819858.52 frames. ], batch size: 84, lr: 3.39e-03, grad_scale: 8.0 2023-04-03 03:17:56,719 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1_rvb from training. Duration: 0.836375 2023-04-03 03:18:01,366 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.063e+02 4.704e+02 6.024e+02 8.376e+02 1.457e+03, threshold=1.205e+03, percent-clipped=5.0 2023-04-03 03:18:11,247 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=4.06 vs. limit=5.0 2023-04-03 03:18:14,474 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=163560.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 03:18:22,440 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=163567.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 03:18:44,305 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=163585.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 03:18:55,199 INFO [train.py:903] (1/4) Epoch 24, batch 6550, loss[loss=0.1588, simple_loss=0.2396, pruned_loss=0.03899, over 19754.00 frames. ], tot_loss[loss=0.2035, simple_loss=0.2844, pruned_loss=0.06126, over 3817056.72 frames. ], batch size: 47, lr: 3.39e-03, grad_scale: 8.0 2023-04-03 03:19:05,569 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=163602.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 03:19:54,268 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3840, 2.0448, 1.6310, 1.4025, 1.8353, 1.3379, 1.3675, 1.7562], device='cuda:1'), covar=tensor([0.0910, 0.0841, 0.1100, 0.0905, 0.0617, 0.1304, 0.0669, 0.0497], device='cuda:1'), in_proj_covar=tensor([0.0302, 0.0315, 0.0337, 0.0267, 0.0248, 0.0341, 0.0290, 0.0273], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 03:19:57,508 INFO [train.py:903] (1/4) Epoch 24, batch 6600, loss[loss=0.2178, simple_loss=0.2954, pruned_loss=0.07013, over 18271.00 frames. ], tot_loss[loss=0.2048, simple_loss=0.2853, pruned_loss=0.06216, over 3822309.16 frames. ], batch size: 84, lr: 3.39e-03, grad_scale: 8.0 2023-04-03 03:20:05,518 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.150e+02 5.011e+02 6.018e+02 7.633e+02 1.393e+03, threshold=1.204e+03, percent-clipped=8.0 2023-04-03 03:20:57,661 INFO [train.py:903] (1/4) Epoch 24, batch 6650, loss[loss=0.1866, simple_loss=0.2644, pruned_loss=0.05437, over 19849.00 frames. ], tot_loss[loss=0.2054, simple_loss=0.2853, pruned_loss=0.06277, over 3826561.82 frames. ], batch size: 52, lr: 3.39e-03, grad_scale: 8.0 2023-04-03 03:20:59,592 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.42 vs. limit=2.0 2023-04-03 03:21:25,820 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=163717.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 03:21:33,086 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.6999, 4.0629, 4.3439, 4.3543, 1.9146, 4.0953, 3.6033, 4.0742], device='cuda:1'), covar=tensor([0.1610, 0.1522, 0.0632, 0.0724, 0.5967, 0.1087, 0.0668, 0.1181], device='cuda:1'), in_proj_covar=tensor([0.0794, 0.0754, 0.0962, 0.0840, 0.0840, 0.0730, 0.0574, 0.0890], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-03 03:21:58,181 INFO [train.py:903] (1/4) Epoch 24, batch 6700, loss[loss=0.168, simple_loss=0.2497, pruned_loss=0.04317, over 19383.00 frames. ], tot_loss[loss=0.2058, simple_loss=0.2856, pruned_loss=0.06295, over 3822244.47 frames. ], batch size: 47, lr: 3.39e-03, grad_scale: 4.0 2023-04-03 03:22:08,773 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.807e+02 4.744e+02 6.010e+02 8.153e+02 1.593e+03, threshold=1.202e+03, percent-clipped=6.0 2023-04-03 03:22:55,950 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.62 vs. limit=5.0 2023-04-03 03:22:57,590 INFO [train.py:903] (1/4) Epoch 24, batch 6750, loss[loss=0.2322, simple_loss=0.307, pruned_loss=0.07869, over 19593.00 frames. ], tot_loss[loss=0.206, simple_loss=0.2863, pruned_loss=0.06291, over 3839167.67 frames. ], batch size: 61, lr: 3.39e-03, grad_scale: 4.0 2023-04-03 03:23:30,592 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=163823.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 03:23:53,915 INFO [train.py:903] (1/4) Epoch 24, batch 6800, loss[loss=0.2039, simple_loss=0.2907, pruned_loss=0.05852, over 19790.00 frames. ], tot_loss[loss=0.2049, simple_loss=0.285, pruned_loss=0.0624, over 3832029.16 frames. ], batch size: 56, lr: 3.39e-03, grad_scale: 8.0 2023-04-03 03:23:58,808 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=163848.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 03:24:03,002 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.305e+02 4.890e+02 5.869e+02 7.347e+02 2.478e+03, threshold=1.174e+03, percent-clipped=2.0 2023-04-03 03:24:39,471 WARNING [train.py:1073] (1/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 2023-04-03 03:24:40,531 WARNING [train.py:1073] (1/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 2023-04-03 03:24:42,791 INFO [train.py:903] (1/4) Epoch 25, batch 0, loss[loss=0.2108, simple_loss=0.2945, pruned_loss=0.06349, over 19600.00 frames. ], tot_loss[loss=0.2108, simple_loss=0.2945, pruned_loss=0.06349, over 19600.00 frames. ], batch size: 57, lr: 3.32e-03, grad_scale: 8.0 2023-04-03 03:24:42,792 INFO [train.py:928] (1/4) Computing validation loss 2023-04-03 03:24:54,385 INFO [train.py:937] (1/4) Epoch 25, validation: loss=0.1672, simple_loss=0.2675, pruned_loss=0.03346, over 944034.00 frames. 2023-04-03 03:24:54,386 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18821MB 2023-04-03 03:25:06,933 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 2023-04-03 03:25:40,923 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.1398, 1.1851, 1.4606, 1.5090, 2.7271, 1.0666, 2.1959, 3.0766], device='cuda:1'), covar=tensor([0.0587, 0.2971, 0.2894, 0.1729, 0.0747, 0.2382, 0.1265, 0.0351], device='cuda:1'), in_proj_covar=tensor([0.0413, 0.0368, 0.0391, 0.0347, 0.0374, 0.0352, 0.0386, 0.0405], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 03:25:57,056 INFO [train.py:903] (1/4) Epoch 25, batch 50, loss[loss=0.1916, simple_loss=0.2821, pruned_loss=0.0506, over 18044.00 frames. ], tot_loss[loss=0.2055, simple_loss=0.2862, pruned_loss=0.06238, over 856953.83 frames. ], batch size: 83, lr: 3.32e-03, grad_scale: 8.0 2023-04-03 03:26:35,504 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.099e+02 4.575e+02 5.589e+02 7.102e+02 2.434e+03, threshold=1.118e+03, percent-clipped=5.0 2023-04-03 03:26:36,713 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1_rvb from training. Duration: 27.0318125 2023-04-03 03:27:00,584 INFO [train.py:903] (1/4) Epoch 25, batch 100, loss[loss=0.1667, simple_loss=0.2491, pruned_loss=0.04218, over 19496.00 frames. ], tot_loss[loss=0.2045, simple_loss=0.2856, pruned_loss=0.06173, over 1514314.29 frames. ], batch size: 49, lr: 3.32e-03, grad_scale: 8.0 2023-04-03 03:27:03,107 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=163973.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 03:27:15,429 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 2023-04-03 03:27:34,449 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=163998.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 03:28:05,191 INFO [train.py:903] (1/4) Epoch 25, batch 150, loss[loss=0.2025, simple_loss=0.288, pruned_loss=0.05844, over 18415.00 frames. ], tot_loss[loss=0.2037, simple_loss=0.2846, pruned_loss=0.06145, over 2037063.57 frames. ], batch size: 83, lr: 3.32e-03, grad_scale: 8.0 2023-04-03 03:28:42,973 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.518e+02 5.520e+02 6.265e+02 7.455e+02 1.438e+03, threshold=1.253e+03, percent-clipped=2.0 2023-04-03 03:29:06,831 INFO [train.py:903] (1/4) Epoch 25, batch 200, loss[loss=0.1963, simple_loss=0.273, pruned_loss=0.05984, over 19774.00 frames. ], tot_loss[loss=0.2034, simple_loss=0.2844, pruned_loss=0.0612, over 2450319.93 frames. ], batch size: 47, lr: 3.31e-03, grad_scale: 8.0 2023-04-03 03:29:09,371 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 2023-04-03 03:29:53,464 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=164108.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 03:30:10,582 INFO [train.py:903] (1/4) Epoch 25, batch 250, loss[loss=0.2005, simple_loss=0.2811, pruned_loss=0.05995, over 19764.00 frames. ], tot_loss[loss=0.2042, simple_loss=0.2844, pruned_loss=0.06195, over 2760793.04 frames. ], batch size: 54, lr: 3.31e-03, grad_scale: 8.0 2023-04-03 03:30:41,653 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 2023-04-03 03:30:48,735 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.799e+02 4.875e+02 6.095e+02 7.386e+02 2.001e+03, threshold=1.219e+03, percent-clipped=3.0 2023-04-03 03:31:07,753 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 2023-04-03 03:31:13,863 INFO [train.py:903] (1/4) Epoch 25, batch 300, loss[loss=0.2212, simple_loss=0.3006, pruned_loss=0.07095, over 19530.00 frames. ], tot_loss[loss=0.2031, simple_loss=0.2833, pruned_loss=0.06143, over 3008289.12 frames. ], batch size: 56, lr: 3.31e-03, grad_scale: 8.0 2023-04-03 03:31:16,571 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=164173.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 03:32:18,086 INFO [train.py:903] (1/4) Epoch 25, batch 350, loss[loss=0.2151, simple_loss=0.2942, pruned_loss=0.068, over 18153.00 frames. ], tot_loss[loss=0.2031, simple_loss=0.2835, pruned_loss=0.06131, over 3190888.59 frames. ], batch size: 83, lr: 3.31e-03, grad_scale: 8.0 2023-04-03 03:32:25,194 WARNING [train.py:1073] (1/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1_rvb from training. Duration: 0.9681875 2023-04-03 03:32:55,003 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.912e+02 5.440e+02 6.552e+02 9.332e+02 1.789e+03, threshold=1.310e+03, percent-clipped=12.0 2023-04-03 03:33:20,727 INFO [train.py:903] (1/4) Epoch 25, batch 400, loss[loss=0.2746, simple_loss=0.335, pruned_loss=0.1071, over 13429.00 frames. ], tot_loss[loss=0.2045, simple_loss=0.2845, pruned_loss=0.06224, over 3322735.40 frames. ], batch size: 136, lr: 3.31e-03, grad_scale: 8.0 2023-04-03 03:33:53,654 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.7524, 1.5236, 1.4490, 1.7712, 1.5338, 1.5301, 1.4955, 1.6625], device='cuda:1'), covar=tensor([0.1085, 0.1354, 0.1510, 0.1009, 0.1180, 0.0590, 0.1420, 0.0774], device='cuda:1'), in_proj_covar=tensor([0.0276, 0.0359, 0.0314, 0.0256, 0.0306, 0.0254, 0.0316, 0.0261], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 03:34:24,825 INFO [train.py:903] (1/4) Epoch 25, batch 450, loss[loss=0.2091, simple_loss=0.2933, pruned_loss=0.06245, over 18976.00 frames. ], tot_loss[loss=0.2053, simple_loss=0.2855, pruned_loss=0.06258, over 3434575.85 frames. ], batch size: 69, lr: 3.31e-03, grad_scale: 8.0 2023-04-03 03:34:48,025 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.3552, 3.8260, 3.9354, 3.9583, 1.5778, 3.7671, 3.2630, 3.6889], device='cuda:1'), covar=tensor([0.1709, 0.1096, 0.0708, 0.0851, 0.5944, 0.1129, 0.0798, 0.1245], device='cuda:1'), in_proj_covar=tensor([0.0807, 0.0767, 0.0971, 0.0852, 0.0848, 0.0736, 0.0580, 0.0903], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-03 03:34:59,265 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 2023-04-03 03:35:00,457 WARNING [train.py:1073] (1/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 2023-04-03 03:35:02,815 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.977e+02 4.778e+02 5.577e+02 7.081e+02 1.680e+03, threshold=1.115e+03, percent-clipped=3.0 2023-04-03 03:35:27,619 INFO [train.py:903] (1/4) Epoch 25, batch 500, loss[loss=0.222, simple_loss=0.3084, pruned_loss=0.06784, over 19531.00 frames. ], tot_loss[loss=0.2042, simple_loss=0.2844, pruned_loss=0.06196, over 3534727.21 frames. ], batch size: 56, lr: 3.31e-03, grad_scale: 8.0 2023-04-03 03:35:35,601 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=164378.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 03:36:31,239 INFO [train.py:903] (1/4) Epoch 25, batch 550, loss[loss=0.1814, simple_loss=0.2617, pruned_loss=0.0505, over 19366.00 frames. ], tot_loss[loss=0.2028, simple_loss=0.2833, pruned_loss=0.06117, over 3590274.96 frames. ], batch size: 47, lr: 3.31e-03, grad_scale: 8.0 2023-04-03 03:37:09,260 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.449e+02 5.163e+02 6.218e+02 7.641e+02 1.675e+03, threshold=1.244e+03, percent-clipped=5.0 2023-04-03 03:37:09,455 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=164452.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 03:37:14,119 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([5.4735, 5.0610, 3.0804, 4.3345, 1.4403, 4.9834, 4.8491, 5.0228], device='cuda:1'), covar=tensor([0.0396, 0.0767, 0.1846, 0.0781, 0.3510, 0.0517, 0.0771, 0.0970], device='cuda:1'), in_proj_covar=tensor([0.0521, 0.0424, 0.0509, 0.0356, 0.0408, 0.0447, 0.0442, 0.0473], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 03:37:34,085 INFO [train.py:903] (1/4) Epoch 25, batch 600, loss[loss=0.2571, simple_loss=0.3171, pruned_loss=0.09853, over 13042.00 frames. ], tot_loss[loss=0.2021, simple_loss=0.2827, pruned_loss=0.0607, over 3647315.28 frames. ], batch size: 136, lr: 3.31e-03, grad_scale: 8.0 2023-04-03 03:38:16,671 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 2023-04-03 03:38:30,626 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=164517.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 03:38:36,205 INFO [train.py:903] (1/4) Epoch 25, batch 650, loss[loss=0.2183, simple_loss=0.2999, pruned_loss=0.06835, over 19532.00 frames. ], tot_loss[loss=0.2037, simple_loss=0.2843, pruned_loss=0.06152, over 3690651.37 frames. ], batch size: 56, lr: 3.31e-03, grad_scale: 8.0 2023-04-03 03:39:15,259 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.327e+02 4.538e+02 5.913e+02 7.820e+02 1.600e+03, threshold=1.183e+03, percent-clipped=2.0 2023-04-03 03:39:33,669 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=164567.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 03:39:39,463 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6271, 1.5299, 1.4411, 2.1647, 1.5690, 1.9345, 1.9560, 1.6576], device='cuda:1'), covar=tensor([0.0855, 0.0946, 0.1068, 0.0758, 0.0920, 0.0757, 0.0870, 0.0723], device='cuda:1'), in_proj_covar=tensor([0.0210, 0.0221, 0.0226, 0.0237, 0.0225, 0.0211, 0.0188, 0.0203], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-03 03:39:40,284 INFO [train.py:903] (1/4) Epoch 25, batch 700, loss[loss=0.2153, simple_loss=0.3008, pruned_loss=0.06495, over 19550.00 frames. ], tot_loss[loss=0.2039, simple_loss=0.2846, pruned_loss=0.06155, over 3717002.44 frames. ], batch size: 56, lr: 3.31e-03, grad_scale: 8.0 2023-04-03 03:39:58,628 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=164586.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 03:40:11,637 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=164596.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 03:40:44,580 INFO [train.py:903] (1/4) Epoch 25, batch 750, loss[loss=0.1753, simple_loss=0.2504, pruned_loss=0.05008, over 19311.00 frames. ], tot_loss[loss=0.204, simple_loss=0.2849, pruned_loss=0.06156, over 3723845.37 frames. ], batch size: 44, lr: 3.31e-03, grad_scale: 8.0 2023-04-03 03:40:57,962 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=164632.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 03:41:21,486 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.304e+02 5.157e+02 6.753e+02 8.219e+02 1.587e+03, threshold=1.351e+03, percent-clipped=10.0 2023-04-03 03:41:32,780 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.76 vs. limit=2.0 2023-04-03 03:41:48,141 INFO [train.py:903] (1/4) Epoch 25, batch 800, loss[loss=0.2292, simple_loss=0.3172, pruned_loss=0.0706, over 18286.00 frames. ], tot_loss[loss=0.204, simple_loss=0.2847, pruned_loss=0.06165, over 3746450.32 frames. ], batch size: 83, lr: 3.31e-03, grad_scale: 8.0 2023-04-03 03:41:52,837 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=164676.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 03:42:03,219 WARNING [train.py:1073] (1/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 2023-04-03 03:42:50,192 INFO [train.py:903] (1/4) Epoch 25, batch 850, loss[loss=0.209, simple_loss=0.2906, pruned_loss=0.06371, over 19663.00 frames. ], tot_loss[loss=0.2046, simple_loss=0.285, pruned_loss=0.06208, over 3759347.00 frames. ], batch size: 55, lr: 3.31e-03, grad_scale: 8.0 2023-04-03 03:42:50,364 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=164722.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 03:43:29,202 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.532e+02 4.928e+02 6.003e+02 7.929e+02 1.446e+03, threshold=1.201e+03, percent-clipped=1.0 2023-04-03 03:43:43,450 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=164764.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 03:43:44,361 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9_rvb from training. Duration: 25.061125 2023-04-03 03:43:52,629 INFO [train.py:903] (1/4) Epoch 25, batch 900, loss[loss=0.2271, simple_loss=0.3055, pruned_loss=0.07432, over 19502.00 frames. ], tot_loss[loss=0.2044, simple_loss=0.2852, pruned_loss=0.06185, over 3777820.34 frames. ], batch size: 64, lr: 3.31e-03, grad_scale: 8.0 2023-04-03 03:44:29,970 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.49 vs. limit=2.0 2023-04-03 03:44:38,680 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9992, 2.0907, 2.3393, 2.6062, 1.9247, 2.4312, 2.3079, 2.1536], device='cuda:1'), covar=tensor([0.4681, 0.4162, 0.2072, 0.2860, 0.4618, 0.2465, 0.5338, 0.3545], device='cuda:1'), in_proj_covar=tensor([0.0915, 0.0986, 0.0726, 0.0936, 0.0894, 0.0826, 0.0852, 0.0791], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-03 03:44:56,069 INFO [train.py:903] (1/4) Epoch 25, batch 950, loss[loss=0.1939, simple_loss=0.2726, pruned_loss=0.05759, over 19803.00 frames. ], tot_loss[loss=0.203, simple_loss=0.2836, pruned_loss=0.06122, over 3789883.77 frames. ], batch size: 48, lr: 3.31e-03, grad_scale: 8.0 2023-04-03 03:44:57,278 WARNING [train.py:1073] (1/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9_rvb from training. Duration: 26.32775 2023-04-03 03:44:57,673 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=164823.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 03:45:15,726 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=164837.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 03:45:28,682 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=164848.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 03:45:32,820 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.080e+02 4.681e+02 5.489e+02 6.922e+02 1.500e+03, threshold=1.098e+03, percent-clipped=4.0 2023-04-03 03:45:43,208 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2023-04-03 03:46:00,724 INFO [train.py:903] (1/4) Epoch 25, batch 1000, loss[loss=0.2154, simple_loss=0.2943, pruned_loss=0.06826, over 19509.00 frames. ], tot_loss[loss=0.2041, simple_loss=0.2844, pruned_loss=0.06193, over 3787091.32 frames. ], batch size: 64, lr: 3.31e-03, grad_scale: 8.0 2023-04-03 03:46:19,855 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=164888.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 03:46:53,177 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=164913.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 03:46:55,222 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 2023-04-03 03:47:03,441 INFO [train.py:903] (1/4) Epoch 25, batch 1050, loss[loss=0.193, simple_loss=0.2811, pruned_loss=0.0524, over 18699.00 frames. ], tot_loss[loss=0.2041, simple_loss=0.2847, pruned_loss=0.0618, over 3794945.62 frames. ], batch size: 74, lr: 3.31e-03, grad_scale: 8.0 2023-04-03 03:47:13,144 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=164930.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 03:47:27,116 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=164940.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 03:47:36,144 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 2023-04-03 03:47:41,879 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.266e+02 4.678e+02 5.688e+02 6.996e+02 1.189e+03, threshold=1.138e+03, percent-clipped=3.0 2023-04-03 03:47:45,578 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.7427, 1.4404, 1.5854, 1.5252, 3.3337, 1.1731, 2.4654, 3.8329], device='cuda:1'), covar=tensor([0.0468, 0.2657, 0.2808, 0.1884, 0.0724, 0.2535, 0.1285, 0.0203], device='cuda:1'), in_proj_covar=tensor([0.0417, 0.0369, 0.0394, 0.0349, 0.0377, 0.0353, 0.0388, 0.0409], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 03:48:06,294 INFO [train.py:903] (1/4) Epoch 25, batch 1100, loss[loss=0.2042, simple_loss=0.2829, pruned_loss=0.06276, over 19835.00 frames. ], tot_loss[loss=0.2058, simple_loss=0.2861, pruned_loss=0.06272, over 3781646.92 frames. ], batch size: 52, lr: 3.31e-03, grad_scale: 8.0 2023-04-03 03:49:07,182 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=165020.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 03:49:09,315 INFO [train.py:903] (1/4) Epoch 25, batch 1150, loss[loss=0.2826, simple_loss=0.3476, pruned_loss=0.1088, over 19518.00 frames. ], tot_loss[loss=0.2058, simple_loss=0.2857, pruned_loss=0.06293, over 3797702.49 frames. ], batch size: 64, lr: 3.31e-03, grad_scale: 8.0 2023-04-03 03:49:40,378 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=165045.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 03:49:48,277 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.006e+02 5.101e+02 6.369e+02 8.453e+02 1.568e+03, threshold=1.274e+03, percent-clipped=10.0 2023-04-03 03:49:52,163 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=165055.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 03:50:14,722 INFO [train.py:903] (1/4) Epoch 25, batch 1200, loss[loss=0.16, simple_loss=0.2376, pruned_loss=0.04118, over 17664.00 frames. ], tot_loss[loss=0.2052, simple_loss=0.2853, pruned_loss=0.0625, over 3800374.03 frames. ], batch size: 39, lr: 3.30e-03, grad_scale: 8.0 2023-04-03 03:50:41,040 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=165093.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 03:50:46,445 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 2023-04-03 03:51:01,321 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=165108.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 03:51:14,662 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=165118.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 03:51:18,442 INFO [train.py:903] (1/4) Epoch 25, batch 1250, loss[loss=0.2614, simple_loss=0.3388, pruned_loss=0.09197, over 19086.00 frames. ], tot_loss[loss=0.2066, simple_loss=0.2867, pruned_loss=0.06325, over 3805597.75 frames. ], batch size: 69, lr: 3.30e-03, grad_scale: 4.0 2023-04-03 03:51:34,074 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=165135.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 03:51:41,200 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.83 vs. limit=2.0 2023-04-03 03:51:43,121 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.8237, 4.2758, 4.4912, 4.5086, 1.6807, 4.2441, 3.7132, 4.1974], device='cuda:1'), covar=tensor([0.1578, 0.1006, 0.0641, 0.0688, 0.6226, 0.1038, 0.0709, 0.1219], device='cuda:1'), in_proj_covar=tensor([0.0802, 0.0766, 0.0974, 0.0850, 0.0856, 0.0736, 0.0579, 0.0903], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-03 03:51:43,140 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=165141.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 03:51:49,266 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1860, 1.2800, 1.6645, 1.3296, 2.7352, 3.8165, 3.5054, 3.9975], device='cuda:1'), covar=tensor([0.1696, 0.3907, 0.3544, 0.2495, 0.0628, 0.0178, 0.0205, 0.0244], device='cuda:1'), in_proj_covar=tensor([0.0276, 0.0329, 0.0359, 0.0267, 0.0248, 0.0191, 0.0218, 0.0266], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-03 03:51:57,598 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.318e+02 4.844e+02 6.322e+02 8.270e+02 1.695e+03, threshold=1.264e+03, percent-clipped=2.0 2023-04-03 03:52:21,098 INFO [train.py:903] (1/4) Epoch 25, batch 1300, loss[loss=0.1661, simple_loss=0.2492, pruned_loss=0.04153, over 19358.00 frames. ], tot_loss[loss=0.2065, simple_loss=0.2864, pruned_loss=0.06333, over 3797822.70 frames. ], batch size: 47, lr: 3.30e-03, grad_scale: 4.0 2023-04-03 03:53:23,839 INFO [train.py:903] (1/4) Epoch 25, batch 1350, loss[loss=0.2673, simple_loss=0.3402, pruned_loss=0.09725, over 18657.00 frames. ], tot_loss[loss=0.2068, simple_loss=0.2867, pruned_loss=0.06342, over 3801339.34 frames. ], batch size: 74, lr: 3.30e-03, grad_scale: 4.0 2023-04-03 03:53:27,617 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=165223.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 03:54:04,597 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.257e+02 4.919e+02 6.162e+02 7.502e+02 1.875e+03, threshold=1.232e+03, percent-clipped=3.0 2023-04-03 03:54:29,446 INFO [train.py:903] (1/4) Epoch 25, batch 1400, loss[loss=0.2049, simple_loss=0.2924, pruned_loss=0.05866, over 19609.00 frames. ], tot_loss[loss=0.2053, simple_loss=0.2855, pruned_loss=0.06257, over 3812820.48 frames. ], batch size: 61, lr: 3.30e-03, grad_scale: 4.0 2023-04-03 03:54:50,993 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.5796, 1.1882, 1.4241, 1.3493, 2.2407, 1.2094, 2.0767, 2.5168], device='cuda:1'), covar=tensor([0.0736, 0.2820, 0.2862, 0.1576, 0.0875, 0.1919, 0.1119, 0.0479], device='cuda:1'), in_proj_covar=tensor([0.0413, 0.0367, 0.0391, 0.0348, 0.0373, 0.0350, 0.0386, 0.0407], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 03:54:53,459 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.7470, 2.6342, 2.1449, 2.1179, 1.9091, 2.3661, 1.2535, 1.9809], device='cuda:1'), covar=tensor([0.0706, 0.0635, 0.0656, 0.1035, 0.1088, 0.0997, 0.1292, 0.1047], device='cuda:1'), in_proj_covar=tensor([0.0363, 0.0359, 0.0363, 0.0388, 0.0465, 0.0394, 0.0341, 0.0347], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 03:55:05,127 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=165301.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 03:55:18,660 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=165311.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 03:55:19,098 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 2023-04-03 03:55:32,124 INFO [train.py:903] (1/4) Epoch 25, batch 1450, loss[loss=0.23, simple_loss=0.3003, pruned_loss=0.07986, over 19678.00 frames. ], tot_loss[loss=0.2056, simple_loss=0.2856, pruned_loss=0.06283, over 3806560.31 frames. ], batch size: 53, lr: 3.30e-03, grad_scale: 4.0 2023-04-03 03:55:33,159 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_rvb from training. Duration: 25.85 2023-04-03 03:55:36,968 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=165326.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 03:55:48,905 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=165336.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 03:56:10,661 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.320e+02 4.584e+02 6.100e+02 8.063e+02 1.924e+03, threshold=1.220e+03, percent-clipped=4.0 2023-04-03 03:56:33,838 INFO [train.py:903] (1/4) Epoch 25, batch 1500, loss[loss=0.1984, simple_loss=0.274, pruned_loss=0.06147, over 19763.00 frames. ], tot_loss[loss=0.2063, simple_loss=0.2865, pruned_loss=0.06309, over 3814492.85 frames. ], batch size: 47, lr: 3.30e-03, grad_scale: 4.0 2023-04-03 03:56:58,446 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=165391.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 03:57:30,045 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=165416.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 03:57:36,921 INFO [train.py:903] (1/4) Epoch 25, batch 1550, loss[loss=0.1675, simple_loss=0.2426, pruned_loss=0.04622, over 19726.00 frames. ], tot_loss[loss=0.2064, simple_loss=0.2865, pruned_loss=0.06311, over 3811384.47 frames. ], batch size: 46, lr: 3.30e-03, grad_scale: 4.0 2023-04-03 03:58:06,783 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4665, 2.2345, 1.7086, 1.5689, 2.0309, 1.4058, 1.3761, 1.8963], device='cuda:1'), covar=tensor([0.1125, 0.0803, 0.1141, 0.0835, 0.0604, 0.1296, 0.0747, 0.0539], device='cuda:1'), in_proj_covar=tensor([0.0304, 0.0318, 0.0340, 0.0268, 0.0249, 0.0343, 0.0293, 0.0276], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 03:58:17,715 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.229e+02 4.957e+02 6.075e+02 6.924e+02 1.069e+03, threshold=1.215e+03, percent-clipped=0.0 2023-04-03 03:58:37,746 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 2023-04-03 03:58:42,875 INFO [train.py:903] (1/4) Epoch 25, batch 1600, loss[loss=0.2085, simple_loss=0.2851, pruned_loss=0.06594, over 19587.00 frames. ], tot_loss[loss=0.2058, simple_loss=0.2859, pruned_loss=0.06281, over 3807481.90 frames. ], batch size: 52, lr: 3.30e-03, grad_scale: 8.0 2023-04-03 03:58:51,603 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=165479.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 03:58:59,594 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=165485.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 03:59:06,532 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9_rvb from training. Duration: 30.1555625 2023-04-03 03:59:21,442 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=165504.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 03:59:44,248 INFO [train.py:903] (1/4) Epoch 25, batch 1650, loss[loss=0.1751, simple_loss=0.2551, pruned_loss=0.04759, over 19741.00 frames. ], tot_loss[loss=0.2058, simple_loss=0.2861, pruned_loss=0.06277, over 3820830.61 frames. ], batch size: 46, lr: 3.30e-03, grad_scale: 8.0 2023-04-03 04:00:23,741 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.160e+02 4.584e+02 6.289e+02 7.621e+02 1.672e+03, threshold=1.258e+03, percent-clipped=5.0 2023-04-03 04:00:44,046 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=165569.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 04:00:47,357 INFO [train.py:903] (1/4) Epoch 25, batch 1700, loss[loss=0.2014, simple_loss=0.2876, pruned_loss=0.0576, over 19676.00 frames. ], tot_loss[loss=0.2049, simple_loss=0.2855, pruned_loss=0.06211, over 3831196.10 frames. ], batch size: 60, lr: 3.30e-03, grad_scale: 8.0 2023-04-03 04:01:22,709 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=165600.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 04:01:29,475 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 2023-04-03 04:01:49,445 INFO [train.py:903] (1/4) Epoch 25, batch 1750, loss[loss=0.1725, simple_loss=0.2636, pruned_loss=0.04069, over 19827.00 frames. ], tot_loss[loss=0.2046, simple_loss=0.2854, pruned_loss=0.06191, over 3832345.47 frames. ], batch size: 52, lr: 3.30e-03, grad_scale: 8.0 2023-04-03 04:01:53,780 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.65 vs. limit=5.0 2023-04-03 04:02:29,216 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.400e+02 4.689e+02 5.610e+02 7.383e+02 2.270e+03, threshold=1.122e+03, percent-clipped=4.0 2023-04-03 04:02:53,370 INFO [train.py:903] (1/4) Epoch 25, batch 1800, loss[loss=0.2074, simple_loss=0.2855, pruned_loss=0.06465, over 19479.00 frames. ], tot_loss[loss=0.2055, simple_loss=0.2865, pruned_loss=0.06222, over 3831053.06 frames. ], batch size: 49, lr: 3.30e-03, grad_scale: 8.0 2023-04-03 04:03:51,789 WARNING [train.py:1073] (1/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9_rvb from training. Duration: 25.0944375 2023-04-03 04:03:56,460 INFO [train.py:903] (1/4) Epoch 25, batch 1850, loss[loss=0.2347, simple_loss=0.3055, pruned_loss=0.082, over 12846.00 frames. ], tot_loss[loss=0.2071, simple_loss=0.2874, pruned_loss=0.06334, over 3784697.13 frames. ], batch size: 135, lr: 3.30e-03, grad_scale: 8.0 2023-04-03 04:04:28,947 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9_rvb from training. Duration: 27.8166875 2023-04-03 04:04:37,056 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.780e+02 4.826e+02 5.776e+02 6.973e+02 2.376e+03, threshold=1.155e+03, percent-clipped=4.0 2023-04-03 04:04:56,243 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=165768.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 04:05:00,658 INFO [train.py:903] (1/4) Epoch 25, batch 1900, loss[loss=0.1774, simple_loss=0.2535, pruned_loss=0.05068, over 18671.00 frames. ], tot_loss[loss=0.2074, simple_loss=0.2877, pruned_loss=0.06352, over 3776914.99 frames. ], batch size: 41, lr: 3.30e-03, grad_scale: 8.0 2023-04-03 04:05:16,877 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 2023-04-03 04:05:21,692 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9_rvb from training. Duration: 27.02225 2023-04-03 04:05:24,248 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.6087, 4.1131, 4.2871, 4.2890, 1.6259, 4.1040, 3.5291, 4.0283], device='cuda:1'), covar=tensor([0.1709, 0.0878, 0.0648, 0.0709, 0.6051, 0.0856, 0.0675, 0.1091], device='cuda:1'), in_proj_covar=tensor([0.0809, 0.0770, 0.0979, 0.0852, 0.0855, 0.0740, 0.0581, 0.0907], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-03 04:05:48,079 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 2023-04-03 04:06:02,690 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=165821.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 04:06:03,528 INFO [train.py:903] (1/4) Epoch 25, batch 1950, loss[loss=0.1778, simple_loss=0.262, pruned_loss=0.04683, over 19732.00 frames. ], tot_loss[loss=0.2059, simple_loss=0.2862, pruned_loss=0.06276, over 3789802.72 frames. ], batch size: 47, lr: 3.30e-03, grad_scale: 8.0 2023-04-03 04:06:20,330 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3132, 1.3059, 1.4428, 1.4367, 1.8639, 1.7479, 1.8546, 1.1111], device='cuda:1'), covar=tensor([0.1823, 0.3242, 0.2068, 0.1461, 0.1224, 0.1748, 0.1167, 0.3873], device='cuda:1'), in_proj_covar=tensor([0.0542, 0.0655, 0.0730, 0.0493, 0.0625, 0.0538, 0.0661, 0.0560], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 04:06:44,201 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.055e+02 4.671e+02 6.009e+02 7.545e+02 1.239e+03, threshold=1.202e+03, percent-clipped=2.0 2023-04-03 04:06:46,988 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.7838, 1.3585, 1.5660, 1.5249, 3.3254, 1.0628, 2.5504, 3.7973], device='cuda:1'), covar=tensor([0.0479, 0.2923, 0.2989, 0.2022, 0.0758, 0.2825, 0.1279, 0.0235], device='cuda:1'), in_proj_covar=tensor([0.0415, 0.0370, 0.0393, 0.0350, 0.0375, 0.0354, 0.0388, 0.0409], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 04:06:48,310 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=165856.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 04:07:08,063 INFO [train.py:903] (1/4) Epoch 25, batch 2000, loss[loss=0.2001, simple_loss=0.2754, pruned_loss=0.06238, over 19741.00 frames. ], tot_loss[loss=0.2058, simple_loss=0.2863, pruned_loss=0.06267, over 3793481.20 frames. ], batch size: 47, lr: 3.30e-03, grad_scale: 8.0 2023-04-03 04:07:20,319 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=165881.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 04:08:00,826 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=165913.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 04:08:07,746 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_rvb from training. Duration: 26.8349375 2023-04-03 04:08:12,467 INFO [train.py:903] (1/4) Epoch 25, batch 2050, loss[loss=0.1879, simple_loss=0.277, pruned_loss=0.04937, over 19533.00 frames. ], tot_loss[loss=0.206, simple_loss=0.2869, pruned_loss=0.06249, over 3784240.92 frames. ], batch size: 54, lr: 3.30e-03, grad_scale: 8.0 2023-04-03 04:08:27,939 WARNING [train.py:1073] (1/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 2023-04-03 04:08:27,984 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 2023-04-03 04:08:38,775 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.0111, 3.6937, 2.5233, 3.2801, 0.8466, 3.6403, 3.4899, 3.5525], device='cuda:1'), covar=tensor([0.0805, 0.1059, 0.1921, 0.0885, 0.3866, 0.0742, 0.0964, 0.1146], device='cuda:1'), in_proj_covar=tensor([0.0514, 0.0421, 0.0503, 0.0351, 0.0402, 0.0444, 0.0437, 0.0468], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 04:08:48,973 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 2023-04-03 04:08:51,219 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.115e+02 5.221e+02 6.289e+02 7.653e+02 1.251e+03, threshold=1.258e+03, percent-clipped=2.0 2023-04-03 04:09:15,839 INFO [train.py:903] (1/4) Epoch 25, batch 2100, loss[loss=0.1776, simple_loss=0.2566, pruned_loss=0.04931, over 19772.00 frames. ], tot_loss[loss=0.2063, simple_loss=0.2871, pruned_loss=0.06274, over 3772656.10 frames. ], batch size: 47, lr: 3.30e-03, grad_scale: 8.0 2023-04-03 04:09:44,601 WARNING [train.py:1073] (1/4) Exclude cut with ID 453-131332-0000-131866_sp0.9_rvb from training. Duration: 25.3333125 2023-04-03 04:10:08,505 WARNING [train.py:1073] (1/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9_rvb from training. Duration: 26.6166875 2023-04-03 04:10:19,021 INFO [train.py:903] (1/4) Epoch 25, batch 2150, loss[loss=0.1922, simple_loss=0.2733, pruned_loss=0.05559, over 19472.00 frames. ], tot_loss[loss=0.2078, simple_loss=0.2883, pruned_loss=0.06365, over 3771109.40 frames. ], batch size: 49, lr: 3.30e-03, grad_scale: 8.0 2023-04-03 04:10:26,463 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0271, 2.1328, 2.4010, 2.6183, 2.0593, 2.5115, 2.3940, 2.2183], device='cuda:1'), covar=tensor([0.4378, 0.4001, 0.1986, 0.2445, 0.4261, 0.2303, 0.5047, 0.3356], device='cuda:1'), in_proj_covar=tensor([0.0917, 0.0992, 0.0729, 0.0937, 0.0895, 0.0829, 0.0855, 0.0793], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-03 04:10:27,624 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=166028.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 04:10:48,718 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.2049, 2.1192, 1.9585, 1.7855, 1.5133, 1.7637, 0.6931, 1.1905], device='cuda:1'), covar=tensor([0.0660, 0.0690, 0.0527, 0.0906, 0.1380, 0.1102, 0.1485, 0.1191], device='cuda:1'), in_proj_covar=tensor([0.0362, 0.0359, 0.0363, 0.0387, 0.0465, 0.0392, 0.0341, 0.0346], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 04:10:58,799 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.324e+02 5.055e+02 6.614e+02 9.413e+02 1.694e+03, threshold=1.323e+03, percent-clipped=9.0 2023-04-03 04:11:21,715 INFO [train.py:903] (1/4) Epoch 25, batch 2200, loss[loss=0.2323, simple_loss=0.314, pruned_loss=0.07533, over 17473.00 frames. ], tot_loss[loss=0.2074, simple_loss=0.2879, pruned_loss=0.06343, over 3781894.57 frames. ], batch size: 101, lr: 3.29e-03, grad_scale: 8.0 2023-04-03 04:11:22,012 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=166072.0, num_to_drop=1, layers_to_drop={0} 2023-04-03 04:11:38,542 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.1031, 1.9575, 1.7558, 2.1056, 1.8988, 1.8165, 1.6478, 1.9853], device='cuda:1'), covar=tensor([0.0962, 0.1392, 0.1380, 0.0989, 0.1294, 0.0539, 0.1480, 0.0714], device='cuda:1'), in_proj_covar=tensor([0.0274, 0.0355, 0.0311, 0.0254, 0.0301, 0.0252, 0.0313, 0.0258], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 04:12:02,236 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3536, 2.0565, 1.6388, 1.4212, 1.8937, 1.3330, 1.3035, 1.8784], device='cuda:1'), covar=tensor([0.0932, 0.0823, 0.1113, 0.0897, 0.0575, 0.1286, 0.0683, 0.0400], device='cuda:1'), in_proj_covar=tensor([0.0304, 0.0318, 0.0340, 0.0267, 0.0249, 0.0341, 0.0293, 0.0274], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 04:12:13,548 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=166112.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 04:12:26,088 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([0.9377, 1.3122, 1.0737, 0.8266, 1.1569, 0.9184, 0.9640, 1.2856], device='cuda:1'), covar=tensor([0.0590, 0.0863, 0.1140, 0.0958, 0.0553, 0.1334, 0.0587, 0.0427], device='cuda:1'), in_proj_covar=tensor([0.0303, 0.0317, 0.0339, 0.0266, 0.0248, 0.0340, 0.0292, 0.0273], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 04:12:26,896 INFO [train.py:903] (1/4) Epoch 25, batch 2250, loss[loss=0.2209, simple_loss=0.3061, pruned_loss=0.06784, over 19694.00 frames. ], tot_loss[loss=0.2062, simple_loss=0.2873, pruned_loss=0.06253, over 3787147.11 frames. ], batch size: 59, lr: 3.29e-03, grad_scale: 8.0 2023-04-03 04:12:34,205 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9518, 1.9324, 1.8217, 1.5881, 1.5294, 1.6370, 0.3997, 0.8593], device='cuda:1'), covar=tensor([0.0659, 0.0662, 0.0422, 0.0765, 0.1265, 0.0855, 0.1447, 0.1201], device='cuda:1'), in_proj_covar=tensor([0.0363, 0.0360, 0.0364, 0.0389, 0.0466, 0.0393, 0.0343, 0.0347], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 04:13:04,899 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.096e+02 4.777e+02 5.716e+02 7.657e+02 1.924e+03, threshold=1.143e+03, percent-clipped=2.0 2023-04-03 04:13:17,345 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6556, 2.2727, 1.7025, 1.6843, 2.1079, 1.4948, 1.5027, 2.0122], device='cuda:1'), covar=tensor([0.1018, 0.0702, 0.1013, 0.0741, 0.0560, 0.1147, 0.0682, 0.0477], device='cuda:1'), in_proj_covar=tensor([0.0303, 0.0317, 0.0340, 0.0267, 0.0248, 0.0341, 0.0292, 0.0273], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 04:13:21,655 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=166165.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 04:13:21,807 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=166165.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 04:13:31,169 INFO [train.py:903] (1/4) Epoch 25, batch 2300, loss[loss=0.1996, simple_loss=0.2818, pruned_loss=0.05875, over 19691.00 frames. ], tot_loss[loss=0.2062, simple_loss=0.287, pruned_loss=0.06274, over 3781868.51 frames. ], batch size: 53, lr: 3.29e-03, grad_scale: 8.0 2023-04-03 04:13:42,668 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_rvb from training. Duration: 26.205 2023-04-03 04:14:00,660 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5403, 1.3086, 1.3752, 2.3069, 1.6555, 1.8053, 1.8572, 1.5625], device='cuda:1'), covar=tensor([0.0995, 0.1223, 0.1204, 0.0768, 0.0980, 0.0941, 0.1024, 0.0870], device='cuda:1'), in_proj_covar=tensor([0.0213, 0.0222, 0.0226, 0.0239, 0.0226, 0.0213, 0.0189, 0.0204], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-04-03 04:14:34,782 INFO [train.py:903] (1/4) Epoch 25, batch 2350, loss[loss=0.2069, simple_loss=0.2941, pruned_loss=0.05981, over 19435.00 frames. ], tot_loss[loss=0.206, simple_loss=0.2866, pruned_loss=0.06272, over 3777896.40 frames. ], batch size: 70, lr: 3.29e-03, grad_scale: 8.0 2023-04-03 04:14:42,124 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=166227.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 04:15:14,902 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.796e+02 5.104e+02 6.317e+02 8.219e+02 1.547e+03, threshold=1.263e+03, percent-clipped=3.0 2023-04-03 04:15:17,167 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_rvb from training. Duration: 25.775 2023-04-03 04:15:19,888 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.1631, 1.2811, 1.4918, 1.4315, 2.8166, 1.1711, 2.2949, 3.1995], device='cuda:1'), covar=tensor([0.0514, 0.2680, 0.2828, 0.1784, 0.0706, 0.2310, 0.1107, 0.0272], device='cuda:1'), in_proj_covar=tensor([0.0414, 0.0370, 0.0391, 0.0348, 0.0375, 0.0353, 0.0389, 0.0408], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 04:15:34,636 WARNING [train.py:1073] (1/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9_rvb from training. Duration: 25.45 2023-04-03 04:15:38,211 INFO [train.py:903] (1/4) Epoch 25, batch 2400, loss[loss=0.1974, simple_loss=0.2818, pruned_loss=0.05647, over 19679.00 frames. ], tot_loss[loss=0.2053, simple_loss=0.2858, pruned_loss=0.06237, over 3761520.05 frames. ], batch size: 53, lr: 3.29e-03, grad_scale: 8.0 2023-04-03 04:15:48,903 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=166280.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 04:15:55,316 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=166284.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 04:15:59,843 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5878, 1.6761, 1.9194, 1.8450, 2.7214, 2.4055, 2.9100, 1.2381], device='cuda:1'), covar=tensor([0.2449, 0.4202, 0.2716, 0.1823, 0.1506, 0.2023, 0.1351, 0.4539], device='cuda:1'), in_proj_covar=tensor([0.0543, 0.0656, 0.0731, 0.0494, 0.0626, 0.0538, 0.0664, 0.0561], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 04:16:25,801 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=166309.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 04:16:41,096 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6742, 1.7928, 2.0436, 1.9976, 1.5505, 1.9730, 2.0390, 1.9115], device='cuda:1'), covar=tensor([0.4154, 0.3750, 0.1986, 0.2231, 0.3842, 0.2152, 0.4965, 0.3419], device='cuda:1'), in_proj_covar=tensor([0.0917, 0.0990, 0.0729, 0.0938, 0.0895, 0.0830, 0.0852, 0.0793], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-03 04:16:41,763 INFO [train.py:903] (1/4) Epoch 25, batch 2450, loss[loss=0.2066, simple_loss=0.2817, pruned_loss=0.06574, over 19372.00 frames. ], tot_loss[loss=0.205, simple_loss=0.2855, pruned_loss=0.06221, over 3775772.08 frames. ], batch size: 47, lr: 3.29e-03, grad_scale: 8.0 2023-04-03 04:16:46,714 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=166325.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 04:17:20,232 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.023e+02 4.903e+02 5.916e+02 7.490e+02 1.353e+03, threshold=1.183e+03, percent-clipped=1.0 2023-04-03 04:17:44,922 INFO [train.py:903] (1/4) Epoch 25, batch 2500, loss[loss=0.1923, simple_loss=0.2683, pruned_loss=0.05815, over 19626.00 frames. ], tot_loss[loss=0.205, simple_loss=0.2858, pruned_loss=0.06214, over 3782815.99 frames. ], batch size: 50, lr: 3.29e-03, grad_scale: 8.0 2023-04-03 04:17:51,989 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.2887, 2.3431, 2.3237, 3.0496, 2.5917, 2.9641, 2.5105, 2.1283], device='cuda:1'), covar=tensor([0.4021, 0.3731, 0.2264, 0.2538, 0.3730, 0.2075, 0.4641, 0.3817], device='cuda:1'), in_proj_covar=tensor([0.0916, 0.0991, 0.0729, 0.0938, 0.0894, 0.0831, 0.0851, 0.0793], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-03 04:18:41,259 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=166416.0, num_to_drop=1, layers_to_drop={0} 2023-04-03 04:18:48,039 INFO [train.py:903] (1/4) Epoch 25, batch 2550, loss[loss=0.1588, simple_loss=0.2379, pruned_loss=0.03985, over 19735.00 frames. ], tot_loss[loss=0.2053, simple_loss=0.286, pruned_loss=0.0623, over 3784321.62 frames. ], batch size: 46, lr: 3.29e-03, grad_scale: 8.0 2023-04-03 04:19:21,986 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.13 vs. limit=2.0 2023-04-03 04:19:28,546 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.584e+02 5.112e+02 6.516e+02 8.109e+02 2.174e+03, threshold=1.303e+03, percent-clipped=8.0 2023-04-03 04:19:40,779 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0838, 2.0024, 1.9156, 1.7249, 1.5862, 1.7099, 0.5008, 1.0141], device='cuda:1'), covar=tensor([0.0626, 0.0652, 0.0462, 0.0820, 0.1265, 0.0936, 0.1456, 0.1148], device='cuda:1'), in_proj_covar=tensor([0.0363, 0.0360, 0.0363, 0.0387, 0.0467, 0.0393, 0.0340, 0.0345], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 04:19:46,425 WARNING [train.py:1073] (1/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 2023-04-03 04:19:53,262 INFO [train.py:903] (1/4) Epoch 25, batch 2600, loss[loss=0.1925, simple_loss=0.2809, pruned_loss=0.05207, over 19278.00 frames. ], tot_loss[loss=0.2037, simple_loss=0.2843, pruned_loss=0.0615, over 3804909.27 frames. ], batch size: 66, lr: 3.29e-03, grad_scale: 8.0 2023-04-03 04:20:08,567 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=166483.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 04:20:29,567 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=166499.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 04:20:34,776 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.42 vs. limit=2.0 2023-04-03 04:20:40,195 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=166508.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 04:20:41,207 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=166509.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 04:20:58,302 INFO [train.py:903] (1/4) Epoch 25, batch 2650, loss[loss=0.2294, simple_loss=0.3047, pruned_loss=0.07705, over 19683.00 frames. ], tot_loss[loss=0.2038, simple_loss=0.2842, pruned_loss=0.06171, over 3804727.46 frames. ], batch size: 60, lr: 3.29e-03, grad_scale: 8.0 2023-04-03 04:21:05,507 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=166527.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 04:21:10,252 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=166531.0, num_to_drop=1, layers_to_drop={1} 2023-04-03 04:21:12,606 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([0.9284, 1.1764, 1.5188, 0.6310, 1.9498, 2.4590, 2.1758, 2.6171], device='cuda:1'), covar=tensor([0.1584, 0.3897, 0.3419, 0.2761, 0.0648, 0.0277, 0.0335, 0.0363], device='cuda:1'), in_proj_covar=tensor([0.0274, 0.0328, 0.0359, 0.0267, 0.0248, 0.0192, 0.0217, 0.0267], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-03 04:21:17,648 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=166536.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 04:21:19,718 WARNING [train.py:1073] (1/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9_rvb from training. Duration: 27.25 2023-04-03 04:21:32,592 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=166549.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 04:21:38,177 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.868e+02 4.612e+02 5.770e+02 6.756e+02 1.610e+03, threshold=1.154e+03, percent-clipped=1.0 2023-04-03 04:21:48,102 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=166561.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 04:22:03,057 INFO [train.py:903] (1/4) Epoch 25, batch 2700, loss[loss=0.1998, simple_loss=0.2886, pruned_loss=0.05556, over 19671.00 frames. ], tot_loss[loss=0.2032, simple_loss=0.2838, pruned_loss=0.06128, over 3819366.62 frames. ], batch size: 55, lr: 3.29e-03, grad_scale: 8.0 2023-04-03 04:22:21,012 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.45 vs. limit=2.0 2023-04-03 04:23:06,175 INFO [train.py:903] (1/4) Epoch 25, batch 2750, loss[loss=0.1905, simple_loss=0.2752, pruned_loss=0.05288, over 19620.00 frames. ], tot_loss[loss=0.2044, simple_loss=0.2853, pruned_loss=0.06175, over 3825230.94 frames. ], batch size: 50, lr: 3.29e-03, grad_scale: 8.0 2023-04-03 04:23:08,990 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=166624.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 04:23:45,202 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.014e+02 4.918e+02 6.109e+02 7.546e+02 1.552e+03, threshold=1.222e+03, percent-clipped=5.0 2023-04-03 04:24:04,562 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=166669.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 04:24:07,645 INFO [train.py:903] (1/4) Epoch 25, batch 2800, loss[loss=0.263, simple_loss=0.3299, pruned_loss=0.09807, over 19776.00 frames. ], tot_loss[loss=0.205, simple_loss=0.2857, pruned_loss=0.06219, over 3842478.59 frames. ], batch size: 56, lr: 3.29e-03, grad_scale: 8.0 2023-04-03 04:24:43,338 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.0425, 4.4508, 4.7636, 4.7716, 1.8329, 4.4614, 3.8813, 4.4659], device='cuda:1'), covar=tensor([0.1602, 0.0851, 0.0614, 0.0655, 0.5884, 0.0972, 0.0606, 0.1084], device='cuda:1'), in_proj_covar=tensor([0.0803, 0.0766, 0.0975, 0.0853, 0.0851, 0.0740, 0.0580, 0.0905], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-03 04:25:11,686 INFO [train.py:903] (1/4) Epoch 25, batch 2850, loss[loss=0.2128, simple_loss=0.2954, pruned_loss=0.06511, over 17752.00 frames. ], tot_loss[loss=0.2057, simple_loss=0.2867, pruned_loss=0.06236, over 3830561.08 frames. ], batch size: 101, lr: 3.29e-03, grad_scale: 8.0 2023-04-03 04:25:50,389 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.797e+02 4.563e+02 5.950e+02 8.060e+02 1.987e+03, threshold=1.190e+03, percent-clipped=10.0 2023-04-03 04:25:50,631 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=166753.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 04:26:11,840 WARNING [train.py:1073] (1/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 2023-04-03 04:26:14,225 INFO [train.py:903] (1/4) Epoch 25, batch 2900, loss[loss=0.1854, simple_loss=0.2717, pruned_loss=0.04953, over 19771.00 frames. ], tot_loss[loss=0.2055, simple_loss=0.2862, pruned_loss=0.06242, over 3831193.79 frames. ], batch size: 54, lr: 3.29e-03, grad_scale: 8.0 2023-04-03 04:26:27,310 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=166782.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 04:26:29,647 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=166784.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 04:26:33,231 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=166787.0, num_to_drop=1, layers_to_drop={1} 2023-04-03 04:27:05,071 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=166812.0, num_to_drop=1, layers_to_drop={0} 2023-04-03 04:27:07,303 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.3329, 2.3500, 2.5197, 2.9860, 2.3751, 2.8387, 2.5294, 2.4265], device='cuda:1'), covar=tensor([0.4154, 0.4091, 0.1956, 0.2670, 0.4272, 0.2327, 0.4782, 0.3186], device='cuda:1'), in_proj_covar=tensor([0.0918, 0.0990, 0.0729, 0.0938, 0.0894, 0.0829, 0.0853, 0.0793], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-03 04:27:17,227 INFO [train.py:903] (1/4) Epoch 25, batch 2950, loss[loss=0.2127, simple_loss=0.2917, pruned_loss=0.06685, over 19694.00 frames. ], tot_loss[loss=0.2048, simple_loss=0.2853, pruned_loss=0.06218, over 3828384.51 frames. ], batch size: 60, lr: 3.29e-03, grad_scale: 8.0 2023-04-03 04:27:44,180 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=166843.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 04:27:56,486 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.385e+02 4.674e+02 5.957e+02 7.713e+02 2.101e+03, threshold=1.191e+03, percent-clipped=6.0 2023-04-03 04:28:20,015 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=166871.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 04:28:21,106 INFO [train.py:903] (1/4) Epoch 25, batch 3000, loss[loss=0.2498, simple_loss=0.3155, pruned_loss=0.09211, over 13203.00 frames. ], tot_loss[loss=0.2043, simple_loss=0.2851, pruned_loss=0.06175, over 3829089.80 frames. ], batch size: 136, lr: 3.29e-03, grad_scale: 8.0 2023-04-03 04:28:21,107 INFO [train.py:928] (1/4) Computing validation loss 2023-04-03 04:28:33,779 INFO [train.py:937] (1/4) Epoch 25, validation: loss=0.1677, simple_loss=0.2674, pruned_loss=0.034, over 944034.00 frames. 2023-04-03 04:28:33,780 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18821MB 2023-04-03 04:28:35,137 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_rvb from training. Duration: 29.735 2023-04-03 04:28:44,952 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=166880.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 04:29:01,373 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=166893.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 04:29:17,265 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=166905.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 04:29:38,485 INFO [train.py:903] (1/4) Epoch 25, batch 3050, loss[loss=0.2283, simple_loss=0.3071, pruned_loss=0.0747, over 17435.00 frames. ], tot_loss[loss=0.2041, simple_loss=0.2848, pruned_loss=0.06169, over 3818022.45 frames. ], batch size: 101, lr: 3.29e-03, grad_scale: 8.0 2023-04-03 04:30:09,922 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=166947.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 04:30:17,432 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.386e+02 4.942e+02 6.233e+02 8.295e+02 1.859e+03, threshold=1.247e+03, percent-clipped=9.0 2023-04-03 04:30:23,331 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=166958.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 04:30:37,564 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0965, 1.9534, 1.7287, 2.0864, 1.8203, 1.7459, 1.6486, 1.9622], device='cuda:1'), covar=tensor([0.1024, 0.1469, 0.1461, 0.1049, 0.1446, 0.0584, 0.1527, 0.0726], device='cuda:1'), in_proj_covar=tensor([0.0275, 0.0355, 0.0314, 0.0254, 0.0304, 0.0254, 0.0314, 0.0259], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 04:30:41,892 INFO [train.py:903] (1/4) Epoch 25, batch 3100, loss[loss=0.2573, simple_loss=0.3192, pruned_loss=0.09774, over 13496.00 frames. ], tot_loss[loss=0.204, simple_loss=0.2846, pruned_loss=0.0617, over 3817779.87 frames. ], batch size: 135, lr: 3.29e-03, grad_scale: 8.0 2023-04-03 04:30:58,982 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=166986.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 04:31:27,808 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=167008.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 04:31:44,381 INFO [train.py:903] (1/4) Epoch 25, batch 3150, loss[loss=0.1855, simple_loss=0.2638, pruned_loss=0.05363, over 19795.00 frames. ], tot_loss[loss=0.2042, simple_loss=0.285, pruned_loss=0.06165, over 3819425.85 frames. ], batch size: 47, lr: 3.29e-03, grad_scale: 8.0 2023-04-03 04:32:08,006 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=167040.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 04:32:08,791 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 2023-04-03 04:32:24,691 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.516e+02 5.039e+02 6.011e+02 8.459e+02 2.094e+03, threshold=1.202e+03, percent-clipped=4.0 2023-04-03 04:32:33,686 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 2023-04-03 04:32:39,339 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=167065.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 04:32:48,328 INFO [train.py:903] (1/4) Epoch 25, batch 3200, loss[loss=0.2148, simple_loss=0.291, pruned_loss=0.06927, over 19667.00 frames. ], tot_loss[loss=0.2047, simple_loss=0.2857, pruned_loss=0.06188, over 3828713.76 frames. ], batch size: 53, lr: 3.29e-03, grad_scale: 8.0 2023-04-03 04:33:20,650 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=167097.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 04:33:23,568 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.53 vs. limit=5.0 2023-04-03 04:33:47,231 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=167118.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 04:33:50,767 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.7905, 4.4190, 2.8377, 3.7917, 1.0414, 4.3674, 4.1837, 4.2808], device='cuda:1'), covar=tensor([0.0566, 0.0852, 0.1732, 0.0817, 0.3784, 0.0629, 0.0853, 0.1097], device='cuda:1'), in_proj_covar=tensor([0.0517, 0.0419, 0.0504, 0.0354, 0.0403, 0.0446, 0.0438, 0.0470], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 04:33:52,591 INFO [train.py:903] (1/4) Epoch 25, batch 3250, loss[loss=0.2145, simple_loss=0.2996, pruned_loss=0.06474, over 19615.00 frames. ], tot_loss[loss=0.2041, simple_loss=0.2855, pruned_loss=0.06133, over 3838523.81 frames. ], batch size: 61, lr: 3.28e-03, grad_scale: 8.0 2023-04-03 04:33:52,875 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=167122.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 04:33:57,478 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=167126.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 04:33:59,209 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.52 vs. limit=2.0 2023-04-03 04:34:23,561 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5835, 1.8150, 2.2439, 1.9490, 3.2117, 2.6290, 3.5200, 1.7984], device='cuda:1'), covar=tensor([0.2583, 0.4357, 0.2684, 0.1882, 0.1548, 0.2192, 0.1563, 0.4315], device='cuda:1'), in_proj_covar=tensor([0.0542, 0.0656, 0.0730, 0.0494, 0.0622, 0.0537, 0.0663, 0.0561], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 04:34:33,313 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.514e+02 4.897e+02 6.248e+02 7.764e+02 1.427e+03, threshold=1.250e+03, percent-clipped=4.0 2023-04-03 04:34:56,637 INFO [train.py:903] (1/4) Epoch 25, batch 3300, loss[loss=0.2261, simple_loss=0.3026, pruned_loss=0.07475, over 17518.00 frames. ], tot_loss[loss=0.2045, simple_loss=0.2861, pruned_loss=0.06144, over 3827166.82 frames. ], batch size: 101, lr: 3.28e-03, grad_scale: 8.0 2023-04-03 04:34:56,669 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 2023-04-03 04:35:15,534 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1777, 1.1607, 1.7579, 1.6228, 2.7111, 4.4206, 4.2694, 5.0141], device='cuda:1'), covar=tensor([0.1861, 0.5576, 0.4761, 0.2611, 0.0820, 0.0262, 0.0258, 0.0212], device='cuda:1'), in_proj_covar=tensor([0.0276, 0.0328, 0.0359, 0.0268, 0.0250, 0.0192, 0.0219, 0.0268], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-03 04:35:47,900 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=167212.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 04:35:50,081 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=167214.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 04:35:59,019 INFO [train.py:903] (1/4) Epoch 25, batch 3350, loss[loss=0.2285, simple_loss=0.3086, pruned_loss=0.07418, over 18748.00 frames. ], tot_loss[loss=0.2051, simple_loss=0.2862, pruned_loss=0.06196, over 3836535.58 frames. ], batch size: 74, lr: 3.28e-03, grad_scale: 8.0 2023-04-03 04:36:21,845 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=167239.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 04:36:24,232 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=167241.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 04:36:25,482 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=167242.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 04:36:41,238 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.112e+02 4.821e+02 5.770e+02 7.565e+02 1.496e+03, threshold=1.154e+03, percent-clipped=2.0 2023-04-03 04:36:53,570 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=167264.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 04:36:56,878 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=167267.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 04:37:02,309 INFO [train.py:903] (1/4) Epoch 25, batch 3400, loss[loss=0.2224, simple_loss=0.3099, pruned_loss=0.06739, over 19526.00 frames. ], tot_loss[loss=0.2044, simple_loss=0.2856, pruned_loss=0.0616, over 3834058.23 frames. ], batch size: 64, lr: 3.28e-03, grad_scale: 8.0 2023-04-03 04:37:26,115 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=167289.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 04:37:28,365 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=167291.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 04:38:08,135 INFO [train.py:903] (1/4) Epoch 25, batch 3450, loss[loss=0.1942, simple_loss=0.2815, pruned_loss=0.05344, over 19600.00 frames. ], tot_loss[loss=0.2028, simple_loss=0.2836, pruned_loss=0.06097, over 3840823.37 frames. ], batch size: 57, lr: 3.28e-03, grad_scale: 8.0 2023-04-03 04:38:10,559 WARNING [train.py:1073] (1/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9_rvb from training. Duration: 25.2444375 2023-04-03 04:38:35,943 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6396, 1.6418, 1.5979, 1.3797, 1.3523, 1.4491, 0.3595, 0.7287], device='cuda:1'), covar=tensor([0.0676, 0.0638, 0.0446, 0.0670, 0.1167, 0.0737, 0.1329, 0.1141], device='cuda:1'), in_proj_covar=tensor([0.0363, 0.0360, 0.0362, 0.0386, 0.0465, 0.0393, 0.0341, 0.0344], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 04:38:49,770 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.112e+02 4.608e+02 5.631e+02 6.989e+02 1.333e+03, threshold=1.126e+03, percent-clipped=1.0 2023-04-03 04:39:06,955 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.3060, 2.1706, 2.1165, 1.9915, 1.6549, 1.9523, 0.6499, 1.3122], device='cuda:1'), covar=tensor([0.0642, 0.0607, 0.0433, 0.0784, 0.1193, 0.0877, 0.1368, 0.1092], device='cuda:1'), in_proj_covar=tensor([0.0361, 0.0358, 0.0361, 0.0385, 0.0464, 0.0392, 0.0340, 0.0343], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 04:39:12,369 INFO [train.py:903] (1/4) Epoch 25, batch 3500, loss[loss=0.2283, simple_loss=0.3103, pruned_loss=0.07312, over 19662.00 frames. ], tot_loss[loss=0.2045, simple_loss=0.2847, pruned_loss=0.06214, over 3836518.84 frames. ], batch size: 60, lr: 3.28e-03, grad_scale: 8.0 2023-04-03 04:39:42,200 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.2763, 3.5857, 2.2810, 2.1595, 3.3738, 1.9527, 1.8115, 2.4647], device='cuda:1'), covar=tensor([0.1345, 0.0614, 0.0949, 0.0944, 0.0493, 0.1163, 0.0910, 0.0643], device='cuda:1'), in_proj_covar=tensor([0.0303, 0.0318, 0.0336, 0.0266, 0.0248, 0.0342, 0.0291, 0.0274], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 04:39:56,126 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=167406.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 04:40:15,427 INFO [train.py:903] (1/4) Epoch 25, batch 3550, loss[loss=0.1681, simple_loss=0.2453, pruned_loss=0.0455, over 19750.00 frames. ], tot_loss[loss=0.2044, simple_loss=0.2847, pruned_loss=0.06203, over 3830570.78 frames. ], batch size: 47, lr: 3.28e-03, grad_scale: 8.0 2023-04-03 04:40:51,399 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=167450.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 04:40:55,637 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.122e+02 5.129e+02 6.252e+02 7.981e+02 1.792e+03, threshold=1.250e+03, percent-clipped=6.0 2023-04-03 04:41:06,577 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=167462.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 04:41:11,114 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=167466.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 04:41:13,918 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=167468.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 04:41:17,988 INFO [train.py:903] (1/4) Epoch 25, batch 3600, loss[loss=0.2605, simple_loss=0.3398, pruned_loss=0.09059, over 19751.00 frames. ], tot_loss[loss=0.2033, simple_loss=0.2838, pruned_loss=0.06137, over 3834417.24 frames. ], batch size: 63, lr: 3.28e-03, grad_scale: 8.0 2023-04-03 04:41:44,845 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=167493.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 04:41:50,891 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=167497.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 04:42:01,218 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.6189, 1.2935, 1.5826, 1.5719, 3.1990, 1.1178, 2.2615, 3.6865], device='cuda:1'), covar=tensor([0.0481, 0.2896, 0.2808, 0.1761, 0.0676, 0.2585, 0.1361, 0.0229], device='cuda:1'), in_proj_covar=tensor([0.0417, 0.0373, 0.0394, 0.0349, 0.0377, 0.0353, 0.0391, 0.0411], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 04:42:20,842 INFO [train.py:903] (1/4) Epoch 25, batch 3650, loss[loss=0.2163, simple_loss=0.2942, pruned_loss=0.06916, over 19320.00 frames. ], tot_loss[loss=0.2044, simple_loss=0.2851, pruned_loss=0.0619, over 3823566.40 frames. ], batch size: 66, lr: 3.28e-03, grad_scale: 8.0 2023-04-03 04:42:21,242 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=167522.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 04:42:36,699 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 2023-04-03 04:43:00,360 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.203e+02 5.264e+02 6.466e+02 8.043e+02 1.528e+03, threshold=1.293e+03, percent-clipped=3.0 2023-04-03 04:43:24,181 INFO [train.py:903] (1/4) Epoch 25, batch 3700, loss[loss=0.2179, simple_loss=0.3013, pruned_loss=0.06727, over 19434.00 frames. ], tot_loss[loss=0.205, simple_loss=0.2854, pruned_loss=0.06226, over 3823121.65 frames. ], batch size: 70, lr: 3.28e-03, grad_scale: 8.0 2023-04-03 04:43:31,595 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=167577.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 04:43:36,278 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=167581.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 04:44:28,019 INFO [train.py:903] (1/4) Epoch 25, batch 3750, loss[loss=0.2144, simple_loss=0.2972, pruned_loss=0.06574, over 19560.00 frames. ], tot_loss[loss=0.2054, simple_loss=0.2858, pruned_loss=0.06252, over 3819447.38 frames. ], batch size: 56, lr: 3.28e-03, grad_scale: 8.0 2023-04-03 04:44:57,789 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0190, 2.1086, 2.2977, 2.7296, 2.0706, 2.5806, 2.2745, 2.0621], device='cuda:1'), covar=tensor([0.4347, 0.4066, 0.2041, 0.2469, 0.4276, 0.2215, 0.5367, 0.3691], device='cuda:1'), in_proj_covar=tensor([0.0923, 0.0995, 0.0731, 0.0943, 0.0898, 0.0832, 0.0857, 0.0797], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-03 04:45:08,818 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.966e+02 5.285e+02 6.593e+02 8.150e+02 1.742e+03, threshold=1.319e+03, percent-clipped=4.0 2023-04-03 04:45:19,685 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.3852, 4.0135, 2.6136, 3.5381, 0.9193, 3.9842, 3.8356, 3.9065], device='cuda:1'), covar=tensor([0.0724, 0.1069, 0.2070, 0.0887, 0.4084, 0.0735, 0.0995, 0.1400], device='cuda:1'), in_proj_covar=tensor([0.0521, 0.0422, 0.0509, 0.0356, 0.0407, 0.0450, 0.0441, 0.0473], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 04:45:19,905 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=167662.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 04:45:31,285 INFO [train.py:903] (1/4) Epoch 25, batch 3800, loss[loss=0.1931, simple_loss=0.2709, pruned_loss=0.05758, over 19675.00 frames. ], tot_loss[loss=0.2056, simple_loss=0.2861, pruned_loss=0.06257, over 3809599.62 frames. ], batch size: 53, lr: 3.28e-03, grad_scale: 8.0 2023-04-03 04:45:50,892 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=167687.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 04:46:01,966 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9_rvb from training. Duration: 29.1166875 2023-04-03 04:46:32,954 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.72 vs. limit=2.0 2023-04-03 04:46:35,340 INFO [train.py:903] (1/4) Epoch 25, batch 3850, loss[loss=0.2661, simple_loss=0.3374, pruned_loss=0.09744, over 19598.00 frames. ], tot_loss[loss=0.2052, simple_loss=0.2856, pruned_loss=0.06234, over 3811218.65 frames. ], batch size: 57, lr: 3.28e-03, grad_scale: 4.0 2023-04-03 04:46:54,001 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=167737.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 04:47:16,489 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.257e+02 4.991e+02 5.959e+02 7.597e+02 1.650e+03, threshold=1.192e+03, percent-clipped=2.0 2023-04-03 04:47:39,474 INFO [train.py:903] (1/4) Epoch 25, batch 3900, loss[loss=0.1901, simple_loss=0.2679, pruned_loss=0.05622, over 19484.00 frames. ], tot_loss[loss=0.2025, simple_loss=0.2832, pruned_loss=0.06092, over 3832855.17 frames. ], batch size: 49, lr: 3.28e-03, grad_scale: 4.0 2023-04-03 04:47:55,776 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=167785.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 04:48:06,176 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=167794.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 04:48:43,220 INFO [train.py:903] (1/4) Epoch 25, batch 3950, loss[loss=0.1837, simple_loss=0.2672, pruned_loss=0.05013, over 19571.00 frames. ], tot_loss[loss=0.2024, simple_loss=0.2829, pruned_loss=0.06095, over 3834705.33 frames. ], batch size: 52, lr: 3.28e-03, grad_scale: 4.0 2023-04-03 04:48:45,713 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 2023-04-03 04:48:56,934 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=167833.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 04:49:01,643 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=167837.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 04:49:13,784 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.09 vs. limit=2.0 2023-04-03 04:49:25,121 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.090e+02 4.848e+02 5.890e+02 7.638e+02 1.655e+03, threshold=1.178e+03, percent-clipped=7.0 2023-04-03 04:49:29,115 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=167858.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 04:49:34,629 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=167862.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 04:49:46,053 INFO [train.py:903] (1/4) Epoch 25, batch 4000, loss[loss=0.1997, simple_loss=0.2906, pruned_loss=0.05444, over 19701.00 frames. ], tot_loss[loss=0.2018, simple_loss=0.2825, pruned_loss=0.06054, over 3843194.68 frames. ], batch size: 59, lr: 3.28e-03, grad_scale: 8.0 2023-04-03 04:50:32,172 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1_rvb from training. Duration: 0.7545625 2023-04-03 04:50:33,704 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=167909.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 04:50:50,808 INFO [train.py:903] (1/4) Epoch 25, batch 4050, loss[loss=0.1708, simple_loss=0.2518, pruned_loss=0.04489, over 19391.00 frames. ], tot_loss[loss=0.2021, simple_loss=0.283, pruned_loss=0.06058, over 3847662.63 frames. ], batch size: 47, lr: 3.28e-03, grad_scale: 4.0 2023-04-03 04:51:32,271 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.281e+02 5.057e+02 6.529e+02 8.113e+02 1.821e+03, threshold=1.306e+03, percent-clipped=7.0 2023-04-03 04:51:52,927 INFO [train.py:903] (1/4) Epoch 25, batch 4100, loss[loss=0.2177, simple_loss=0.2968, pruned_loss=0.06925, over 19308.00 frames. ], tot_loss[loss=0.2024, simple_loss=0.2833, pruned_loss=0.06077, over 3850401.50 frames. ], batch size: 44, lr: 3.28e-03, grad_scale: 4.0 2023-04-03 04:52:24,783 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.43 vs. limit=2.0 2023-04-03 04:52:27,605 WARNING [train.py:1073] (1/4) Exclude cut with ID 497-129325-0061-9566_sp1.1_rvb from training. Duration: 0.97725 2023-04-03 04:52:56,842 INFO [train.py:903] (1/4) Epoch 25, batch 4150, loss[loss=0.2045, simple_loss=0.2816, pruned_loss=0.06369, over 19682.00 frames. ], tot_loss[loss=0.2017, simple_loss=0.283, pruned_loss=0.06021, over 3856304.89 frames. ], batch size: 53, lr: 3.28e-03, grad_scale: 4.0 2023-04-03 04:53:36,895 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=168053.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 04:53:39,909 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.124e+02 5.393e+02 6.582e+02 8.080e+02 1.683e+03, threshold=1.316e+03, percent-clipped=2.0 2023-04-03 04:53:59,622 INFO [train.py:903] (1/4) Epoch 25, batch 4200, loss[loss=0.1963, simple_loss=0.2766, pruned_loss=0.05794, over 19793.00 frames. ], tot_loss[loss=0.2028, simple_loss=0.284, pruned_loss=0.06083, over 3844315.25 frames. ], batch size: 49, lr: 3.28e-03, grad_scale: 4.0 2023-04-03 04:54:01,966 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 2023-04-03 04:54:10,942 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=168081.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 04:54:18,321 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 2023-04-03 04:55:03,246 INFO [train.py:903] (1/4) Epoch 25, batch 4250, loss[loss=0.1912, simple_loss=0.2871, pruned_loss=0.04766, over 19615.00 frames. ], tot_loss[loss=0.2025, simple_loss=0.2837, pruned_loss=0.06069, over 3833640.48 frames. ], batch size: 57, lr: 3.27e-03, grad_scale: 4.0 2023-04-03 04:55:13,195 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=168129.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 04:55:17,964 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9_rvb from training. Duration: 29.816625 2023-04-03 04:55:20,782 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.47 vs. limit=5.0 2023-04-03 04:55:29,581 WARNING [train.py:1073] (1/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 2023-04-03 04:55:46,757 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.141e+02 4.961e+02 6.439e+02 7.740e+02 2.119e+03, threshold=1.288e+03, percent-clipped=3.0 2023-04-03 04:55:59,646 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=168165.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 04:56:07,294 INFO [train.py:903] (1/4) Epoch 25, batch 4300, loss[loss=0.1875, simple_loss=0.264, pruned_loss=0.05552, over 19732.00 frames. ], tot_loss[loss=0.2042, simple_loss=0.2853, pruned_loss=0.06151, over 3838108.63 frames. ], batch size: 51, lr: 3.27e-03, grad_scale: 4.0 2023-04-03 04:56:15,588 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=1.97 vs. limit=5.0 2023-04-03 04:56:29,218 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.5679, 1.1439, 1.4625, 1.4498, 2.9593, 1.1762, 2.5341, 3.5316], device='cuda:1'), covar=tensor([0.0692, 0.3843, 0.3367, 0.2302, 0.1172, 0.2981, 0.1279, 0.0353], device='cuda:1'), in_proj_covar=tensor([0.0419, 0.0375, 0.0395, 0.0351, 0.0379, 0.0354, 0.0393, 0.0413], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 04:56:30,430 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=168190.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 04:56:37,473 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=168196.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 04:56:42,773 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.93 vs. limit=5.0 2023-04-03 04:56:57,665 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.42 vs. limit=2.0 2023-04-03 04:57:00,302 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 2023-04-03 04:57:10,764 INFO [train.py:903] (1/4) Epoch 25, batch 4350, loss[loss=0.2006, simple_loss=0.2839, pruned_loss=0.05859, over 19763.00 frames. ], tot_loss[loss=0.2046, simple_loss=0.2856, pruned_loss=0.06183, over 3835015.03 frames. ], batch size: 54, lr: 3.27e-03, grad_scale: 4.0 2023-04-03 04:57:38,627 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=168244.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 04:57:47,857 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=168251.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 04:57:49,660 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 2023-04-03 04:57:53,358 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.429e+02 4.774e+02 5.735e+02 7.211e+02 1.236e+03, threshold=1.147e+03, percent-clipped=0.0 2023-04-03 04:58:13,423 INFO [train.py:903] (1/4) Epoch 25, batch 4400, loss[loss=0.1877, simple_loss=0.2749, pruned_loss=0.05029, over 19501.00 frames. ], tot_loss[loss=0.2042, simple_loss=0.2848, pruned_loss=0.06183, over 3838120.71 frames. ], batch size: 64, lr: 3.27e-03, grad_scale: 8.0 2023-04-03 04:58:40,135 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_rvb from training. Duration: 25.285 2023-04-03 04:58:50,363 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 2023-04-03 04:59:16,254 INFO [train.py:903] (1/4) Epoch 25, batch 4450, loss[loss=0.1759, simple_loss=0.2483, pruned_loss=0.05173, over 19736.00 frames. ], tot_loss[loss=0.203, simple_loss=0.2839, pruned_loss=0.06101, over 3844124.03 frames. ], batch size: 46, lr: 3.27e-03, grad_scale: 8.0 2023-04-03 04:59:59,615 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.939e+02 4.609e+02 5.804e+02 7.720e+02 1.927e+03, threshold=1.161e+03, percent-clipped=7.0 2023-04-03 05:00:17,544 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.85 vs. limit=2.0 2023-04-03 05:00:20,127 INFO [train.py:903] (1/4) Epoch 25, batch 4500, loss[loss=0.209, simple_loss=0.294, pruned_loss=0.06198, over 19751.00 frames. ], tot_loss[loss=0.2039, simple_loss=0.285, pruned_loss=0.06136, over 3836774.33 frames. ], batch size: 63, lr: 3.27e-03, grad_scale: 8.0 2023-04-03 05:00:44,722 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 2023-04-03 05:00:52,068 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=168397.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 05:01:23,378 INFO [train.py:903] (1/4) Epoch 25, batch 4550, loss[loss=0.1886, simple_loss=0.2694, pruned_loss=0.05393, over 19485.00 frames. ], tot_loss[loss=0.2029, simple_loss=0.2842, pruned_loss=0.06078, over 3846288.42 frames. ], batch size: 49, lr: 3.27e-03, grad_scale: 8.0 2023-04-03 05:01:34,645 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9_rvb from training. Duration: 28.72225 2023-04-03 05:01:59,985 WARNING [train.py:1073] (1/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 2023-04-03 05:02:02,740 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=168452.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 05:02:08,836 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.350e+02 4.803e+02 5.692e+02 6.651e+02 1.392e+03, threshold=1.138e+03, percent-clipped=3.0 2023-04-03 05:02:15,291 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1418, 1.3056, 1.6998, 0.9545, 2.3241, 3.0869, 2.7825, 3.2762], device='cuda:1'), covar=tensor([0.1521, 0.3898, 0.3296, 0.2756, 0.0628, 0.0213, 0.0260, 0.0301], device='cuda:1'), in_proj_covar=tensor([0.0273, 0.0327, 0.0357, 0.0267, 0.0248, 0.0191, 0.0217, 0.0266], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-03 05:02:27,737 INFO [train.py:903] (1/4) Epoch 25, batch 4600, loss[loss=0.2067, simple_loss=0.2915, pruned_loss=0.06101, over 18303.00 frames. ], tot_loss[loss=0.2029, simple_loss=0.2847, pruned_loss=0.06058, over 3842900.47 frames. ], batch size: 83, lr: 3.27e-03, grad_scale: 4.0 2023-04-03 05:02:35,462 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=168477.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 05:02:59,881 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.3820, 4.0138, 2.6330, 3.4945, 0.8875, 3.9510, 3.8441, 3.9177], device='cuda:1'), covar=tensor([0.0668, 0.0985, 0.1934, 0.0931, 0.4004, 0.0756, 0.0911, 0.1001], device='cuda:1'), in_proj_covar=tensor([0.0517, 0.0420, 0.0504, 0.0353, 0.0405, 0.0447, 0.0441, 0.0471], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 05:03:04,788 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=168500.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 05:03:20,621 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=168512.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 05:03:27,496 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.14 vs. limit=2.0 2023-04-03 05:03:31,423 INFO [train.py:903] (1/4) Epoch 25, batch 4650, loss[loss=0.2293, simple_loss=0.3071, pruned_loss=0.07572, over 19751.00 frames. ], tot_loss[loss=0.202, simple_loss=0.2838, pruned_loss=0.06007, over 3837931.25 frames. ], batch size: 63, lr: 3.27e-03, grad_scale: 4.0 2023-04-03 05:03:35,306 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=168525.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 05:03:50,730 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_rvb from training. Duration: 0.92 2023-04-03 05:04:02,131 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_rvb from training. Duration: 0.83 2023-04-03 05:04:02,404 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.9841, 2.1243, 2.2306, 1.9903, 3.6560, 1.6814, 2.9994, 3.7821], device='cuda:1'), covar=tensor([0.0558, 0.2301, 0.2260, 0.1807, 0.0640, 0.2302, 0.1739, 0.0298], device='cuda:1'), in_proj_covar=tensor([0.0420, 0.0373, 0.0394, 0.0351, 0.0378, 0.0352, 0.0393, 0.0414], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 05:04:16,034 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.438e+02 5.060e+02 6.015e+02 7.632e+02 1.453e+03, threshold=1.203e+03, percent-clipped=3.0 2023-04-03 05:04:34,761 INFO [train.py:903] (1/4) Epoch 25, batch 4700, loss[loss=0.226, simple_loss=0.3034, pruned_loss=0.07433, over 12821.00 frames. ], tot_loss[loss=0.2026, simple_loss=0.2842, pruned_loss=0.06046, over 3824345.65 frames. ], batch size: 138, lr: 3.27e-03, grad_scale: 4.0 2023-04-03 05:04:57,646 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9_rvb from training. Duration: 0.92225 2023-04-03 05:05:04,386 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=168595.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 05:05:38,974 INFO [train.py:903] (1/4) Epoch 25, batch 4750, loss[loss=0.187, simple_loss=0.2731, pruned_loss=0.05045, over 19837.00 frames. ], tot_loss[loss=0.2034, simple_loss=0.2847, pruned_loss=0.06103, over 3822673.75 frames. ], batch size: 52, lr: 3.27e-03, grad_scale: 4.0 2023-04-03 05:06:06,418 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.08 vs. limit=2.0 2023-04-03 05:06:09,939 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 2023-04-03 05:06:22,551 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.376e+02 5.029e+02 6.051e+02 7.124e+02 1.309e+03, threshold=1.210e+03, percent-clipped=1.0 2023-04-03 05:06:40,981 INFO [train.py:903] (1/4) Epoch 25, batch 4800, loss[loss=0.2718, simple_loss=0.3354, pruned_loss=0.1041, over 12776.00 frames. ], tot_loss[loss=0.2042, simple_loss=0.2855, pruned_loss=0.06148, over 3806108.58 frames. ], batch size: 136, lr: 3.27e-03, grad_scale: 8.0 2023-04-03 05:06:59,395 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0583, 1.8005, 1.6648, 1.9907, 1.8714, 1.7156, 1.5397, 1.9269], device='cuda:1'), covar=tensor([0.1043, 0.1270, 0.1462, 0.1001, 0.1152, 0.0571, 0.1536, 0.0757], device='cuda:1'), in_proj_covar=tensor([0.0276, 0.0356, 0.0315, 0.0253, 0.0303, 0.0254, 0.0314, 0.0260], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 05:07:29,195 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=168710.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 05:07:43,821 INFO [train.py:903] (1/4) Epoch 25, batch 4850, loss[loss=0.1997, simple_loss=0.287, pruned_loss=0.05624, over 19761.00 frames. ], tot_loss[loss=0.2035, simple_loss=0.2849, pruned_loss=0.06111, over 3805375.08 frames. ], batch size: 63, lr: 3.27e-03, grad_scale: 8.0 2023-04-03 05:08:10,303 WARNING [train.py:1073] (1/4) Exclude cut with ID 774-127930-0014-48411_sp1.1_rvb from training. Duration: 0.95 2023-04-03 05:08:29,303 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.525e+02 5.536e+02 6.758e+02 9.275e+02 1.787e+03, threshold=1.352e+03, percent-clipped=12.0 2023-04-03 05:08:30,524 WARNING [train.py:1073] (1/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1_rvb from training. Duration: 0.9409375 2023-04-03 05:08:36,436 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 2023-04-03 05:08:36,485 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 2023-04-03 05:08:43,795 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=168768.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 05:08:47,079 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1_rvb from training. Duration: 25.3818125 2023-04-03 05:08:48,232 INFO [train.py:903] (1/4) Epoch 25, batch 4900, loss[loss=0.1676, simple_loss=0.245, pruned_loss=0.04508, over 19753.00 frames. ], tot_loss[loss=0.2032, simple_loss=0.2844, pruned_loss=0.06098, over 3794679.44 frames. ], batch size: 46, lr: 3.27e-03, grad_scale: 8.0 2023-04-03 05:09:06,553 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 2023-04-03 05:09:10,239 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9468, 1.2749, 1.6943, 2.9155, 2.0807, 1.6493, 2.1991, 1.6729], device='cuda:1'), covar=tensor([0.1053, 0.1754, 0.1316, 0.0885, 0.1126, 0.1437, 0.1313, 0.1043], device='cuda:1'), in_proj_covar=tensor([0.0212, 0.0223, 0.0226, 0.0238, 0.0226, 0.0213, 0.0188, 0.0205], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-04-03 05:09:15,962 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=168793.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 05:09:52,873 INFO [train.py:903] (1/4) Epoch 25, batch 4950, loss[loss=0.1947, simple_loss=0.2743, pruned_loss=0.05761, over 19780.00 frames. ], tot_loss[loss=0.2036, simple_loss=0.2844, pruned_loss=0.06143, over 3796986.41 frames. ], batch size: 54, lr: 3.27e-03, grad_scale: 8.0 2023-04-03 05:10:04,496 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 2023-04-03 05:10:30,186 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 2023-04-03 05:10:36,919 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.098e+02 4.693e+02 5.645e+02 7.417e+02 1.662e+03, threshold=1.129e+03, percent-clipped=1.0 2023-04-03 05:10:49,018 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6321, 1.7559, 2.1076, 1.8449, 3.1193, 2.6125, 3.2307, 1.6745], device='cuda:1'), covar=tensor([0.2618, 0.4502, 0.2778, 0.2094, 0.1650, 0.2306, 0.1750, 0.4569], device='cuda:1'), in_proj_covar=tensor([0.0548, 0.0666, 0.0739, 0.0501, 0.0630, 0.0544, 0.0670, 0.0567], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-03 05:10:50,517 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 2023-04-03 05:10:55,825 INFO [train.py:903] (1/4) Epoch 25, batch 5000, loss[loss=0.1873, simple_loss=0.2742, pruned_loss=0.05022, over 18215.00 frames. ], tot_loss[loss=0.2021, simple_loss=0.2832, pruned_loss=0.06052, over 3804555.51 frames. ], batch size: 84, lr: 3.27e-03, grad_scale: 8.0 2023-04-03 05:11:02,558 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_rvb from training. Duration: 27.14 2023-04-03 05:11:13,717 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 2023-04-03 05:11:58,384 INFO [train.py:903] (1/4) Epoch 25, batch 5050, loss[loss=0.2161, simple_loss=0.2772, pruned_loss=0.07747, over 19365.00 frames. ], tot_loss[loss=0.2019, simple_loss=0.2826, pruned_loss=0.06055, over 3810778.31 frames. ], batch size: 47, lr: 3.27e-03, grad_scale: 8.0 2023-04-03 05:12:06,806 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.2639, 2.3023, 2.5518, 2.9864, 2.2851, 2.8406, 2.5795, 2.2832], device='cuda:1'), covar=tensor([0.4371, 0.4267, 0.1886, 0.2822, 0.4640, 0.2342, 0.4910, 0.3488], device='cuda:1'), in_proj_covar=tensor([0.0920, 0.0994, 0.0731, 0.0941, 0.0896, 0.0833, 0.0851, 0.0796], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-03 05:12:18,628 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9317, 2.0201, 2.3122, 2.4609, 1.8944, 2.3773, 2.2945, 2.0583], device='cuda:1'), covar=tensor([0.4070, 0.3820, 0.1846, 0.2602, 0.4244, 0.2265, 0.4891, 0.3416], device='cuda:1'), in_proj_covar=tensor([0.0920, 0.0994, 0.0731, 0.0942, 0.0897, 0.0833, 0.0851, 0.0796], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-03 05:12:33,918 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9_rvb from training. Duration: 26.62775 2023-04-03 05:12:41,885 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.044e+02 4.713e+02 5.534e+02 7.099e+02 1.364e+03, threshold=1.107e+03, percent-clipped=2.0 2023-04-03 05:12:55,417 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=168966.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 05:13:02,129 INFO [train.py:903] (1/4) Epoch 25, batch 5100, loss[loss=0.1844, simple_loss=0.2613, pruned_loss=0.05381, over 19844.00 frames. ], tot_loss[loss=0.2013, simple_loss=0.2823, pruned_loss=0.06011, over 3810823.49 frames. ], batch size: 52, lr: 3.27e-03, grad_scale: 8.0 2023-04-03 05:13:11,308 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 2023-04-03 05:13:14,755 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_rvb from training. Duration: 25.035 2023-04-03 05:13:19,307 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=168985.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 05:13:20,102 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_rvb from training. Duration: 27.92 2023-04-03 05:13:21,526 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=168987.0, num_to_drop=1, layers_to_drop={1} 2023-04-03 05:13:26,024 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=168991.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 05:13:31,792 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.7959, 1.6378, 1.7036, 2.3249, 1.9118, 2.0803, 2.1193, 1.8693], device='cuda:1'), covar=tensor([0.0797, 0.0887, 0.0960, 0.0667, 0.0817, 0.0719, 0.0811, 0.0642], device='cuda:1'), in_proj_covar=tensor([0.0213, 0.0224, 0.0226, 0.0239, 0.0226, 0.0214, 0.0189, 0.0205], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-04-03 05:14:05,074 INFO [train.py:903] (1/4) Epoch 25, batch 5150, loss[loss=0.1919, simple_loss=0.2852, pruned_loss=0.04933, over 19684.00 frames. ], tot_loss[loss=0.201, simple_loss=0.282, pruned_loss=0.05998, over 3812641.34 frames. ], batch size: 59, lr: 3.27e-03, grad_scale: 8.0 2023-04-03 05:14:16,406 WARNING [train.py:1073] (1/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 2023-04-03 05:14:23,763 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.6332, 4.2364, 2.9211, 3.7499, 1.1702, 4.2046, 4.0584, 4.1698], device='cuda:1'), covar=tensor([0.0626, 0.0868, 0.1788, 0.0802, 0.3725, 0.0678, 0.0933, 0.1253], device='cuda:1'), in_proj_covar=tensor([0.0524, 0.0427, 0.0511, 0.0358, 0.0412, 0.0454, 0.0446, 0.0477], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 05:14:48,108 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.896e+02 5.295e+02 6.704e+02 8.101e+02 2.101e+03, threshold=1.341e+03, percent-clipped=6.0 2023-04-03 05:14:52,541 WARNING [train.py:1073] (1/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1_rvb from training. Duration: 0.9681875 2023-04-03 05:15:08,233 INFO [train.py:903] (1/4) Epoch 25, batch 5200, loss[loss=0.2264, simple_loss=0.3107, pruned_loss=0.07109, over 19622.00 frames. ], tot_loss[loss=0.202, simple_loss=0.2828, pruned_loss=0.06057, over 3789903.05 frames. ], batch size: 57, lr: 3.27e-03, grad_scale: 8.0 2023-04-03 05:15:23,456 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 2023-04-03 05:15:44,163 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=169100.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 05:16:08,636 WARNING [train.py:1073] (1/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 2023-04-03 05:16:13,024 INFO [train.py:903] (1/4) Epoch 25, batch 5250, loss[loss=0.2358, simple_loss=0.3124, pruned_loss=0.07962, over 19654.00 frames. ], tot_loss[loss=0.2027, simple_loss=0.284, pruned_loss=0.06063, over 3802271.04 frames. ], batch size: 60, lr: 3.27e-03, grad_scale: 8.0 2023-04-03 05:16:27,842 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=169134.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 05:16:56,981 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.150e+02 5.211e+02 5.791e+02 7.204e+02 1.532e+03, threshold=1.158e+03, percent-clipped=2.0 2023-04-03 05:17:16,201 INFO [train.py:903] (1/4) Epoch 25, batch 5300, loss[loss=0.1718, simple_loss=0.2651, pruned_loss=0.03925, over 19667.00 frames. ], tot_loss[loss=0.2032, simple_loss=0.2844, pruned_loss=0.06098, over 3816242.26 frames. ], batch size: 55, lr: 3.26e-03, grad_scale: 8.0 2023-04-03 05:17:22,416 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=169177.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 05:17:34,466 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 2023-04-03 05:18:18,931 INFO [train.py:903] (1/4) Epoch 25, batch 5350, loss[loss=0.2073, simple_loss=0.2718, pruned_loss=0.07133, over 19753.00 frames. ], tot_loss[loss=0.2029, simple_loss=0.284, pruned_loss=0.0609, over 3830803.30 frames. ], batch size: 45, lr: 3.26e-03, grad_scale: 8.0 2023-04-03 05:18:56,377 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9_rvb from training. Duration: 26.438875 2023-04-03 05:19:04,350 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.964e+02 4.992e+02 6.506e+02 8.048e+02 1.510e+03, threshold=1.301e+03, percent-clipped=6.0 2023-04-03 05:19:07,176 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.7358, 4.2355, 4.4733, 4.4442, 1.8315, 4.2278, 3.6183, 4.2139], device='cuda:1'), covar=tensor([0.1723, 0.0996, 0.0647, 0.0753, 0.5927, 0.0959, 0.0756, 0.1116], device='cuda:1'), in_proj_covar=tensor([0.0805, 0.0774, 0.0978, 0.0861, 0.0850, 0.0741, 0.0583, 0.0909], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-03 05:19:24,336 INFO [train.py:903] (1/4) Epoch 25, batch 5400, loss[loss=0.2385, simple_loss=0.3164, pruned_loss=0.08033, over 19690.00 frames. ], tot_loss[loss=0.2041, simple_loss=0.2853, pruned_loss=0.06146, over 3817909.12 frames. ], batch size: 59, lr: 3.26e-03, grad_scale: 8.0 2023-04-03 05:19:38,033 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=4.05 vs. limit=5.0 2023-04-03 05:19:38,683 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.3836, 2.3883, 2.2340, 2.5769, 2.3443, 2.1177, 2.2073, 2.3540], device='cuda:1'), covar=tensor([0.0842, 0.1212, 0.1049, 0.0817, 0.1066, 0.0451, 0.1113, 0.0589], device='cuda:1'), in_proj_covar=tensor([0.0275, 0.0355, 0.0313, 0.0251, 0.0302, 0.0254, 0.0314, 0.0259], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 05:20:27,258 INFO [train.py:903] (1/4) Epoch 25, batch 5450, loss[loss=0.201, simple_loss=0.2773, pruned_loss=0.06235, over 19407.00 frames. ], tot_loss[loss=0.2039, simple_loss=0.2852, pruned_loss=0.06135, over 3821839.06 frames. ], batch size: 48, lr: 3.26e-03, grad_scale: 8.0 2023-04-03 05:20:33,628 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.37 vs. limit=5.0 2023-04-03 05:20:36,085 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 2023-04-03 05:20:36,562 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=169329.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 05:20:39,943 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=169331.0, num_to_drop=1, layers_to_drop={1} 2023-04-03 05:20:46,792 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2365, 1.2783, 1.2259, 1.0167, 1.0491, 1.1232, 0.0717, 0.3340], device='cuda:1'), covar=tensor([0.0619, 0.0626, 0.0429, 0.0583, 0.1212, 0.0614, 0.1375, 0.1087], device='cuda:1'), in_proj_covar=tensor([0.0364, 0.0363, 0.0364, 0.0391, 0.0467, 0.0399, 0.0344, 0.0346], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 05:20:50,641 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 2023-04-03 05:21:11,460 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.767e+02 4.467e+02 5.289e+02 7.172e+02 1.661e+03, threshold=1.058e+03, percent-clipped=2.0 2023-04-03 05:21:29,438 INFO [train.py:903] (1/4) Epoch 25, batch 5500, loss[loss=0.2002, simple_loss=0.2838, pruned_loss=0.05833, over 19535.00 frames. ], tot_loss[loss=0.2041, simple_loss=0.285, pruned_loss=0.0616, over 3836513.10 frames. ], batch size: 56, lr: 3.26e-03, grad_scale: 8.0 2023-04-03 05:21:57,703 WARNING [train.py:1073] (1/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9_rvb from training. Duration: 27.47775 2023-04-03 05:22:24,178 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 2023-04-03 05:22:33,001 INFO [train.py:903] (1/4) Epoch 25, batch 5550, loss[loss=0.225, simple_loss=0.3058, pruned_loss=0.07211, over 13436.00 frames. ], tot_loss[loss=0.2047, simple_loss=0.2855, pruned_loss=0.06194, over 3812521.35 frames. ], batch size: 136, lr: 3.26e-03, grad_scale: 8.0 2023-04-03 05:22:43,786 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 2023-04-03 05:23:00,564 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=169444.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 05:23:00,776 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=169444.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 05:23:04,001 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=169446.0, num_to_drop=1, layers_to_drop={1} 2023-04-03 05:23:17,172 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.302e+02 5.192e+02 6.082e+02 7.576e+02 1.216e+03, threshold=1.216e+03, percent-clipped=3.0 2023-04-03 05:23:33,244 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 2023-04-03 05:23:36,938 INFO [train.py:903] (1/4) Epoch 25, batch 5600, loss[loss=0.2098, simple_loss=0.293, pruned_loss=0.06332, over 19479.00 frames. ], tot_loss[loss=0.2043, simple_loss=0.2855, pruned_loss=0.06156, over 3811720.06 frames. ], batch size: 64, lr: 3.26e-03, grad_scale: 8.0 2023-04-03 05:23:44,272 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=169478.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 05:24:38,814 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=169521.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 05:24:39,769 INFO [train.py:903] (1/4) Epoch 25, batch 5650, loss[loss=0.2167, simple_loss=0.2989, pruned_loss=0.06723, over 19684.00 frames. ], tot_loss[loss=0.2034, simple_loss=0.2845, pruned_loss=0.06111, over 3818158.74 frames. ], batch size: 60, lr: 3.26e-03, grad_scale: 8.0 2023-04-03 05:25:24,924 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.766e+02 4.961e+02 6.158e+02 8.288e+02 1.627e+03, threshold=1.232e+03, percent-clipped=5.0 2023-04-03 05:25:27,491 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=169559.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 05:25:30,433 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9_rvb from training. Duration: 28.0944375 2023-04-03 05:25:37,669 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3693, 1.2999, 1.3441, 1.4074, 0.9803, 1.4324, 1.3530, 1.4024], device='cuda:1'), covar=tensor([0.0860, 0.0947, 0.0983, 0.0639, 0.0848, 0.0799, 0.0851, 0.0728], device='cuda:1'), in_proj_covar=tensor([0.0212, 0.0222, 0.0225, 0.0238, 0.0225, 0.0212, 0.0188, 0.0203], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-03 05:25:43,210 INFO [train.py:903] (1/4) Epoch 25, batch 5700, loss[loss=0.2401, simple_loss=0.3225, pruned_loss=0.07887, over 19673.00 frames. ], tot_loss[loss=0.2044, simple_loss=0.2855, pruned_loss=0.0617, over 3816442.60 frames. ], batch size: 55, lr: 3.26e-03, grad_scale: 8.0 2023-04-03 05:26:11,280 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=169593.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 05:26:11,668 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.80 vs. limit=2.0 2023-04-03 05:26:12,794 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=4.49 vs. limit=5.0 2023-04-03 05:26:22,649 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.51 vs. limit=2.0 2023-04-03 05:26:29,605 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0000, 2.0683, 2.2833, 2.7212, 2.0815, 2.5616, 2.2661, 2.0442], device='cuda:1'), covar=tensor([0.4497, 0.3976, 0.1891, 0.2503, 0.4344, 0.2235, 0.5304, 0.3577], device='cuda:1'), in_proj_covar=tensor([0.0920, 0.0995, 0.0730, 0.0942, 0.0898, 0.0833, 0.0853, 0.0795], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-03 05:26:47,653 INFO [train.py:903] (1/4) Epoch 25, batch 5750, loss[loss=0.1864, simple_loss=0.2763, pruned_loss=0.04819, over 19664.00 frames. ], tot_loss[loss=0.2046, simple_loss=0.2858, pruned_loss=0.06173, over 3813078.19 frames. ], batch size: 55, lr: 3.26e-03, grad_scale: 8.0 2023-04-03 05:26:48,806 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9_rvb from training. Duration: 33.038875 2023-04-03 05:26:59,216 WARNING [train.py:1073] (1/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 2023-04-03 05:27:04,031 WARNING [train.py:1073] (1/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 2023-04-03 05:27:06,674 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=169636.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 05:27:09,124 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3060, 1.4308, 2.0747, 1.6076, 3.1648, 4.7566, 4.5556, 5.0709], device='cuda:1'), covar=tensor([0.1701, 0.3912, 0.3274, 0.2349, 0.0598, 0.0188, 0.0183, 0.0194], device='cuda:1'), in_proj_covar=tensor([0.0275, 0.0326, 0.0356, 0.0267, 0.0248, 0.0192, 0.0217, 0.0267], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-03 05:27:10,245 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=169639.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 05:27:32,988 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.159e+02 4.765e+02 5.968e+02 8.362e+02 1.575e+03, threshold=1.194e+03, percent-clipped=5.0 2023-04-03 05:27:52,476 INFO [train.py:903] (1/4) Epoch 25, batch 5800, loss[loss=0.2026, simple_loss=0.2894, pruned_loss=0.0579, over 19401.00 frames. ], tot_loss[loss=0.2048, simple_loss=0.2861, pruned_loss=0.06173, over 3829186.83 frames. ], batch size: 70, lr: 3.26e-03, grad_scale: 8.0 2023-04-03 05:28:20,515 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1836, 1.3026, 1.9135, 1.5338, 2.8993, 4.5907, 4.3907, 4.9083], device='cuda:1'), covar=tensor([0.1759, 0.3934, 0.3401, 0.2405, 0.0715, 0.0208, 0.0181, 0.0234], device='cuda:1'), in_proj_covar=tensor([0.0275, 0.0326, 0.0357, 0.0267, 0.0248, 0.0192, 0.0217, 0.0267], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-03 05:28:27,879 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=169700.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 05:28:30,466 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=169702.0, num_to_drop=1, layers_to_drop={1} 2023-04-03 05:28:55,625 INFO [train.py:903] (1/4) Epoch 25, batch 5850, loss[loss=0.2145, simple_loss=0.3043, pruned_loss=0.06233, over 19297.00 frames. ], tot_loss[loss=0.205, simple_loss=0.2862, pruned_loss=0.06183, over 3812530.37 frames. ], batch size: 66, lr: 3.26e-03, grad_scale: 8.0 2023-04-03 05:29:00,327 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=169725.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 05:29:02,815 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=169727.0, num_to_drop=1, layers_to_drop={1} 2023-04-03 05:29:08,752 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.7011, 1.7847, 2.0540, 2.0563, 3.3412, 2.8799, 3.4936, 1.6671], device='cuda:1'), covar=tensor([0.2492, 0.4537, 0.2998, 0.1924, 0.1526, 0.1998, 0.1698, 0.4581], device='cuda:1'), in_proj_covar=tensor([0.0548, 0.0663, 0.0736, 0.0501, 0.0630, 0.0542, 0.0668, 0.0565], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 05:29:28,428 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=169747.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 05:29:41,209 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.382e+02 5.003e+02 6.037e+02 8.413e+02 1.989e+03, threshold=1.207e+03, percent-clipped=6.0 2023-04-03 05:29:51,800 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=169765.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 05:29:59,816 INFO [train.py:903] (1/4) Epoch 25, batch 5900, loss[loss=0.1612, simple_loss=0.2361, pruned_loss=0.04312, over 19740.00 frames. ], tot_loss[loss=0.2044, simple_loss=0.2857, pruned_loss=0.06162, over 3803667.86 frames. ], batch size: 46, lr: 3.26e-03, grad_scale: 8.0 2023-04-03 05:30:04,519 WARNING [train.py:1073] (1/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 2023-04-03 05:30:27,865 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9_rvb from training. Duration: 27.511125 2023-04-03 05:30:41,016 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=169804.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 05:30:56,655 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=169815.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 05:31:04,220 INFO [train.py:903] (1/4) Epoch 25, batch 5950, loss[loss=0.2141, simple_loss=0.2956, pruned_loss=0.06632, over 17420.00 frames. ], tot_loss[loss=0.2038, simple_loss=0.2851, pruned_loss=0.06126, over 3812211.01 frames. ], batch size: 101, lr: 3.26e-03, grad_scale: 8.0 2023-04-03 05:31:28,366 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=169840.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 05:31:39,983 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=169849.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 05:31:49,610 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.314e+02 4.837e+02 6.128e+02 7.395e+02 1.765e+03, threshold=1.226e+03, percent-clipped=3.0 2023-04-03 05:32:09,287 INFO [train.py:903] (1/4) Epoch 25, batch 6000, loss[loss=0.2189, simple_loss=0.2991, pruned_loss=0.06931, over 18189.00 frames. ], tot_loss[loss=0.2021, simple_loss=0.2834, pruned_loss=0.06037, over 3811439.75 frames. ], batch size: 83, lr: 3.26e-03, grad_scale: 8.0 2023-04-03 05:32:09,287 INFO [train.py:928] (1/4) Computing validation loss 2023-04-03 05:32:21,943 INFO [train.py:937] (1/4) Epoch 25, validation: loss=0.1675, simple_loss=0.2674, pruned_loss=0.03383, over 944034.00 frames. 2023-04-03 05:32:21,945 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18821MB 2023-04-03 05:32:25,781 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=169874.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 05:32:44,059 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.52 vs. limit=2.0 2023-04-03 05:32:48,283 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=169892.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 05:33:20,601 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=169917.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 05:33:21,670 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.1325, 1.2905, 1.5708, 1.2861, 2.7671, 1.1291, 2.1795, 3.0801], device='cuda:1'), covar=tensor([0.0567, 0.2885, 0.2756, 0.2001, 0.0700, 0.2406, 0.1246, 0.0335], device='cuda:1'), in_proj_covar=tensor([0.0414, 0.0370, 0.0390, 0.0350, 0.0376, 0.0351, 0.0389, 0.0409], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 05:33:26,747 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 2023-04-03 05:33:26,974 INFO [train.py:903] (1/4) Epoch 25, batch 6050, loss[loss=0.2015, simple_loss=0.2817, pruned_loss=0.06065, over 19669.00 frames. ], tot_loss[loss=0.203, simple_loss=0.284, pruned_loss=0.06101, over 3822527.54 frames. ], batch size: 58, lr: 3.26e-03, grad_scale: 4.0 2023-04-03 05:34:00,504 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.1084, 1.9933, 1.8708, 1.6703, 1.5954, 1.7017, 0.5499, 1.0642], device='cuda:1'), covar=tensor([0.0672, 0.0682, 0.0505, 0.0844, 0.1192, 0.0994, 0.1426, 0.1154], device='cuda:1'), in_proj_covar=tensor([0.0362, 0.0360, 0.0362, 0.0387, 0.0464, 0.0396, 0.0342, 0.0344], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 05:34:12,786 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.356e+02 4.971e+02 6.200e+02 7.955e+02 1.563e+03, threshold=1.240e+03, percent-clipped=4.0 2023-04-03 05:34:16,767 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4477, 1.2932, 1.3150, 1.9607, 1.4605, 1.5769, 1.7110, 1.4308], device='cuda:1'), covar=tensor([0.1016, 0.1216, 0.1245, 0.0793, 0.0970, 0.1000, 0.1017, 0.0916], device='cuda:1'), in_proj_covar=tensor([0.0211, 0.0222, 0.0224, 0.0238, 0.0225, 0.0213, 0.0188, 0.0203], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-04-03 05:34:30,155 INFO [train.py:903] (1/4) Epoch 25, batch 6100, loss[loss=0.1945, simple_loss=0.2665, pruned_loss=0.06128, over 19735.00 frames. ], tot_loss[loss=0.2026, simple_loss=0.2837, pruned_loss=0.06075, over 3826300.49 frames. ], batch size: 51, lr: 3.26e-03, grad_scale: 4.0 2023-04-03 05:34:44,073 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=169983.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 05:35:07,103 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.9697, 4.5158, 2.8397, 3.9393, 0.6231, 4.5489, 4.2907, 4.4723], device='cuda:1'), covar=tensor([0.0519, 0.0921, 0.1878, 0.0778, 0.4541, 0.0566, 0.0820, 0.0987], device='cuda:1'), in_proj_covar=tensor([0.0517, 0.0420, 0.0503, 0.0353, 0.0406, 0.0444, 0.0441, 0.0470], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 05:35:20,547 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.8048, 1.9267, 2.1853, 2.3422, 1.7649, 2.2105, 2.2060, 2.0150], device='cuda:1'), covar=tensor([0.4266, 0.3834, 0.2005, 0.2483, 0.4180, 0.2333, 0.4849, 0.3412], device='cuda:1'), in_proj_covar=tensor([0.0919, 0.0994, 0.0731, 0.0942, 0.0898, 0.0832, 0.0853, 0.0796], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-03 05:35:35,275 INFO [train.py:903] (1/4) Epoch 25, batch 6150, loss[loss=0.1656, simple_loss=0.2421, pruned_loss=0.04449, over 19089.00 frames. ], tot_loss[loss=0.2013, simple_loss=0.2825, pruned_loss=0.06009, over 3837283.27 frames. ], batch size: 42, lr: 3.26e-03, grad_scale: 4.0 2023-04-03 05:36:07,480 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9_rvb from training. Duration: 31.02225 2023-04-03 05:36:18,154 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.8871, 2.6270, 2.4700, 2.7896, 2.6518, 2.4905, 2.3666, 2.8447], device='cuda:1'), covar=tensor([0.0893, 0.1590, 0.1363, 0.1123, 0.1370, 0.0499, 0.1410, 0.0606], device='cuda:1'), in_proj_covar=tensor([0.0273, 0.0353, 0.0311, 0.0251, 0.0300, 0.0252, 0.0312, 0.0260], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 05:36:22,371 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.332e+02 4.967e+02 5.743e+02 7.363e+02 2.013e+03, threshold=1.149e+03, percent-clipped=2.0 2023-04-03 05:36:28,744 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=170063.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 05:36:40,353 INFO [train.py:903] (1/4) Epoch 25, batch 6200, loss[loss=0.2317, simple_loss=0.3114, pruned_loss=0.07597, over 18039.00 frames. ], tot_loss[loss=0.2024, simple_loss=0.2835, pruned_loss=0.0607, over 3837884.76 frames. ], batch size: 83, lr: 3.26e-03, grad_scale: 4.0 2023-04-03 05:37:04,431 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=170091.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 05:37:13,786 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=170098.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 05:37:26,466 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=170109.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 05:37:43,644 INFO [train.py:903] (1/4) Epoch 25, batch 6250, loss[loss=0.1751, simple_loss=0.2671, pruned_loss=0.04151, over 19671.00 frames. ], tot_loss[loss=0.202, simple_loss=0.2831, pruned_loss=0.06049, over 3840707.56 frames. ], batch size: 59, lr: 3.26e-03, grad_scale: 4.0 2023-04-03 05:38:16,020 WARNING [train.py:1073] (1/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9_rvb from training. Duration: 25.988875 2023-04-03 05:38:16,212 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=170148.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 05:38:29,526 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.163e+02 5.199e+02 6.170e+02 7.836e+02 1.706e+03, threshold=1.234e+03, percent-clipped=7.0 2023-04-03 05:38:47,597 INFO [train.py:903] (1/4) Epoch 25, batch 6300, loss[loss=0.2614, simple_loss=0.3251, pruned_loss=0.09883, over 13630.00 frames. ], tot_loss[loss=0.2023, simple_loss=0.2832, pruned_loss=0.06071, over 3835190.13 frames. ], batch size: 136, lr: 3.26e-03, grad_scale: 4.0 2023-04-03 05:39:32,202 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=170206.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 05:39:51,010 INFO [train.py:903] (1/4) Epoch 25, batch 6350, loss[loss=0.1963, simple_loss=0.272, pruned_loss=0.06029, over 19485.00 frames. ], tot_loss[loss=0.2025, simple_loss=0.2836, pruned_loss=0.06075, over 3844728.67 frames. ], batch size: 49, lr: 3.25e-03, grad_scale: 4.0 2023-04-03 05:39:53,842 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=170224.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 05:40:36,695 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.666e+02 4.417e+02 5.475e+02 7.140e+02 1.571e+03, threshold=1.095e+03, percent-clipped=3.0 2023-04-03 05:40:42,896 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=170263.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 05:40:54,621 INFO [train.py:903] (1/4) Epoch 25, batch 6400, loss[loss=0.203, simple_loss=0.2803, pruned_loss=0.06285, over 19794.00 frames. ], tot_loss[loss=0.2026, simple_loss=0.284, pruned_loss=0.06058, over 3846666.31 frames. ], batch size: 49, lr: 3.25e-03, grad_scale: 8.0 2023-04-03 05:41:40,119 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.66 vs. limit=2.0 2023-04-03 05:41:46,479 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1306, 1.2916, 1.6679, 1.2535, 2.7488, 3.5947, 3.2266, 3.7904], device='cuda:1'), covar=tensor([0.1699, 0.3955, 0.3471, 0.2635, 0.0574, 0.0198, 0.0220, 0.0261], device='cuda:1'), in_proj_covar=tensor([0.0278, 0.0330, 0.0360, 0.0269, 0.0251, 0.0193, 0.0218, 0.0270], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-03 05:41:59,021 INFO [train.py:903] (1/4) Epoch 25, batch 6450, loss[loss=0.2041, simple_loss=0.2974, pruned_loss=0.05542, over 19685.00 frames. ], tot_loss[loss=0.2015, simple_loss=0.283, pruned_loss=0.05999, over 3840021.20 frames. ], batch size: 60, lr: 3.25e-03, grad_scale: 8.0 2023-04-03 05:42:40,350 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=170354.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 05:42:44,564 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.061e+02 4.688e+02 6.468e+02 8.318e+02 2.178e+03, threshold=1.294e+03, percent-clipped=13.0 2023-04-03 05:42:45,723 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9_rvb from training. Duration: 28.638875 2023-04-03 05:43:03,216 INFO [train.py:903] (1/4) Epoch 25, batch 6500, loss[loss=0.2196, simple_loss=0.2887, pruned_loss=0.07522, over 19392.00 frames. ], tot_loss[loss=0.2021, simple_loss=0.2832, pruned_loss=0.06048, over 3842542.03 frames. ], batch size: 47, lr: 3.25e-03, grad_scale: 8.0 2023-04-03 05:43:07,934 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1_rvb from training. Duration: 0.836375 2023-04-03 05:43:12,742 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=170379.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 05:43:48,128 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=170407.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 05:44:06,265 INFO [train.py:903] (1/4) Epoch 25, batch 6550, loss[loss=0.2035, simple_loss=0.2787, pruned_loss=0.06413, over 19728.00 frames. ], tot_loss[loss=0.2035, simple_loss=0.2844, pruned_loss=0.06131, over 3846582.29 frames. ], batch size: 51, lr: 3.25e-03, grad_scale: 8.0 2023-04-03 05:44:52,745 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.983e+02 4.881e+02 6.158e+02 7.799e+02 1.457e+03, threshold=1.232e+03, percent-clipped=1.0 2023-04-03 05:44:57,807 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=170462.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 05:45:09,732 INFO [train.py:903] (1/4) Epoch 25, batch 6600, loss[loss=0.1749, simple_loss=0.2585, pruned_loss=0.0457, over 19751.00 frames. ], tot_loss[loss=0.2034, simple_loss=0.2842, pruned_loss=0.06135, over 3831631.90 frames. ], batch size: 51, lr: 3.25e-03, grad_scale: 8.0 2023-04-03 05:45:19,102 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=170479.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 05:45:20,451 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=170480.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 05:45:28,700 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=170487.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 05:45:51,415 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=170505.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 05:46:09,045 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=170519.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 05:46:13,015 INFO [train.py:903] (1/4) Epoch 25, batch 6650, loss[loss=0.218, simple_loss=0.2981, pruned_loss=0.06891, over 19683.00 frames. ], tot_loss[loss=0.2034, simple_loss=0.2844, pruned_loss=0.0612, over 3835619.35 frames. ], batch size: 60, lr: 3.25e-03, grad_scale: 8.0 2023-04-03 05:46:13,384 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=170522.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 05:46:41,290 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=170544.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 05:46:58,409 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.718e+02 4.827e+02 5.978e+02 7.974e+02 2.215e+03, threshold=1.196e+03, percent-clipped=6.0 2023-04-03 05:47:16,988 INFO [train.py:903] (1/4) Epoch 25, batch 6700, loss[loss=0.2148, simple_loss=0.2943, pruned_loss=0.06762, over 19608.00 frames. ], tot_loss[loss=0.2021, simple_loss=0.283, pruned_loss=0.06056, over 3830550.83 frames. ], batch size: 61, lr: 3.25e-03, grad_scale: 8.0 2023-04-03 05:47:36,685 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=170587.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 05:48:14,054 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 2023-04-03 05:48:17,850 INFO [train.py:903] (1/4) Epoch 25, batch 6750, loss[loss=0.1874, simple_loss=0.2722, pruned_loss=0.05126, over 19663.00 frames. ], tot_loss[loss=0.2032, simple_loss=0.2838, pruned_loss=0.06127, over 3824848.29 frames. ], batch size: 53, lr: 3.25e-03, grad_scale: 8.0 2023-04-03 05:48:59,806 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.047e+02 4.526e+02 5.801e+02 6.899e+02 1.670e+03, threshold=1.160e+03, percent-clipped=2.0 2023-04-03 05:49:15,798 INFO [train.py:903] (1/4) Epoch 25, batch 6800, loss[loss=0.2242, simple_loss=0.3128, pruned_loss=0.06779, over 19299.00 frames. ], tot_loss[loss=0.2026, simple_loss=0.2837, pruned_loss=0.06072, over 3822854.67 frames. ], batch size: 66, lr: 3.25e-03, grad_scale: 8.0 2023-04-03 05:50:02,603 WARNING [train.py:1073] (1/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 2023-04-03 05:50:03,063 WARNING [train.py:1073] (1/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 2023-04-03 05:50:07,362 INFO [train.py:903] (1/4) Epoch 26, batch 0, loss[loss=0.1801, simple_loss=0.2603, pruned_loss=0.04994, over 19374.00 frames. ], tot_loss[loss=0.1801, simple_loss=0.2603, pruned_loss=0.04994, over 19374.00 frames. ], batch size: 48, lr: 3.19e-03, grad_scale: 8.0 2023-04-03 05:50:07,362 INFO [train.py:928] (1/4) Computing validation loss 2023-04-03 05:50:19,302 INFO [train.py:937] (1/4) Epoch 26, validation: loss=0.1673, simple_loss=0.2675, pruned_loss=0.03355, over 944034.00 frames. 2023-04-03 05:50:19,303 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18821MB 2023-04-03 05:50:26,498 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.5884, 2.5801, 2.1485, 2.6053, 2.6123, 2.2960, 2.1129, 2.5499], device='cuda:1'), covar=tensor([0.1036, 0.1631, 0.1560, 0.1107, 0.1285, 0.0546, 0.1527, 0.0696], device='cuda:1'), in_proj_covar=tensor([0.0274, 0.0355, 0.0314, 0.0253, 0.0303, 0.0254, 0.0314, 0.0261], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 05:50:32,194 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 2023-04-03 05:51:20,626 INFO [train.py:903] (1/4) Epoch 26, batch 50, loss[loss=0.2164, simple_loss=0.2969, pruned_loss=0.06798, over 18030.00 frames. ], tot_loss[loss=0.202, simple_loss=0.2827, pruned_loss=0.06063, over 871316.83 frames. ], batch size: 83, lr: 3.19e-03, grad_scale: 8.0 2023-04-03 05:51:30,196 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.374e+02 5.268e+02 6.216e+02 7.845e+02 1.668e+03, threshold=1.243e+03, percent-clipped=9.0 2023-04-03 05:51:54,944 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=170778.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 05:51:55,738 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1_rvb from training. Duration: 27.0318125 2023-04-03 05:52:21,933 INFO [train.py:903] (1/4) Epoch 26, batch 100, loss[loss=0.1778, simple_loss=0.2571, pruned_loss=0.04922, over 19760.00 frames. ], tot_loss[loss=0.2005, simple_loss=0.2819, pruned_loss=0.05958, over 1540647.77 frames. ], batch size: 46, lr: 3.18e-03, grad_scale: 4.0 2023-04-03 05:52:26,035 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=170803.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 05:52:32,353 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 2023-04-03 05:52:50,719 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=170823.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 05:53:12,614 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 2023-04-03 05:53:24,753 INFO [train.py:903] (1/4) Epoch 26, batch 150, loss[loss=0.2204, simple_loss=0.3045, pruned_loss=0.06821, over 19771.00 frames. ], tot_loss[loss=0.2025, simple_loss=0.2837, pruned_loss=0.06064, over 2044313.45 frames. ], batch size: 63, lr: 3.18e-03, grad_scale: 4.0 2023-04-03 05:53:36,325 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.111e+02 4.908e+02 6.470e+02 7.917e+02 1.560e+03, threshold=1.294e+03, percent-clipped=6.0 2023-04-03 05:54:20,353 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.1257, 1.9395, 1.7848, 2.0737, 1.7603, 1.8424, 1.6995, 2.0119], device='cuda:1'), covar=tensor([0.0979, 0.1391, 0.1419, 0.0950, 0.1427, 0.0566, 0.1495, 0.0717], device='cuda:1'), in_proj_covar=tensor([0.0272, 0.0352, 0.0312, 0.0252, 0.0302, 0.0253, 0.0312, 0.0259], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 05:54:25,520 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 2023-04-03 05:54:26,701 INFO [train.py:903] (1/4) Epoch 26, batch 200, loss[loss=0.2216, simple_loss=0.3009, pruned_loss=0.07116, over 19436.00 frames. ], tot_loss[loss=0.2037, simple_loss=0.2851, pruned_loss=0.06116, over 2439130.34 frames. ], batch size: 64, lr: 3.18e-03, grad_scale: 4.0 2023-04-03 05:55:05,046 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=170931.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 05:55:14,493 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=170938.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 05:55:24,845 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=170946.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 05:55:30,410 INFO [train.py:903] (1/4) Epoch 26, batch 250, loss[loss=0.2018, simple_loss=0.2859, pruned_loss=0.05883, over 18803.00 frames. ], tot_loss[loss=0.2055, simple_loss=0.2868, pruned_loss=0.06209, over 2748096.70 frames. ], batch size: 74, lr: 3.18e-03, grad_scale: 4.0 2023-04-03 05:55:31,162 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.76 vs. limit=2.0 2023-04-03 05:55:39,620 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([0.8349, 1.3225, 1.0627, 0.9272, 1.1381, 0.9970, 0.9315, 1.2085], device='cuda:1'), covar=tensor([0.0689, 0.0935, 0.1193, 0.0844, 0.0594, 0.1386, 0.0677, 0.0552], device='cuda:1'), in_proj_covar=tensor([0.0304, 0.0318, 0.0336, 0.0270, 0.0249, 0.0342, 0.0294, 0.0275], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 05:55:42,603 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.950e+02 4.848e+02 5.950e+02 8.014e+02 1.769e+03, threshold=1.190e+03, percent-clipped=1.0 2023-04-03 05:56:34,946 INFO [train.py:903] (1/4) Epoch 26, batch 300, loss[loss=0.1903, simple_loss=0.2839, pruned_loss=0.04832, over 18860.00 frames. ], tot_loss[loss=0.2039, simple_loss=0.2849, pruned_loss=0.06145, over 2995287.27 frames. ], batch size: 74, lr: 3.18e-03, grad_scale: 4.0 2023-04-03 05:57:34,240 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=171046.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 05:57:34,578 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.45 vs. limit=5.0 2023-04-03 05:57:38,573 INFO [train.py:903] (1/4) Epoch 26, batch 350, loss[loss=0.1951, simple_loss=0.2699, pruned_loss=0.06018, over 19367.00 frames. ], tot_loss[loss=0.2016, simple_loss=0.2828, pruned_loss=0.06022, over 3186422.36 frames. ], batch size: 47, lr: 3.18e-03, grad_scale: 4.0 2023-04-03 05:57:45,661 WARNING [train.py:1073] (1/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1_rvb from training. Duration: 0.9681875 2023-04-03 05:57:49,078 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.001e+02 4.764e+02 6.020e+02 7.720e+02 1.602e+03, threshold=1.204e+03, percent-clipped=4.0 2023-04-03 05:58:42,186 INFO [train.py:903] (1/4) Epoch 26, batch 400, loss[loss=0.2095, simple_loss=0.2853, pruned_loss=0.06682, over 19592.00 frames. ], tot_loss[loss=0.2027, simple_loss=0.2839, pruned_loss=0.06072, over 3319470.87 frames. ], batch size: 52, lr: 3.18e-03, grad_scale: 8.0 2023-04-03 05:58:55,397 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=171110.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 05:59:43,979 INFO [train.py:903] (1/4) Epoch 26, batch 450, loss[loss=0.2141, simple_loss=0.2957, pruned_loss=0.06622, over 19717.00 frames. ], tot_loss[loss=0.2035, simple_loss=0.2842, pruned_loss=0.06143, over 3435385.63 frames. ], batch size: 63, lr: 3.18e-03, grad_scale: 8.0 2023-04-03 05:59:56,340 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.853e+02 5.217e+02 6.577e+02 9.059e+02 2.566e+03, threshold=1.315e+03, percent-clipped=7.0 2023-04-03 06:00:19,593 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 2023-04-03 06:00:20,813 WARNING [train.py:1073] (1/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 2023-04-03 06:00:30,492 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.2272, 1.9978, 1.8275, 2.1446, 1.8767, 1.8711, 1.7166, 2.1179], device='cuda:1'), covar=tensor([0.0960, 0.1395, 0.1419, 0.0948, 0.1392, 0.0574, 0.1470, 0.0694], device='cuda:1'), in_proj_covar=tensor([0.0275, 0.0354, 0.0314, 0.0254, 0.0304, 0.0255, 0.0315, 0.0262], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 06:00:38,896 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=171194.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 06:00:47,725 INFO [train.py:903] (1/4) Epoch 26, batch 500, loss[loss=0.1755, simple_loss=0.2644, pruned_loss=0.04334, over 19740.00 frames. ], tot_loss[loss=0.2043, simple_loss=0.2853, pruned_loss=0.0616, over 3532348.41 frames. ], batch size: 51, lr: 3.18e-03, grad_scale: 8.0 2023-04-03 06:01:05,600 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=171214.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 06:01:11,706 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=171219.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 06:01:52,229 INFO [train.py:903] (1/4) Epoch 26, batch 550, loss[loss=0.1829, simple_loss=0.2665, pruned_loss=0.04964, over 16009.00 frames. ], tot_loss[loss=0.203, simple_loss=0.2842, pruned_loss=0.06089, over 3594833.59 frames. ], batch size: 35, lr: 3.18e-03, grad_scale: 8.0 2023-04-03 06:02:03,093 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.195e+02 4.798e+02 6.471e+02 7.842e+02 1.459e+03, threshold=1.294e+03, percent-clipped=3.0 2023-04-03 06:02:12,074 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.14 vs. limit=2.0 2023-04-03 06:02:44,063 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=171290.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 06:02:55,680 INFO [train.py:903] (1/4) Epoch 26, batch 600, loss[loss=0.2114, simple_loss=0.2929, pruned_loss=0.06492, over 19484.00 frames. ], tot_loss[loss=0.2025, simple_loss=0.2838, pruned_loss=0.0606, over 3635273.34 frames. ], batch size: 64, lr: 3.18e-03, grad_scale: 8.0 2023-04-03 06:02:58,477 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=171302.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 06:03:31,076 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=171327.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 06:03:36,404 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 2023-04-03 06:03:57,571 INFO [train.py:903] (1/4) Epoch 26, batch 650, loss[loss=0.2056, simple_loss=0.2883, pruned_loss=0.06141, over 19799.00 frames. ], tot_loss[loss=0.2019, simple_loss=0.2828, pruned_loss=0.06048, over 3684938.91 frames. ], batch size: 56, lr: 3.18e-03, grad_scale: 8.0 2023-04-03 06:04:09,344 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.689e+02 4.759e+02 5.860e+02 7.918e+02 1.260e+03, threshold=1.172e+03, percent-clipped=0.0 2023-04-03 06:04:15,260 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=171363.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 06:05:01,713 INFO [train.py:903] (1/4) Epoch 26, batch 700, loss[loss=0.2012, simple_loss=0.2894, pruned_loss=0.05648, over 19710.00 frames. ], tot_loss[loss=0.2033, simple_loss=0.2842, pruned_loss=0.06117, over 3701612.22 frames. ], batch size: 59, lr: 3.18e-03, grad_scale: 8.0 2023-04-03 06:05:09,285 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=171405.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 06:06:08,113 INFO [train.py:903] (1/4) Epoch 26, batch 750, loss[loss=0.1766, simple_loss=0.2581, pruned_loss=0.04751, over 19492.00 frames. ], tot_loss[loss=0.2046, simple_loss=0.2855, pruned_loss=0.06189, over 3724352.22 frames. ], batch size: 49, lr: 3.18e-03, grad_scale: 8.0 2023-04-03 06:06:11,881 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.4649, 1.3051, 1.5487, 1.5433, 3.0429, 1.2253, 2.2472, 3.4240], device='cuda:1'), covar=tensor([0.0505, 0.2916, 0.2944, 0.1854, 0.0689, 0.2384, 0.1345, 0.0274], device='cuda:1'), in_proj_covar=tensor([0.0420, 0.0374, 0.0394, 0.0354, 0.0379, 0.0354, 0.0392, 0.0413], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 06:06:12,969 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=171454.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 06:06:18,672 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.178e+02 5.123e+02 6.550e+02 8.686e+02 2.549e+03, threshold=1.310e+03, percent-clipped=11.0 2023-04-03 06:07:12,528 INFO [train.py:903] (1/4) Epoch 26, batch 800, loss[loss=0.1862, simple_loss=0.2708, pruned_loss=0.05081, over 19692.00 frames. ], tot_loss[loss=0.2036, simple_loss=0.2845, pruned_loss=0.06137, over 3746973.95 frames. ], batch size: 53, lr: 3.18e-03, grad_scale: 8.0 2023-04-03 06:07:25,360 WARNING [train.py:1073] (1/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 2023-04-03 06:07:32,343 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=171516.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 06:08:14,615 INFO [train.py:903] (1/4) Epoch 26, batch 850, loss[loss=0.1891, simple_loss=0.2876, pruned_loss=0.04526, over 19532.00 frames. ], tot_loss[loss=0.2043, simple_loss=0.2856, pruned_loss=0.06148, over 3772401.39 frames. ], batch size: 56, lr: 3.18e-03, grad_scale: 8.0 2023-04-03 06:08:24,827 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=171558.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 06:08:25,743 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.196e+02 4.689e+02 5.699e+02 7.261e+02 1.636e+03, threshold=1.140e+03, percent-clipped=2.0 2023-04-03 06:08:40,921 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=171569.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 06:09:05,344 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9_rvb from training. Duration: 25.061125 2023-04-03 06:09:18,238 INFO [train.py:903] (1/4) Epoch 26, batch 900, loss[loss=0.1982, simple_loss=0.2811, pruned_loss=0.05764, over 19661.00 frames. ], tot_loss[loss=0.2041, simple_loss=0.2852, pruned_loss=0.06147, over 3787000.43 frames. ], batch size: 55, lr: 3.18e-03, grad_scale: 8.0 2023-04-03 06:10:17,102 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.54 vs. limit=5.0 2023-04-03 06:10:22,218 WARNING [train.py:1073] (1/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9_rvb from training. Duration: 26.32775 2023-04-03 06:10:23,374 INFO [train.py:903] (1/4) Epoch 26, batch 950, loss[loss=0.1549, simple_loss=0.2344, pruned_loss=0.03771, over 19748.00 frames. ], tot_loss[loss=0.2044, simple_loss=0.2855, pruned_loss=0.06168, over 3798821.85 frames. ], batch size: 45, lr: 3.18e-03, grad_scale: 8.0 2023-04-03 06:10:34,908 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.376e+02 5.213e+02 6.211e+02 8.451e+02 1.981e+03, threshold=1.242e+03, percent-clipped=10.0 2023-04-03 06:10:37,583 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=171661.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 06:10:45,899 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([0.9000, 1.3480, 1.0904, 0.8676, 1.1850, 0.9975, 0.8284, 1.2662], device='cuda:1'), covar=tensor([0.0641, 0.0755, 0.1043, 0.0907, 0.0561, 0.1273, 0.0639, 0.0441], device='cuda:1'), in_proj_covar=tensor([0.0304, 0.0318, 0.0337, 0.0270, 0.0250, 0.0343, 0.0295, 0.0274], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 06:10:51,862 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=171673.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 06:11:00,267 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5022, 1.4825, 1.4564, 1.8363, 1.3523, 1.6530, 1.7021, 1.5772], device='cuda:1'), covar=tensor([0.0830, 0.0918, 0.1001, 0.0658, 0.0840, 0.0765, 0.0796, 0.0700], device='cuda:1'), in_proj_covar=tensor([0.0212, 0.0223, 0.0227, 0.0239, 0.0226, 0.0214, 0.0189, 0.0204], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-04-03 06:11:09,442 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=171686.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 06:11:27,632 INFO [train.py:903] (1/4) Epoch 26, batch 1000, loss[loss=0.1939, simple_loss=0.2621, pruned_loss=0.06284, over 19738.00 frames. ], tot_loss[loss=0.2042, simple_loss=0.285, pruned_loss=0.06167, over 3811255.54 frames. ], batch size: 46, lr: 3.18e-03, grad_scale: 8.0 2023-04-03 06:11:36,041 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=171707.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 06:12:19,800 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 2023-04-03 06:12:31,840 INFO [train.py:903] (1/4) Epoch 26, batch 1050, loss[loss=0.1879, simple_loss=0.2603, pruned_loss=0.05776, over 19770.00 frames. ], tot_loss[loss=0.2037, simple_loss=0.2845, pruned_loss=0.06148, over 3800248.97 frames. ], batch size: 48, lr: 3.18e-03, grad_scale: 8.0 2023-04-03 06:12:42,477 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.548e+02 5.465e+02 6.383e+02 7.807e+02 1.569e+03, threshold=1.277e+03, percent-clipped=7.0 2023-04-03 06:12:55,279 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1620, 1.4470, 1.9028, 1.5014, 2.7833, 3.8274, 3.5289, 4.0256], device='cuda:1'), covar=tensor([0.1681, 0.3832, 0.3312, 0.2401, 0.0620, 0.0188, 0.0214, 0.0290], device='cuda:1'), in_proj_covar=tensor([0.0278, 0.0330, 0.0361, 0.0270, 0.0252, 0.0192, 0.0219, 0.0270], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-03 06:13:01,870 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 2023-04-03 06:13:34,460 INFO [train.py:903] (1/4) Epoch 26, batch 1100, loss[loss=0.2348, simple_loss=0.3136, pruned_loss=0.07802, over 19346.00 frames. ], tot_loss[loss=0.2048, simple_loss=0.2853, pruned_loss=0.06217, over 3794197.90 frames. ], batch size: 66, lr: 3.18e-03, grad_scale: 8.0 2023-04-03 06:14:05,449 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=171822.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 06:14:09,038 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=171825.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 06:14:40,985 INFO [train.py:903] (1/4) Epoch 26, batch 1150, loss[loss=0.1984, simple_loss=0.2891, pruned_loss=0.05382, over 18816.00 frames. ], tot_loss[loss=0.2041, simple_loss=0.285, pruned_loss=0.06157, over 3805610.85 frames. ], batch size: 74, lr: 3.18e-03, grad_scale: 8.0 2023-04-03 06:14:41,470 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=171850.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 06:14:54,274 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.309e+02 5.298e+02 6.784e+02 8.276e+02 1.649e+03, threshold=1.357e+03, percent-clipped=5.0 2023-04-03 06:14:55,497 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=171860.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 06:15:46,128 INFO [train.py:903] (1/4) Epoch 26, batch 1200, loss[loss=0.185, simple_loss=0.2586, pruned_loss=0.05573, over 18577.00 frames. ], tot_loss[loss=0.2044, simple_loss=0.2852, pruned_loss=0.06176, over 3813877.96 frames. ], batch size: 41, lr: 3.17e-03, grad_scale: 8.0 2023-04-03 06:16:12,809 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 2023-04-03 06:16:22,262 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=171929.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 06:16:48,465 INFO [train.py:903] (1/4) Epoch 26, batch 1250, loss[loss=0.1951, simple_loss=0.2687, pruned_loss=0.06069, over 19483.00 frames. ], tot_loss[loss=0.2036, simple_loss=0.2841, pruned_loss=0.06149, over 3829539.64 frames. ], batch size: 49, lr: 3.17e-03, grad_scale: 8.0 2023-04-03 06:16:53,688 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=171954.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 06:16:54,744 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=171955.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 06:16:58,918 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.249e+02 5.182e+02 6.196e+02 7.754e+02 1.405e+03, threshold=1.239e+03, percent-clipped=1.0 2023-04-03 06:17:21,574 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=171975.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 06:17:34,210 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0060, 2.0871, 2.3158, 2.6571, 2.0541, 2.5775, 2.3432, 2.1386], device='cuda:1'), covar=tensor([0.4414, 0.4219, 0.2085, 0.2634, 0.4493, 0.2384, 0.5083, 0.3637], device='cuda:1'), in_proj_covar=tensor([0.0925, 0.1000, 0.0734, 0.0946, 0.0903, 0.0838, 0.0857, 0.0801], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-03 06:17:51,746 INFO [train.py:903] (1/4) Epoch 26, batch 1300, loss[loss=0.1766, simple_loss=0.2608, pruned_loss=0.04626, over 19848.00 frames. ], tot_loss[loss=0.2036, simple_loss=0.2844, pruned_loss=0.06137, over 3818406.22 frames. ], batch size: 52, lr: 3.17e-03, grad_scale: 8.0 2023-04-03 06:18:24,709 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 2023-04-03 06:18:56,754 INFO [train.py:903] (1/4) Epoch 26, batch 1350, loss[loss=0.2155, simple_loss=0.3017, pruned_loss=0.06461, over 19711.00 frames. ], tot_loss[loss=0.204, simple_loss=0.2848, pruned_loss=0.06165, over 3811873.97 frames. ], batch size: 59, lr: 3.17e-03, grad_scale: 8.0 2023-04-03 06:19:09,470 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.423e+02 5.487e+02 6.725e+02 8.152e+02 1.378e+03, threshold=1.345e+03, percent-clipped=4.0 2023-04-03 06:19:25,619 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.3339, 3.5761, 2.1326, 2.3201, 3.2255, 1.9861, 1.5882, 2.5434], device='cuda:1'), covar=tensor([0.1296, 0.0612, 0.1104, 0.0860, 0.0527, 0.1209, 0.1079, 0.0654], device='cuda:1'), in_proj_covar=tensor([0.0304, 0.0317, 0.0336, 0.0270, 0.0249, 0.0341, 0.0292, 0.0274], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 06:19:34,030 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=172078.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 06:20:01,388 INFO [train.py:903] (1/4) Epoch 26, batch 1400, loss[loss=0.199, simple_loss=0.2812, pruned_loss=0.05842, over 19679.00 frames. ], tot_loss[loss=0.2032, simple_loss=0.2841, pruned_loss=0.06118, over 3815891.66 frames. ], batch size: 53, lr: 3.17e-03, grad_scale: 8.0 2023-04-03 06:20:07,094 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=172103.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 06:20:47,864 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=172136.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 06:21:05,098 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_rvb from training. Duration: 25.85 2023-04-03 06:21:06,113 INFO [train.py:903] (1/4) Epoch 26, batch 1450, loss[loss=0.1744, simple_loss=0.2552, pruned_loss=0.04684, over 19726.00 frames. ], tot_loss[loss=0.2035, simple_loss=0.2845, pruned_loss=0.06128, over 3814564.14 frames. ], batch size: 51, lr: 3.17e-03, grad_scale: 8.0 2023-04-03 06:21:16,562 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.245e+02 4.778e+02 6.066e+02 7.310e+02 2.231e+03, threshold=1.213e+03, percent-clipped=2.0 2023-04-03 06:21:26,484 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.67 vs. limit=2.0 2023-04-03 06:22:09,970 INFO [train.py:903] (1/4) Epoch 26, batch 1500, loss[loss=0.1674, simple_loss=0.2476, pruned_loss=0.04358, over 19822.00 frames. ], tot_loss[loss=0.2034, simple_loss=0.2844, pruned_loss=0.06115, over 3822185.71 frames. ], batch size: 52, lr: 3.17e-03, grad_scale: 8.0 2023-04-03 06:22:51,448 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=172231.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 06:23:14,775 INFO [train.py:903] (1/4) Epoch 26, batch 1550, loss[loss=0.1758, simple_loss=0.2501, pruned_loss=0.05079, over 19766.00 frames. ], tot_loss[loss=0.2036, simple_loss=0.2845, pruned_loss=0.06129, over 3806966.20 frames. ], batch size: 48, lr: 3.17e-03, grad_scale: 8.0 2023-04-03 06:23:23,379 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=172256.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 06:23:26,537 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.787e+02 4.585e+02 5.725e+02 7.044e+02 1.122e+03, threshold=1.145e+03, percent-clipped=0.0 2023-04-03 06:23:59,131 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.2998, 1.1963, 1.3819, 1.4151, 2.8309, 1.2333, 2.3623, 3.3552], device='cuda:1'), covar=tensor([0.0742, 0.3267, 0.3348, 0.2251, 0.0955, 0.2729, 0.1498, 0.0364], device='cuda:1'), in_proj_covar=tensor([0.0420, 0.0374, 0.0395, 0.0354, 0.0380, 0.0354, 0.0393, 0.0414], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 06:24:17,338 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=172299.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 06:24:18,407 INFO [train.py:903] (1/4) Epoch 26, batch 1600, loss[loss=0.1785, simple_loss=0.2531, pruned_loss=0.0519, over 19126.00 frames. ], tot_loss[loss=0.2034, simple_loss=0.2845, pruned_loss=0.06112, over 3821482.20 frames. ], batch size: 42, lr: 3.17e-03, grad_scale: 8.0 2023-04-03 06:24:30,670 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 2023-04-03 06:24:45,016 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9_rvb from training. Duration: 30.1555625 2023-04-03 06:24:49,249 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.1686, 3.1751, 1.9070, 2.0631, 2.8333, 1.6572, 1.5413, 2.4036], device='cuda:1'), covar=tensor([0.1339, 0.0757, 0.1123, 0.0911, 0.0626, 0.1360, 0.1048, 0.0671], device='cuda:1'), in_proj_covar=tensor([0.0307, 0.0320, 0.0340, 0.0272, 0.0251, 0.0345, 0.0294, 0.0275], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 06:25:22,891 INFO [train.py:903] (1/4) Epoch 26, batch 1650, loss[loss=0.2353, simple_loss=0.3164, pruned_loss=0.07706, over 19533.00 frames. ], tot_loss[loss=0.2022, simple_loss=0.2834, pruned_loss=0.06048, over 3835642.56 frames. ], batch size: 54, lr: 3.17e-03, grad_scale: 8.0 2023-04-03 06:25:32,972 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.237e+02 4.806e+02 6.116e+02 7.815e+02 1.931e+03, threshold=1.223e+03, percent-clipped=6.0 2023-04-03 06:26:24,989 INFO [train.py:903] (1/4) Epoch 26, batch 1700, loss[loss=0.1508, simple_loss=0.2292, pruned_loss=0.03619, over 19743.00 frames. ], tot_loss[loss=0.2031, simple_loss=0.284, pruned_loss=0.06114, over 3822233.05 frames. ], batch size: 45, lr: 3.17e-03, grad_scale: 8.0 2023-04-03 06:26:27,915 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 2023-04-03 06:26:43,085 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=172414.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 06:27:07,050 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 2023-04-03 06:27:27,208 INFO [train.py:903] (1/4) Epoch 26, batch 1750, loss[loss=0.1703, simple_loss=0.2522, pruned_loss=0.04425, over 19378.00 frames. ], tot_loss[loss=0.2039, simple_loss=0.2845, pruned_loss=0.0616, over 3826105.16 frames. ], batch size: 48, lr: 3.17e-03, grad_scale: 8.0 2023-04-03 06:27:39,753 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.268e+02 4.883e+02 5.717e+02 7.034e+02 1.807e+03, threshold=1.143e+03, percent-clipped=3.0 2023-04-03 06:27:52,712 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0697, 1.9457, 2.0061, 2.5093, 2.1777, 2.1981, 2.1819, 2.1215], device='cuda:1'), covar=tensor([0.0694, 0.0751, 0.0845, 0.0688, 0.0806, 0.0722, 0.0865, 0.0587], device='cuda:1'), in_proj_covar=tensor([0.0212, 0.0223, 0.0227, 0.0240, 0.0226, 0.0214, 0.0190, 0.0204], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-04-03 06:28:06,467 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=172480.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 06:28:09,144 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2306, 1.4685, 2.0381, 1.4676, 3.1963, 4.6673, 4.5058, 5.0965], device='cuda:1'), covar=tensor([0.1674, 0.3882, 0.3322, 0.2494, 0.0562, 0.0188, 0.0167, 0.0159], device='cuda:1'), in_proj_covar=tensor([0.0278, 0.0330, 0.0361, 0.0270, 0.0252, 0.0193, 0.0219, 0.0270], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-03 06:28:26,592 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([6.3195, 5.7249, 3.1647, 5.0927, 1.0845, 5.9135, 5.7776, 5.9090], device='cuda:1'), covar=tensor([0.0332, 0.0829, 0.1821, 0.0684, 0.3881, 0.0463, 0.0696, 0.0830], device='cuda:1'), in_proj_covar=tensor([0.0518, 0.0420, 0.0507, 0.0355, 0.0406, 0.0447, 0.0441, 0.0472], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 06:28:32,027 INFO [train.py:903] (1/4) Epoch 26, batch 1800, loss[loss=0.2359, simple_loss=0.3115, pruned_loss=0.08018, over 17557.00 frames. ], tot_loss[loss=0.2032, simple_loss=0.2843, pruned_loss=0.06104, over 3822676.35 frames. ], batch size: 101, lr: 3.17e-03, grad_scale: 8.0 2023-04-03 06:29:31,750 WARNING [train.py:1073] (1/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9_rvb from training. Duration: 25.0944375 2023-04-03 06:29:36,214 INFO [train.py:903] (1/4) Epoch 26, batch 1850, loss[loss=0.2254, simple_loss=0.3144, pruned_loss=0.06825, over 19731.00 frames. ], tot_loss[loss=0.2034, simple_loss=0.2843, pruned_loss=0.06127, over 3822332.02 frames. ], batch size: 63, lr: 3.17e-03, grad_scale: 8.0 2023-04-03 06:29:46,976 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.860e+02 4.880e+02 5.794e+02 7.257e+02 1.575e+03, threshold=1.159e+03, percent-clipped=2.0 2023-04-03 06:30:11,354 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9_rvb from training. Duration: 27.8166875 2023-04-03 06:30:20,705 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.3627, 3.1056, 2.2219, 2.8126, 0.8815, 3.0567, 2.9228, 2.9893], device='cuda:1'), covar=tensor([0.1036, 0.1246, 0.2030, 0.1030, 0.3653, 0.0897, 0.1103, 0.1337], device='cuda:1'), in_proj_covar=tensor([0.0521, 0.0423, 0.0511, 0.0357, 0.0407, 0.0450, 0.0443, 0.0475], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 06:30:34,172 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=172595.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 06:30:39,841 INFO [train.py:903] (1/4) Epoch 26, batch 1900, loss[loss=0.2011, simple_loss=0.2859, pruned_loss=0.05809, over 19288.00 frames. ], tot_loss[loss=0.2033, simple_loss=0.2841, pruned_loss=0.06128, over 3824437.27 frames. ], batch size: 66, lr: 3.17e-03, grad_scale: 4.0 2023-04-03 06:30:59,226 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 2023-04-03 06:31:01,796 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=172616.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 06:31:05,030 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9_rvb from training. Duration: 27.02225 2023-04-03 06:31:23,037 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([5.0016, 5.0361, 5.8055, 5.8433, 1.9076, 5.4243, 4.5970, 5.4429], device='cuda:1'), covar=tensor([0.1989, 0.0974, 0.0597, 0.0713, 0.6847, 0.0834, 0.0673, 0.1309], device='cuda:1'), in_proj_covar=tensor([0.0806, 0.0774, 0.0978, 0.0859, 0.0852, 0.0745, 0.0584, 0.0906], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-03 06:31:30,028 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 2023-04-03 06:31:36,140 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=172643.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 06:31:44,209 INFO [train.py:903] (1/4) Epoch 26, batch 1950, loss[loss=0.1915, simple_loss=0.2844, pruned_loss=0.04926, over 19581.00 frames. ], tot_loss[loss=0.2034, simple_loss=0.2839, pruned_loss=0.06139, over 3819349.09 frames. ], batch size: 61, lr: 3.17e-03, grad_scale: 4.0 2023-04-03 06:31:57,691 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.017e+02 5.105e+02 6.247e+02 7.487e+02 1.872e+03, threshold=1.249e+03, percent-clipped=7.0 2023-04-03 06:32:12,308 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=172670.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 06:32:43,403 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=172695.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 06:32:48,408 INFO [train.py:903] (1/4) Epoch 26, batch 2000, loss[loss=0.2472, simple_loss=0.3205, pruned_loss=0.08695, over 18042.00 frames. ], tot_loss[loss=0.2029, simple_loss=0.2835, pruned_loss=0.06116, over 3804691.89 frames. ], batch size: 83, lr: 3.17e-03, grad_scale: 8.0 2023-04-03 06:33:14,447 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 2023-04-03 06:33:39,031 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([5.2444, 5.2871, 6.0643, 6.0845, 1.9866, 5.6879, 4.7750, 5.7438], device='cuda:1'), covar=tensor([0.1819, 0.0758, 0.0591, 0.0621, 0.6426, 0.0818, 0.0656, 0.1155], device='cuda:1'), in_proj_covar=tensor([0.0803, 0.0774, 0.0978, 0.0859, 0.0852, 0.0744, 0.0584, 0.0905], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-03 06:33:46,802 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_rvb from training. Duration: 26.8349375 2023-04-03 06:33:51,653 INFO [train.py:903] (1/4) Epoch 26, batch 2050, loss[loss=0.1759, simple_loss=0.2502, pruned_loss=0.05084, over 19289.00 frames. ], tot_loss[loss=0.2031, simple_loss=0.2838, pruned_loss=0.06121, over 3820338.63 frames. ], batch size: 44, lr: 3.17e-03, grad_scale: 8.0 2023-04-03 06:33:55,269 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=172752.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 06:34:03,944 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.396e+02 5.217e+02 6.186e+02 8.512e+02 2.102e+03, threshold=1.237e+03, percent-clipped=6.0 2023-04-03 06:34:06,492 WARNING [train.py:1073] (1/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 2023-04-03 06:34:07,796 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 2023-04-03 06:34:27,707 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 2023-04-03 06:34:54,263 INFO [train.py:903] (1/4) Epoch 26, batch 2100, loss[loss=0.2078, simple_loss=0.298, pruned_loss=0.0588, over 19531.00 frames. ], tot_loss[loss=0.2048, simple_loss=0.2853, pruned_loss=0.06219, over 3826095.53 frames. ], batch size: 56, lr: 3.17e-03, grad_scale: 8.0 2023-04-03 06:35:25,202 WARNING [train.py:1073] (1/4) Exclude cut with ID 453-131332-0000-131866_sp0.9_rvb from training. Duration: 25.3333125 2023-04-03 06:35:47,314 WARNING [train.py:1073] (1/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9_rvb from training. Duration: 26.6166875 2023-04-03 06:35:56,669 INFO [train.py:903] (1/4) Epoch 26, batch 2150, loss[loss=0.1774, simple_loss=0.2552, pruned_loss=0.04978, over 19470.00 frames. ], tot_loss[loss=0.205, simple_loss=0.2853, pruned_loss=0.06234, over 3812734.06 frames. ], batch size: 49, lr: 3.17e-03, grad_scale: 8.0 2023-04-03 06:35:58,269 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=172851.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 06:36:02,860 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.1064, 1.4436, 1.6120, 1.5289, 2.7291, 1.1796, 2.2326, 3.1252], device='cuda:1'), covar=tensor([0.0569, 0.2665, 0.2666, 0.1758, 0.0744, 0.2227, 0.1203, 0.0290], device='cuda:1'), in_proj_covar=tensor([0.0417, 0.0371, 0.0392, 0.0351, 0.0378, 0.0352, 0.0389, 0.0410], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 06:36:10,067 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.604e+02 4.749e+02 6.121e+02 8.086e+02 1.727e+03, threshold=1.224e+03, percent-clipped=6.0 2023-04-03 06:36:14,759 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=172863.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 06:36:21,374 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.45 vs. limit=2.0 2023-04-03 06:36:31,065 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=172876.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 06:36:59,983 INFO [train.py:903] (1/4) Epoch 26, batch 2200, loss[loss=0.2054, simple_loss=0.2986, pruned_loss=0.05614, over 19620.00 frames. ], tot_loss[loss=0.2043, simple_loss=0.2847, pruned_loss=0.06198, over 3820592.16 frames. ], batch size: 50, lr: 3.17e-03, grad_scale: 4.0 2023-04-03 06:38:04,010 INFO [train.py:903] (1/4) Epoch 26, batch 2250, loss[loss=0.1603, simple_loss=0.2413, pruned_loss=0.03965, over 19761.00 frames. ], tot_loss[loss=0.2044, simple_loss=0.2847, pruned_loss=0.06207, over 3812258.79 frames. ], batch size: 48, lr: 3.16e-03, grad_scale: 4.0 2023-04-03 06:38:16,970 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=172960.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 06:38:17,987 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.182e+02 4.806e+02 5.924e+02 7.729e+02 1.368e+03, threshold=1.185e+03, percent-clipped=2.0 2023-04-03 06:38:51,868 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=172987.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 06:39:07,879 INFO [train.py:903] (1/4) Epoch 26, batch 2300, loss[loss=0.1823, simple_loss=0.2756, pruned_loss=0.04452, over 19657.00 frames. ], tot_loss[loss=0.2042, simple_loss=0.2848, pruned_loss=0.06181, over 3805842.17 frames. ], batch size: 58, lr: 3.16e-03, grad_scale: 4.0 2023-04-03 06:39:13,393 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 2023-04-03 06:39:19,744 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_rvb from training. Duration: 26.205 2023-04-03 06:40:11,049 INFO [train.py:903] (1/4) Epoch 26, batch 2350, loss[loss=0.1823, simple_loss=0.274, pruned_loss=0.04526, over 19546.00 frames. ], tot_loss[loss=0.2039, simple_loss=0.2849, pruned_loss=0.06148, over 3818338.38 frames. ], batch size: 56, lr: 3.16e-03, grad_scale: 4.0 2023-04-03 06:40:25,951 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.095e+02 5.175e+02 6.518e+02 7.971e+02 2.695e+03, threshold=1.304e+03, percent-clipped=8.0 2023-04-03 06:40:43,706 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=173075.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 06:40:54,471 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_rvb from training. Duration: 25.775 2023-04-03 06:41:08,278 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=173096.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 06:41:10,573 WARNING [train.py:1073] (1/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9_rvb from training. Duration: 25.45 2023-04-03 06:41:14,043 INFO [train.py:903] (1/4) Epoch 26, batch 2400, loss[loss=0.2442, simple_loss=0.3208, pruned_loss=0.08382, over 19664.00 frames. ], tot_loss[loss=0.2049, simple_loss=0.2858, pruned_loss=0.06201, over 3819210.33 frames. ], batch size: 55, lr: 3.16e-03, grad_scale: 8.0 2023-04-03 06:41:18,165 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=173102.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 06:41:48,388 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5563, 1.4432, 2.0252, 1.6242, 3.1482, 4.7380, 4.5259, 5.1565], device='cuda:1'), covar=tensor([0.1560, 0.3921, 0.3388, 0.2424, 0.0601, 0.0203, 0.0177, 0.0183], device='cuda:1'), in_proj_covar=tensor([0.0277, 0.0330, 0.0362, 0.0271, 0.0252, 0.0193, 0.0220, 0.0270], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-03 06:42:19,064 INFO [train.py:903] (1/4) Epoch 26, batch 2450, loss[loss=0.1968, simple_loss=0.2755, pruned_loss=0.05909, over 19598.00 frames. ], tot_loss[loss=0.2048, simple_loss=0.2856, pruned_loss=0.06202, over 3809984.91 frames. ], batch size: 52, lr: 3.16e-03, grad_scale: 8.0 2023-04-03 06:42:32,917 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.093e+02 4.972e+02 5.973e+02 7.732e+02 1.743e+03, threshold=1.195e+03, percent-clipped=3.0 2023-04-03 06:42:41,751 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=173168.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 06:43:22,673 INFO [train.py:903] (1/4) Epoch 26, batch 2500, loss[loss=0.1674, simple_loss=0.2576, pruned_loss=0.03856, over 19542.00 frames. ], tot_loss[loss=0.2055, simple_loss=0.2859, pruned_loss=0.06258, over 3795773.00 frames. ], batch size: 54, lr: 3.16e-03, grad_scale: 8.0 2023-04-03 06:43:24,454 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.8906, 2.0018, 2.2342, 2.4632, 1.8873, 2.3669, 2.2243, 2.0537], device='cuda:1'), covar=tensor([0.4123, 0.4053, 0.1975, 0.2353, 0.4141, 0.2252, 0.5002, 0.3445], device='cuda:1'), in_proj_covar=tensor([0.0927, 0.1004, 0.0736, 0.0948, 0.0906, 0.0842, 0.0859, 0.0804], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-03 06:43:31,260 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=173207.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 06:43:36,118 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=173211.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 06:44:25,903 INFO [train.py:903] (1/4) Epoch 26, batch 2550, loss[loss=0.2055, simple_loss=0.288, pruned_loss=0.06145, over 19506.00 frames. ], tot_loss[loss=0.2054, simple_loss=0.286, pruned_loss=0.06241, over 3798432.78 frames. ], batch size: 64, lr: 3.16e-03, grad_scale: 8.0 2023-04-03 06:44:40,264 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3371, 2.0553, 1.6038, 1.3689, 1.8315, 1.2867, 1.3213, 1.8462], device='cuda:1'), covar=tensor([0.1040, 0.0810, 0.1142, 0.0897, 0.0607, 0.1366, 0.0735, 0.0498], device='cuda:1'), in_proj_covar=tensor([0.0303, 0.0317, 0.0337, 0.0269, 0.0248, 0.0341, 0.0291, 0.0273], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 06:44:40,981 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.733e+02 4.982e+02 5.984e+02 8.594e+02 2.255e+03, threshold=1.197e+03, percent-clipped=6.0 2023-04-03 06:45:23,058 WARNING [train.py:1073] (1/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 2023-04-03 06:45:30,060 INFO [train.py:903] (1/4) Epoch 26, batch 2600, loss[loss=0.2142, simple_loss=0.2998, pruned_loss=0.06434, over 19603.00 frames. ], tot_loss[loss=0.205, simple_loss=0.2858, pruned_loss=0.06206, over 3814512.24 frames. ], batch size: 57, lr: 3.16e-03, grad_scale: 8.0 2023-04-03 06:45:59,666 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=173322.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 06:46:05,371 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=173327.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 06:46:05,465 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4732, 1.4224, 1.4039, 1.7491, 1.3145, 1.6022, 1.6397, 1.5079], device='cuda:1'), covar=tensor([0.0915, 0.0995, 0.1068, 0.0754, 0.0937, 0.0815, 0.0860, 0.0786], device='cuda:1'), in_proj_covar=tensor([0.0213, 0.0226, 0.0228, 0.0242, 0.0228, 0.0214, 0.0190, 0.0206], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-04-03 06:46:10,050 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=173331.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 06:46:35,579 INFO [train.py:903] (1/4) Epoch 26, batch 2650, loss[loss=0.1921, simple_loss=0.2764, pruned_loss=0.05387, over 19757.00 frames. ], tot_loss[loss=0.2056, simple_loss=0.2863, pruned_loss=0.06246, over 3801478.68 frames. ], batch size: 54, lr: 3.16e-03, grad_scale: 8.0 2023-04-03 06:46:41,847 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.1956, 3.4154, 2.0661, 2.1263, 3.1475, 1.8532, 1.6332, 2.3623], device='cuda:1'), covar=tensor([0.1354, 0.0689, 0.1128, 0.0895, 0.0547, 0.1241, 0.0991, 0.0673], device='cuda:1'), in_proj_covar=tensor([0.0302, 0.0316, 0.0336, 0.0269, 0.0248, 0.0340, 0.0290, 0.0272], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 06:46:43,004 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=173356.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 06:46:45,487 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=173358.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 06:46:49,589 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.269e+02 5.033e+02 5.962e+02 7.363e+02 1.395e+03, threshold=1.192e+03, percent-clipped=2.0 2023-04-03 06:46:55,544 WARNING [train.py:1073] (1/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9_rvb from training. Duration: 27.25 2023-04-03 06:47:17,823 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=173383.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 06:47:39,153 INFO [train.py:903] (1/4) Epoch 26, batch 2700, loss[loss=0.1942, simple_loss=0.2833, pruned_loss=0.05259, over 19125.00 frames. ], tot_loss[loss=0.2056, simple_loss=0.2862, pruned_loss=0.06249, over 3790292.28 frames. ], batch size: 69, lr: 3.16e-03, grad_scale: 8.0 2023-04-03 06:48:00,265 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.0378, 4.4564, 4.7824, 4.7931, 1.8876, 4.4868, 3.8553, 4.4789], device='cuda:1'), covar=tensor([0.1712, 0.0869, 0.0582, 0.0698, 0.5989, 0.0887, 0.0689, 0.1146], device='cuda:1'), in_proj_covar=tensor([0.0806, 0.0773, 0.0978, 0.0861, 0.0856, 0.0742, 0.0584, 0.0909], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-03 06:48:32,623 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.3083, 2.3346, 2.5789, 3.0308, 2.3880, 2.9197, 2.5814, 2.3459], device='cuda:1'), covar=tensor([0.4353, 0.4328, 0.1987, 0.2692, 0.4584, 0.2277, 0.4996, 0.3498], device='cuda:1'), in_proj_covar=tensor([0.0928, 0.1004, 0.0736, 0.0949, 0.0908, 0.0841, 0.0859, 0.0805], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-03 06:48:39,711 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.42 vs. limit=2.0 2023-04-03 06:48:41,213 INFO [train.py:903] (1/4) Epoch 26, batch 2750, loss[loss=0.214, simple_loss=0.2953, pruned_loss=0.06632, over 19544.00 frames. ], tot_loss[loss=0.2047, simple_loss=0.2855, pruned_loss=0.06198, over 3793475.14 frames. ], batch size: 56, lr: 3.16e-03, grad_scale: 8.0 2023-04-03 06:48:54,807 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.847e+02 4.981e+02 6.163e+02 7.639e+02 1.916e+03, threshold=1.233e+03, percent-clipped=5.0 2023-04-03 06:49:03,530 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=173467.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 06:49:34,160 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=173492.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 06:49:44,190 INFO [train.py:903] (1/4) Epoch 26, batch 2800, loss[loss=0.1615, simple_loss=0.2366, pruned_loss=0.04316, over 19759.00 frames. ], tot_loss[loss=0.204, simple_loss=0.2847, pruned_loss=0.06169, over 3809661.83 frames. ], batch size: 47, lr: 3.16e-03, grad_scale: 8.0 2023-04-03 06:49:53,017 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5004, 1.5445, 1.7292, 1.7276, 2.3913, 2.2222, 2.3493, 1.1240], device='cuda:1'), covar=tensor([0.2511, 0.4455, 0.2873, 0.1983, 0.1447, 0.2166, 0.1422, 0.4532], device='cuda:1'), in_proj_covar=tensor([0.0545, 0.0661, 0.0739, 0.0500, 0.0629, 0.0539, 0.0666, 0.0566], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 06:49:56,421 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=173509.0, num_to_drop=1, layers_to_drop={1} 2023-04-03 06:49:56,870 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.71 vs. limit=5.0 2023-04-03 06:50:00,274 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=173512.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 06:50:48,001 INFO [train.py:903] (1/4) Epoch 26, batch 2850, loss[loss=0.2535, simple_loss=0.3317, pruned_loss=0.08762, over 18220.00 frames. ], tot_loss[loss=0.2047, simple_loss=0.2852, pruned_loss=0.06204, over 3793112.47 frames. ], batch size: 83, lr: 3.16e-03, grad_scale: 8.0 2023-04-03 06:51:01,767 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.227e+02 4.954e+02 6.202e+02 8.183e+02 1.932e+03, threshold=1.240e+03, percent-clipped=7.0 2023-04-03 06:51:23,144 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=173578.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 06:51:51,093 WARNING [train.py:1073] (1/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 2023-04-03 06:51:52,153 INFO [train.py:903] (1/4) Epoch 26, batch 2900, loss[loss=0.2167, simple_loss=0.2947, pruned_loss=0.06937, over 19677.00 frames. ], tot_loss[loss=0.2032, simple_loss=0.2838, pruned_loss=0.0613, over 3809913.46 frames. ], batch size: 60, lr: 3.16e-03, grad_scale: 8.0 2023-04-03 06:51:57,234 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=173603.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 06:52:08,009 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9149, 1.9718, 2.2554, 2.5498, 1.9277, 2.4830, 2.2412, 2.1013], device='cuda:1'), covar=tensor([0.4426, 0.4037, 0.2055, 0.2581, 0.4240, 0.2258, 0.5408, 0.3595], device='cuda:1'), in_proj_covar=tensor([0.0928, 0.1003, 0.0736, 0.0946, 0.0907, 0.0840, 0.0859, 0.0805], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-03 06:52:27,669 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=173627.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 06:52:33,491 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=173632.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 06:52:56,769 INFO [train.py:903] (1/4) Epoch 26, batch 2950, loss[loss=0.1853, simple_loss=0.2643, pruned_loss=0.05313, over 19589.00 frames. ], tot_loss[loss=0.203, simple_loss=0.284, pruned_loss=0.06103, over 3815379.90 frames. ], batch size: 52, lr: 3.16e-03, grad_scale: 8.0 2023-04-03 06:53:10,747 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.057e+02 5.287e+02 6.719e+02 8.706e+02 2.181e+03, threshold=1.344e+03, percent-clipped=5.0 2023-04-03 06:53:23,943 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=173671.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 06:53:59,611 INFO [train.py:903] (1/4) Epoch 26, batch 3000, loss[loss=0.2225, simple_loss=0.3064, pruned_loss=0.06928, over 18722.00 frames. ], tot_loss[loss=0.2023, simple_loss=0.2833, pruned_loss=0.06063, over 3811088.01 frames. ], batch size: 74, lr: 3.16e-03, grad_scale: 8.0 2023-04-03 06:53:59,611 INFO [train.py:928] (1/4) Computing validation loss 2023-04-03 06:54:12,258 INFO [train.py:937] (1/4) Epoch 26, validation: loss=0.1681, simple_loss=0.2675, pruned_loss=0.03435, over 944034.00 frames. 2023-04-03 06:54:12,259 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18821MB 2023-04-03 06:54:17,249 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_rvb from training. Duration: 29.735 2023-04-03 06:55:16,493 INFO [train.py:903] (1/4) Epoch 26, batch 3050, loss[loss=0.1727, simple_loss=0.2532, pruned_loss=0.04608, over 19408.00 frames. ], tot_loss[loss=0.2025, simple_loss=0.2835, pruned_loss=0.06081, over 3801707.48 frames. ], batch size: 48, lr: 3.16e-03, grad_scale: 8.0 2023-04-03 06:55:25,128 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.8307, 4.4159, 2.5823, 3.8659, 1.2882, 4.4377, 4.2594, 4.3184], device='cuda:1'), covar=tensor([0.0543, 0.0882, 0.2120, 0.0854, 0.3523, 0.0649, 0.0896, 0.1216], device='cuda:1'), in_proj_covar=tensor([0.0522, 0.0424, 0.0510, 0.0357, 0.0408, 0.0450, 0.0447, 0.0473], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 06:55:30,852 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.198e+02 4.825e+02 5.950e+02 7.456e+02 1.374e+03, threshold=1.190e+03, percent-clipped=1.0 2023-04-03 06:56:02,590 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=173786.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 06:56:20,031 INFO [train.py:903] (1/4) Epoch 26, batch 3100, loss[loss=0.1893, simple_loss=0.2802, pruned_loss=0.04924, over 19661.00 frames. ], tot_loss[loss=0.2027, simple_loss=0.2839, pruned_loss=0.06073, over 3813755.83 frames. ], batch size: 58, lr: 3.16e-03, grad_scale: 8.0 2023-04-03 06:57:23,203 INFO [train.py:903] (1/4) Epoch 26, batch 3150, loss[loss=0.1811, simple_loss=0.2637, pruned_loss=0.04925, over 19723.00 frames. ], tot_loss[loss=0.2017, simple_loss=0.2824, pruned_loss=0.06044, over 3824624.44 frames. ], batch size: 51, lr: 3.16e-03, grad_scale: 8.0 2023-04-03 06:57:26,872 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=173853.0, num_to_drop=1, layers_to_drop={0} 2023-04-03 06:57:37,134 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.975e+02 5.081e+02 6.233e+02 7.406e+02 2.417e+03, threshold=1.247e+03, percent-clipped=3.0 2023-04-03 06:57:50,105 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 2023-04-03 06:58:05,127 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=173883.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 06:58:25,317 INFO [train.py:903] (1/4) Epoch 26, batch 3200, loss[loss=0.2128, simple_loss=0.2866, pruned_loss=0.06954, over 19724.00 frames. ], tot_loss[loss=0.2028, simple_loss=0.2832, pruned_loss=0.06119, over 3815302.35 frames. ], batch size: 51, lr: 3.16e-03, grad_scale: 8.0 2023-04-03 06:58:35,898 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=173908.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 06:59:25,935 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.8243, 4.2302, 4.4923, 4.5212, 1.7265, 4.2598, 3.6566, 4.2097], device='cuda:1'), covar=tensor([0.1559, 0.0926, 0.0599, 0.0687, 0.6012, 0.1042, 0.0737, 0.1171], device='cuda:1'), in_proj_covar=tensor([0.0804, 0.0771, 0.0974, 0.0856, 0.0851, 0.0742, 0.0579, 0.0902], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-03 06:59:27,976 INFO [train.py:903] (1/4) Epoch 26, batch 3250, loss[loss=0.1936, simple_loss=0.2776, pruned_loss=0.05482, over 19677.00 frames. ], tot_loss[loss=0.2038, simple_loss=0.2841, pruned_loss=0.06176, over 3830558.78 frames. ], batch size: 55, lr: 3.16e-03, grad_scale: 8.0 2023-04-03 06:59:42,854 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.716e+02 4.798e+02 6.185e+02 8.155e+02 2.789e+03, threshold=1.237e+03, percent-clipped=7.0 2023-04-03 06:59:52,489 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=173968.0, num_to_drop=1, layers_to_drop={0} 2023-04-03 06:59:55,190 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.55 vs. limit=5.0 2023-04-03 07:00:01,778 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=173976.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:00:31,320 INFO [train.py:903] (1/4) Epoch 26, batch 3300, loss[loss=0.2172, simple_loss=0.2796, pruned_loss=0.07743, over 19034.00 frames. ], tot_loss[loss=0.2041, simple_loss=0.2847, pruned_loss=0.06177, over 3821958.96 frames. ], batch size: 42, lr: 3.16e-03, grad_scale: 8.0 2023-04-03 07:00:35,708 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 2023-04-03 07:01:08,019 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=4.05 vs. limit=5.0 2023-04-03 07:01:09,845 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.8591, 1.9286, 1.4629, 1.8722, 1.9376, 1.5318, 1.5284, 1.7468], device='cuda:1'), covar=tensor([0.1240, 0.1662, 0.1956, 0.1202, 0.1403, 0.0983, 0.1962, 0.1039], device='cuda:1'), in_proj_covar=tensor([0.0279, 0.0358, 0.0316, 0.0256, 0.0306, 0.0255, 0.0319, 0.0265], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 07:01:14,201 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=174033.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:01:17,513 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=174035.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:01:25,974 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=174042.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:01:35,837 INFO [train.py:903] (1/4) Epoch 26, batch 3350, loss[loss=0.2309, simple_loss=0.3085, pruned_loss=0.07671, over 19120.00 frames. ], tot_loss[loss=0.2038, simple_loss=0.2844, pruned_loss=0.06162, over 3828431.68 frames. ], batch size: 69, lr: 3.15e-03, grad_scale: 8.0 2023-04-03 07:01:39,779 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.0899, 1.0334, 1.4482, 1.2933, 2.5041, 1.1655, 2.3061, 2.8823], device='cuda:1'), covar=tensor([0.0758, 0.3898, 0.3308, 0.2272, 0.1175, 0.2665, 0.1182, 0.0487], device='cuda:1'), in_proj_covar=tensor([0.0419, 0.0374, 0.0393, 0.0352, 0.0378, 0.0354, 0.0392, 0.0412], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 07:01:49,564 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.212e+02 4.999e+02 5.843e+02 7.881e+02 1.777e+03, threshold=1.169e+03, percent-clipped=4.0 2023-04-03 07:01:56,840 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=174067.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:02:01,253 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.2209, 2.0636, 2.1348, 2.9786, 1.9098, 2.4341, 2.3411, 2.2842], device='cuda:1'), covar=tensor([0.0792, 0.0865, 0.0884, 0.0727, 0.0866, 0.0704, 0.0880, 0.0632], device='cuda:1'), in_proj_covar=tensor([0.0214, 0.0224, 0.0227, 0.0241, 0.0226, 0.0213, 0.0189, 0.0206], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-04-03 07:02:25,285 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 2023-04-03 07:02:28,229 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=174091.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:02:39,919 INFO [train.py:903] (1/4) Epoch 26, batch 3400, loss[loss=0.2127, simple_loss=0.297, pruned_loss=0.06418, over 19543.00 frames. ], tot_loss[loss=0.2026, simple_loss=0.2834, pruned_loss=0.06092, over 3824857.66 frames. ], batch size: 56, lr: 3.15e-03, grad_scale: 8.0 2023-04-03 07:02:49,781 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9300, 1.7954, 1.5978, 1.9594, 1.6764, 1.6914, 1.5336, 1.8208], device='cuda:1'), covar=tensor([0.1052, 0.1352, 0.1499, 0.0956, 0.1334, 0.0585, 0.1559, 0.0826], device='cuda:1'), in_proj_covar=tensor([0.0279, 0.0357, 0.0316, 0.0255, 0.0305, 0.0255, 0.0318, 0.0264], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 07:03:41,115 WARNING [train.py:1073] (1/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9_rvb from training. Duration: 25.2444375 2023-04-03 07:03:42,256 INFO [train.py:903] (1/4) Epoch 26, batch 3450, loss[loss=0.2519, simple_loss=0.3213, pruned_loss=0.09122, over 18188.00 frames. ], tot_loss[loss=0.2035, simple_loss=0.2843, pruned_loss=0.06135, over 3827180.70 frames. ], batch size: 83, lr: 3.15e-03, grad_scale: 8.0 2023-04-03 07:03:53,283 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=174157.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:03:57,805 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.388e+02 4.666e+02 5.900e+02 7.449e+02 1.550e+03, threshold=1.180e+03, percent-clipped=6.0 2023-04-03 07:04:47,402 INFO [train.py:903] (1/4) Epoch 26, batch 3500, loss[loss=0.2002, simple_loss=0.2787, pruned_loss=0.06083, over 19532.00 frames. ], tot_loss[loss=0.203, simple_loss=0.2838, pruned_loss=0.06105, over 3836604.47 frames. ], batch size: 56, lr: 3.15e-03, grad_scale: 8.0 2023-04-03 07:05:18,028 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=174224.0, num_to_drop=1, layers_to_drop={1} 2023-04-03 07:05:49,660 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=174249.0, num_to_drop=1, layers_to_drop={0} 2023-04-03 07:05:50,379 INFO [train.py:903] (1/4) Epoch 26, batch 3550, loss[loss=0.1974, simple_loss=0.2774, pruned_loss=0.05875, over 19861.00 frames. ], tot_loss[loss=0.2039, simple_loss=0.2845, pruned_loss=0.0617, over 3828996.56 frames. ], batch size: 52, lr: 3.15e-03, grad_scale: 8.0 2023-04-03 07:06:03,174 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.093e+02 4.676e+02 6.063e+02 7.972e+02 1.969e+03, threshold=1.213e+03, percent-clipped=7.0 2023-04-03 07:06:53,384 INFO [train.py:903] (1/4) Epoch 26, batch 3600, loss[loss=0.1607, simple_loss=0.2441, pruned_loss=0.03865, over 19636.00 frames. ], tot_loss[loss=0.2036, simple_loss=0.2846, pruned_loss=0.06133, over 3831076.86 frames. ], batch size: 50, lr: 3.15e-03, grad_scale: 8.0 2023-04-03 07:06:54,849 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=174301.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:07:05,346 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=174309.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:07:46,744 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 2023-04-03 07:07:54,107 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=174347.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:07:57,125 INFO [train.py:903] (1/4) Epoch 26, batch 3650, loss[loss=0.2927, simple_loss=0.3458, pruned_loss=0.1198, over 13091.00 frames. ], tot_loss[loss=0.203, simple_loss=0.2837, pruned_loss=0.06112, over 3826225.18 frames. ], batch size: 136, lr: 3.15e-03, grad_scale: 8.0 2023-04-03 07:08:12,054 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.231e+02 5.064e+02 6.896e+02 8.623e+02 2.807e+03, threshold=1.379e+03, percent-clipped=9.0 2023-04-03 07:08:26,840 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=174372.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:08:32,710 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=174377.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:08:35,144 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=174379.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:09:02,314 INFO [train.py:903] (1/4) Epoch 26, batch 3700, loss[loss=0.2098, simple_loss=0.2947, pruned_loss=0.0625, over 19580.00 frames. ], tot_loss[loss=0.2024, simple_loss=0.2833, pruned_loss=0.06079, over 3837715.84 frames. ], batch size: 61, lr: 3.15e-03, grad_scale: 8.0 2023-04-03 07:09:10,627 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=174406.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:10:07,623 INFO [train.py:903] (1/4) Epoch 26, batch 3750, loss[loss=0.222, simple_loss=0.3002, pruned_loss=0.07197, over 19538.00 frames. ], tot_loss[loss=0.2025, simple_loss=0.2836, pruned_loss=0.06072, over 3828404.21 frames. ], batch size: 64, lr: 3.15e-03, grad_scale: 8.0 2023-04-03 07:10:20,549 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.722e+02 4.995e+02 5.757e+02 7.069e+02 1.518e+03, threshold=1.151e+03, percent-clipped=1.0 2023-04-03 07:11:01,649 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=174492.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:11:04,044 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=174494.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:11:10,959 INFO [train.py:903] (1/4) Epoch 26, batch 3800, loss[loss=0.1652, simple_loss=0.2457, pruned_loss=0.04231, over 19844.00 frames. ], tot_loss[loss=0.2024, simple_loss=0.2837, pruned_loss=0.06057, over 3828466.02 frames. ], batch size: 52, lr: 3.15e-03, grad_scale: 8.0 2023-04-03 07:11:12,311 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=174501.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:11:42,936 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9_rvb from training. Duration: 29.1166875 2023-04-03 07:12:14,058 INFO [train.py:903] (1/4) Epoch 26, batch 3850, loss[loss=0.2594, simple_loss=0.3421, pruned_loss=0.0883, over 19528.00 frames. ], tot_loss[loss=0.2034, simple_loss=0.2843, pruned_loss=0.06131, over 3827210.03 frames. ], batch size: 54, lr: 3.15e-03, grad_scale: 8.0 2023-04-03 07:12:27,531 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.991e+02 5.118e+02 6.382e+02 8.415e+02 1.605e+03, threshold=1.276e+03, percent-clipped=5.0 2023-04-03 07:13:15,072 INFO [train.py:903] (1/4) Epoch 26, batch 3900, loss[loss=0.1916, simple_loss=0.2729, pruned_loss=0.05516, over 19869.00 frames. ], tot_loss[loss=0.2044, simple_loss=0.285, pruned_loss=0.06194, over 3818372.01 frames. ], batch size: 52, lr: 3.15e-03, grad_scale: 8.0 2023-04-03 07:13:29,239 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.9085, 5.0907, 5.7353, 5.7178, 2.1669, 5.4426, 4.6697, 5.4133], device='cuda:1'), covar=tensor([0.1700, 0.0815, 0.0529, 0.0612, 0.6262, 0.0893, 0.0600, 0.1070], device='cuda:1'), in_proj_covar=tensor([0.0813, 0.0775, 0.0978, 0.0866, 0.0856, 0.0745, 0.0584, 0.0911], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-03 07:13:36,311 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=174616.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:14:13,851 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=174645.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:14:20,760 INFO [train.py:903] (1/4) Epoch 26, batch 3950, loss[loss=0.1974, simple_loss=0.2903, pruned_loss=0.05228, over 19315.00 frames. ], tot_loss[loss=0.2036, simple_loss=0.2844, pruned_loss=0.06142, over 3825606.33 frames. ], batch size: 66, lr: 3.15e-03, grad_scale: 8.0 2023-04-03 07:14:24,354 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 2023-04-03 07:14:24,466 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=174653.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:14:34,032 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.720e+02 4.927e+02 6.085e+02 7.650e+02 1.385e+03, threshold=1.217e+03, percent-clipped=3.0 2023-04-03 07:15:24,383 INFO [train.py:903] (1/4) Epoch 26, batch 4000, loss[loss=0.1759, simple_loss=0.2573, pruned_loss=0.04726, over 19789.00 frames. ], tot_loss[loss=0.2035, simple_loss=0.2843, pruned_loss=0.06132, over 3822892.43 frames. ], batch size: 48, lr: 3.15e-03, grad_scale: 8.0 2023-04-03 07:15:54,764 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=174723.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:16:13,009 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1_rvb from training. Duration: 0.7545625 2023-04-03 07:16:16,808 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4393, 1.5352, 1.8312, 1.7201, 2.6844, 2.3390, 2.8341, 1.3976], device='cuda:1'), covar=tensor([0.2519, 0.4293, 0.2682, 0.1905, 0.1486, 0.2076, 0.1425, 0.4346], device='cuda:1'), in_proj_covar=tensor([0.0549, 0.0666, 0.0742, 0.0502, 0.0629, 0.0542, 0.0666, 0.0567], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 07:16:23,821 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4695, 1.5098, 1.8504, 1.7073, 2.6979, 2.3934, 2.8348, 1.2232], device='cuda:1'), covar=tensor([0.2479, 0.4452, 0.2773, 0.1974, 0.1586, 0.2047, 0.1486, 0.4716], device='cuda:1'), in_proj_covar=tensor([0.0548, 0.0665, 0.0741, 0.0502, 0.0629, 0.0542, 0.0665, 0.0567], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 07:16:24,949 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=174748.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:16:27,000 INFO [train.py:903] (1/4) Epoch 26, batch 4050, loss[loss=0.1521, simple_loss=0.2345, pruned_loss=0.03482, over 14636.00 frames. ], tot_loss[loss=0.2031, simple_loss=0.2841, pruned_loss=0.06104, over 3816085.43 frames. ], batch size: 32, lr: 3.15e-03, grad_scale: 8.0 2023-04-03 07:16:27,156 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=174750.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:16:27,412 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=174750.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:16:40,322 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=174760.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:16:41,099 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.277e+02 4.947e+02 6.251e+02 7.600e+02 1.203e+03, threshold=1.250e+03, percent-clipped=0.0 2023-04-03 07:16:52,104 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=174768.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:16:59,292 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=174773.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:17:01,624 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=174775.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:17:32,067 INFO [train.py:903] (1/4) Epoch 26, batch 4100, loss[loss=0.2238, simple_loss=0.3064, pruned_loss=0.07063, over 19664.00 frames. ], tot_loss[loss=0.2024, simple_loss=0.2835, pruned_loss=0.06063, over 3805262.32 frames. ], batch size: 60, lr: 3.15e-03, grad_scale: 4.0 2023-04-03 07:18:07,612 WARNING [train.py:1073] (1/4) Exclude cut with ID 497-129325-0061-9566_sp1.1_rvb from training. Duration: 0.97725 2023-04-03 07:18:22,222 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.3434, 3.0709, 2.4431, 2.7632, 0.8089, 3.0728, 2.9366, 2.9676], device='cuda:1'), covar=tensor([0.1121, 0.1418, 0.1911, 0.1098, 0.4081, 0.1030, 0.1235, 0.1462], device='cuda:1'), in_proj_covar=tensor([0.0521, 0.0421, 0.0506, 0.0354, 0.0409, 0.0447, 0.0444, 0.0471], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 07:18:38,182 INFO [train.py:903] (1/4) Epoch 26, batch 4150, loss[loss=0.1907, simple_loss=0.2711, pruned_loss=0.0551, over 19537.00 frames. ], tot_loss[loss=0.2036, simple_loss=0.2845, pruned_loss=0.06137, over 3804368.77 frames. ], batch size: 56, lr: 3.15e-03, grad_scale: 4.0 2023-04-03 07:18:53,316 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.134e+02 5.046e+02 6.484e+02 8.542e+02 2.236e+03, threshold=1.297e+03, percent-clipped=8.0 2023-04-03 07:18:57,234 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=174865.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:19:05,601 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=174872.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:19:11,135 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.3242, 3.0555, 2.3360, 2.4441, 2.3145, 2.7730, 1.1011, 2.2318], device='cuda:1'), covar=tensor([0.0676, 0.0622, 0.0753, 0.1141, 0.1059, 0.1132, 0.1520, 0.1080], device='cuda:1'), in_proj_covar=tensor([0.0365, 0.0362, 0.0366, 0.0389, 0.0469, 0.0396, 0.0345, 0.0347], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 07:19:38,547 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=174897.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:19:41,589 INFO [train.py:903] (1/4) Epoch 26, batch 4200, loss[loss=0.1847, simple_loss=0.259, pruned_loss=0.05521, over 19358.00 frames. ], tot_loss[loss=0.2008, simple_loss=0.282, pruned_loss=0.0598, over 3813114.18 frames. ], batch size: 47, lr: 3.15e-03, grad_scale: 4.0 2023-04-03 07:19:42,856 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 2023-04-03 07:20:14,720 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0887, 2.8747, 1.8219, 1.8983, 2.6368, 1.7281, 1.6031, 2.2732], device='cuda:1'), covar=tensor([0.1234, 0.0824, 0.1113, 0.0875, 0.0591, 0.1243, 0.0913, 0.0665], device='cuda:1'), in_proj_covar=tensor([0.0304, 0.0320, 0.0339, 0.0272, 0.0250, 0.0345, 0.0293, 0.0276], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 07:20:36,869 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6909, 1.6305, 1.6116, 2.2045, 1.5772, 2.1046, 2.0342, 1.7755], device='cuda:1'), covar=tensor([0.0841, 0.0895, 0.0979, 0.0692, 0.0935, 0.0684, 0.0809, 0.0662], device='cuda:1'), in_proj_covar=tensor([0.0213, 0.0223, 0.0227, 0.0240, 0.0225, 0.0212, 0.0188, 0.0204], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-04-03 07:20:44,729 INFO [train.py:903] (1/4) Epoch 26, batch 4250, loss[loss=0.2349, simple_loss=0.3229, pruned_loss=0.0735, over 19741.00 frames. ], tot_loss[loss=0.2022, simple_loss=0.2837, pruned_loss=0.06028, over 3809316.61 frames. ], batch size: 63, lr: 3.15e-03, grad_scale: 4.0 2023-04-03 07:20:55,195 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9_rvb from training. Duration: 29.816625 2023-04-03 07:21:01,859 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.522e+02 4.804e+02 5.687e+02 7.207e+02 1.586e+03, threshold=1.137e+03, percent-clipped=5.0 2023-04-03 07:21:07,828 WARNING [train.py:1073] (1/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 2023-04-03 07:21:16,333 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 2023-04-03 07:21:30,486 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6935, 1.5718, 1.6844, 2.3080, 1.6183, 1.9924, 2.0398, 1.7926], device='cuda:1'), covar=tensor([0.0874, 0.0980, 0.1007, 0.0723, 0.0899, 0.0801, 0.0876, 0.0704], device='cuda:1'), in_proj_covar=tensor([0.0213, 0.0223, 0.0227, 0.0240, 0.0225, 0.0213, 0.0188, 0.0204], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-04-03 07:21:49,465 INFO [train.py:903] (1/4) Epoch 26, batch 4300, loss[loss=0.1665, simple_loss=0.2408, pruned_loss=0.04616, over 19742.00 frames. ], tot_loss[loss=0.202, simple_loss=0.2834, pruned_loss=0.06033, over 3810570.48 frames. ], batch size: 46, lr: 3.15e-03, grad_scale: 4.0 2023-04-03 07:22:11,471 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=175016.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:22:20,965 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=175024.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:22:38,878 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 2023-04-03 07:22:41,527 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=175041.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:22:52,842 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=175049.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:22:53,607 INFO [train.py:903] (1/4) Epoch 26, batch 4350, loss[loss=0.2015, simple_loss=0.2768, pruned_loss=0.06308, over 19763.00 frames. ], tot_loss[loss=0.2023, simple_loss=0.2831, pruned_loss=0.06073, over 3811737.95 frames. ], batch size: 54, lr: 3.15e-03, grad_scale: 4.0 2023-04-03 07:23:08,536 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.209e+02 4.980e+02 6.586e+02 7.974e+02 2.340e+03, threshold=1.317e+03, percent-clipped=8.0 2023-04-03 07:23:14,390 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=175067.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:23:33,061 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.2765, 2.3156, 2.0052, 2.3837, 2.1252, 2.1267, 1.9551, 2.3653], device='cuda:1'), covar=tensor([0.1100, 0.1533, 0.1564, 0.1110, 0.1572, 0.0543, 0.1520, 0.0758], device='cuda:1'), in_proj_covar=tensor([0.0278, 0.0355, 0.0314, 0.0254, 0.0304, 0.0255, 0.0316, 0.0262], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 07:23:57,001 INFO [train.py:903] (1/4) Epoch 26, batch 4400, loss[loss=0.1981, simple_loss=0.2813, pruned_loss=0.05743, over 18829.00 frames. ], tot_loss[loss=0.2032, simple_loss=0.2839, pruned_loss=0.0612, over 3813711.20 frames. ], batch size: 74, lr: 3.15e-03, grad_scale: 8.0 2023-04-03 07:24:17,244 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_rvb from training. Duration: 25.285 2023-04-03 07:24:23,101 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=175121.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:24:27,366 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 2023-04-03 07:24:41,394 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.9943, 3.6712, 2.5920, 3.2829, 0.8709, 3.6200, 3.4876, 3.5191], device='cuda:1'), covar=tensor([0.0862, 0.1052, 0.1822, 0.0881, 0.3981, 0.0806, 0.1055, 0.1291], device='cuda:1'), in_proj_covar=tensor([0.0524, 0.0423, 0.0509, 0.0356, 0.0409, 0.0449, 0.0447, 0.0473], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 07:24:55,270 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=175146.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:24:59,532 INFO [train.py:903] (1/4) Epoch 26, batch 4450, loss[loss=0.2013, simple_loss=0.2932, pruned_loss=0.05467, over 19682.00 frames. ], tot_loss[loss=0.2033, simple_loss=0.2842, pruned_loss=0.06122, over 3807471.44 frames. ], batch size: 60, lr: 3.14e-03, grad_scale: 8.0 2023-04-03 07:25:08,169 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=175157.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:25:14,967 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.113e+02 4.830e+02 6.413e+02 8.619e+02 2.132e+03, threshold=1.283e+03, percent-clipped=8.0 2023-04-03 07:25:41,735 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=175182.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:26:02,757 INFO [train.py:903] (1/4) Epoch 26, batch 4500, loss[loss=0.2528, simple_loss=0.3187, pruned_loss=0.09348, over 19290.00 frames. ], tot_loss[loss=0.2033, simple_loss=0.284, pruned_loss=0.06135, over 3809193.31 frames. ], batch size: 66, lr: 3.14e-03, grad_scale: 8.0 2023-04-03 07:26:06,570 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1857, 1.8382, 1.4819, 1.2912, 1.6234, 1.2757, 1.1993, 1.6372], device='cuda:1'), covar=tensor([0.0829, 0.0830, 0.1105, 0.0865, 0.0588, 0.1319, 0.0648, 0.0455], device='cuda:1'), in_proj_covar=tensor([0.0303, 0.0319, 0.0338, 0.0271, 0.0249, 0.0344, 0.0292, 0.0273], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 07:26:36,001 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=175225.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:26:38,420 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4086, 1.3242, 1.3654, 1.8209, 1.3435, 1.5591, 1.7349, 1.4587], device='cuda:1'), covar=tensor([0.0904, 0.0986, 0.1072, 0.0653, 0.0905, 0.0818, 0.0816, 0.0769], device='cuda:1'), in_proj_covar=tensor([0.0215, 0.0225, 0.0229, 0.0242, 0.0227, 0.0214, 0.0190, 0.0206], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-04-03 07:27:08,199 INFO [train.py:903] (1/4) Epoch 26, batch 4550, loss[loss=0.207, simple_loss=0.3046, pruned_loss=0.05468, over 19611.00 frames. ], tot_loss[loss=0.204, simple_loss=0.2851, pruned_loss=0.06148, over 3825496.64 frames. ], batch size: 57, lr: 3.14e-03, grad_scale: 8.0 2023-04-03 07:27:15,027 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9_rvb from training. Duration: 28.72225 2023-04-03 07:27:23,519 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.068e+02 4.834e+02 6.204e+02 7.660e+02 2.009e+03, threshold=1.241e+03, percent-clipped=3.0 2023-04-03 07:27:28,795 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5067, 1.4983, 1.7882, 1.7579, 2.6468, 2.2693, 2.8187, 1.1402], device='cuda:1'), covar=tensor([0.2504, 0.4569, 0.2834, 0.2009, 0.1469, 0.2261, 0.1418, 0.4736], device='cuda:1'), in_proj_covar=tensor([0.0545, 0.0661, 0.0738, 0.0500, 0.0626, 0.0540, 0.0662, 0.0564], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 07:27:40,258 WARNING [train.py:1073] (1/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 2023-04-03 07:28:13,307 INFO [train.py:903] (1/4) Epoch 26, batch 4600, loss[loss=0.1807, simple_loss=0.2758, pruned_loss=0.04284, over 19616.00 frames. ], tot_loss[loss=0.2028, simple_loss=0.2842, pruned_loss=0.06069, over 3834014.52 frames. ], batch size: 57, lr: 3.14e-03, grad_scale: 8.0 2023-04-03 07:29:16,039 INFO [train.py:903] (1/4) Epoch 26, batch 4650, loss[loss=0.1865, simple_loss=0.2734, pruned_loss=0.04976, over 19691.00 frames. ], tot_loss[loss=0.2018, simple_loss=0.2833, pruned_loss=0.06013, over 3837562.53 frames. ], batch size: 59, lr: 3.14e-03, grad_scale: 8.0 2023-04-03 07:29:29,903 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.676e+02 4.879e+02 5.869e+02 7.462e+02 1.692e+03, threshold=1.174e+03, percent-clipped=3.0 2023-04-03 07:29:31,145 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_rvb from training. Duration: 0.92 2023-04-03 07:29:44,489 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_rvb from training. Duration: 0.83 2023-04-03 07:30:00,525 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.0389, 1.2710, 1.4832, 1.4094, 2.5732, 1.1941, 2.2243, 3.0258], device='cuda:1'), covar=tensor([0.0739, 0.3206, 0.3094, 0.2062, 0.1017, 0.2575, 0.1382, 0.0406], device='cuda:1'), in_proj_covar=tensor([0.0418, 0.0372, 0.0392, 0.0350, 0.0380, 0.0355, 0.0391, 0.0412], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 07:30:03,972 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=175388.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:30:17,583 INFO [train.py:903] (1/4) Epoch 26, batch 4700, loss[loss=0.1969, simple_loss=0.2732, pruned_loss=0.06029, over 19373.00 frames. ], tot_loss[loss=0.2033, simple_loss=0.2844, pruned_loss=0.0611, over 3816551.02 frames. ], batch size: 47, lr: 3.14e-03, grad_scale: 8.0 2023-04-03 07:30:35,694 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=175412.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:30:36,042 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.75 vs. limit=5.0 2023-04-03 07:30:40,246 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2036, 1.2921, 1.2559, 1.0524, 1.0771, 1.1241, 0.0920, 0.3719], device='cuda:1'), covar=tensor([0.0714, 0.0710, 0.0477, 0.0651, 0.1388, 0.0646, 0.1417, 0.1211], device='cuda:1'), in_proj_covar=tensor([0.0363, 0.0362, 0.0367, 0.0389, 0.0468, 0.0395, 0.0344, 0.0346], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 07:30:41,001 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9_rvb from training. Duration: 0.92225 2023-04-03 07:30:43,610 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.2085, 1.9914, 2.1241, 2.5926, 2.1776, 2.3442, 2.3709, 2.1569], device='cuda:1'), covar=tensor([0.0612, 0.0719, 0.0723, 0.0584, 0.0807, 0.0579, 0.0728, 0.0558], device='cuda:1'), in_proj_covar=tensor([0.0213, 0.0224, 0.0228, 0.0240, 0.0226, 0.0213, 0.0188, 0.0205], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-04-03 07:31:07,022 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=175438.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:31:18,763 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.8148, 1.5166, 1.4291, 1.7686, 1.4471, 1.5603, 1.4417, 1.6800], device='cuda:1'), covar=tensor([0.1131, 0.1362, 0.1681, 0.1100, 0.1383, 0.0630, 0.1614, 0.0868], device='cuda:1'), in_proj_covar=tensor([0.0277, 0.0356, 0.0315, 0.0255, 0.0303, 0.0255, 0.0317, 0.0262], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 07:31:21,522 INFO [train.py:903] (1/4) Epoch 26, batch 4750, loss[loss=0.2131, simple_loss=0.2944, pruned_loss=0.0659, over 19685.00 frames. ], tot_loss[loss=0.2035, simple_loss=0.2847, pruned_loss=0.06116, over 3824308.04 frames. ], batch size: 60, lr: 3.14e-03, grad_scale: 8.0 2023-04-03 07:31:37,186 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.357e+02 4.945e+02 6.030e+02 7.655e+02 2.128e+03, threshold=1.206e+03, percent-clipped=6.0 2023-04-03 07:31:38,753 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=175463.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:32:22,893 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=175499.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:32:23,789 INFO [train.py:903] (1/4) Epoch 26, batch 4800, loss[loss=0.1919, simple_loss=0.273, pruned_loss=0.05534, over 19769.00 frames. ], tot_loss[loss=0.2035, simple_loss=0.2847, pruned_loss=0.06119, over 3818026.98 frames. ], batch size: 56, lr: 3.14e-03, grad_scale: 8.0 2023-04-03 07:32:25,139 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=175501.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:32:29,427 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0 2023-04-03 07:32:43,064 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.0369, 1.2598, 1.6466, 0.9261, 2.2802, 3.0470, 2.7794, 3.2556], device='cuda:1'), covar=tensor([0.1693, 0.4010, 0.3523, 0.2795, 0.0666, 0.0223, 0.0276, 0.0334], device='cuda:1'), in_proj_covar=tensor([0.0278, 0.0329, 0.0363, 0.0270, 0.0253, 0.0195, 0.0219, 0.0272], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-03 07:33:28,511 INFO [train.py:903] (1/4) Epoch 26, batch 4850, loss[loss=0.1785, simple_loss=0.2633, pruned_loss=0.04687, over 19836.00 frames. ], tot_loss[loss=0.2037, simple_loss=0.2849, pruned_loss=0.06124, over 3834807.96 frames. ], batch size: 52, lr: 3.14e-03, grad_scale: 8.0 2023-04-03 07:33:42,551 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.071e+02 4.553e+02 5.462e+02 6.461e+02 1.537e+03, threshold=1.092e+03, percent-clipped=2.0 2023-04-03 07:33:49,696 WARNING [train.py:1073] (1/4) Exclude cut with ID 774-127930-0014-48411_sp1.1_rvb from training. Duration: 0.95 2023-04-03 07:33:51,022 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=175569.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:34:12,009 WARNING [train.py:1073] (1/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1_rvb from training. Duration: 0.9409375 2023-04-03 07:34:17,974 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 2023-04-03 07:34:17,999 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 2023-04-03 07:34:19,635 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4086, 1.4601, 1.7670, 1.6544, 2.5122, 2.0731, 2.6330, 1.0714], device='cuda:1'), covar=tensor([0.2601, 0.4485, 0.2792, 0.2026, 0.1591, 0.2391, 0.1454, 0.4831], device='cuda:1'), in_proj_covar=tensor([0.0547, 0.0662, 0.0741, 0.0503, 0.0629, 0.0542, 0.0664, 0.0565], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 07:34:27,164 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1_rvb from training. Duration: 25.3818125 2023-04-03 07:34:30,597 INFO [train.py:903] (1/4) Epoch 26, batch 4900, loss[loss=0.1744, simple_loss=0.2586, pruned_loss=0.04512, over 19753.00 frames. ], tot_loss[loss=0.2039, simple_loss=0.285, pruned_loss=0.06143, over 3823492.74 frames. ], batch size: 51, lr: 3.14e-03, grad_scale: 8.0 2023-04-03 07:34:46,808 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 2023-04-03 07:34:50,728 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=175616.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:35:31,952 INFO [train.py:903] (1/4) Epoch 26, batch 4950, loss[loss=0.1994, simple_loss=0.2751, pruned_loss=0.06184, over 19484.00 frames. ], tot_loss[loss=0.2035, simple_loss=0.2844, pruned_loss=0.06129, over 3821513.17 frames. ], batch size: 49, lr: 3.14e-03, grad_scale: 8.0 2023-04-03 07:35:49,946 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.056e+02 4.878e+02 6.108e+02 7.489e+02 1.803e+03, threshold=1.222e+03, percent-clipped=10.0 2023-04-03 07:35:49,998 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 2023-04-03 07:36:13,145 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 2023-04-03 07:36:15,582 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=175684.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:36:36,317 INFO [train.py:903] (1/4) Epoch 26, batch 5000, loss[loss=0.2491, simple_loss=0.3175, pruned_loss=0.09032, over 12834.00 frames. ], tot_loss[loss=0.2031, simple_loss=0.2841, pruned_loss=0.06099, over 3818104.99 frames. ], batch size: 136, lr: 3.14e-03, grad_scale: 8.0 2023-04-03 07:36:46,117 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_rvb from training. Duration: 27.14 2023-04-03 07:36:56,359 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 2023-04-03 07:37:10,893 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=175728.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:37:16,379 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=175732.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:37:33,271 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.43 vs. limit=2.0 2023-04-03 07:37:39,577 INFO [train.py:903] (1/4) Epoch 26, batch 5050, loss[loss=0.1818, simple_loss=0.2633, pruned_loss=0.05015, over 19770.00 frames. ], tot_loss[loss=0.2025, simple_loss=0.2835, pruned_loss=0.06073, over 3818437.15 frames. ], batch size: 47, lr: 3.14e-03, grad_scale: 8.0 2023-04-03 07:37:46,797 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=175756.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:37:53,679 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.913e+02 4.916e+02 5.717e+02 6.994e+02 1.273e+03, threshold=1.143e+03, percent-clipped=1.0 2023-04-03 07:38:14,300 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9_rvb from training. Duration: 26.62775 2023-04-03 07:38:37,156 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.72 vs. limit=2.0 2023-04-03 07:38:42,432 INFO [train.py:903] (1/4) Epoch 26, batch 5100, loss[loss=0.2093, simple_loss=0.2873, pruned_loss=0.06566, over 19407.00 frames. ], tot_loss[loss=0.204, simple_loss=0.2848, pruned_loss=0.06158, over 3813796.65 frames. ], batch size: 48, lr: 3.14e-03, grad_scale: 8.0 2023-04-03 07:38:44,298 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9207, 2.0468, 2.3070, 2.5893, 1.9920, 2.5149, 2.2714, 2.1010], device='cuda:1'), covar=tensor([0.4317, 0.4113, 0.2002, 0.2480, 0.4318, 0.2278, 0.5034, 0.3558], device='cuda:1'), in_proj_covar=tensor([0.0928, 0.1004, 0.0737, 0.0949, 0.0907, 0.0843, 0.0857, 0.0803], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-03 07:38:49,478 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 2023-04-03 07:38:49,795 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.1031, 2.7112, 2.5043, 3.0020, 2.6939, 2.4971, 2.3285, 2.9718], device='cuda:1'), covar=tensor([0.0799, 0.1422, 0.1409, 0.0974, 0.1307, 0.0489, 0.1377, 0.0597], device='cuda:1'), in_proj_covar=tensor([0.0280, 0.0360, 0.0318, 0.0257, 0.0307, 0.0259, 0.0321, 0.0265], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 07:38:52,913 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_rvb from training. Duration: 25.035 2023-04-03 07:38:57,493 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_rvb from training. Duration: 27.92 2023-04-03 07:39:37,110 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=175843.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:39:40,217 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.46 vs. limit=2.0 2023-04-03 07:39:41,975 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=175847.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:39:45,159 INFO [train.py:903] (1/4) Epoch 26, batch 5150, loss[loss=0.1661, simple_loss=0.2527, pruned_loss=0.03974, over 19767.00 frames. ], tot_loss[loss=0.2048, simple_loss=0.2857, pruned_loss=0.06196, over 3804878.15 frames. ], batch size: 54, lr: 3.14e-03, grad_scale: 8.0 2023-04-03 07:39:55,549 WARNING [train.py:1073] (1/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 2023-04-03 07:39:58,800 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.82 vs. limit=2.0 2023-04-03 07:40:01,362 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.182e+02 5.003e+02 6.332e+02 8.398e+02 1.509e+03, threshold=1.266e+03, percent-clipped=8.0 2023-04-03 07:40:14,018 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=175871.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:40:15,270 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=175872.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:40:30,109 WARNING [train.py:1073] (1/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1_rvb from training. Duration: 0.9681875 2023-04-03 07:40:47,068 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=175897.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:40:49,313 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=175899.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:40:50,207 INFO [train.py:903] (1/4) Epoch 26, batch 5200, loss[loss=0.1836, simple_loss=0.2762, pruned_loss=0.04544, over 19290.00 frames. ], tot_loss[loss=0.2051, simple_loss=0.2861, pruned_loss=0.06207, over 3799210.57 frames. ], batch size: 66, lr: 3.14e-03, grad_scale: 8.0 2023-04-03 07:41:02,260 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 2023-04-03 07:41:12,161 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0813, 2.0037, 1.9175, 1.8088, 1.5618, 1.7290, 0.6602, 1.0897], device='cuda:1'), covar=tensor([0.0632, 0.0668, 0.0486, 0.0768, 0.1305, 0.0975, 0.1413, 0.1125], device='cuda:1'), in_proj_covar=tensor([0.0363, 0.0361, 0.0366, 0.0389, 0.0469, 0.0396, 0.0344, 0.0346], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 07:41:41,498 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=175940.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:41:45,595 WARNING [train.py:1073] (1/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 2023-04-03 07:41:53,683 INFO [train.py:903] (1/4) Epoch 26, batch 5250, loss[loss=0.2399, simple_loss=0.314, pruned_loss=0.0829, over 18862.00 frames. ], tot_loss[loss=0.2058, simple_loss=0.2868, pruned_loss=0.06234, over 3805653.12 frames. ], batch size: 74, lr: 3.14e-03, grad_scale: 8.0 2023-04-03 07:42:03,331 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=175958.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:42:07,497 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.944e+02 4.675e+02 5.849e+02 7.641e+02 1.436e+03, threshold=1.170e+03, percent-clipped=2.0 2023-04-03 07:42:08,950 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=175963.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:42:11,312 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=175965.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:42:54,948 INFO [train.py:903] (1/4) Epoch 26, batch 5300, loss[loss=0.2313, simple_loss=0.3146, pruned_loss=0.07399, over 19518.00 frames. ], tot_loss[loss=0.2056, simple_loss=0.2865, pruned_loss=0.06234, over 3814866.40 frames. ], batch size: 54, lr: 3.14e-03, grad_scale: 8.0 2023-04-03 07:43:08,900 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 2023-04-03 07:43:57,533 INFO [train.py:903] (1/4) Epoch 26, batch 5350, loss[loss=0.2761, simple_loss=0.34, pruned_loss=0.1061, over 12762.00 frames. ], tot_loss[loss=0.2059, simple_loss=0.2864, pruned_loss=0.06268, over 3806218.20 frames. ], batch size: 136, lr: 3.14e-03, grad_scale: 8.0 2023-04-03 07:44:14,891 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.150e+02 4.809e+02 5.990e+02 7.353e+02 1.688e+03, threshold=1.198e+03, percent-clipped=9.0 2023-04-03 07:44:27,815 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=176072.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:44:30,135 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9_rvb from training. Duration: 26.438875 2023-04-03 07:45:03,470 INFO [train.py:903] (1/4) Epoch 26, batch 5400, loss[loss=0.2102, simple_loss=0.2942, pruned_loss=0.06308, over 19594.00 frames. ], tot_loss[loss=0.2048, simple_loss=0.2855, pruned_loss=0.062, over 3804257.35 frames. ], batch size: 52, lr: 3.14e-03, grad_scale: 4.0 2023-04-03 07:45:07,733 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=176103.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:45:36,747 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=176127.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:45:37,877 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=176128.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:46:08,011 INFO [train.py:903] (1/4) Epoch 26, batch 5450, loss[loss=0.2085, simple_loss=0.2924, pruned_loss=0.06227, over 19672.00 frames. ], tot_loss[loss=0.2041, simple_loss=0.2848, pruned_loss=0.06173, over 3820350.70 frames. ], batch size: 53, lr: 3.14e-03, grad_scale: 4.0 2023-04-03 07:46:10,684 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=176152.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:46:23,252 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.815e+02 4.882e+02 5.855e+02 6.912e+02 1.680e+03, threshold=1.171e+03, percent-clipped=1.0 2023-04-03 07:46:55,112 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=176187.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:47:11,318 INFO [train.py:903] (1/4) Epoch 26, batch 5500, loss[loss=0.2045, simple_loss=0.2784, pruned_loss=0.06527, over 19713.00 frames. ], tot_loss[loss=0.2049, simple_loss=0.2857, pruned_loss=0.06208, over 3813439.09 frames. ], batch size: 51, lr: 3.14e-03, grad_scale: 4.0 2023-04-03 07:47:28,795 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=176214.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:47:30,704 WARNING [train.py:1073] (1/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9_rvb from training. Duration: 27.47775 2023-04-03 07:47:48,449 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.1297, 1.9093, 1.7932, 2.0610, 1.8230, 1.8323, 1.7081, 2.0457], device='cuda:1'), covar=tensor([0.1086, 0.1484, 0.1555, 0.1128, 0.1416, 0.0582, 0.1591, 0.0750], device='cuda:1'), in_proj_covar=tensor([0.0277, 0.0355, 0.0315, 0.0255, 0.0304, 0.0255, 0.0317, 0.0262], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 07:48:02,309 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=176239.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:48:06,792 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=176243.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:48:14,672 INFO [train.py:903] (1/4) Epoch 26, batch 5550, loss[loss=0.2115, simple_loss=0.2883, pruned_loss=0.06733, over 19772.00 frames. ], tot_loss[loss=0.2057, simple_loss=0.2862, pruned_loss=0.06256, over 3807155.59 frames. ], batch size: 54, lr: 3.14e-03, grad_scale: 4.0 2023-04-03 07:48:17,143 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 2023-04-03 07:48:29,063 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=176260.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:48:32,237 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.168e+02 4.802e+02 5.930e+02 7.241e+02 1.738e+03, threshold=1.186e+03, percent-clipped=4.0 2023-04-03 07:48:37,816 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.10 vs. limit=2.0 2023-04-03 07:48:47,672 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=176275.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:49:08,010 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 2023-04-03 07:49:18,894 INFO [train.py:903] (1/4) Epoch 26, batch 5600, loss[loss=0.1864, simple_loss=0.2753, pruned_loss=0.04872, over 19744.00 frames. ], tot_loss[loss=0.2047, simple_loss=0.2855, pruned_loss=0.06198, over 3809389.93 frames. ], batch size: 63, lr: 3.13e-03, grad_scale: 8.0 2023-04-03 07:49:28,226 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=176307.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:49:48,579 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.49 vs. limit=2.0 2023-04-03 07:50:23,052 INFO [train.py:903] (1/4) Epoch 26, batch 5650, loss[loss=0.1855, simple_loss=0.2673, pruned_loss=0.05182, over 19726.00 frames. ], tot_loss[loss=0.2053, simple_loss=0.2862, pruned_loss=0.06221, over 3807757.57 frames. ], batch size: 51, lr: 3.13e-03, grad_scale: 4.0 2023-04-03 07:50:26,195 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.14 vs. limit=2.0 2023-04-03 07:50:32,669 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=176358.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:50:39,208 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.976e+02 4.381e+02 5.619e+02 7.137e+02 2.187e+03, threshold=1.124e+03, percent-clipped=4.0 2023-04-03 07:51:02,581 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9_rvb from training. Duration: 28.0944375 2023-04-03 07:51:24,798 INFO [train.py:903] (1/4) Epoch 26, batch 5700, loss[loss=0.2147, simple_loss=0.2963, pruned_loss=0.06657, over 19537.00 frames. ], tot_loss[loss=0.206, simple_loss=0.2867, pruned_loss=0.06263, over 3805528.39 frames. ], batch size: 54, lr: 3.13e-03, grad_scale: 4.0 2023-04-03 07:51:52,416 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=176422.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:52:19,138 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=176443.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:52:22,361 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9_rvb from training. Duration: 33.038875 2023-04-03 07:52:26,807 INFO [train.py:903] (1/4) Epoch 26, batch 5750, loss[loss=0.1756, simple_loss=0.2458, pruned_loss=0.05273, over 18654.00 frames. ], tot_loss[loss=0.2047, simple_loss=0.2849, pruned_loss=0.0622, over 3803336.86 frames. ], batch size: 41, lr: 3.13e-03, grad_scale: 4.0 2023-04-03 07:52:30,362 WARNING [train.py:1073] (1/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 2023-04-03 07:52:33,963 WARNING [train.py:1073] (1/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 2023-04-03 07:52:44,005 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.352e+02 5.112e+02 6.171e+02 7.862e+02 1.795e+03, threshold=1.234e+03, percent-clipped=7.0 2023-04-03 07:52:50,040 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=176468.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:53:09,531 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=176484.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:53:28,519 INFO [train.py:903] (1/4) Epoch 26, batch 5800, loss[loss=0.2344, simple_loss=0.3078, pruned_loss=0.08043, over 19683.00 frames. ], tot_loss[loss=0.2042, simple_loss=0.2844, pruned_loss=0.06198, over 3806824.82 frames. ], batch size: 59, lr: 3.13e-03, grad_scale: 4.0 2023-04-03 07:53:57,698 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.60 vs. limit=2.0 2023-04-03 07:54:32,079 INFO [train.py:903] (1/4) Epoch 26, batch 5850, loss[loss=0.1786, simple_loss=0.2587, pruned_loss=0.04923, over 19726.00 frames. ], tot_loss[loss=0.203, simple_loss=0.2832, pruned_loss=0.06142, over 3820429.63 frames. ], batch size: 51, lr: 3.13e-03, grad_scale: 4.0 2023-04-03 07:54:48,249 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.299e+02 4.827e+02 6.114e+02 8.553e+02 2.097e+03, threshold=1.223e+03, percent-clipped=6.0 2023-04-03 07:55:29,979 WARNING [train.py:1073] (1/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 2023-04-03 07:55:33,108 INFO [train.py:903] (1/4) Epoch 26, batch 5900, loss[loss=0.1985, simple_loss=0.2839, pruned_loss=0.05657, over 19593.00 frames. ], tot_loss[loss=0.2037, simple_loss=0.2841, pruned_loss=0.06171, over 3827066.28 frames. ], batch size: 61, lr: 3.13e-03, grad_scale: 4.0 2023-04-03 07:55:37,898 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=176604.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:55:50,019 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=176614.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:55:51,779 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9_rvb from training. Duration: 27.511125 2023-04-03 07:55:55,452 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=176619.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:56:22,773 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=176639.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:56:27,331 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=176643.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:56:35,021 INFO [train.py:903] (1/4) Epoch 26, batch 5950, loss[loss=0.1694, simple_loss=0.2478, pruned_loss=0.04549, over 19733.00 frames. ], tot_loss[loss=0.204, simple_loss=0.2845, pruned_loss=0.06171, over 3834305.91 frames. ], batch size: 47, lr: 3.13e-03, grad_scale: 4.0 2023-04-03 07:56:51,411 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.421e+02 5.079e+02 6.315e+02 7.489e+02 1.732e+03, threshold=1.263e+03, percent-clipped=4.0 2023-04-03 07:57:12,913 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=176678.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:57:18,431 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=176683.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:57:38,447 INFO [train.py:903] (1/4) Epoch 26, batch 6000, loss[loss=0.169, simple_loss=0.2586, pruned_loss=0.03976, over 19841.00 frames. ], tot_loss[loss=0.2031, simple_loss=0.2839, pruned_loss=0.06117, over 3826537.95 frames. ], batch size: 52, lr: 3.13e-03, grad_scale: 8.0 2023-04-03 07:57:38,448 INFO [train.py:928] (1/4) Computing validation loss 2023-04-03 07:57:51,359 INFO [train.py:937] (1/4) Epoch 26, validation: loss=0.1675, simple_loss=0.2672, pruned_loss=0.03393, over 944034.00 frames. 2023-04-03 07:57:51,360 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18821MB 2023-04-03 07:57:55,496 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=176703.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:58:16,475 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=176719.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:58:36,066 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=176734.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 07:58:54,245 INFO [train.py:903] (1/4) Epoch 26, batch 6050, loss[loss=0.2059, simple_loss=0.2989, pruned_loss=0.05648, over 19688.00 frames. ], tot_loss[loss=0.2034, simple_loss=0.2842, pruned_loss=0.06127, over 3807733.64 frames. ], batch size: 59, lr: 3.13e-03, grad_scale: 8.0 2023-04-03 07:59:11,767 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.381e+02 5.094e+02 6.239e+02 7.368e+02 1.384e+03, threshold=1.248e+03, percent-clipped=1.0 2023-04-03 07:59:28,834 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 2023-04-03 07:59:57,922 INFO [train.py:903] (1/4) Epoch 26, batch 6100, loss[loss=0.2035, simple_loss=0.2872, pruned_loss=0.05988, over 19355.00 frames. ], tot_loss[loss=0.2026, simple_loss=0.2837, pruned_loss=0.06076, over 3816456.02 frames. ], batch size: 66, lr: 3.13e-03, grad_scale: 8.0 2023-04-03 08:00:03,356 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.27 vs. limit=5.0 2023-04-03 08:00:33,075 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=176828.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 08:01:00,302 INFO [train.py:903] (1/4) Epoch 26, batch 6150, loss[loss=0.2309, simple_loss=0.3073, pruned_loss=0.07722, over 19479.00 frames. ], tot_loss[loss=0.2038, simple_loss=0.2844, pruned_loss=0.06167, over 3806903.40 frames. ], batch size: 64, lr: 3.13e-03, grad_scale: 4.0 2023-04-03 08:01:18,087 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.177e+02 4.899e+02 5.851e+02 7.446e+02 2.190e+03, threshold=1.170e+03, percent-clipped=4.0 2023-04-03 08:01:21,486 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9_rvb from training. Duration: 31.02225 2023-04-03 08:01:25,351 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=176871.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 08:02:01,464 INFO [train.py:903] (1/4) Epoch 26, batch 6200, loss[loss=0.2216, simple_loss=0.2822, pruned_loss=0.08048, over 16436.00 frames. ], tot_loss[loss=0.2033, simple_loss=0.2835, pruned_loss=0.06155, over 3809325.13 frames. ], batch size: 36, lr: 3.13e-03, grad_scale: 4.0 2023-04-03 08:02:54,333 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=176943.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 08:03:02,383 INFO [train.py:903] (1/4) Epoch 26, batch 6250, loss[loss=0.234, simple_loss=0.3052, pruned_loss=0.08136, over 19740.00 frames. ], tot_loss[loss=0.2051, simple_loss=0.2855, pruned_loss=0.06234, over 3815157.84 frames. ], batch size: 54, lr: 3.13e-03, grad_scale: 4.0 2023-04-03 08:03:20,781 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.905e+02 4.899e+02 6.025e+02 7.517e+02 2.005e+03, threshold=1.205e+03, percent-clipped=5.0 2023-04-03 08:03:29,821 WARNING [train.py:1073] (1/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9_rvb from training. Duration: 25.988875 2023-04-03 08:03:33,668 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=176975.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 08:03:47,539 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=176987.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 08:03:51,237 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=176990.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 08:04:03,339 INFO [train.py:903] (1/4) Epoch 26, batch 6300, loss[loss=0.2075, simple_loss=0.294, pruned_loss=0.06046, over 19290.00 frames. ], tot_loss[loss=0.2059, simple_loss=0.2865, pruned_loss=0.06264, over 3812746.91 frames. ], batch size: 66, lr: 3.13e-03, grad_scale: 4.0 2023-04-03 08:04:03,955 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=177000.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 08:04:23,642 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=177015.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 08:04:37,551 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=177027.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 08:04:40,417 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.71 vs. limit=2.0 2023-04-03 08:05:06,346 INFO [train.py:903] (1/4) Epoch 26, batch 6350, loss[loss=0.2, simple_loss=0.2876, pruned_loss=0.05614, over 19527.00 frames. ], tot_loss[loss=0.2053, simple_loss=0.2858, pruned_loss=0.06239, over 3794026.48 frames. ], batch size: 54, lr: 3.13e-03, grad_scale: 4.0 2023-04-03 08:05:14,765 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.8140, 1.9039, 2.1143, 2.2998, 1.7381, 2.2070, 2.1222, 1.9391], device='cuda:1'), covar=tensor([0.4305, 0.3955, 0.2129, 0.2557, 0.4045, 0.2349, 0.4991, 0.3612], device='cuda:1'), in_proj_covar=tensor([0.0930, 0.1008, 0.0739, 0.0952, 0.0909, 0.0845, 0.0861, 0.0806], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-03 08:05:26,132 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.448e+02 5.061e+02 6.104e+02 7.524e+02 1.291e+03, threshold=1.221e+03, percent-clipped=2.0 2023-04-03 08:06:11,969 INFO [train.py:903] (1/4) Epoch 26, batch 6400, loss[loss=0.2047, simple_loss=0.2982, pruned_loss=0.05556, over 19611.00 frames. ], tot_loss[loss=0.2046, simple_loss=0.2854, pruned_loss=0.06189, over 3799248.90 frames. ], batch size: 57, lr: 3.13e-03, grad_scale: 8.0 2023-04-03 08:06:14,727 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=177102.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 08:06:45,641 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=177127.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 08:06:48,154 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=177129.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 08:07:05,548 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=177142.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 08:07:14,951 INFO [train.py:903] (1/4) Epoch 26, batch 6450, loss[loss=0.2012, simple_loss=0.2849, pruned_loss=0.05874, over 19756.00 frames. ], tot_loss[loss=0.2039, simple_loss=0.2851, pruned_loss=0.06133, over 3802594.05 frames. ], batch size: 56, lr: 3.13e-03, grad_scale: 8.0 2023-04-03 08:07:33,748 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.467e+02 4.651e+02 5.846e+02 7.696e+02 2.286e+03, threshold=1.169e+03, percent-clipped=3.0 2023-04-03 08:07:56,648 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9_rvb from training. Duration: 28.638875 2023-04-03 08:08:15,579 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=177199.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 08:08:16,324 INFO [train.py:903] (1/4) Epoch 26, batch 6500, loss[loss=0.2646, simple_loss=0.3283, pruned_loss=0.1004, over 13247.00 frames. ], tot_loss[loss=0.2032, simple_loss=0.2844, pruned_loss=0.06095, over 3809077.48 frames. ], batch size: 135, lr: 3.13e-03, grad_scale: 8.0 2023-04-03 08:08:17,671 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1_rvb from training. Duration: 0.836375 2023-04-03 08:08:38,037 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=177215.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 08:08:49,213 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=177224.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 08:09:20,907 INFO [train.py:903] (1/4) Epoch 26, batch 6550, loss[loss=0.2157, simple_loss=0.3033, pruned_loss=0.06406, over 19670.00 frames. ], tot_loss[loss=0.2023, simple_loss=0.2836, pruned_loss=0.06047, over 3825022.51 frames. ], batch size: 60, lr: 3.13e-03, grad_scale: 8.0 2023-04-03 08:09:28,586 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.5286, 2.1566, 2.2863, 3.0892, 2.0152, 2.7161, 2.5214, 2.4401], device='cuda:1'), covar=tensor([0.0722, 0.0859, 0.0865, 0.0710, 0.0919, 0.0721, 0.0884, 0.0616], device='cuda:1'), in_proj_covar=tensor([0.0212, 0.0224, 0.0227, 0.0239, 0.0225, 0.0212, 0.0188, 0.0205], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-04-03 08:09:39,901 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.007e+02 4.796e+02 6.270e+02 7.966e+02 1.683e+03, threshold=1.254e+03, percent-clipped=5.0 2023-04-03 08:10:25,210 INFO [train.py:903] (1/4) Epoch 26, batch 6600, loss[loss=0.2245, simple_loss=0.2999, pruned_loss=0.07454, over 18354.00 frames. ], tot_loss[loss=0.2033, simple_loss=0.2848, pruned_loss=0.06087, over 3834214.27 frames. ], batch size: 83, lr: 3.13e-03, grad_scale: 8.0 2023-04-03 08:11:02,477 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=177330.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 08:11:27,590 INFO [train.py:903] (1/4) Epoch 26, batch 6650, loss[loss=0.1816, simple_loss=0.2675, pruned_loss=0.04785, over 19623.00 frames. ], tot_loss[loss=0.2034, simple_loss=0.2847, pruned_loss=0.06101, over 3820250.08 frames. ], batch size: 50, lr: 3.13e-03, grad_scale: 8.0 2023-04-03 08:11:37,298 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=177358.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 08:11:47,334 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.622e+02 4.339e+02 5.694e+02 7.782e+02 1.307e+03, threshold=1.139e+03, percent-clipped=1.0 2023-04-03 08:12:11,067 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=177383.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 08:12:28,604 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=177398.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 08:12:30,470 INFO [train.py:903] (1/4) Epoch 26, batch 6700, loss[loss=0.2638, simple_loss=0.3302, pruned_loss=0.09867, over 13692.00 frames. ], tot_loss[loss=0.2026, simple_loss=0.284, pruned_loss=0.06058, over 3820131.06 frames. ], batch size: 136, lr: 3.12e-03, grad_scale: 8.0 2023-04-03 08:12:49,345 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4283, 1.4860, 1.6748, 1.6277, 2.2331, 2.0786, 2.3386, 0.9825], device='cuda:1'), covar=tensor([0.2593, 0.4444, 0.2755, 0.2047, 0.1645, 0.2357, 0.1504, 0.4821], device='cuda:1'), in_proj_covar=tensor([0.0549, 0.0665, 0.0743, 0.0502, 0.0630, 0.0543, 0.0664, 0.0567], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 08:13:01,365 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=177423.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 08:13:02,832 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 2023-04-03 08:13:31,550 INFO [train.py:903] (1/4) Epoch 26, batch 6750, loss[loss=0.18, simple_loss=0.2604, pruned_loss=0.04976, over 19750.00 frames. ], tot_loss[loss=0.2025, simple_loss=0.284, pruned_loss=0.06054, over 3825244.50 frames. ], batch size: 51, lr: 3.12e-03, grad_scale: 8.0 2023-04-03 08:13:48,518 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.769e+02 4.721e+02 5.880e+02 7.244e+02 1.873e+03, threshold=1.176e+03, percent-clipped=5.0 2023-04-03 08:13:55,565 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=177471.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 08:13:57,770 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=177473.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 08:14:07,413 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 2023-04-03 08:14:28,174 INFO [train.py:903] (1/4) Epoch 26, batch 6800, loss[loss=0.1945, simple_loss=0.2868, pruned_loss=0.05107, over 19740.00 frames. ], tot_loss[loss=0.2023, simple_loss=0.2839, pruned_loss=0.06037, over 3821418.00 frames. ], batch size: 63, lr: 3.12e-03, grad_scale: 8.0 2023-04-03 08:15:14,924 WARNING [train.py:1073] (1/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 2023-04-03 08:15:15,415 WARNING [train.py:1073] (1/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 2023-04-03 08:15:18,382 INFO [train.py:903] (1/4) Epoch 27, batch 0, loss[loss=0.1994, simple_loss=0.2691, pruned_loss=0.0648, over 18621.00 frames. ], tot_loss[loss=0.1994, simple_loss=0.2691, pruned_loss=0.0648, over 18621.00 frames. ], batch size: 41, lr: 3.06e-03, grad_scale: 8.0 2023-04-03 08:15:18,383 INFO [train.py:928] (1/4) Computing validation loss 2023-04-03 08:15:30,265 INFO [train.py:937] (1/4) Epoch 27, validation: loss=0.1666, simple_loss=0.2668, pruned_loss=0.03317, over 944034.00 frames. 2023-04-03 08:15:30,266 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18821MB 2023-04-03 08:15:35,648 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9483, 1.8870, 1.7744, 1.6160, 1.4138, 1.5531, 0.4871, 0.8582], device='cuda:1'), covar=tensor([0.0709, 0.0678, 0.0469, 0.0788, 0.1336, 0.0919, 0.1434, 0.1229], device='cuda:1'), in_proj_covar=tensor([0.0361, 0.0360, 0.0364, 0.0387, 0.0468, 0.0392, 0.0342, 0.0344], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 08:15:42,925 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 2023-04-03 08:16:15,175 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.146e+02 4.975e+02 6.244e+02 7.696e+02 2.158e+03, threshold=1.249e+03, percent-clipped=8.0 2023-04-03 08:16:24,654 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=177571.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 08:16:33,685 INFO [train.py:903] (1/4) Epoch 27, batch 50, loss[loss=0.1797, simple_loss=0.2636, pruned_loss=0.04788, over 19781.00 frames. ], tot_loss[loss=0.2031, simple_loss=0.2837, pruned_loss=0.06123, over 848799.99 frames. ], batch size: 49, lr: 3.06e-03, grad_scale: 8.0 2023-04-03 08:16:43,177 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=177586.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 08:16:43,239 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=177586.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 08:16:46,416 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=177588.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 08:17:06,250 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1_rvb from training. Duration: 27.0318125 2023-04-03 08:17:15,801 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=177611.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 08:17:35,992 INFO [train.py:903] (1/4) Epoch 27, batch 100, loss[loss=0.2617, simple_loss=0.3342, pruned_loss=0.09454, over 19777.00 frames. ], tot_loss[loss=0.2047, simple_loss=0.2842, pruned_loss=0.06257, over 1510887.62 frames. ], batch size: 54, lr: 3.06e-03, grad_scale: 8.0 2023-04-03 08:17:47,473 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 2023-04-03 08:17:47,841 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3660, 1.4955, 1.8546, 1.6396, 2.9611, 4.5799, 4.4523, 5.0078], device='cuda:1'), covar=tensor([0.1583, 0.3587, 0.3304, 0.2255, 0.0663, 0.0195, 0.0182, 0.0228], device='cuda:1'), in_proj_covar=tensor([0.0277, 0.0328, 0.0361, 0.0269, 0.0251, 0.0194, 0.0218, 0.0273], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-03 08:18:23,247 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.041e+02 5.210e+02 6.703e+02 8.227e+02 2.617e+03, threshold=1.341e+03, percent-clipped=11.0 2023-04-03 08:18:39,596 INFO [train.py:903] (1/4) Epoch 27, batch 150, loss[loss=0.1512, simple_loss=0.2315, pruned_loss=0.03542, over 19757.00 frames. ], tot_loss[loss=0.2017, simple_loss=0.2815, pruned_loss=0.061, over 2031770.09 frames. ], batch size: 47, lr: 3.06e-03, grad_scale: 8.0 2023-04-03 08:18:44,578 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=177682.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 08:19:40,062 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 2023-04-03 08:19:42,199 INFO [train.py:903] (1/4) Epoch 27, batch 200, loss[loss=0.2195, simple_loss=0.3002, pruned_loss=0.0694, over 19356.00 frames. ], tot_loss[loss=0.2031, simple_loss=0.2831, pruned_loss=0.06151, over 2417055.89 frames. ], batch size: 70, lr: 3.06e-03, grad_scale: 8.0 2023-04-03 08:20:29,446 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.575e+02 4.383e+02 5.368e+02 7.234e+02 1.640e+03, threshold=1.074e+03, percent-clipped=1.0 2023-04-03 08:20:46,585 INFO [train.py:903] (1/4) Epoch 27, batch 250, loss[loss=0.2261, simple_loss=0.2911, pruned_loss=0.0805, over 19731.00 frames. ], tot_loss[loss=0.2018, simple_loss=0.2822, pruned_loss=0.06074, over 2742415.91 frames. ], batch size: 45, lr: 3.06e-03, grad_scale: 8.0 2023-04-03 08:21:12,220 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=177798.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 08:21:32,171 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=177815.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 08:21:43,032 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.01 vs. limit=5.0 2023-04-03 08:21:50,874 INFO [train.py:903] (1/4) Epoch 27, batch 300, loss[loss=0.1939, simple_loss=0.2848, pruned_loss=0.05147, over 19437.00 frames. ], tot_loss[loss=0.2013, simple_loss=0.2819, pruned_loss=0.0603, over 2991837.47 frames. ], batch size: 70, lr: 3.06e-03, grad_scale: 8.0 2023-04-03 08:22:08,570 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=177842.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 08:22:10,836 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=177844.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 08:22:36,375 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.100e+02 5.015e+02 6.251e+02 7.839e+02 1.329e+03, threshold=1.250e+03, percent-clipped=8.0 2023-04-03 08:22:40,049 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=177867.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 08:22:42,336 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=177869.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 08:22:52,700 INFO [train.py:903] (1/4) Epoch 27, batch 350, loss[loss=0.1993, simple_loss=0.2796, pruned_loss=0.05949, over 19669.00 frames. ], tot_loss[loss=0.2028, simple_loss=0.2831, pruned_loss=0.06126, over 3173513.52 frames. ], batch size: 58, lr: 3.06e-03, grad_scale: 8.0 2023-04-03 08:23:00,611 WARNING [train.py:1073] (1/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1_rvb from training. Duration: 0.9681875 2023-04-03 08:23:40,481 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=177915.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 08:23:47,963 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0019, 1.9307, 1.8257, 1.6664, 1.5830, 1.6413, 0.4195, 0.9031], device='cuda:1'), covar=tensor([0.0698, 0.0691, 0.0475, 0.0767, 0.1256, 0.0870, 0.1452, 0.1178], device='cuda:1'), in_proj_covar=tensor([0.0364, 0.0363, 0.0367, 0.0390, 0.0472, 0.0395, 0.0346, 0.0347], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 08:23:56,874 INFO [train.py:903] (1/4) Epoch 27, batch 400, loss[loss=0.2068, simple_loss=0.2865, pruned_loss=0.06354, over 19673.00 frames. ], tot_loss[loss=0.2022, simple_loss=0.2828, pruned_loss=0.06082, over 3331125.95 frames. ], batch size: 59, lr: 3.06e-03, grad_scale: 8.0 2023-04-03 08:24:43,713 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.926e+02 4.795e+02 5.555e+02 6.674e+02 1.146e+03, threshold=1.111e+03, percent-clipped=0.0 2023-04-03 08:24:58,338 INFO [train.py:903] (1/4) Epoch 27, batch 450, loss[loss=0.1819, simple_loss=0.2538, pruned_loss=0.05497, over 19783.00 frames. ], tot_loss[loss=0.2015, simple_loss=0.2823, pruned_loss=0.06037, over 3443594.76 frames. ], batch size: 46, lr: 3.06e-03, grad_scale: 8.0 2023-04-03 08:25:39,729 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 2023-04-03 08:25:40,951 WARNING [train.py:1073] (1/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 2023-04-03 08:25:42,669 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.7077, 1.7228, 1.6249, 1.4078, 1.3898, 1.4355, 0.3303, 0.7205], device='cuda:1'), covar=tensor([0.0730, 0.0684, 0.0457, 0.0763, 0.1334, 0.0799, 0.1476, 0.1197], device='cuda:1'), in_proj_covar=tensor([0.0364, 0.0363, 0.0366, 0.0390, 0.0472, 0.0394, 0.0346, 0.0347], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 08:25:59,701 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=178026.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 08:26:02,523 INFO [train.py:903] (1/4) Epoch 27, batch 500, loss[loss=0.2381, simple_loss=0.3147, pruned_loss=0.08078, over 18825.00 frames. ], tot_loss[loss=0.2018, simple_loss=0.2827, pruned_loss=0.06048, over 3536138.91 frames. ], batch size: 74, lr: 3.06e-03, grad_scale: 8.0 2023-04-03 08:26:06,495 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=178030.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 08:26:48,570 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.047e+02 5.151e+02 6.325e+02 8.134e+02 1.856e+03, threshold=1.265e+03, percent-clipped=5.0 2023-04-03 08:27:07,119 INFO [train.py:903] (1/4) Epoch 27, batch 550, loss[loss=0.1975, simple_loss=0.2853, pruned_loss=0.05486, over 19533.00 frames. ], tot_loss[loss=0.2007, simple_loss=0.2817, pruned_loss=0.05981, over 3609358.69 frames. ], batch size: 56, lr: 3.06e-03, grad_scale: 8.0 2023-04-03 08:27:22,529 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.0746, 1.3471, 1.8641, 1.2392, 2.6620, 3.6477, 3.3566, 3.9057], device='cuda:1'), covar=tensor([0.1800, 0.3977, 0.3288, 0.2723, 0.0673, 0.0210, 0.0228, 0.0265], device='cuda:1'), in_proj_covar=tensor([0.0278, 0.0329, 0.0361, 0.0269, 0.0252, 0.0195, 0.0218, 0.0273], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-03 08:28:09,845 INFO [train.py:903] (1/4) Epoch 27, batch 600, loss[loss=0.2334, simple_loss=0.3172, pruned_loss=0.07482, over 18444.00 frames. ], tot_loss[loss=0.2028, simple_loss=0.2836, pruned_loss=0.06099, over 3650615.48 frames. ], batch size: 84, lr: 3.06e-03, grad_scale: 8.0 2023-04-03 08:28:25,323 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=178141.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 08:28:26,321 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=178142.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 08:28:48,728 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=178159.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 08:28:54,534 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 2023-04-03 08:28:55,502 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.215e+02 5.259e+02 6.338e+02 7.640e+02 1.730e+03, threshold=1.268e+03, percent-clipped=4.0 2023-04-03 08:29:11,517 INFO [train.py:903] (1/4) Epoch 27, batch 650, loss[loss=0.1779, simple_loss=0.2541, pruned_loss=0.05081, over 19740.00 frames. ], tot_loss[loss=0.2041, simple_loss=0.2846, pruned_loss=0.06182, over 3668577.86 frames. ], batch size: 48, lr: 3.06e-03, grad_scale: 8.0 2023-04-03 08:30:11,957 INFO [train.py:903] (1/4) Epoch 27, batch 700, loss[loss=0.1981, simple_loss=0.29, pruned_loss=0.05314, over 19680.00 frames. ], tot_loss[loss=0.2042, simple_loss=0.2848, pruned_loss=0.06177, over 3681334.05 frames. ], batch size: 55, lr: 3.06e-03, grad_scale: 8.0 2023-04-03 08:30:49,322 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=178257.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 08:30:50,587 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9636, 1.8392, 1.6558, 1.9214, 1.7218, 1.6685, 1.5379, 1.8219], device='cuda:1'), covar=tensor([0.1155, 0.1405, 0.1576, 0.1040, 0.1411, 0.0620, 0.1637, 0.0819], device='cuda:1'), in_proj_covar=tensor([0.0279, 0.0361, 0.0321, 0.0258, 0.0309, 0.0259, 0.0323, 0.0266], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-04-03 08:30:58,527 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.944e+02 4.694e+02 6.359e+02 8.615e+02 1.569e+03, threshold=1.272e+03, percent-clipped=7.0 2023-04-03 08:31:11,181 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=178274.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 08:31:16,251 INFO [train.py:903] (1/4) Epoch 27, batch 750, loss[loss=0.2212, simple_loss=0.2985, pruned_loss=0.07194, over 19308.00 frames. ], tot_loss[loss=0.2034, simple_loss=0.2841, pruned_loss=0.06141, over 3703694.98 frames. ], batch size: 70, lr: 3.06e-03, grad_scale: 8.0 2023-04-03 08:31:26,151 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=178286.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 08:31:56,987 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=178311.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 08:32:16,079 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.1040, 2.0984, 1.8811, 1.6754, 1.5035, 1.6261, 0.9048, 1.2916], device='cuda:1'), covar=tensor([0.0830, 0.0783, 0.0573, 0.1071, 0.1419, 0.1235, 0.1586, 0.1239], device='cuda:1'), in_proj_covar=tensor([0.0366, 0.0365, 0.0368, 0.0392, 0.0473, 0.0395, 0.0347, 0.0349], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 08:32:17,794 INFO [train.py:903] (1/4) Epoch 27, batch 800, loss[loss=0.2185, simple_loss=0.2976, pruned_loss=0.06964, over 17494.00 frames. ], tot_loss[loss=0.2039, simple_loss=0.2847, pruned_loss=0.06157, over 3733458.07 frames. ], batch size: 101, lr: 3.06e-03, grad_scale: 8.0 2023-04-03 08:32:31,930 WARNING [train.py:1073] (1/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 2023-04-03 08:32:32,894 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.78 vs. limit=2.0 2023-04-03 08:32:58,404 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.8240, 3.2921, 3.3440, 3.3496, 1.4907, 3.2123, 2.7997, 3.1346], device='cuda:1'), covar=tensor([0.1788, 0.1112, 0.0857, 0.0979, 0.5309, 0.1212, 0.0843, 0.1326], device='cuda:1'), in_proj_covar=tensor([0.0820, 0.0784, 0.0989, 0.0871, 0.0863, 0.0754, 0.0582, 0.0917], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-03 08:33:04,021 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.236e+02 4.845e+02 6.131e+02 7.065e+02 2.334e+03, threshold=1.226e+03, percent-clipped=1.0 2023-04-03 08:33:20,261 INFO [train.py:903] (1/4) Epoch 27, batch 850, loss[loss=0.1767, simple_loss=0.2581, pruned_loss=0.04766, over 19683.00 frames. ], tot_loss[loss=0.2042, simple_loss=0.2853, pruned_loss=0.06158, over 3754363.85 frames. ], batch size: 53, lr: 3.06e-03, grad_scale: 8.0 2023-04-03 08:33:45,280 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=178397.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 08:33:57,599 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([6.2592, 5.6959, 3.2172, 4.9620, 1.1297, 5.9109, 5.6507, 5.8578], device='cuda:1'), covar=tensor([0.0417, 0.0810, 0.1691, 0.0699, 0.3928, 0.0438, 0.0771, 0.0909], device='cuda:1'), in_proj_covar=tensor([0.0528, 0.0427, 0.0515, 0.0359, 0.0410, 0.0452, 0.0447, 0.0476], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 08:34:12,690 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9_rvb from training. Duration: 25.061125 2023-04-03 08:34:16,587 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=178422.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 08:34:22,832 INFO [train.py:903] (1/4) Epoch 27, batch 900, loss[loss=0.1795, simple_loss=0.2573, pruned_loss=0.05084, over 16418.00 frames. ], tot_loss[loss=0.2033, simple_loss=0.2847, pruned_loss=0.06095, over 3741850.92 frames. ], batch size: 36, lr: 3.06e-03, grad_scale: 8.0 2023-04-03 08:35:10,868 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.108e+02 4.416e+02 5.422e+02 6.593e+02 1.258e+03, threshold=1.084e+03, percent-clipped=1.0 2023-04-03 08:35:17,031 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=178470.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 08:35:28,147 INFO [train.py:903] (1/4) Epoch 27, batch 950, loss[loss=0.1951, simple_loss=0.27, pruned_loss=0.06008, over 19580.00 frames. ], tot_loss[loss=0.2032, simple_loss=0.2843, pruned_loss=0.06099, over 3759722.15 frames. ], batch size: 52, lr: 3.06e-03, grad_scale: 8.0 2023-04-03 08:35:30,648 WARNING [train.py:1073] (1/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9_rvb from training. Duration: 26.32775 2023-04-03 08:36:11,746 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=178513.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 08:36:32,191 INFO [train.py:903] (1/4) Epoch 27, batch 1000, loss[loss=0.2142, simple_loss=0.2967, pruned_loss=0.06591, over 18767.00 frames. ], tot_loss[loss=0.2031, simple_loss=0.2843, pruned_loss=0.06099, over 3777989.44 frames. ], batch size: 74, lr: 3.06e-03, grad_scale: 8.0 2023-04-03 08:36:35,040 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=178530.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 08:36:45,498 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=178538.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 08:37:01,977 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 2023-04-03 08:37:06,213 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=178555.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 08:37:19,294 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.158e+02 4.971e+02 6.338e+02 8.822e+02 2.004e+03, threshold=1.268e+03, percent-clipped=12.0 2023-04-03 08:37:25,282 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 2023-04-03 08:37:35,818 INFO [train.py:903] (1/4) Epoch 27, batch 1050, loss[loss=0.1962, simple_loss=0.2892, pruned_loss=0.05154, over 18070.00 frames. ], tot_loss[loss=0.2033, simple_loss=0.2844, pruned_loss=0.06108, over 3772496.87 frames. ], batch size: 83, lr: 3.06e-03, grad_scale: 8.0 2023-04-03 08:38:09,368 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 2023-04-03 08:38:38,440 INFO [train.py:903] (1/4) Epoch 27, batch 1100, loss[loss=0.2713, simple_loss=0.3289, pruned_loss=0.1069, over 13192.00 frames. ], tot_loss[loss=0.2033, simple_loss=0.2842, pruned_loss=0.06121, over 3779812.23 frames. ], batch size: 136, lr: 3.05e-03, grad_scale: 8.0 2023-04-03 08:38:50,308 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=178637.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 08:39:07,821 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.9903, 4.5704, 2.7463, 3.9858, 0.9219, 4.5654, 4.3902, 4.4848], device='cuda:1'), covar=tensor([0.0509, 0.0873, 0.2049, 0.0838, 0.4135, 0.0603, 0.0883, 0.1171], device='cuda:1'), in_proj_covar=tensor([0.0528, 0.0428, 0.0515, 0.0361, 0.0410, 0.0453, 0.0448, 0.0476], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 08:39:27,052 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.060e+02 4.758e+02 5.833e+02 7.524e+02 1.653e+03, threshold=1.167e+03, percent-clipped=2.0 2023-04-03 08:39:42,276 INFO [train.py:903] (1/4) Epoch 27, batch 1150, loss[loss=0.2135, simple_loss=0.2952, pruned_loss=0.06587, over 19613.00 frames. ], tot_loss[loss=0.2026, simple_loss=0.2837, pruned_loss=0.06076, over 3786525.42 frames. ], batch size: 57, lr: 3.05e-03, grad_scale: 4.0 2023-04-03 08:40:25,279 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=178712.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 08:40:30,189 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=178715.0, num_to_drop=1, layers_to_drop={1} 2023-04-03 08:40:47,648 INFO [train.py:903] (1/4) Epoch 27, batch 1200, loss[loss=0.2715, simple_loss=0.3401, pruned_loss=0.1015, over 13505.00 frames. ], tot_loss[loss=0.2028, simple_loss=0.2839, pruned_loss=0.06088, over 3784756.27 frames. ], batch size: 137, lr: 3.05e-03, grad_scale: 8.0 2023-04-03 08:41:18,720 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 2023-04-03 08:41:38,084 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.626e+02 4.854e+02 6.177e+02 8.249e+02 1.593e+03, threshold=1.235e+03, percent-clipped=5.0 2023-04-03 08:41:53,217 INFO [train.py:903] (1/4) Epoch 27, batch 1250, loss[loss=0.1668, simple_loss=0.2424, pruned_loss=0.04554, over 19740.00 frames. ], tot_loss[loss=0.202, simple_loss=0.2833, pruned_loss=0.0603, over 3802528.40 frames. ], batch size: 46, lr: 3.05e-03, grad_scale: 8.0 2023-04-03 08:42:05,335 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=178788.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 08:42:39,170 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=178814.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 08:42:56,286 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 2023-04-03 08:42:56,586 INFO [train.py:903] (1/4) Epoch 27, batch 1300, loss[loss=0.2001, simple_loss=0.2863, pruned_loss=0.05701, over 19742.00 frames. ], tot_loss[loss=0.2017, simple_loss=0.2829, pruned_loss=0.06026, over 3804311.87 frames. ], batch size: 63, lr: 3.05e-03, grad_scale: 8.0 2023-04-03 08:43:44,690 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.050e+02 5.004e+02 6.015e+02 7.494e+02 1.649e+03, threshold=1.203e+03, percent-clipped=2.0 2023-04-03 08:43:58,885 INFO [train.py:903] (1/4) Epoch 27, batch 1350, loss[loss=0.1997, simple_loss=0.2846, pruned_loss=0.05742, over 19531.00 frames. ], tot_loss[loss=0.202, simple_loss=0.2834, pruned_loss=0.06032, over 3792928.80 frames. ], batch size: 54, lr: 3.05e-03, grad_scale: 8.0 2023-04-03 08:44:41,880 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=178912.0, num_to_drop=1, layers_to_drop={0} 2023-04-03 08:45:03,464 INFO [train.py:903] (1/4) Epoch 27, batch 1400, loss[loss=0.2417, simple_loss=0.3069, pruned_loss=0.08827, over 13356.00 frames. ], tot_loss[loss=0.2021, simple_loss=0.2832, pruned_loss=0.06052, over 3788216.69 frames. ], batch size: 135, lr: 3.05e-03, grad_scale: 8.0 2023-04-03 08:45:05,013 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=178929.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 08:45:09,899 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6682, 1.5055, 1.5827, 2.2093, 1.6762, 1.8939, 1.9648, 1.6546], device='cuda:1'), covar=tensor([0.0852, 0.0970, 0.1032, 0.0717, 0.0863, 0.0819, 0.0831, 0.0765], device='cuda:1'), in_proj_covar=tensor([0.0212, 0.0225, 0.0228, 0.0240, 0.0227, 0.0214, 0.0188, 0.0206], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-04-03 08:45:50,934 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.415e+02 4.777e+02 6.195e+02 8.137e+02 1.607e+03, threshold=1.239e+03, percent-clipped=6.0 2023-04-03 08:46:03,730 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_rvb from training. Duration: 25.85 2023-04-03 08:46:04,762 INFO [train.py:903] (1/4) Epoch 27, batch 1450, loss[loss=0.2233, simple_loss=0.3027, pruned_loss=0.07196, over 19604.00 frames. ], tot_loss[loss=0.2024, simple_loss=0.2832, pruned_loss=0.0608, over 3789778.10 frames. ], batch size: 57, lr: 3.05e-03, grad_scale: 8.0 2023-04-03 08:46:09,315 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=178981.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 08:46:43,746 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 2023-04-03 08:47:07,119 INFO [train.py:903] (1/4) Epoch 27, batch 1500, loss[loss=0.2044, simple_loss=0.2904, pruned_loss=0.05921, over 18700.00 frames. ], tot_loss[loss=0.2022, simple_loss=0.2831, pruned_loss=0.06064, over 3805984.28 frames. ], batch size: 74, lr: 3.05e-03, grad_scale: 8.0 2023-04-03 08:47:42,382 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=179056.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 08:47:45,635 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=179059.0, num_to_drop=1, layers_to_drop={0} 2023-04-03 08:47:54,377 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.273e+02 4.749e+02 5.669e+02 7.445e+02 1.551e+03, threshold=1.134e+03, percent-clipped=3.0 2023-04-03 08:48:08,107 INFO [train.py:903] (1/4) Epoch 27, batch 1550, loss[loss=0.2102, simple_loss=0.2934, pruned_loss=0.06351, over 19567.00 frames. ], tot_loss[loss=0.2029, simple_loss=0.2836, pruned_loss=0.06112, over 3809174.99 frames. ], batch size: 61, lr: 3.05e-03, grad_scale: 8.0 2023-04-03 08:48:32,740 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=179096.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 08:49:12,291 INFO [train.py:903] (1/4) Epoch 27, batch 1600, loss[loss=0.2047, simple_loss=0.2917, pruned_loss=0.05882, over 17281.00 frames. ], tot_loss[loss=0.2035, simple_loss=0.2842, pruned_loss=0.0614, over 3816878.79 frames. ], batch size: 101, lr: 3.05e-03, grad_scale: 8.0 2023-04-03 08:49:18,226 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=179132.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 08:49:37,310 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9_rvb from training. Duration: 30.1555625 2023-04-03 08:49:59,057 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.135e+02 4.994e+02 6.186e+02 7.700e+02 1.707e+03, threshold=1.237e+03, percent-clipped=5.0 2023-04-03 08:50:06,421 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=179171.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 08:50:07,878 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.13 vs. limit=2.0 2023-04-03 08:50:09,878 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=179174.0, num_to_drop=1, layers_to_drop={1} 2023-04-03 08:50:14,088 INFO [train.py:903] (1/4) Epoch 27, batch 1650, loss[loss=0.2023, simple_loss=0.2863, pruned_loss=0.05909, over 19307.00 frames. ], tot_loss[loss=0.2042, simple_loss=0.2849, pruned_loss=0.06176, over 3811725.41 frames. ], batch size: 66, lr: 3.05e-03, grad_scale: 8.0 2023-04-03 08:50:21,677 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4060, 1.3725, 1.5130, 1.5289, 1.8459, 1.8810, 1.8445, 0.5497], device='cuda:1'), covar=tensor([0.2547, 0.4507, 0.2891, 0.2040, 0.1678, 0.2438, 0.1457, 0.5159], device='cuda:1'), in_proj_covar=tensor([0.0550, 0.0662, 0.0743, 0.0501, 0.0631, 0.0541, 0.0665, 0.0567], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 08:50:23,740 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=179185.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 08:50:51,064 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2178, 1.4180, 2.1084, 1.5662, 3.1550, 4.7901, 4.6067, 5.2559], device='cuda:1'), covar=tensor([0.1738, 0.3977, 0.3229, 0.2432, 0.0553, 0.0189, 0.0161, 0.0187], device='cuda:1'), in_proj_covar=tensor([0.0280, 0.0330, 0.0363, 0.0271, 0.0253, 0.0196, 0.0219, 0.0275], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-03 08:50:55,267 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=179210.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 08:51:17,041 INFO [train.py:903] (1/4) Epoch 27, batch 1700, loss[loss=0.1648, simple_loss=0.2544, pruned_loss=0.03758, over 19747.00 frames. ], tot_loss[loss=0.2037, simple_loss=0.2846, pruned_loss=0.06143, over 3824588.63 frames. ], batch size: 51, lr: 3.05e-03, grad_scale: 8.0 2023-04-03 08:51:41,283 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=179247.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 08:51:52,244 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=179256.0, num_to_drop=1, layers_to_drop={0} 2023-04-03 08:51:58,258 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 2023-04-03 08:52:04,047 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.943e+02 5.162e+02 6.407e+02 7.875e+02 2.392e+03, threshold=1.281e+03, percent-clipped=6.0 2023-04-03 08:52:18,199 INFO [train.py:903] (1/4) Epoch 27, batch 1750, loss[loss=0.17, simple_loss=0.2563, pruned_loss=0.04189, over 19854.00 frames. ], tot_loss[loss=0.2048, simple_loss=0.2853, pruned_loss=0.06212, over 3819007.46 frames. ], batch size: 52, lr: 3.05e-03, grad_scale: 8.0 2023-04-03 08:52:47,237 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.8185, 1.5791, 1.4405, 1.7859, 1.4988, 1.5868, 1.4841, 1.6753], device='cuda:1'), covar=tensor([0.1215, 0.1423, 0.1680, 0.1101, 0.1372, 0.0621, 0.1622, 0.0887], device='cuda:1'), in_proj_covar=tensor([0.0279, 0.0360, 0.0319, 0.0257, 0.0309, 0.0258, 0.0322, 0.0265], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 08:52:53,860 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2080, 1.3907, 1.7749, 1.2584, 2.5826, 3.5823, 3.3093, 3.7907], device='cuda:1'), covar=tensor([0.1605, 0.3785, 0.3411, 0.2566, 0.0640, 0.0187, 0.0210, 0.0280], device='cuda:1'), in_proj_covar=tensor([0.0279, 0.0330, 0.0363, 0.0271, 0.0253, 0.0195, 0.0218, 0.0274], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-03 08:53:04,994 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 2023-04-03 08:53:21,041 INFO [train.py:903] (1/4) Epoch 27, batch 1800, loss[loss=0.1755, simple_loss=0.2539, pruned_loss=0.04855, over 19356.00 frames. ], tot_loss[loss=0.2039, simple_loss=0.2845, pruned_loss=0.06169, over 3825833.89 frames. ], batch size: 47, lr: 3.05e-03, grad_scale: 8.0 2023-04-03 08:53:51,524 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=179352.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 08:54:08,231 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.712e+02 4.924e+02 6.082e+02 7.444e+02 1.664e+03, threshold=1.216e+03, percent-clipped=4.0 2023-04-03 08:54:15,340 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=179371.0, num_to_drop=1, layers_to_drop={1} 2023-04-03 08:54:21,585 WARNING [train.py:1073] (1/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9_rvb from training. Duration: 25.0944375 2023-04-03 08:54:23,149 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=179377.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 08:54:23,937 INFO [train.py:903] (1/4) Epoch 27, batch 1850, loss[loss=0.2119, simple_loss=0.2934, pruned_loss=0.0652, over 19714.00 frames. ], tot_loss[loss=0.2039, simple_loss=0.2845, pruned_loss=0.06169, over 3821319.66 frames. ], batch size: 59, lr: 3.05e-03, grad_scale: 8.0 2023-04-03 08:54:25,265 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=179379.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 08:55:00,545 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9_rvb from training. Duration: 27.8166875 2023-04-03 08:55:25,754 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3404, 1.5085, 1.9788, 1.4266, 2.7858, 3.7110, 3.4414, 3.9838], device='cuda:1'), covar=tensor([0.1560, 0.3575, 0.3094, 0.2490, 0.0553, 0.0189, 0.0215, 0.0248], device='cuda:1'), in_proj_covar=tensor([0.0280, 0.0330, 0.0363, 0.0270, 0.0253, 0.0195, 0.0218, 0.0274], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-03 08:55:25,836 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=179427.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 08:55:26,556 INFO [train.py:903] (1/4) Epoch 27, batch 1900, loss[loss=0.1632, simple_loss=0.2484, pruned_loss=0.03896, over 19743.00 frames. ], tot_loss[loss=0.2052, simple_loss=0.2858, pruned_loss=0.06229, over 3812170.76 frames. ], batch size: 45, lr: 3.05e-03, grad_scale: 8.0 2023-04-03 08:55:29,215 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=179430.0, num_to_drop=1, layers_to_drop={1} 2023-04-03 08:55:45,919 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 2023-04-03 08:55:49,550 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9_rvb from training. Duration: 27.02225 2023-04-03 08:55:56,529 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=179452.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 08:56:00,928 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=179455.0, num_to_drop=1, layers_to_drop={1} 2023-04-03 08:56:09,140 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=179462.0, num_to_drop=1, layers_to_drop={0} 2023-04-03 08:56:13,659 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.4072, 1.3974, 1.6518, 1.7328, 3.0518, 1.3959, 2.4081, 3.3888], device='cuda:1'), covar=tensor([0.0554, 0.2755, 0.2758, 0.1684, 0.0681, 0.2230, 0.1321, 0.0268], device='cuda:1'), in_proj_covar=tensor([0.0420, 0.0376, 0.0394, 0.0352, 0.0380, 0.0357, 0.0392, 0.0414], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 08:56:14,507 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.488e+02 5.214e+02 6.120e+02 7.486e+02 1.853e+03, threshold=1.224e+03, percent-clipped=1.0 2023-04-03 08:56:15,789 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 2023-04-03 08:56:28,548 INFO [train.py:903] (1/4) Epoch 27, batch 1950, loss[loss=0.1868, simple_loss=0.2712, pruned_loss=0.05121, over 19657.00 frames. ], tot_loss[loss=0.2043, simple_loss=0.2849, pruned_loss=0.0618, over 3800938.67 frames. ], batch size: 53, lr: 3.05e-03, grad_scale: 8.0 2023-04-03 08:57:01,354 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.62 vs. limit=2.0 2023-04-03 08:57:02,016 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=179503.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 08:57:31,833 INFO [train.py:903] (1/4) Epoch 27, batch 2000, loss[loss=0.2369, simple_loss=0.3084, pruned_loss=0.08265, over 13047.00 frames. ], tot_loss[loss=0.2043, simple_loss=0.285, pruned_loss=0.06182, over 3794045.42 frames. ], batch size: 137, lr: 3.05e-03, grad_scale: 8.0 2023-04-03 08:57:32,248 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=179528.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 08:58:19,140 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.349e+02 5.376e+02 6.611e+02 8.547e+02 2.231e+03, threshold=1.322e+03, percent-clipped=11.0 2023-04-03 08:58:33,840 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_rvb from training. Duration: 26.8349375 2023-04-03 08:58:34,817 INFO [train.py:903] (1/4) Epoch 27, batch 2050, loss[loss=0.244, simple_loss=0.3174, pruned_loss=0.08524, over 19605.00 frames. ], tot_loss[loss=0.2027, simple_loss=0.2835, pruned_loss=0.061, over 3820685.03 frames. ], batch size: 61, lr: 3.05e-03, grad_scale: 8.0 2023-04-03 08:58:53,478 WARNING [train.py:1073] (1/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 2023-04-03 08:58:54,682 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 2023-04-03 08:59:14,232 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 2023-04-03 08:59:37,294 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=179627.0, num_to_drop=1, layers_to_drop={1} 2023-04-03 08:59:37,899 INFO [train.py:903] (1/4) Epoch 27, batch 2100, loss[loss=0.2398, simple_loss=0.3288, pruned_loss=0.07541, over 19508.00 frames. ], tot_loss[loss=0.203, simple_loss=0.284, pruned_loss=0.061, over 3818019.51 frames. ], batch size: 64, lr: 3.05e-03, grad_scale: 8.0 2023-04-03 09:00:07,543 WARNING [train.py:1073] (1/4) Exclude cut with ID 453-131332-0000-131866_sp0.9_rvb from training. Duration: 25.3333125 2023-04-03 09:00:08,046 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=179652.0, num_to_drop=1, layers_to_drop={1} 2023-04-03 09:00:10,145 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=179654.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 09:00:13,435 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([0.8433, 1.3400, 1.0751, 0.9826, 1.1370, 1.0022, 0.9236, 1.2020], device='cuda:1'), covar=tensor([0.0727, 0.0972, 0.1237, 0.0882, 0.0652, 0.1499, 0.0753, 0.0592], device='cuda:1'), in_proj_covar=tensor([0.0304, 0.0320, 0.0337, 0.0273, 0.0250, 0.0345, 0.0294, 0.0277], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 09:00:21,264 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1976, 1.4276, 1.7757, 1.5856, 2.9261, 4.6299, 4.5252, 5.2578], device='cuda:1'), covar=tensor([0.1809, 0.4814, 0.4417, 0.2510, 0.0707, 0.0212, 0.0201, 0.0194], device='cuda:1'), in_proj_covar=tensor([0.0279, 0.0330, 0.0363, 0.0270, 0.0253, 0.0196, 0.0219, 0.0275], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-03 09:00:25,323 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.840e+02 4.609e+02 5.792e+02 6.874e+02 1.495e+03, threshold=1.158e+03, percent-clipped=2.0 2023-04-03 09:00:28,927 WARNING [train.py:1073] (1/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9_rvb from training. Duration: 26.6166875 2023-04-03 09:00:39,310 INFO [train.py:903] (1/4) Epoch 27, batch 2150, loss[loss=0.182, simple_loss=0.276, pruned_loss=0.04406, over 19694.00 frames. ], tot_loss[loss=0.2043, simple_loss=0.2849, pruned_loss=0.06179, over 3803735.25 frames. ], batch size: 59, lr: 3.05e-03, grad_scale: 8.0 2023-04-03 09:01:04,778 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 2023-04-03 09:01:37,660 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=179723.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 09:01:43,442 INFO [train.py:903] (1/4) Epoch 27, batch 2200, loss[loss=0.1852, simple_loss=0.267, pruned_loss=0.05172, over 19488.00 frames. ], tot_loss[loss=0.2038, simple_loss=0.2844, pruned_loss=0.06157, over 3790401.54 frames. ], batch size: 49, lr: 3.05e-03, grad_scale: 8.0 2023-04-03 09:02:30,599 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.069e+02 5.235e+02 6.025e+02 7.394e+02 1.358e+03, threshold=1.205e+03, percent-clipped=1.0 2023-04-03 09:02:32,032 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.0963, 3.5644, 3.8308, 3.8520, 1.6584, 3.5999, 3.0445, 3.3589], device='cuda:1'), covar=tensor([0.2633, 0.1544, 0.1077, 0.1466, 0.7436, 0.2098, 0.1424, 0.2077], device='cuda:1'), in_proj_covar=tensor([0.0808, 0.0777, 0.0983, 0.0863, 0.0859, 0.0747, 0.0576, 0.0909], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-03 09:02:46,335 INFO [train.py:903] (1/4) Epoch 27, batch 2250, loss[loss=0.2535, simple_loss=0.3183, pruned_loss=0.09438, over 13115.00 frames. ], tot_loss[loss=0.2034, simple_loss=0.2841, pruned_loss=0.06141, over 3789900.32 frames. ], batch size: 136, lr: 3.05e-03, grad_scale: 8.0 2023-04-03 09:03:21,838 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=179806.0, num_to_drop=1, layers_to_drop={1} 2023-04-03 09:03:33,117 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=179814.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 09:03:50,157 INFO [train.py:903] (1/4) Epoch 27, batch 2300, loss[loss=0.196, simple_loss=0.2839, pruned_loss=0.05406, over 19724.00 frames. ], tot_loss[loss=0.2023, simple_loss=0.2831, pruned_loss=0.06074, over 3813361.12 frames. ], batch size: 63, lr: 3.04e-03, grad_scale: 8.0 2023-04-03 09:03:58,676 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.0229, 3.6601, 2.6541, 3.2743, 0.8711, 3.6476, 3.5272, 3.6111], device='cuda:1'), covar=tensor([0.0718, 0.1110, 0.1914, 0.0948, 0.3906, 0.0753, 0.0944, 0.1168], device='cuda:1'), in_proj_covar=tensor([0.0526, 0.0427, 0.0512, 0.0360, 0.0407, 0.0454, 0.0446, 0.0475], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 09:04:02,220 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=179838.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 09:04:03,018 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_rvb from training. Duration: 26.205 2023-04-03 09:04:05,830 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3252, 1.4834, 2.1435, 1.6010, 3.2213, 4.7710, 4.6091, 5.1964], device='cuda:1'), covar=tensor([0.1706, 0.3853, 0.3105, 0.2384, 0.0578, 0.0177, 0.0166, 0.0184], device='cuda:1'), in_proj_covar=tensor([0.0281, 0.0331, 0.0364, 0.0272, 0.0254, 0.0197, 0.0220, 0.0276], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-03 09:04:20,344 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.1833, 2.8827, 2.2984, 2.2590, 1.9152, 2.5290, 1.1965, 2.0825], device='cuda:1'), covar=tensor([0.0743, 0.0651, 0.0707, 0.1218, 0.1308, 0.1160, 0.1414, 0.1200], device='cuda:1'), in_proj_covar=tensor([0.0364, 0.0363, 0.0367, 0.0391, 0.0469, 0.0396, 0.0345, 0.0347], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 09:04:32,015 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.13 vs. limit=2.0 2023-04-03 09:04:38,196 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.712e+02 5.173e+02 6.408e+02 8.011e+02 2.024e+03, threshold=1.282e+03, percent-clipped=7.0 2023-04-03 09:04:51,779 INFO [train.py:903] (1/4) Epoch 27, batch 2350, loss[loss=0.1955, simple_loss=0.2825, pruned_loss=0.05428, over 19244.00 frames. ], tot_loss[loss=0.2027, simple_loss=0.2834, pruned_loss=0.06097, over 3822959.25 frames. ], batch size: 66, lr: 3.04e-03, grad_scale: 8.0 2023-04-03 09:05:34,711 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_rvb from training. Duration: 25.775 2023-04-03 09:05:35,358 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.14 vs. limit=2.0 2023-04-03 09:05:45,230 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=179921.0, num_to_drop=1, layers_to_drop={1} 2023-04-03 09:05:48,602 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=179924.0, num_to_drop=1, layers_to_drop={1} 2023-04-03 09:05:51,525 WARNING [train.py:1073] (1/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9_rvb from training. Duration: 25.45 2023-04-03 09:05:53,778 INFO [train.py:903] (1/4) Epoch 27, batch 2400, loss[loss=0.2155, simple_loss=0.2992, pruned_loss=0.0659, over 19563.00 frames. ], tot_loss[loss=0.203, simple_loss=0.2836, pruned_loss=0.06115, over 3817598.04 frames. ], batch size: 61, lr: 3.04e-03, grad_scale: 8.0 2023-04-03 09:06:11,775 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=179942.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 09:06:28,955 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.7560, 4.1771, 4.4085, 4.4229, 2.0213, 4.1447, 3.6460, 4.1418], device='cuda:1'), covar=tensor([0.1722, 0.1169, 0.0651, 0.0735, 0.5679, 0.1100, 0.0709, 0.1135], device='cuda:1'), in_proj_covar=tensor([0.0817, 0.0780, 0.0990, 0.0869, 0.0864, 0.0753, 0.0582, 0.0916], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-03 09:06:41,354 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.340e+02 4.979e+02 6.417e+02 8.310e+02 2.182e+03, threshold=1.283e+03, percent-clipped=4.0 2023-04-03 09:06:43,106 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.61 vs. limit=2.0 2023-04-03 09:06:57,082 INFO [train.py:903] (1/4) Epoch 27, batch 2450, loss[loss=0.2168, simple_loss=0.3016, pruned_loss=0.06599, over 17660.00 frames. ], tot_loss[loss=0.2027, simple_loss=0.2835, pruned_loss=0.06098, over 3821077.91 frames. ], batch size: 101, lr: 3.04e-03, grad_scale: 8.0 2023-04-03 09:07:21,328 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=179998.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 09:08:00,985 INFO [train.py:903] (1/4) Epoch 27, batch 2500, loss[loss=0.1581, simple_loss=0.2426, pruned_loss=0.03679, over 19393.00 frames. ], tot_loss[loss=0.2017, simple_loss=0.2827, pruned_loss=0.06036, over 3813271.68 frames. ], batch size: 48, lr: 3.04e-03, grad_scale: 8.0 2023-04-03 09:08:19,336 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=180044.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 09:08:48,293 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.631e+02 4.714e+02 5.802e+02 6.784e+02 1.958e+03, threshold=1.160e+03, percent-clipped=2.0 2023-04-03 09:09:02,315 INFO [train.py:903] (1/4) Epoch 27, batch 2550, loss[loss=0.2014, simple_loss=0.2815, pruned_loss=0.06066, over 19521.00 frames. ], tot_loss[loss=0.202, simple_loss=0.2828, pruned_loss=0.06059, over 3820459.56 frames. ], batch size: 54, lr: 3.04e-03, grad_scale: 8.0 2023-04-03 09:09:22,081 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=180094.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 09:09:41,387 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0559, 2.1980, 2.3899, 2.6419, 2.1395, 2.5823, 2.3508, 2.2074], device='cuda:1'), covar=tensor([0.4155, 0.3800, 0.1947, 0.2471, 0.4018, 0.2204, 0.4796, 0.3269], device='cuda:1'), in_proj_covar=tensor([0.0932, 0.1008, 0.0741, 0.0951, 0.0908, 0.0850, 0.0857, 0.0806], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-03 09:09:47,038 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=180113.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 09:09:54,371 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=180119.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 09:09:58,714 WARNING [train.py:1073] (1/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 2023-04-03 09:10:04,539 INFO [train.py:903] (1/4) Epoch 27, batch 2600, loss[loss=0.1959, simple_loss=0.2707, pruned_loss=0.06052, over 19616.00 frames. ], tot_loss[loss=0.2022, simple_loss=0.2829, pruned_loss=0.06071, over 3828208.66 frames. ], batch size: 50, lr: 3.04e-03, grad_scale: 8.0 2023-04-03 09:10:42,778 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=180158.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 09:10:52,188 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.132e+02 5.258e+02 6.240e+02 7.647e+02 1.495e+03, threshold=1.248e+03, percent-clipped=4.0 2023-04-03 09:11:07,734 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=180177.0, num_to_drop=1, layers_to_drop={1} 2023-04-03 09:11:08,384 INFO [train.py:903] (1/4) Epoch 27, batch 2650, loss[loss=0.2602, simple_loss=0.335, pruned_loss=0.09268, over 19580.00 frames. ], tot_loss[loss=0.202, simple_loss=0.283, pruned_loss=0.0605, over 3830124.08 frames. ], batch size: 61, lr: 3.04e-03, grad_scale: 8.0 2023-04-03 09:11:29,718 WARNING [train.py:1073] (1/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9_rvb from training. Duration: 27.25 2023-04-03 09:11:38,037 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=180202.0, num_to_drop=1, layers_to_drop={1} 2023-04-03 09:12:04,733 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=180223.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 09:12:11,778 INFO [train.py:903] (1/4) Epoch 27, batch 2700, loss[loss=0.2015, simple_loss=0.2936, pruned_loss=0.05467, over 19675.00 frames. ], tot_loss[loss=0.2034, simple_loss=0.2843, pruned_loss=0.06122, over 3840003.75 frames. ], batch size: 59, lr: 3.04e-03, grad_scale: 8.0 2023-04-03 09:13:00,987 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.426e+02 5.233e+02 6.139e+02 8.955e+02 2.009e+03, threshold=1.228e+03, percent-clipped=9.0 2023-04-03 09:13:02,194 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=180268.0, num_to_drop=1, layers_to_drop={0} 2023-04-03 09:13:08,017 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=180273.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 09:13:13,279 INFO [train.py:903] (1/4) Epoch 27, batch 2750, loss[loss=0.1698, simple_loss=0.239, pruned_loss=0.05035, over 19298.00 frames. ], tot_loss[loss=0.2037, simple_loss=0.2847, pruned_loss=0.06138, over 3837416.06 frames. ], batch size: 44, lr: 3.04e-03, grad_scale: 4.0 2023-04-03 09:13:21,776 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.0955, 1.1574, 1.1282, 0.9664, 0.9612, 0.9930, 0.0945, 0.3478], device='cuda:1'), covar=tensor([0.0683, 0.0670, 0.0492, 0.0603, 0.1193, 0.0681, 0.1438, 0.1139], device='cuda:1'), in_proj_covar=tensor([0.0362, 0.0361, 0.0367, 0.0387, 0.0466, 0.0394, 0.0344, 0.0344], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 09:13:22,804 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=180286.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 09:13:54,499 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1784, 1.3179, 1.7038, 1.3094, 2.6284, 3.4125, 3.1691, 3.6657], device='cuda:1'), covar=tensor([0.1773, 0.4081, 0.3685, 0.2602, 0.0706, 0.0255, 0.0262, 0.0325], device='cuda:1'), in_proj_covar=tensor([0.0279, 0.0330, 0.0363, 0.0269, 0.0253, 0.0196, 0.0219, 0.0275], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-03 09:14:15,236 INFO [train.py:903] (1/4) Epoch 27, batch 2800, loss[loss=0.2046, simple_loss=0.2894, pruned_loss=0.05994, over 19763.00 frames. ], tot_loss[loss=0.2028, simple_loss=0.2841, pruned_loss=0.06077, over 3831429.16 frames. ], batch size: 63, lr: 3.04e-03, grad_scale: 8.0 2023-04-03 09:14:31,378 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4250, 1.3966, 1.5540, 1.5685, 1.6818, 1.9257, 1.7409, 0.5408], device='cuda:1'), covar=tensor([0.2489, 0.4459, 0.2733, 0.1993, 0.1744, 0.2354, 0.1542, 0.5106], device='cuda:1'), in_proj_covar=tensor([0.0553, 0.0669, 0.0750, 0.0505, 0.0635, 0.0547, 0.0670, 0.0572], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-03 09:14:49,743 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=180354.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 09:15:04,817 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.309e+02 4.838e+02 6.554e+02 8.151e+02 1.805e+03, threshold=1.311e+03, percent-clipped=3.0 2023-04-03 09:15:08,550 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=180369.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 09:15:18,463 INFO [train.py:903] (1/4) Epoch 27, batch 2850, loss[loss=0.223, simple_loss=0.3071, pruned_loss=0.06941, over 19612.00 frames. ], tot_loss[loss=0.2037, simple_loss=0.285, pruned_loss=0.06123, over 3833504.16 frames. ], batch size: 61, lr: 3.04e-03, grad_scale: 8.0 2023-04-03 09:15:26,248 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=180383.0, num_to_drop=1, layers_to_drop={0} 2023-04-03 09:15:33,049 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=180388.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 09:15:40,221 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=180394.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 09:15:48,644 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=180401.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 09:15:48,714 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.1336, 1.9431, 1.9307, 1.7126, 1.5971, 1.6960, 0.4673, 1.0564], device='cuda:1'), covar=tensor([0.0705, 0.0695, 0.0495, 0.0857, 0.1218, 0.0905, 0.1453, 0.1147], device='cuda:1'), in_proj_covar=tensor([0.0364, 0.0364, 0.0369, 0.0390, 0.0469, 0.0395, 0.0346, 0.0347], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 09:16:20,688 WARNING [train.py:1073] (1/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 2023-04-03 09:16:22,956 INFO [train.py:903] (1/4) Epoch 27, batch 2900, loss[loss=0.1732, simple_loss=0.2553, pruned_loss=0.04552, over 19615.00 frames. ], tot_loss[loss=0.2036, simple_loss=0.2847, pruned_loss=0.06123, over 3824330.18 frames. ], batch size: 50, lr: 3.04e-03, grad_scale: 8.0 2023-04-03 09:16:42,351 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.2079, 1.7373, 2.0802, 3.0105, 1.9091, 2.3290, 2.5792, 1.9463], device='cuda:1'), covar=tensor([0.0810, 0.0958, 0.0946, 0.0739, 0.0895, 0.0812, 0.0843, 0.0763], device='cuda:1'), in_proj_covar=tensor([0.0212, 0.0225, 0.0228, 0.0240, 0.0225, 0.0213, 0.0189, 0.0207], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-04-03 09:17:13,071 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.869e+02 4.922e+02 6.355e+02 8.136e+02 1.738e+03, threshold=1.271e+03, percent-clipped=4.0 2023-04-03 09:17:26,071 INFO [train.py:903] (1/4) Epoch 27, batch 2950, loss[loss=0.2206, simple_loss=0.299, pruned_loss=0.07104, over 19669.00 frames. ], tot_loss[loss=0.2026, simple_loss=0.2841, pruned_loss=0.0606, over 3821808.65 frames. ], batch size: 53, lr: 3.04e-03, grad_scale: 8.0 2023-04-03 09:17:36,104 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=4.18 vs. limit=5.0 2023-04-03 09:17:56,619 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=180503.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 09:18:27,181 INFO [train.py:903] (1/4) Epoch 27, batch 3000, loss[loss=0.1909, simple_loss=0.2683, pruned_loss=0.05674, over 19852.00 frames. ], tot_loss[loss=0.2016, simple_loss=0.2833, pruned_loss=0.05997, over 3829399.38 frames. ], batch size: 52, lr: 3.04e-03, grad_scale: 8.0 2023-04-03 09:18:27,181 INFO [train.py:928] (1/4) Computing validation loss 2023-04-03 09:18:39,751 INFO [train.py:937] (1/4) Epoch 27, validation: loss=0.1667, simple_loss=0.2664, pruned_loss=0.03355, over 944034.00 frames. 2023-04-03 09:18:39,751 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18821MB 2023-04-03 09:18:41,249 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=180529.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 09:18:43,432 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_rvb from training. Duration: 29.735 2023-04-03 09:19:11,464 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=180554.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 09:19:29,335 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=180567.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 09:19:30,289 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.939e+02 4.587e+02 5.827e+02 7.595e+02 1.750e+03, threshold=1.165e+03, percent-clipped=2.0 2023-04-03 09:19:41,901 INFO [train.py:903] (1/4) Epoch 27, batch 3050, loss[loss=0.2316, simple_loss=0.3091, pruned_loss=0.07702, over 17178.00 frames. ], tot_loss[loss=0.202, simple_loss=0.2836, pruned_loss=0.06015, over 3817409.59 frames. ], batch size: 101, lr: 3.04e-03, grad_scale: 4.0 2023-04-03 09:19:54,671 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=180588.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 09:20:12,196 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.59 vs. limit=5.0 2023-04-03 09:20:44,773 INFO [train.py:903] (1/4) Epoch 27, batch 3100, loss[loss=0.277, simple_loss=0.3365, pruned_loss=0.1088, over 13804.00 frames. ], tot_loss[loss=0.2023, simple_loss=0.2839, pruned_loss=0.06041, over 3823829.44 frames. ], batch size: 137, lr: 3.04e-03, grad_scale: 4.0 2023-04-03 09:20:59,858 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=180639.0, num_to_drop=1, layers_to_drop={1} 2023-04-03 09:21:14,929 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.96 vs. limit=5.0 2023-04-03 09:21:20,343 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=180657.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 09:21:29,206 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=180664.0, num_to_drop=1, layers_to_drop={0} 2023-04-03 09:21:33,520 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.182e+02 4.719e+02 5.890e+02 7.652e+02 1.677e+03, threshold=1.178e+03, percent-clipped=3.0 2023-04-03 09:21:46,247 INFO [train.py:903] (1/4) Epoch 27, batch 3150, loss[loss=0.2079, simple_loss=0.2903, pruned_loss=0.06277, over 19616.00 frames. ], tot_loss[loss=0.2016, simple_loss=0.2832, pruned_loss=0.06003, over 3834100.85 frames. ], batch size: 57, lr: 3.04e-03, grad_scale: 4.0 2023-04-03 09:21:52,998 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=180682.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 09:21:53,038 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=180682.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 09:21:54,386 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.21 vs. limit=5.0 2023-04-03 09:21:57,971 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 2023-04-03 09:22:11,319 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=180698.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 09:22:16,785 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 2023-04-03 09:22:49,869 INFO [train.py:903] (1/4) Epoch 27, batch 3200, loss[loss=0.2339, simple_loss=0.3102, pruned_loss=0.07882, over 18708.00 frames. ], tot_loss[loss=0.2014, simple_loss=0.2828, pruned_loss=0.05998, over 3830354.25 frames. ], batch size: 74, lr: 3.04e-03, grad_scale: 8.0 2023-04-03 09:23:16,646 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=180750.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 09:23:27,986 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=180759.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 09:23:30,252 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.14 vs. limit=2.0 2023-04-03 09:23:39,759 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.293e+02 5.255e+02 6.624e+02 9.042e+02 3.460e+03, threshold=1.325e+03, percent-clipped=12.0 2023-04-03 09:23:51,466 INFO [train.py:903] (1/4) Epoch 27, batch 3250, loss[loss=0.207, simple_loss=0.2929, pruned_loss=0.06057, over 19752.00 frames. ], tot_loss[loss=0.2017, simple_loss=0.2831, pruned_loss=0.06021, over 3826154.49 frames. ], batch size: 63, lr: 3.04e-03, grad_scale: 8.0 2023-04-03 09:23:59,357 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=180784.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 09:24:07,814 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 2023-04-03 09:24:36,408 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=180813.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 09:24:44,424 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([5.2465, 5.1581, 5.9661, 6.0600, 2.1865, 5.6439, 4.6802, 5.6619], device='cuda:1'), covar=tensor([0.1708, 0.0796, 0.0598, 0.0652, 0.6039, 0.0894, 0.0690, 0.1165], device='cuda:1'), in_proj_covar=tensor([0.0819, 0.0781, 0.0994, 0.0874, 0.0864, 0.0753, 0.0585, 0.0919], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-03 09:24:54,214 INFO [train.py:903] (1/4) Epoch 27, batch 3300, loss[loss=0.219, simple_loss=0.2959, pruned_loss=0.07106, over 19338.00 frames. ], tot_loss[loss=0.2023, simple_loss=0.2836, pruned_loss=0.06047, over 3828947.35 frames. ], batch size: 70, lr: 3.04e-03, grad_scale: 8.0 2023-04-03 09:25:00,021 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 2023-04-03 09:25:13,373 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=180843.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 09:25:43,900 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.436e+02 5.181e+02 6.566e+02 8.708e+02 1.878e+03, threshold=1.313e+03, percent-clipped=7.0 2023-04-03 09:25:56,106 INFO [train.py:903] (1/4) Epoch 27, batch 3350, loss[loss=0.1907, simple_loss=0.2787, pruned_loss=0.05134, over 19333.00 frames. ], tot_loss[loss=0.2021, simple_loss=0.283, pruned_loss=0.06061, over 3823912.26 frames. ], batch size: 66, lr: 3.04e-03, grad_scale: 8.0 2023-04-03 09:26:37,264 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.43 vs. limit=2.0 2023-04-03 09:27:00,223 INFO [train.py:903] (1/4) Epoch 27, batch 3400, loss[loss=0.209, simple_loss=0.2951, pruned_loss=0.06143, over 19786.00 frames. ], tot_loss[loss=0.2018, simple_loss=0.283, pruned_loss=0.06036, over 3822435.00 frames. ], batch size: 56, lr: 3.04e-03, grad_scale: 8.0 2023-04-03 09:27:05,039 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=180932.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 09:27:11,973 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=180938.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 09:27:44,017 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=180963.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 09:27:50,455 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.175e+02 4.823e+02 5.841e+02 8.287e+02 1.627e+03, threshold=1.168e+03, percent-clipped=5.0 2023-04-03 09:28:02,289 INFO [train.py:903] (1/4) Epoch 27, batch 3450, loss[loss=0.1919, simple_loss=0.2818, pruned_loss=0.05094, over 19531.00 frames. ], tot_loss[loss=0.2015, simple_loss=0.2825, pruned_loss=0.0602, over 3831159.83 frames. ], batch size: 56, lr: 3.03e-03, grad_scale: 8.0 2023-04-03 09:28:07,890 WARNING [train.py:1073] (1/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9_rvb from training. Duration: 25.2444375 2023-04-03 09:29:04,453 INFO [train.py:903] (1/4) Epoch 27, batch 3500, loss[loss=0.1829, simple_loss=0.2758, pruned_loss=0.04501, over 19460.00 frames. ], tot_loss[loss=0.2009, simple_loss=0.2819, pruned_loss=0.05997, over 3835900.93 frames. ], batch size: 64, lr: 3.03e-03, grad_scale: 8.0 2023-04-03 09:29:28,491 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=181047.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 09:29:53,844 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.012e+02 5.177e+02 6.114e+02 8.226e+02 1.424e+03, threshold=1.223e+03, percent-clipped=6.0 2023-04-03 09:29:55,469 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=181069.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 09:30:06,529 INFO [train.py:903] (1/4) Epoch 27, batch 3550, loss[loss=0.1986, simple_loss=0.2843, pruned_loss=0.05651, over 19615.00 frames. ], tot_loss[loss=0.2018, simple_loss=0.283, pruned_loss=0.06031, over 3829562.01 frames. ], batch size: 50, lr: 3.03e-03, grad_scale: 8.0 2023-04-03 09:30:27,210 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=181094.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 09:30:27,457 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=181094.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 09:31:09,215 INFO [train.py:903] (1/4) Epoch 27, batch 3600, loss[loss=0.1906, simple_loss=0.2709, pruned_loss=0.0551, over 19398.00 frames. ], tot_loss[loss=0.2019, simple_loss=0.2828, pruned_loss=0.06045, over 3815654.70 frames. ], batch size: 48, lr: 3.03e-03, grad_scale: 8.0 2023-04-03 09:31:54,790 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=4.06 vs. limit=5.0 2023-04-03 09:32:00,761 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.317e+02 4.881e+02 5.766e+02 7.270e+02 1.755e+03, threshold=1.153e+03, percent-clipped=5.0 2023-04-03 09:32:12,506 INFO [train.py:903] (1/4) Epoch 27, batch 3650, loss[loss=0.1839, simple_loss=0.2717, pruned_loss=0.04807, over 19765.00 frames. ], tot_loss[loss=0.203, simple_loss=0.2839, pruned_loss=0.06104, over 3810019.95 frames. ], batch size: 54, lr: 3.03e-03, grad_scale: 4.0 2023-04-03 09:32:24,266 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=181187.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 09:32:37,952 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.5979, 1.1399, 1.4142, 1.1936, 2.2631, 1.0040, 2.0751, 2.5444], device='cuda:1'), covar=tensor([0.0666, 0.2833, 0.2723, 0.1682, 0.0837, 0.2032, 0.1087, 0.0430], device='cuda:1'), in_proj_covar=tensor([0.0422, 0.0376, 0.0395, 0.0352, 0.0382, 0.0356, 0.0392, 0.0415], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 09:32:52,770 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=181209.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 09:33:15,874 INFO [train.py:903] (1/4) Epoch 27, batch 3700, loss[loss=0.1582, simple_loss=0.237, pruned_loss=0.03971, over 19743.00 frames. ], tot_loss[loss=0.2038, simple_loss=0.2845, pruned_loss=0.06158, over 3807018.73 frames. ], batch size: 45, lr: 3.03e-03, grad_scale: 4.0 2023-04-03 09:34:06,268 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.101e+02 5.357e+02 6.375e+02 8.389e+02 2.807e+03, threshold=1.275e+03, percent-clipped=10.0 2023-04-03 09:34:17,479 INFO [train.py:903] (1/4) Epoch 27, batch 3750, loss[loss=0.2275, simple_loss=0.3094, pruned_loss=0.07274, over 18802.00 frames. ], tot_loss[loss=0.2027, simple_loss=0.2834, pruned_loss=0.06098, over 3822571.16 frames. ], batch size: 74, lr: 3.03e-03, grad_scale: 4.0 2023-04-03 09:34:40,814 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1169, 1.2984, 1.7113, 1.1587, 2.4177, 3.3109, 3.0271, 3.5565], device='cuda:1'), covar=tensor([0.1710, 0.3951, 0.3528, 0.2586, 0.0678, 0.0225, 0.0245, 0.0312], device='cuda:1'), in_proj_covar=tensor([0.0280, 0.0332, 0.0365, 0.0270, 0.0255, 0.0196, 0.0219, 0.0276], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-03 09:34:48,381 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=181302.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 09:34:49,692 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=181303.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 09:35:04,219 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=181315.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 09:35:20,124 INFO [train.py:903] (1/4) Epoch 27, batch 3800, loss[loss=0.2188, simple_loss=0.2963, pruned_loss=0.07066, over 18122.00 frames. ], tot_loss[loss=0.2019, simple_loss=0.2829, pruned_loss=0.06047, over 3824340.00 frames. ], batch size: 83, lr: 3.03e-03, grad_scale: 4.0 2023-04-03 09:35:20,547 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=181328.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 09:35:48,531 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9_rvb from training. Duration: 29.1166875 2023-04-03 09:36:10,638 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.827e+02 5.403e+02 6.773e+02 8.761e+02 1.536e+03, threshold=1.355e+03, percent-clipped=8.0 2023-04-03 09:36:22,057 INFO [train.py:903] (1/4) Epoch 27, batch 3850, loss[loss=0.2256, simple_loss=0.304, pruned_loss=0.07363, over 19654.00 frames. ], tot_loss[loss=0.2035, simple_loss=0.2846, pruned_loss=0.06121, over 3817091.95 frames. ], batch size: 55, lr: 3.03e-03, grad_scale: 4.0 2023-04-03 09:36:58,904 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.12 vs. limit=2.0 2023-04-03 09:37:25,666 INFO [train.py:903] (1/4) Epoch 27, batch 3900, loss[loss=0.2118, simple_loss=0.2922, pruned_loss=0.06567, over 19536.00 frames. ], tot_loss[loss=0.2026, simple_loss=0.2838, pruned_loss=0.06073, over 3828988.79 frames. ], batch size: 54, lr: 3.03e-03, grad_scale: 4.0 2023-04-03 09:37:30,186 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.91 vs. limit=2.0 2023-04-03 09:38:13,088 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=181465.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 09:38:17,131 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.394e+02 5.123e+02 6.226e+02 8.139e+02 1.802e+03, threshold=1.245e+03, percent-clipped=4.0 2023-04-03 09:38:28,447 INFO [train.py:903] (1/4) Epoch 27, batch 3950, loss[loss=0.1943, simple_loss=0.2725, pruned_loss=0.05806, over 19730.00 frames. ], tot_loss[loss=0.201, simple_loss=0.2824, pruned_loss=0.05982, over 3840395.81 frames. ], batch size: 51, lr: 3.03e-03, grad_scale: 4.0 2023-04-03 09:38:30,650 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 2023-04-03 09:38:43,806 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=181490.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 09:39:12,745 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 2023-04-03 09:39:30,607 INFO [train.py:903] (1/4) Epoch 27, batch 4000, loss[loss=0.1917, simple_loss=0.2773, pruned_loss=0.05305, over 19762.00 frames. ], tot_loss[loss=0.2021, simple_loss=0.283, pruned_loss=0.06064, over 3825856.37 frames. ], batch size: 54, lr: 3.03e-03, grad_scale: 8.0 2023-04-03 09:39:55,706 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=181548.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 09:40:08,352 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=181558.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 09:40:14,697 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1_rvb from training. Duration: 0.7545625 2023-04-03 09:40:18,992 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 2023-04-03 09:40:21,514 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.200e+02 5.331e+02 6.990e+02 9.796e+02 2.756e+03, threshold=1.398e+03, percent-clipped=11.0 2023-04-03 09:40:32,612 INFO [train.py:903] (1/4) Epoch 27, batch 4050, loss[loss=0.1794, simple_loss=0.262, pruned_loss=0.04841, over 19842.00 frames. ], tot_loss[loss=0.2016, simple_loss=0.2827, pruned_loss=0.06023, over 3833134.48 frames. ], batch size: 52, lr: 3.03e-03, grad_scale: 4.0 2023-04-03 09:40:38,780 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=181583.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 09:41:35,048 INFO [train.py:903] (1/4) Epoch 27, batch 4100, loss[loss=0.2649, simple_loss=0.3257, pruned_loss=0.102, over 13029.00 frames. ], tot_loss[loss=0.2024, simple_loss=0.2833, pruned_loss=0.06072, over 3816468.83 frames. ], batch size: 136, lr: 3.03e-03, grad_scale: 4.0 2023-04-03 09:42:07,723 WARNING [train.py:1073] (1/4) Exclude cut with ID 497-129325-0061-9566_sp1.1_rvb from training. Duration: 0.97725 2023-04-03 09:42:13,763 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=181659.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 09:42:26,979 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.417e+02 5.142e+02 6.045e+02 7.647e+02 1.406e+03, threshold=1.209e+03, percent-clipped=1.0 2023-04-03 09:42:35,663 INFO [train.py:903] (1/4) Epoch 27, batch 4150, loss[loss=0.1705, simple_loss=0.2549, pruned_loss=0.04302, over 19732.00 frames. ], tot_loss[loss=0.2037, simple_loss=0.2846, pruned_loss=0.06138, over 3818699.62 frames. ], batch size: 51, lr: 3.03e-03, grad_scale: 4.0 2023-04-03 09:42:58,886 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0573, 2.2368, 2.5344, 2.8350, 2.2027, 2.7305, 2.5016, 2.2071], device='cuda:1'), covar=tensor([0.4456, 0.4084, 0.1867, 0.2424, 0.4185, 0.2222, 0.4899, 0.3517], device='cuda:1'), in_proj_covar=tensor([0.0930, 0.1007, 0.0737, 0.0948, 0.0905, 0.0847, 0.0857, 0.0806], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-03 09:42:59,065 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.75 vs. limit=2.0 2023-04-03 09:43:05,786 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.5584, 1.2585, 1.5497, 1.5164, 3.0887, 1.3182, 2.3417, 3.5472], device='cuda:1'), covar=tensor([0.0610, 0.3077, 0.2993, 0.2011, 0.0878, 0.2425, 0.1434, 0.0299], device='cuda:1'), in_proj_covar=tensor([0.0420, 0.0375, 0.0392, 0.0351, 0.0381, 0.0354, 0.0390, 0.0413], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 09:43:38,714 INFO [train.py:903] (1/4) Epoch 27, batch 4200, loss[loss=0.1886, simple_loss=0.2799, pruned_loss=0.04859, over 19671.00 frames. ], tot_loss[loss=0.2027, simple_loss=0.2837, pruned_loss=0.06084, over 3832342.57 frames. ], batch size: 58, lr: 3.03e-03, grad_scale: 4.0 2023-04-03 09:43:42,099 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 2023-04-03 09:43:55,203 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.4040, 1.7069, 1.8996, 1.8747, 4.0225, 1.4445, 2.8668, 4.2340], device='cuda:1'), covar=tensor([0.0517, 0.2764, 0.2744, 0.1919, 0.0707, 0.2537, 0.1497, 0.0215], device='cuda:1'), in_proj_covar=tensor([0.0422, 0.0377, 0.0394, 0.0352, 0.0382, 0.0356, 0.0392, 0.0415], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 09:44:30,943 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.172e+02 5.083e+02 6.482e+02 7.948e+02 1.533e+03, threshold=1.296e+03, percent-clipped=4.0 2023-04-03 09:44:36,211 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=181774.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 09:44:40,470 INFO [train.py:903] (1/4) Epoch 27, batch 4250, loss[loss=0.2065, simple_loss=0.2841, pruned_loss=0.06444, over 19488.00 frames. ], tot_loss[loss=0.2017, simple_loss=0.2827, pruned_loss=0.06029, over 3838202.24 frames. ], batch size: 49, lr: 3.03e-03, grad_scale: 4.0 2023-04-03 09:44:56,632 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9_rvb from training. Duration: 29.816625 2023-04-03 09:45:08,439 WARNING [train.py:1073] (1/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 2023-04-03 09:45:43,725 INFO [train.py:903] (1/4) Epoch 27, batch 4300, loss[loss=0.1666, simple_loss=0.2533, pruned_loss=0.03997, over 19836.00 frames. ], tot_loss[loss=0.2022, simple_loss=0.2832, pruned_loss=0.06055, over 3836794.81 frames. ], batch size: 52, lr: 3.03e-03, grad_scale: 4.0 2023-04-03 09:45:46,607 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([5.4680, 5.0720, 3.1830, 4.4532, 1.3453, 5.0324, 4.8778, 5.0293], device='cuda:1'), covar=tensor([0.0408, 0.0770, 0.1761, 0.0699, 0.3726, 0.0605, 0.0867, 0.1023], device='cuda:1'), in_proj_covar=tensor([0.0527, 0.0427, 0.0512, 0.0357, 0.0406, 0.0452, 0.0448, 0.0477], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 09:46:16,691 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.0365, 4.4609, 4.7285, 4.7273, 1.8313, 4.4336, 3.9064, 4.4854], device='cuda:1'), covar=tensor([0.1626, 0.0772, 0.0611, 0.0675, 0.5812, 0.0792, 0.0682, 0.1059], device='cuda:1'), in_proj_covar=tensor([0.0823, 0.0784, 0.0994, 0.0870, 0.0864, 0.0755, 0.0588, 0.0922], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-03 09:46:36,879 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.524e+02 4.917e+02 6.450e+02 8.224e+02 1.543e+03, threshold=1.290e+03, percent-clipped=3.0 2023-04-03 09:46:40,234 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 2023-04-03 09:46:47,397 INFO [train.py:903] (1/4) Epoch 27, batch 4350, loss[loss=0.1812, simple_loss=0.2629, pruned_loss=0.04974, over 19721.00 frames. ], tot_loss[loss=0.2012, simple_loss=0.2822, pruned_loss=0.06008, over 3839372.20 frames. ], batch size: 51, lr: 3.03e-03, grad_scale: 4.0 2023-04-03 09:46:59,906 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.4659, 4.1164, 2.6971, 3.6077, 1.2668, 4.0018, 3.8708, 4.0032], device='cuda:1'), covar=tensor([0.0693, 0.0962, 0.2063, 0.0848, 0.3734, 0.0735, 0.1043, 0.1286], device='cuda:1'), in_proj_covar=tensor([0.0528, 0.0428, 0.0513, 0.0358, 0.0408, 0.0454, 0.0450, 0.0479], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 09:47:05,497 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=181892.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 09:47:25,481 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.77 vs. limit=2.0 2023-04-03 09:47:49,881 INFO [train.py:903] (1/4) Epoch 27, batch 4400, loss[loss=0.1744, simple_loss=0.2552, pruned_loss=0.0468, over 19618.00 frames. ], tot_loss[loss=0.2016, simple_loss=0.2826, pruned_loss=0.06032, over 3820629.92 frames. ], batch size: 50, lr: 3.03e-03, grad_scale: 8.0 2023-04-03 09:48:15,011 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_rvb from training. Duration: 25.285 2023-04-03 09:48:24,205 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 2023-04-03 09:48:42,710 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.593e+02 4.919e+02 6.507e+02 9.105e+02 1.976e+03, threshold=1.301e+03, percent-clipped=10.0 2023-04-03 09:48:52,990 INFO [train.py:903] (1/4) Epoch 27, batch 4450, loss[loss=0.2369, simple_loss=0.3068, pruned_loss=0.08351, over 18793.00 frames. ], tot_loss[loss=0.2032, simple_loss=0.284, pruned_loss=0.06124, over 3822602.31 frames. ], batch size: 74, lr: 3.03e-03, grad_scale: 8.0 2023-04-03 09:48:53,752 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 2023-04-03 09:48:59,257 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0383, 2.1427, 2.3376, 2.6338, 2.0486, 2.5387, 2.2745, 2.1595], device='cuda:1'), covar=tensor([0.4212, 0.4124, 0.1999, 0.2732, 0.4408, 0.2360, 0.4820, 0.3430], device='cuda:1'), in_proj_covar=tensor([0.0930, 0.1006, 0.0739, 0.0948, 0.0907, 0.0848, 0.0855, 0.0805], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-03 09:49:16,012 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=181996.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 09:49:30,783 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=182007.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 09:49:36,523 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.6931, 4.2774, 2.8778, 3.7604, 0.9706, 4.2387, 4.1276, 4.1813], device='cuda:1'), covar=tensor([0.0576, 0.1027, 0.1869, 0.0840, 0.4032, 0.0642, 0.0925, 0.1091], device='cuda:1'), in_proj_covar=tensor([0.0528, 0.0427, 0.0513, 0.0357, 0.0407, 0.0453, 0.0449, 0.0477], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 09:49:56,884 INFO [train.py:903] (1/4) Epoch 27, batch 4500, loss[loss=0.1891, simple_loss=0.2689, pruned_loss=0.05459, over 19743.00 frames. ], tot_loss[loss=0.2033, simple_loss=0.2839, pruned_loss=0.06128, over 3816894.07 frames. ], batch size: 51, lr: 3.03e-03, grad_scale: 8.0 2023-04-03 09:49:59,816 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=182030.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 09:50:31,945 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=182055.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 09:50:49,974 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.312e+02 5.059e+02 6.218e+02 7.785e+02 2.105e+03, threshold=1.244e+03, percent-clipped=5.0 2023-04-03 09:51:00,204 INFO [train.py:903] (1/4) Epoch 27, batch 4550, loss[loss=0.203, simple_loss=0.2837, pruned_loss=0.06113, over 19858.00 frames. ], tot_loss[loss=0.2043, simple_loss=0.2851, pruned_loss=0.06171, over 3815250.30 frames. ], batch size: 52, lr: 3.03e-03, grad_scale: 8.0 2023-04-03 09:51:09,775 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9_rvb from training. Duration: 28.72225 2023-04-03 09:51:32,217 WARNING [train.py:1073] (1/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 2023-04-03 09:51:35,765 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6252, 2.2853, 1.6534, 1.5188, 2.1252, 1.3706, 1.3856, 1.9826], device='cuda:1'), covar=tensor([0.1108, 0.0835, 0.1120, 0.0991, 0.0558, 0.1399, 0.0838, 0.0538], device='cuda:1'), in_proj_covar=tensor([0.0301, 0.0321, 0.0337, 0.0273, 0.0251, 0.0345, 0.0292, 0.0277], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 09:52:02,911 INFO [train.py:903] (1/4) Epoch 27, batch 4600, loss[loss=0.2, simple_loss=0.2826, pruned_loss=0.05866, over 17226.00 frames. ], tot_loss[loss=0.2035, simple_loss=0.2842, pruned_loss=0.06135, over 3826101.67 frames. ], batch size: 101, lr: 3.03e-03, grad_scale: 8.0 2023-04-03 09:52:54,758 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.336e+02 4.830e+02 5.724e+02 7.323e+02 1.391e+03, threshold=1.145e+03, percent-clipped=2.0 2023-04-03 09:53:05,189 INFO [train.py:903] (1/4) Epoch 27, batch 4650, loss[loss=0.1757, simple_loss=0.2647, pruned_loss=0.04336, over 19670.00 frames. ], tot_loss[loss=0.2019, simple_loss=0.2834, pruned_loss=0.06018, over 3837075.61 frames. ], batch size: 53, lr: 3.02e-03, grad_scale: 8.0 2023-04-03 09:53:22,617 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_rvb from training. Duration: 0.92 2023-04-03 09:53:34,185 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_rvb from training. Duration: 0.83 2023-04-03 09:53:57,746 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5693, 1.8231, 2.1292, 1.8817, 3.1267, 2.5745, 3.4678, 1.7016], device='cuda:1'), covar=tensor([0.2537, 0.4216, 0.2635, 0.1947, 0.1584, 0.2213, 0.1513, 0.4232], device='cuda:1'), in_proj_covar=tensor([0.0552, 0.0667, 0.0751, 0.0505, 0.0636, 0.0544, 0.0668, 0.0570], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 09:54:07,706 INFO [train.py:903] (1/4) Epoch 27, batch 4700, loss[loss=0.2226, simple_loss=0.2964, pruned_loss=0.07443, over 19629.00 frames. ], tot_loss[loss=0.2014, simple_loss=0.2832, pruned_loss=0.05976, over 3835622.13 frames. ], batch size: 50, lr: 3.02e-03, grad_scale: 8.0 2023-04-03 09:54:30,876 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9_rvb from training. Duration: 0.92225 2023-04-03 09:54:51,517 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=182263.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 09:54:59,130 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.994e+02 4.730e+02 5.895e+02 7.660e+02 1.174e+03, threshold=1.179e+03, percent-clipped=2.0 2023-04-03 09:55:10,405 INFO [train.py:903] (1/4) Epoch 27, batch 4750, loss[loss=0.2057, simple_loss=0.2896, pruned_loss=0.06093, over 19510.00 frames. ], tot_loss[loss=0.2, simple_loss=0.282, pruned_loss=0.059, over 3846733.15 frames. ], batch size: 56, lr: 3.02e-03, grad_scale: 8.0 2023-04-03 09:55:22,518 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=182288.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 09:55:36,008 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=182299.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 09:56:12,226 INFO [train.py:903] (1/4) Epoch 27, batch 4800, loss[loss=0.1798, simple_loss=0.2673, pruned_loss=0.04613, over 19674.00 frames. ], tot_loss[loss=0.2017, simple_loss=0.2834, pruned_loss=0.06004, over 3816024.35 frames. ], batch size: 53, lr: 3.02e-03, grad_scale: 8.0 2023-04-03 09:56:26,913 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=182340.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 09:56:27,639 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 2023-04-03 09:56:42,892 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.6110, 2.8542, 2.3199, 2.8182, 2.6899, 2.3786, 2.1970, 2.7778], device='cuda:1'), covar=tensor([0.1051, 0.1405, 0.1461, 0.0976, 0.1236, 0.0500, 0.1480, 0.0626], device='cuda:1'), in_proj_covar=tensor([0.0277, 0.0358, 0.0317, 0.0257, 0.0307, 0.0256, 0.0321, 0.0264], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 09:57:03,570 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.957e+02 4.862e+02 5.780e+02 7.243e+02 1.108e+03, threshold=1.156e+03, percent-clipped=0.0 2023-04-03 09:57:13,598 INFO [train.py:903] (1/4) Epoch 27, batch 4850, loss[loss=0.2112, simple_loss=0.2921, pruned_loss=0.06517, over 19332.00 frames. ], tot_loss[loss=0.203, simple_loss=0.2845, pruned_loss=0.06076, over 3807071.28 frames. ], batch size: 66, lr: 3.02e-03, grad_scale: 8.0 2023-04-03 09:57:36,912 WARNING [train.py:1073] (1/4) Exclude cut with ID 774-127930-0014-48411_sp1.1_rvb from training. Duration: 0.95 2023-04-03 09:57:58,336 WARNING [train.py:1073] (1/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1_rvb from training. Duration: 0.9409375 2023-04-03 09:58:03,902 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 2023-04-03 09:58:03,927 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 2023-04-03 09:58:13,254 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1_rvb from training. Duration: 25.3818125 2023-04-03 09:58:14,420 INFO [train.py:903] (1/4) Epoch 27, batch 4900, loss[loss=0.2275, simple_loss=0.3031, pruned_loss=0.07592, over 19765.00 frames. ], tot_loss[loss=0.203, simple_loss=0.2844, pruned_loss=0.06081, over 3798843.53 frames. ], batch size: 63, lr: 3.02e-03, grad_scale: 8.0 2023-04-03 09:58:34,893 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 2023-04-03 09:58:50,325 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=182455.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 09:59:07,216 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.920e+02 4.786e+02 5.871e+02 7.388e+02 1.622e+03, threshold=1.174e+03, percent-clipped=2.0 2023-04-03 09:59:18,868 INFO [train.py:903] (1/4) Epoch 27, batch 4950, loss[loss=0.1882, simple_loss=0.2787, pruned_loss=0.04892, over 19624.00 frames. ], tot_loss[loss=0.2026, simple_loss=0.2842, pruned_loss=0.06055, over 3808917.46 frames. ], batch size: 57, lr: 3.02e-03, grad_scale: 8.0 2023-04-03 09:59:36,557 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 2023-04-03 10:00:00,881 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 2023-04-03 10:00:21,851 INFO [train.py:903] (1/4) Epoch 27, batch 5000, loss[loss=0.1637, simple_loss=0.2427, pruned_loss=0.04231, over 19765.00 frames. ], tot_loss[loss=0.202, simple_loss=0.2832, pruned_loss=0.06036, over 3798385.22 frames. ], batch size: 47, lr: 3.02e-03, grad_scale: 8.0 2023-04-03 10:00:27,430 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=182532.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 10:00:32,550 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_rvb from training. Duration: 27.14 2023-04-03 10:00:44,445 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 2023-04-03 10:01:15,171 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.165e+02 4.878e+02 5.976e+02 7.448e+02 1.686e+03, threshold=1.195e+03, percent-clipped=3.0 2023-04-03 10:01:25,326 INFO [train.py:903] (1/4) Epoch 27, batch 5050, loss[loss=0.21, simple_loss=0.2937, pruned_loss=0.0631, over 19752.00 frames. ], tot_loss[loss=0.2023, simple_loss=0.2834, pruned_loss=0.06059, over 3793867.86 frames. ], batch size: 63, lr: 3.02e-03, grad_scale: 8.0 2023-04-03 10:01:41,295 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.49 vs. limit=5.0 2023-04-03 10:02:02,774 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9_rvb from training. Duration: 26.62775 2023-04-03 10:02:27,484 INFO [train.py:903] (1/4) Epoch 27, batch 5100, loss[loss=0.2074, simple_loss=0.2894, pruned_loss=0.06269, over 18765.00 frames. ], tot_loss[loss=0.2019, simple_loss=0.2833, pruned_loss=0.06029, over 3787586.03 frames. ], batch size: 74, lr: 3.02e-03, grad_scale: 8.0 2023-04-03 10:02:37,796 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 2023-04-03 10:02:42,003 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_rvb from training. Duration: 25.035 2023-04-03 10:02:46,621 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_rvb from training. Duration: 27.92 2023-04-03 10:02:46,774 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=182643.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 10:03:19,758 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.602e+02 5.110e+02 6.408e+02 8.268e+02 2.195e+03, threshold=1.282e+03, percent-clipped=9.0 2023-04-03 10:03:30,281 INFO [train.py:903] (1/4) Epoch 27, batch 5150, loss[loss=0.205, simple_loss=0.2955, pruned_loss=0.05728, over 19665.00 frames. ], tot_loss[loss=0.2023, simple_loss=0.2836, pruned_loss=0.06051, over 3776794.79 frames. ], batch size: 58, lr: 3.02e-03, grad_scale: 4.0 2023-04-03 10:03:44,254 WARNING [train.py:1073] (1/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 2023-04-03 10:04:12,442 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=182711.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 10:04:21,109 WARNING [train.py:1073] (1/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1_rvb from training. Duration: 0.9681875 2023-04-03 10:04:23,045 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.08 vs. limit=2.0 2023-04-03 10:04:34,488 INFO [train.py:903] (1/4) Epoch 27, batch 5200, loss[loss=0.2003, simple_loss=0.2768, pruned_loss=0.06191, over 19853.00 frames. ], tot_loss[loss=0.2013, simple_loss=0.2826, pruned_loss=0.06, over 3787707.95 frames. ], batch size: 52, lr: 3.02e-03, grad_scale: 8.0 2023-04-03 10:04:45,073 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=182736.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 10:04:50,601 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 2023-04-03 10:05:11,541 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=182758.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 10:05:17,511 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.7402, 1.8588, 2.1811, 2.2362, 1.6761, 2.1448, 2.1552, 1.9440], device='cuda:1'), covar=tensor([0.4289, 0.3973, 0.1959, 0.2569, 0.4248, 0.2292, 0.4918, 0.3573], device='cuda:1'), in_proj_covar=tensor([0.0927, 0.1005, 0.0738, 0.0946, 0.0905, 0.0844, 0.0854, 0.0803], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-03 10:05:28,503 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.060e+02 4.726e+02 5.796e+02 7.282e+02 1.371e+03, threshold=1.159e+03, percent-clipped=1.0 2023-04-03 10:05:35,290 WARNING [train.py:1073] (1/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 2023-04-03 10:05:37,678 INFO [train.py:903] (1/4) Epoch 27, batch 5250, loss[loss=0.1868, simple_loss=0.2718, pruned_loss=0.05095, over 19538.00 frames. ], tot_loss[loss=0.2015, simple_loss=0.2828, pruned_loss=0.06012, over 3785982.39 frames. ], batch size: 54, lr: 3.02e-03, grad_scale: 8.0 2023-04-03 10:06:39,522 INFO [train.py:903] (1/4) Epoch 27, batch 5300, loss[loss=0.166, simple_loss=0.2415, pruned_loss=0.04524, over 19751.00 frames. ], tot_loss[loss=0.201, simple_loss=0.2825, pruned_loss=0.05979, over 3788949.52 frames. ], batch size: 46, lr: 3.02e-03, grad_scale: 4.0 2023-04-03 10:06:57,455 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 2023-04-03 10:07:31,280 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4783, 1.5698, 1.8562, 1.7512, 2.6789, 2.2657, 2.7618, 1.1905], device='cuda:1'), covar=tensor([0.2572, 0.4465, 0.2861, 0.1975, 0.1471, 0.2219, 0.1458, 0.4822], device='cuda:1'), in_proj_covar=tensor([0.0553, 0.0668, 0.0753, 0.0505, 0.0635, 0.0545, 0.0669, 0.0571], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 10:07:34,153 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.440e+02 4.985e+02 5.848e+02 7.587e+02 2.195e+03, threshold=1.170e+03, percent-clipped=4.0 2023-04-03 10:07:37,804 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=182874.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 10:07:39,953 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=182876.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 10:07:42,167 INFO [train.py:903] (1/4) Epoch 27, batch 5350, loss[loss=0.1847, simple_loss=0.2613, pruned_loss=0.05406, over 19391.00 frames. ], tot_loss[loss=0.2013, simple_loss=0.2825, pruned_loss=0.06001, over 3787889.54 frames. ], batch size: 48, lr: 3.02e-03, grad_scale: 4.0 2023-04-03 10:07:44,841 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=182880.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 10:08:16,929 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9_rvb from training. Duration: 26.438875 2023-04-03 10:08:46,439 INFO [train.py:903] (1/4) Epoch 27, batch 5400, loss[loss=0.1769, simple_loss=0.2523, pruned_loss=0.05081, over 19767.00 frames. ], tot_loss[loss=0.2004, simple_loss=0.2818, pruned_loss=0.05952, over 3806409.20 frames. ], batch size: 45, lr: 3.02e-03, grad_scale: 4.0 2023-04-03 10:09:18,292 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5246, 1.6075, 1.8818, 1.7760, 2.8149, 2.2927, 3.0070, 1.3085], device='cuda:1'), covar=tensor([0.2609, 0.4485, 0.2935, 0.2055, 0.1508, 0.2291, 0.1393, 0.4840], device='cuda:1'), in_proj_covar=tensor([0.0554, 0.0669, 0.0755, 0.0507, 0.0636, 0.0546, 0.0670, 0.0573], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-03 10:09:41,330 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.624e+02 4.932e+02 6.009e+02 7.341e+02 1.388e+03, threshold=1.202e+03, percent-clipped=2.0 2023-04-03 10:09:49,110 INFO [train.py:903] (1/4) Epoch 27, batch 5450, loss[loss=0.2323, simple_loss=0.3058, pruned_loss=0.0794, over 19504.00 frames. ], tot_loss[loss=0.2012, simple_loss=0.2826, pruned_loss=0.05986, over 3812506.13 frames. ], batch size: 64, lr: 3.02e-03, grad_scale: 4.0 2023-04-03 10:10:04,459 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=182991.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 10:10:35,589 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=183014.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 10:10:50,992 INFO [train.py:903] (1/4) Epoch 27, batch 5500, loss[loss=0.2902, simple_loss=0.3534, pruned_loss=0.1135, over 12932.00 frames. ], tot_loss[loss=0.2002, simple_loss=0.2814, pruned_loss=0.05946, over 3814219.43 frames. ], batch size: 137, lr: 3.02e-03, grad_scale: 4.0 2023-04-03 10:11:05,209 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=183039.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 10:11:13,896 WARNING [train.py:1073] (1/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9_rvb from training. Duration: 27.47775 2023-04-03 10:11:26,361 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0768, 2.0632, 1.7536, 2.1040, 1.8543, 1.7689, 1.7579, 2.0100], device='cuda:1'), covar=tensor([0.1147, 0.1527, 0.1621, 0.1128, 0.1476, 0.0624, 0.1545, 0.0826], device='cuda:1'), in_proj_covar=tensor([0.0278, 0.0357, 0.0317, 0.0258, 0.0307, 0.0256, 0.0320, 0.0265], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 10:11:45,355 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.090e+02 5.032e+02 6.184e+02 7.491e+02 1.557e+03, threshold=1.237e+03, percent-clipped=5.0 2023-04-03 10:11:52,115 INFO [train.py:903] (1/4) Epoch 27, batch 5550, loss[loss=0.2015, simple_loss=0.2937, pruned_loss=0.05465, over 19515.00 frames. ], tot_loss[loss=0.2011, simple_loss=0.2827, pruned_loss=0.05974, over 3810089.17 frames. ], batch size: 56, lr: 3.02e-03, grad_scale: 4.0 2023-04-03 10:11:57,938 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 2023-04-03 10:12:19,477 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.3295, 3.8721, 3.9717, 3.9733, 1.6358, 3.7897, 3.3133, 3.7429], device='cuda:1'), covar=tensor([0.1727, 0.0930, 0.0766, 0.0852, 0.6102, 0.1065, 0.0738, 0.1301], device='cuda:1'), in_proj_covar=tensor([0.0818, 0.0782, 0.0991, 0.0871, 0.0862, 0.0753, 0.0586, 0.0919], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-03 10:12:39,053 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.5039, 2.0582, 2.3029, 3.1162, 2.1724, 2.4879, 2.5295, 2.3283], device='cuda:1'), covar=tensor([0.0744, 0.0924, 0.0907, 0.0715, 0.0847, 0.0773, 0.0920, 0.0686], device='cuda:1'), in_proj_covar=tensor([0.0213, 0.0225, 0.0229, 0.0241, 0.0226, 0.0213, 0.0188, 0.0207], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-04-03 10:12:46,687 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 2023-04-03 10:12:50,105 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.0423, 3.7099, 2.5622, 3.2890, 0.8012, 3.6629, 3.5415, 3.5544], device='cuda:1'), covar=tensor([0.0816, 0.1073, 0.1974, 0.0955, 0.4013, 0.0767, 0.1096, 0.1356], device='cuda:1'), in_proj_covar=tensor([0.0529, 0.0426, 0.0515, 0.0358, 0.0410, 0.0453, 0.0450, 0.0478], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 10:12:56,188 INFO [train.py:903] (1/4) Epoch 27, batch 5600, loss[loss=0.2841, simple_loss=0.3594, pruned_loss=0.1044, over 19709.00 frames. ], tot_loss[loss=0.2016, simple_loss=0.2829, pruned_loss=0.06015, over 3799034.83 frames. ], batch size: 63, lr: 3.02e-03, grad_scale: 8.0 2023-04-03 10:13:25,412 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.0672, 4.5248, 4.8470, 4.8158, 1.7517, 4.5120, 3.8646, 4.5642], device='cuda:1'), covar=tensor([0.1668, 0.0808, 0.0547, 0.0659, 0.6363, 0.0884, 0.0696, 0.1086], device='cuda:1'), in_proj_covar=tensor([0.0816, 0.0780, 0.0987, 0.0868, 0.0859, 0.0750, 0.0585, 0.0917], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-03 10:13:51,540 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.195e+02 4.790e+02 6.069e+02 7.863e+02 1.689e+03, threshold=1.214e+03, percent-clipped=3.0 2023-04-03 10:13:59,319 INFO [train.py:903] (1/4) Epoch 27, batch 5650, loss[loss=0.1541, simple_loss=0.2311, pruned_loss=0.03853, over 19732.00 frames. ], tot_loss[loss=0.2023, simple_loss=0.2833, pruned_loss=0.06065, over 3804004.90 frames. ], batch size: 46, lr: 3.02e-03, grad_scale: 8.0 2023-04-03 10:14:44,837 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9_rvb from training. Duration: 28.0944375 2023-04-03 10:14:49,511 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=183218.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 10:14:56,356 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=183224.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 10:15:00,999 INFO [train.py:903] (1/4) Epoch 27, batch 5700, loss[loss=0.2041, simple_loss=0.2877, pruned_loss=0.0603, over 19679.00 frames. ], tot_loss[loss=0.2026, simple_loss=0.2836, pruned_loss=0.06083, over 3787406.35 frames. ], batch size: 55, lr: 3.02e-03, grad_scale: 8.0 2023-04-03 10:15:25,105 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=183247.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 10:15:40,177 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.4296, 2.4822, 2.6297, 3.0861, 2.5410, 2.9286, 2.6051, 2.4885], device='cuda:1'), covar=tensor([0.3686, 0.3527, 0.1781, 0.2343, 0.3735, 0.2104, 0.4083, 0.2992], device='cuda:1'), in_proj_covar=tensor([0.0930, 0.1009, 0.0740, 0.0946, 0.0908, 0.0847, 0.0857, 0.0804], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-03 10:15:54,498 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.213e+02 5.103e+02 5.954e+02 7.593e+02 1.308e+03, threshold=1.191e+03, percent-clipped=1.0 2023-04-03 10:15:54,918 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=183272.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 10:15:55,553 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 2023-04-03 10:15:59,212 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9_rvb from training. Duration: 33.038875 2023-04-03 10:16:01,590 INFO [train.py:903] (1/4) Epoch 27, batch 5750, loss[loss=0.1689, simple_loss=0.241, pruned_loss=0.04837, over 19785.00 frames. ], tot_loss[loss=0.2035, simple_loss=0.2842, pruned_loss=0.06139, over 3785374.16 frames. ], batch size: 48, lr: 3.02e-03, grad_scale: 8.0 2023-04-03 10:16:08,296 WARNING [train.py:1073] (1/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 2023-04-03 10:16:12,815 WARNING [train.py:1073] (1/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 2023-04-03 10:16:32,338 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.4517, 1.3238, 1.5604, 1.3894, 3.0476, 1.1328, 2.4026, 3.4811], device='cuda:1'), covar=tensor([0.0574, 0.2878, 0.2940, 0.1976, 0.0730, 0.2512, 0.1149, 0.0271], device='cuda:1'), in_proj_covar=tensor([0.0422, 0.0375, 0.0395, 0.0351, 0.0382, 0.0356, 0.0392, 0.0414], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 10:17:05,413 INFO [train.py:903] (1/4) Epoch 27, batch 5800, loss[loss=0.1602, simple_loss=0.2334, pruned_loss=0.0435, over 19775.00 frames. ], tot_loss[loss=0.2025, simple_loss=0.2834, pruned_loss=0.06078, over 3791238.77 frames. ], batch size: 45, lr: 3.02e-03, grad_scale: 8.0 2023-04-03 10:17:11,539 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.8613, 1.2462, 1.6233, 1.5457, 4.2488, 1.2253, 2.7123, 4.6886], device='cuda:1'), covar=tensor([0.0596, 0.3822, 0.3470, 0.2441, 0.1160, 0.3145, 0.1519, 0.0255], device='cuda:1'), in_proj_covar=tensor([0.0422, 0.0376, 0.0395, 0.0352, 0.0383, 0.0357, 0.0394, 0.0415], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 10:17:11,585 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=183333.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 10:17:19,543 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=183339.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 10:17:59,574 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.172e+02 5.364e+02 7.027e+02 9.052e+02 1.891e+03, threshold=1.405e+03, percent-clipped=4.0 2023-04-03 10:18:07,743 INFO [train.py:903] (1/4) Epoch 27, batch 5850, loss[loss=0.184, simple_loss=0.2707, pruned_loss=0.04866, over 19307.00 frames. ], tot_loss[loss=0.2028, simple_loss=0.2836, pruned_loss=0.06101, over 3808757.98 frames. ], batch size: 66, lr: 3.02e-03, grad_scale: 8.0 2023-04-03 10:19:01,473 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4430, 2.0500, 1.6105, 1.4477, 1.9760, 1.2323, 1.3797, 1.8211], device='cuda:1'), covar=tensor([0.1050, 0.0939, 0.1131, 0.0888, 0.0561, 0.1370, 0.0781, 0.0531], device='cuda:1'), in_proj_covar=tensor([0.0300, 0.0318, 0.0335, 0.0272, 0.0249, 0.0342, 0.0291, 0.0276], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 10:19:07,141 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=183426.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 10:19:08,994 INFO [train.py:903] (1/4) Epoch 27, batch 5900, loss[loss=0.2196, simple_loss=0.3045, pruned_loss=0.06729, over 19412.00 frames. ], tot_loss[loss=0.2029, simple_loss=0.284, pruned_loss=0.06093, over 3805551.87 frames. ], batch size: 70, lr: 3.01e-03, grad_scale: 8.0 2023-04-03 10:19:09,042 WARNING [train.py:1073] (1/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 2023-04-03 10:19:21,567 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 2023-04-03 10:19:32,071 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9_rvb from training. Duration: 27.511125 2023-04-03 10:20:00,754 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.13 vs. limit=2.0 2023-04-03 10:20:03,306 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.958e+02 4.868e+02 5.663e+02 7.388e+02 1.454e+03, threshold=1.133e+03, percent-clipped=1.0 2023-04-03 10:20:10,248 INFO [train.py:903] (1/4) Epoch 27, batch 5950, loss[loss=0.2219, simple_loss=0.3051, pruned_loss=0.06939, over 17344.00 frames. ], tot_loss[loss=0.2031, simple_loss=0.2842, pruned_loss=0.06098, over 3810691.51 frames. ], batch size: 101, lr: 3.01e-03, grad_scale: 8.0 2023-04-03 10:21:12,625 INFO [train.py:903] (1/4) Epoch 27, batch 6000, loss[loss=0.1783, simple_loss=0.2699, pruned_loss=0.04336, over 19678.00 frames. ], tot_loss[loss=0.2032, simple_loss=0.2844, pruned_loss=0.06101, over 3805882.31 frames. ], batch size: 59, lr: 3.01e-03, grad_scale: 8.0 2023-04-03 10:21:12,625 INFO [train.py:928] (1/4) Computing validation loss 2023-04-03 10:21:19,204 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6008, 1.6353, 1.6530, 1.4016, 1.4188, 1.3997, 0.3653, 0.7263], device='cuda:1'), covar=tensor([0.0867, 0.0813, 0.0509, 0.0799, 0.1572, 0.0957, 0.1518, 0.1385], device='cuda:1'), in_proj_covar=tensor([0.0364, 0.0361, 0.0368, 0.0391, 0.0469, 0.0394, 0.0344, 0.0347], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 10:21:25,589 INFO [train.py:937] (1/4) Epoch 27, validation: loss=0.1675, simple_loss=0.2669, pruned_loss=0.03401, over 944034.00 frames. 2023-04-03 10:21:25,590 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18821MB 2023-04-03 10:22:22,434 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.170e+02 5.376e+02 6.105e+02 7.773e+02 1.848e+03, threshold=1.221e+03, percent-clipped=6.0 2023-04-03 10:22:28,411 INFO [train.py:903] (1/4) Epoch 27, batch 6050, loss[loss=0.2252, simple_loss=0.3069, pruned_loss=0.07179, over 18837.00 frames. ], tot_loss[loss=0.2034, simple_loss=0.2844, pruned_loss=0.0612, over 3813203.56 frames. ], batch size: 74, lr: 3.01e-03, grad_scale: 4.0 2023-04-03 10:22:43,111 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=183589.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 10:22:50,991 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=183595.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 10:23:14,128 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=183614.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 10:23:20,871 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=183620.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 10:23:31,578 INFO [train.py:903] (1/4) Epoch 27, batch 6100, loss[loss=0.277, simple_loss=0.3372, pruned_loss=0.1084, over 13326.00 frames. ], tot_loss[loss=0.2016, simple_loss=0.2827, pruned_loss=0.0603, over 3826447.50 frames. ], batch size: 135, lr: 3.01e-03, grad_scale: 4.0 2023-04-03 10:23:37,180 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=183632.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 10:23:37,353 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5468, 1.7285, 2.0421, 1.8332, 3.3511, 2.6157, 3.6291, 1.6478], device='cuda:1'), covar=tensor([0.2555, 0.4325, 0.2826, 0.1959, 0.1379, 0.2158, 0.1359, 0.4333], device='cuda:1'), in_proj_covar=tensor([0.0554, 0.0670, 0.0756, 0.0509, 0.0638, 0.0547, 0.0670, 0.0575], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-03 10:24:27,767 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.311e+02 5.176e+02 6.035e+02 7.763e+02 1.396e+03, threshold=1.207e+03, percent-clipped=4.0 2023-04-03 10:24:33,563 INFO [train.py:903] (1/4) Epoch 27, batch 6150, loss[loss=0.1901, simple_loss=0.2772, pruned_loss=0.05153, over 19527.00 frames. ], tot_loss[loss=0.2011, simple_loss=0.2824, pruned_loss=0.05994, over 3823809.05 frames. ], batch size: 54, lr: 3.01e-03, grad_scale: 4.0 2023-04-03 10:25:01,325 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9_rvb from training. Duration: 31.02225 2023-04-03 10:25:34,990 INFO [train.py:903] (1/4) Epoch 27, batch 6200, loss[loss=0.2284, simple_loss=0.3123, pruned_loss=0.07224, over 18058.00 frames. ], tot_loss[loss=0.2004, simple_loss=0.2818, pruned_loss=0.05947, over 3831973.21 frames. ], batch size: 83, lr: 3.01e-03, grad_scale: 4.0 2023-04-03 10:26:28,519 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=183770.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 10:26:31,939 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.388e+02 4.839e+02 5.826e+02 7.637e+02 1.855e+03, threshold=1.165e+03, percent-clipped=4.0 2023-04-03 10:26:37,752 INFO [train.py:903] (1/4) Epoch 27, batch 6250, loss[loss=0.1932, simple_loss=0.2726, pruned_loss=0.05693, over 19602.00 frames. ], tot_loss[loss=0.2017, simple_loss=0.2829, pruned_loss=0.06023, over 3834820.38 frames. ], batch size: 52, lr: 3.01e-03, grad_scale: 4.0 2023-04-03 10:27:08,077 WARNING [train.py:1073] (1/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9_rvb from training. Duration: 25.988875 2023-04-03 10:27:40,120 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5801, 1.2436, 1.2429, 1.5136, 1.1444, 1.3491, 1.2467, 1.4131], device='cuda:1'), covar=tensor([0.1215, 0.1282, 0.1662, 0.1111, 0.1377, 0.0664, 0.1666, 0.0944], device='cuda:1'), in_proj_covar=tensor([0.0277, 0.0358, 0.0317, 0.0256, 0.0306, 0.0256, 0.0320, 0.0264], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 10:27:40,819 INFO [train.py:903] (1/4) Epoch 27, batch 6300, loss[loss=0.2455, simple_loss=0.3273, pruned_loss=0.08187, over 19265.00 frames. ], tot_loss[loss=0.2017, simple_loss=0.283, pruned_loss=0.06017, over 3836378.74 frames. ], batch size: 66, lr: 3.01e-03, grad_scale: 4.0 2023-04-03 10:28:20,269 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=183860.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 10:28:36,678 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.559e+02 4.928e+02 5.882e+02 7.199e+02 1.705e+03, threshold=1.176e+03, percent-clipped=2.0 2023-04-03 10:28:43,631 INFO [train.py:903] (1/4) Epoch 27, batch 6350, loss[loss=0.22, simple_loss=0.2913, pruned_loss=0.07438, over 19748.00 frames. ], tot_loss[loss=0.202, simple_loss=0.2831, pruned_loss=0.06041, over 3822468.30 frames. ], batch size: 51, lr: 3.01e-03, grad_scale: 4.0 2023-04-03 10:28:51,881 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=183885.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 10:29:43,990 INFO [train.py:903] (1/4) Epoch 27, batch 6400, loss[loss=0.192, simple_loss=0.2799, pruned_loss=0.05206, over 19775.00 frames. ], tot_loss[loss=0.2027, simple_loss=0.2839, pruned_loss=0.06079, over 3818002.89 frames. ], batch size: 54, lr: 3.01e-03, grad_scale: 8.0 2023-04-03 10:30:39,967 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.119e+02 4.757e+02 6.035e+02 7.575e+02 1.901e+03, threshold=1.207e+03, percent-clipped=7.0 2023-04-03 10:30:43,676 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=183976.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 10:30:45,040 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2936, 1.3240, 1.6916, 1.2464, 2.5180, 3.4097, 3.1219, 3.5910], device='cuda:1'), covar=tensor([0.1543, 0.3837, 0.3453, 0.2626, 0.0627, 0.0194, 0.0221, 0.0256], device='cuda:1'), in_proj_covar=tensor([0.0279, 0.0331, 0.0363, 0.0270, 0.0253, 0.0195, 0.0219, 0.0277], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-03 10:30:45,878 INFO [train.py:903] (1/4) Epoch 27, batch 6450, loss[loss=0.1908, simple_loss=0.2675, pruned_loss=0.05709, over 19479.00 frames. ], tot_loss[loss=0.2033, simple_loss=0.2845, pruned_loss=0.06105, over 3821994.90 frames. ], batch size: 49, lr: 3.01e-03, grad_scale: 8.0 2023-04-03 10:31:28,488 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9_rvb from training. Duration: 28.638875 2023-04-03 10:31:42,577 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3188, 1.9732, 1.5130, 1.3976, 1.8606, 1.2492, 1.3233, 1.8282], device='cuda:1'), covar=tensor([0.0952, 0.0834, 0.1133, 0.0892, 0.0496, 0.1343, 0.0709, 0.0476], device='cuda:1'), in_proj_covar=tensor([0.0302, 0.0320, 0.0337, 0.0273, 0.0251, 0.0344, 0.0293, 0.0277], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 10:31:50,547 INFO [train.py:903] (1/4) Epoch 27, batch 6500, loss[loss=0.1959, simple_loss=0.2877, pruned_loss=0.05209, over 19607.00 frames. ], tot_loss[loss=0.2015, simple_loss=0.2829, pruned_loss=0.06006, over 3828363.62 frames. ], batch size: 61, lr: 3.01e-03, grad_scale: 8.0 2023-04-03 10:31:52,981 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1_rvb from training. Duration: 0.836375 2023-04-03 10:32:04,489 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([5.0555, 1.7774, 2.0164, 1.7500, 4.5867, 1.2976, 2.6933, 4.9979], device='cuda:1'), covar=tensor([0.0478, 0.2848, 0.2886, 0.2007, 0.0723, 0.2636, 0.1463, 0.0171], device='cuda:1'), in_proj_covar=tensor([0.0424, 0.0378, 0.0397, 0.0354, 0.0385, 0.0357, 0.0395, 0.0415], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 10:32:11,857 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.63 vs. limit=2.0 2023-04-03 10:32:40,167 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=184068.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 10:32:46,399 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.099e+02 4.542e+02 5.565e+02 7.286e+02 1.442e+03, threshold=1.113e+03, percent-clipped=3.0 2023-04-03 10:32:53,091 INFO [train.py:903] (1/4) Epoch 27, batch 6550, loss[loss=0.1832, simple_loss=0.2756, pruned_loss=0.04537, over 19767.00 frames. ], tot_loss[loss=0.2017, simple_loss=0.2834, pruned_loss=0.05998, over 3827495.95 frames. ], batch size: 56, lr: 3.01e-03, grad_scale: 8.0 2023-04-03 10:33:08,661 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=184091.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 10:33:24,683 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6105, 1.6427, 1.8486, 1.8163, 2.6559, 2.3556, 2.7898, 1.1913], device='cuda:1'), covar=tensor([0.2458, 0.4460, 0.2857, 0.1986, 0.1495, 0.2143, 0.1422, 0.4804], device='cuda:1'), in_proj_covar=tensor([0.0552, 0.0669, 0.0751, 0.0507, 0.0636, 0.0544, 0.0666, 0.0571], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 10:33:51,928 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2240, 1.2989, 1.6058, 1.2367, 2.5503, 3.3747, 3.1092, 3.5867], device='cuda:1'), covar=tensor([0.1588, 0.3887, 0.3557, 0.2586, 0.0646, 0.0213, 0.0259, 0.0299], device='cuda:1'), in_proj_covar=tensor([0.0279, 0.0332, 0.0363, 0.0271, 0.0253, 0.0195, 0.0220, 0.0277], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-03 10:33:55,011 INFO [train.py:903] (1/4) Epoch 27, batch 6600, loss[loss=0.2346, simple_loss=0.3124, pruned_loss=0.07845, over 19276.00 frames. ], tot_loss[loss=0.2015, simple_loss=0.2833, pruned_loss=0.05987, over 3833650.07 frames. ], batch size: 66, lr: 3.01e-03, grad_scale: 8.0 2023-04-03 10:34:06,002 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.4136, 2.4027, 2.5763, 3.2195, 2.4472, 3.0515, 2.4937, 2.4608], device='cuda:1'), covar=tensor([0.4250, 0.4371, 0.1921, 0.2469, 0.4593, 0.2156, 0.5354, 0.3360], device='cuda:1'), in_proj_covar=tensor([0.0933, 0.1009, 0.0740, 0.0947, 0.0909, 0.0846, 0.0859, 0.0803], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-03 10:34:11,902 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=184141.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 10:34:43,108 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=184166.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 10:34:46,826 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 2023-04-03 10:34:46,856 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 2023-04-03 10:34:50,802 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.027e+02 4.833e+02 5.798e+02 7.178e+02 1.551e+03, threshold=1.160e+03, percent-clipped=2.0 2023-04-03 10:34:58,088 INFO [train.py:903] (1/4) Epoch 27, batch 6650, loss[loss=0.1877, simple_loss=0.2805, pruned_loss=0.04749, over 19455.00 frames. ], tot_loss[loss=0.2011, simple_loss=0.283, pruned_loss=0.05963, over 3824897.13 frames. ], batch size: 64, lr: 3.01e-03, grad_scale: 8.0 2023-04-03 10:35:30,806 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=184204.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 10:35:59,896 INFO [train.py:903] (1/4) Epoch 27, batch 6700, loss[loss=0.2087, simple_loss=0.2854, pruned_loss=0.06605, over 19691.00 frames. ], tot_loss[loss=0.2017, simple_loss=0.2832, pruned_loss=0.06004, over 3821091.67 frames. ], batch size: 53, lr: 3.01e-03, grad_scale: 8.0 2023-04-03 10:36:52,206 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.305e+02 5.122e+02 6.084e+02 8.596e+02 2.606e+03, threshold=1.217e+03, percent-clipped=9.0 2023-04-03 10:36:57,977 INFO [train.py:903] (1/4) Epoch 27, batch 6750, loss[loss=0.1902, simple_loss=0.275, pruned_loss=0.05266, over 18253.00 frames. ], tot_loss[loss=0.2022, simple_loss=0.2834, pruned_loss=0.06048, over 3816114.45 frames. ], batch size: 83, lr: 3.01e-03, grad_scale: 8.0 2023-04-03 10:37:31,820 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.3486, 4.0099, 2.5543, 3.5200, 0.8463, 3.9258, 3.8118, 3.9233], device='cuda:1'), covar=tensor([0.0752, 0.1050, 0.2053, 0.0858, 0.4027, 0.0740, 0.1017, 0.1180], device='cuda:1'), in_proj_covar=tensor([0.0526, 0.0425, 0.0516, 0.0357, 0.0407, 0.0455, 0.0451, 0.0479], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 10:37:44,317 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=184319.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 10:37:54,076 INFO [train.py:903] (1/4) Epoch 27, batch 6800, loss[loss=0.2136, simple_loss=0.2911, pruned_loss=0.06805, over 19614.00 frames. ], tot_loss[loss=0.204, simple_loss=0.285, pruned_loss=0.06149, over 3802813.14 frames. ], batch size: 57, lr: 3.01e-03, grad_scale: 8.0 2023-04-03 10:38:15,929 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=184347.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 10:38:39,949 WARNING [train.py:1073] (1/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 2023-04-03 10:38:41,605 WARNING [train.py:1073] (1/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 2023-04-03 10:38:43,614 INFO [train.py:903] (1/4) Epoch 28, batch 0, loss[loss=0.2313, simple_loss=0.3111, pruned_loss=0.07572, over 19657.00 frames. ], tot_loss[loss=0.2313, simple_loss=0.3111, pruned_loss=0.07572, over 19657.00 frames. ], batch size: 55, lr: 2.95e-03, grad_scale: 8.0 2023-04-03 10:38:43,615 INFO [train.py:928] (1/4) Computing validation loss 2023-04-03 10:38:54,475 INFO [train.py:937] (1/4) Epoch 28, validation: loss=0.1665, simple_loss=0.2666, pruned_loss=0.03316, over 944034.00 frames. 2023-04-03 10:38:54,476 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18821MB 2023-04-03 10:39:08,348 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 2023-04-03 10:39:14,468 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=184372.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 10:39:15,177 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.715e+02 5.190e+02 6.304e+02 8.212e+02 1.288e+03, threshold=1.261e+03, percent-clipped=2.0 2023-04-03 10:39:21,702 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=184377.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 10:39:57,733 INFO [train.py:903] (1/4) Epoch 28, batch 50, loss[loss=0.166, simple_loss=0.2465, pruned_loss=0.04271, over 19756.00 frames. ], tot_loss[loss=0.1984, simple_loss=0.2806, pruned_loss=0.05811, over 866180.12 frames. ], batch size: 45, lr: 2.95e-03, grad_scale: 8.0 2023-04-03 10:40:04,863 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=184412.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 10:40:25,864 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4914, 1.6389, 1.8717, 1.7859, 2.5738, 2.2698, 2.7757, 1.1652], device='cuda:1'), covar=tensor([0.2541, 0.4426, 0.2847, 0.1960, 0.1590, 0.2224, 0.1488, 0.4853], device='cuda:1'), in_proj_covar=tensor([0.0553, 0.0668, 0.0751, 0.0506, 0.0636, 0.0544, 0.0668, 0.0571], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 10:40:32,268 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1_rvb from training. Duration: 27.0318125 2023-04-03 10:40:58,079 INFO [train.py:903] (1/4) Epoch 28, batch 100, loss[loss=0.2012, simple_loss=0.2874, pruned_loss=0.0575, over 18126.00 frames. ], tot_loss[loss=0.1971, simple_loss=0.2794, pruned_loss=0.05739, over 1535157.98 frames. ], batch size: 83, lr: 2.95e-03, grad_scale: 8.0 2023-04-03 10:41:08,321 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 2023-04-03 10:41:18,601 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.133e+02 4.572e+02 5.776e+02 7.316e+02 1.195e+03, threshold=1.155e+03, percent-clipped=0.0 2023-04-03 10:41:56,020 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=184504.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 10:41:58,099 INFO [train.py:903] (1/4) Epoch 28, batch 150, loss[loss=0.1602, simple_loss=0.2393, pruned_loss=0.0406, over 19719.00 frames. ], tot_loss[loss=0.1999, simple_loss=0.2813, pruned_loss=0.05925, over 2041962.95 frames. ], batch size: 45, lr: 2.95e-03, grad_scale: 8.0 2023-04-03 10:42:24,557 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=184527.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 10:42:57,418 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 2023-04-03 10:42:58,575 INFO [train.py:903] (1/4) Epoch 28, batch 200, loss[loss=0.1852, simple_loss=0.2792, pruned_loss=0.0456, over 18123.00 frames. ], tot_loss[loss=0.201, simple_loss=0.2821, pruned_loss=0.05992, over 2438615.72 frames. ], batch size: 83, lr: 2.95e-03, grad_scale: 8.0 2023-04-03 10:43:19,481 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.222e+02 4.969e+02 6.258e+02 7.516e+02 2.266e+03, threshold=1.252e+03, percent-clipped=4.0 2023-04-03 10:43:22,260 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=184575.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 10:43:24,687 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9835, 1.7777, 1.6953, 2.0188, 1.6869, 1.6669, 1.6536, 1.8746], device='cuda:1'), covar=tensor([0.1126, 0.1641, 0.1522, 0.1162, 0.1499, 0.0626, 0.1559, 0.0811], device='cuda:1'), in_proj_covar=tensor([0.0279, 0.0360, 0.0320, 0.0258, 0.0307, 0.0258, 0.0321, 0.0266], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 10:43:51,887 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=184600.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 10:43:59,861 INFO [train.py:903] (1/4) Epoch 28, batch 250, loss[loss=0.2497, simple_loss=0.3192, pruned_loss=0.0901, over 19543.00 frames. ], tot_loss[loss=0.2026, simple_loss=0.2836, pruned_loss=0.06079, over 2744999.45 frames. ], batch size: 56, lr: 2.95e-03, grad_scale: 8.0 2023-04-03 10:45:01,674 INFO [train.py:903] (1/4) Epoch 28, batch 300, loss[loss=0.249, simple_loss=0.3192, pruned_loss=0.0894, over 19620.00 frames. ], tot_loss[loss=0.2018, simple_loss=0.2831, pruned_loss=0.06029, over 2991804.47 frames. ], batch size: 61, lr: 2.95e-03, grad_scale: 8.0 2023-04-03 10:45:22,244 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.089e+02 4.936e+02 6.396e+02 8.049e+02 1.564e+03, threshold=1.279e+03, percent-clipped=7.0 2023-04-03 10:45:29,098 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.78 vs. limit=2.0 2023-04-03 10:45:40,684 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1231, 1.2596, 1.6732, 1.0147, 2.3646, 3.0327, 2.7887, 3.2614], device='cuda:1'), covar=tensor([0.1702, 0.4136, 0.3571, 0.2884, 0.0671, 0.0236, 0.0276, 0.0339], device='cuda:1'), in_proj_covar=tensor([0.0280, 0.0333, 0.0364, 0.0272, 0.0255, 0.0196, 0.0220, 0.0278], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-03 10:46:02,770 INFO [train.py:903] (1/4) Epoch 28, batch 350, loss[loss=0.1996, simple_loss=0.2911, pruned_loss=0.05405, over 19692.00 frames. ], tot_loss[loss=0.2023, simple_loss=0.2835, pruned_loss=0.06057, over 3189196.01 frames. ], batch size: 59, lr: 2.95e-03, grad_scale: 8.0 2023-04-03 10:46:06,321 WARNING [train.py:1073] (1/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1_rvb from training. Duration: 0.9681875 2023-04-03 10:46:11,121 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1449, 1.3508, 1.6474, 1.0353, 2.3244, 2.9847, 2.7268, 3.1848], device='cuda:1'), covar=tensor([0.1687, 0.3890, 0.3568, 0.2924, 0.0717, 0.0297, 0.0287, 0.0369], device='cuda:1'), in_proj_covar=tensor([0.0280, 0.0334, 0.0364, 0.0272, 0.0255, 0.0196, 0.0220, 0.0278], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-03 10:46:20,186 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=184721.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 10:47:04,866 INFO [train.py:903] (1/4) Epoch 28, batch 400, loss[loss=0.1953, simple_loss=0.2743, pruned_loss=0.0581, over 19469.00 frames. ], tot_loss[loss=0.2025, simple_loss=0.2837, pruned_loss=0.06061, over 3342713.13 frames. ], batch size: 49, lr: 2.95e-03, grad_scale: 8.0 2023-04-03 10:47:07,990 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0 2023-04-03 10:47:24,994 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.223e+02 4.910e+02 5.909e+02 7.518e+02 1.907e+03, threshold=1.182e+03, percent-clipped=3.0 2023-04-03 10:47:33,359 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.5847, 1.3135, 1.5246, 1.1990, 2.1929, 1.0520, 2.2181, 2.5752], device='cuda:1'), covar=tensor([0.0760, 0.2804, 0.2781, 0.1804, 0.0932, 0.2192, 0.0964, 0.0437], device='cuda:1'), in_proj_covar=tensor([0.0426, 0.0380, 0.0398, 0.0354, 0.0385, 0.0359, 0.0397, 0.0417], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 10:47:38,186 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=184783.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 10:47:54,349 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.8396, 1.6872, 1.4862, 1.8047, 1.5985, 1.5508, 1.4879, 1.7090], device='cuda:1'), covar=tensor([0.1242, 0.1232, 0.1690, 0.1082, 0.1258, 0.0692, 0.1655, 0.0853], device='cuda:1'), in_proj_covar=tensor([0.0278, 0.0359, 0.0318, 0.0258, 0.0306, 0.0257, 0.0321, 0.0265], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 10:48:05,123 INFO [train.py:903] (1/4) Epoch 28, batch 450, loss[loss=0.2555, simple_loss=0.3304, pruned_loss=0.09028, over 18118.00 frames. ], tot_loss[loss=0.2022, simple_loss=0.2836, pruned_loss=0.0604, over 3438655.55 frames. ], batch size: 83, lr: 2.95e-03, grad_scale: 8.0 2023-04-03 10:48:07,819 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=184808.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 10:48:20,498 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0677, 1.9006, 1.7813, 2.0702, 1.7032, 1.7316, 1.7491, 2.0164], device='cuda:1'), covar=tensor([0.1125, 0.1481, 0.1568, 0.1049, 0.1487, 0.0612, 0.1529, 0.0764], device='cuda:1'), in_proj_covar=tensor([0.0277, 0.0358, 0.0318, 0.0257, 0.0306, 0.0256, 0.0320, 0.0264], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 10:48:32,961 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.1295, 1.3123, 1.4395, 1.3511, 2.7601, 1.0712, 2.1969, 3.1462], device='cuda:1'), covar=tensor([0.0568, 0.2848, 0.3029, 0.1942, 0.0717, 0.2501, 0.1258, 0.0298], device='cuda:1'), in_proj_covar=tensor([0.0424, 0.0379, 0.0398, 0.0353, 0.0384, 0.0358, 0.0396, 0.0417], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 10:48:38,475 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 2023-04-03 10:48:39,491 WARNING [train.py:1073] (1/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 2023-04-03 10:48:43,493 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=184836.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 10:48:56,807 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=184848.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 10:49:06,861 INFO [train.py:903] (1/4) Epoch 28, batch 500, loss[loss=0.1801, simple_loss=0.2594, pruned_loss=0.05044, over 19635.00 frames. ], tot_loss[loss=0.2019, simple_loss=0.2831, pruned_loss=0.06034, over 3507846.05 frames. ], batch size: 50, lr: 2.95e-03, grad_scale: 8.0 2023-04-03 10:49:28,424 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.110e+02 5.109e+02 6.404e+02 8.256e+02 1.456e+03, threshold=1.281e+03, percent-clipped=5.0 2023-04-03 10:49:33,517 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4252, 1.4913, 1.7477, 1.6540, 2.6295, 2.2165, 2.7761, 1.2321], device='cuda:1'), covar=tensor([0.2587, 0.4428, 0.2867, 0.2061, 0.1548, 0.2269, 0.1435, 0.4544], device='cuda:1'), in_proj_covar=tensor([0.0551, 0.0666, 0.0750, 0.0505, 0.0635, 0.0543, 0.0666, 0.0569], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 10:49:41,467 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=184884.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 10:50:02,645 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.0958, 2.8406, 2.2238, 2.1183, 1.8505, 2.3968, 0.9726, 2.0449], device='cuda:1'), covar=tensor([0.0756, 0.0739, 0.0875, 0.1374, 0.1409, 0.1324, 0.1657, 0.1219], device='cuda:1'), in_proj_covar=tensor([0.0368, 0.0365, 0.0371, 0.0395, 0.0472, 0.0397, 0.0347, 0.0351], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 10:50:09,152 INFO [train.py:903] (1/4) Epoch 28, batch 550, loss[loss=0.2018, simple_loss=0.2888, pruned_loss=0.05745, over 19114.00 frames. ], tot_loss[loss=0.2016, simple_loss=0.2831, pruned_loss=0.06001, over 3582490.85 frames. ], batch size: 69, lr: 2.95e-03, grad_scale: 8.0 2023-04-03 10:51:08,833 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.35 vs. limit=5.0 2023-04-03 10:51:11,433 INFO [train.py:903] (1/4) Epoch 28, batch 600, loss[loss=0.2351, simple_loss=0.2911, pruned_loss=0.08953, over 18212.00 frames. ], tot_loss[loss=0.2019, simple_loss=0.2835, pruned_loss=0.06012, over 3639147.53 frames. ], batch size: 40, lr: 2.95e-03, grad_scale: 8.0 2023-04-03 10:51:12,910 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=184957.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 10:51:19,892 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=184963.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 10:51:31,268 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.601e+02 5.007e+02 6.276e+02 8.222e+02 1.849e+03, threshold=1.255e+03, percent-clipped=3.0 2023-04-03 10:51:50,596 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 2023-04-03 10:52:14,345 INFO [train.py:903] (1/4) Epoch 28, batch 650, loss[loss=0.1993, simple_loss=0.2902, pruned_loss=0.05417, over 19533.00 frames. ], tot_loss[loss=0.2006, simple_loss=0.2825, pruned_loss=0.05934, over 3686580.85 frames. ], batch size: 54, lr: 2.95e-03, grad_scale: 8.0 2023-04-03 10:53:16,128 INFO [train.py:903] (1/4) Epoch 28, batch 700, loss[loss=0.1735, simple_loss=0.2486, pruned_loss=0.0492, over 19716.00 frames. ], tot_loss[loss=0.2001, simple_loss=0.2818, pruned_loss=0.05919, over 3703572.96 frames. ], batch size: 46, lr: 2.95e-03, grad_scale: 8.0 2023-04-03 10:53:38,016 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.143e+02 4.589e+02 5.550e+02 7.126e+02 1.317e+03, threshold=1.110e+03, percent-clipped=1.0 2023-04-03 10:53:45,166 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=185078.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 10:54:01,331 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=185092.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 10:54:19,784 INFO [train.py:903] (1/4) Epoch 28, batch 750, loss[loss=0.2043, simple_loss=0.2811, pruned_loss=0.06369, over 19520.00 frames. ], tot_loss[loss=0.2017, simple_loss=0.2828, pruned_loss=0.06026, over 3723036.72 frames. ], batch size: 56, lr: 2.95e-03, grad_scale: 8.0 2023-04-03 10:54:34,001 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=185117.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 10:55:10,733 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=185147.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 10:55:20,946 INFO [train.py:903] (1/4) Epoch 28, batch 800, loss[loss=0.204, simple_loss=0.2885, pruned_loss=0.05977, over 19662.00 frames. ], tot_loss[loss=0.2031, simple_loss=0.2844, pruned_loss=0.06084, over 3742571.71 frames. ], batch size: 55, lr: 2.95e-03, grad_scale: 8.0 2023-04-03 10:55:30,807 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.49 vs. limit=2.0 2023-04-03 10:55:34,902 WARNING [train.py:1073] (1/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 2023-04-03 10:55:41,840 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.136e+02 5.493e+02 6.558e+02 7.858e+02 2.224e+03, threshold=1.312e+03, percent-clipped=8.0 2023-04-03 10:56:22,074 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([6.2452, 5.7099, 3.1233, 5.0825, 1.2028, 5.9469, 5.6681, 5.8471], device='cuda:1'), covar=tensor([0.0366, 0.0779, 0.1820, 0.0699, 0.3859, 0.0500, 0.0731, 0.0971], device='cuda:1'), in_proj_covar=tensor([0.0527, 0.0426, 0.0517, 0.0359, 0.0409, 0.0456, 0.0451, 0.0480], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 10:56:24,246 INFO [train.py:903] (1/4) Epoch 28, batch 850, loss[loss=0.18, simple_loss=0.2576, pruned_loss=0.05122, over 19379.00 frames. ], tot_loss[loss=0.2024, simple_loss=0.2842, pruned_loss=0.06024, over 3755156.39 frames. ], batch size: 47, lr: 2.95e-03, grad_scale: 8.0 2023-04-03 10:56:39,446 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=185219.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 10:56:51,614 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=185228.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 10:57:11,558 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=185244.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 10:57:15,606 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9_rvb from training. Duration: 25.061125 2023-04-03 10:57:24,911 INFO [train.py:903] (1/4) Epoch 28, batch 900, loss[loss=0.1747, simple_loss=0.2478, pruned_loss=0.0508, over 19750.00 frames. ], tot_loss[loss=0.2035, simple_loss=0.285, pruned_loss=0.06098, over 3764862.18 frames. ], batch size: 47, lr: 2.94e-03, grad_scale: 8.0 2023-04-03 10:57:37,059 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 2023-04-03 10:57:47,702 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.300e+02 4.798e+02 5.781e+02 7.336e+02 1.381e+03, threshold=1.156e+03, percent-clipped=1.0 2023-04-03 10:58:21,555 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=185301.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 10:58:28,121 INFO [train.py:903] (1/4) Epoch 28, batch 950, loss[loss=0.1995, simple_loss=0.2829, pruned_loss=0.05806, over 17374.00 frames. ], tot_loss[loss=0.2023, simple_loss=0.2838, pruned_loss=0.06034, over 3784636.38 frames. ], batch size: 101, lr: 2.94e-03, grad_scale: 8.0 2023-04-03 10:58:29,317 WARNING [train.py:1073] (1/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9_rvb from training. Duration: 26.32775 2023-04-03 10:59:11,094 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.4176, 3.7308, 2.2006, 2.4227, 3.3857, 2.0678, 1.7603, 2.4860], device='cuda:1'), covar=tensor([0.1101, 0.0460, 0.0974, 0.0699, 0.0439, 0.1059, 0.0811, 0.0588], device='cuda:1'), in_proj_covar=tensor([0.0299, 0.0317, 0.0335, 0.0271, 0.0249, 0.0341, 0.0290, 0.0274], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 10:59:15,343 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=185343.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 10:59:32,239 INFO [train.py:903] (1/4) Epoch 28, batch 1000, loss[loss=0.1901, simple_loss=0.2763, pruned_loss=0.052, over 18139.00 frames. ], tot_loss[loss=0.2022, simple_loss=0.2838, pruned_loss=0.06029, over 3795212.82 frames. ], batch size: 83, lr: 2.94e-03, grad_scale: 8.0 2023-04-03 10:59:49,500 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([5.1521, 1.9013, 2.1616, 1.8983, 4.6959, 1.4259, 2.7098, 5.1566], device='cuda:1'), covar=tensor([0.0440, 0.2522, 0.2459, 0.1869, 0.0705, 0.2511, 0.1475, 0.0142], device='cuda:1'), in_proj_covar=tensor([0.0421, 0.0376, 0.0395, 0.0350, 0.0381, 0.0356, 0.0393, 0.0414], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 10:59:53,731 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.982e+02 4.891e+02 5.853e+02 7.878e+02 2.572e+03, threshold=1.171e+03, percent-clipped=6.0 2023-04-03 10:59:57,745 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 2023-04-03 11:00:23,369 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 2023-04-03 11:00:34,046 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.4558, 3.1517, 2.6114, 2.6728, 2.5509, 2.7566, 1.2811, 2.3391], device='cuda:1'), covar=tensor([0.0617, 0.0615, 0.0639, 0.0982, 0.0943, 0.0999, 0.1430, 0.1060], device='cuda:1'), in_proj_covar=tensor([0.0364, 0.0363, 0.0369, 0.0391, 0.0470, 0.0394, 0.0346, 0.0347], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 11:00:34,737 INFO [train.py:903] (1/4) Epoch 28, batch 1050, loss[loss=0.1702, simple_loss=0.2456, pruned_loss=0.04736, over 18654.00 frames. ], tot_loss[loss=0.2008, simple_loss=0.2823, pruned_loss=0.05968, over 3814698.02 frames. ], batch size: 41, lr: 2.94e-03, grad_scale: 8.0 2023-04-03 11:00:47,025 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=185416.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:00:53,766 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=185422.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:01:02,627 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 2023-04-03 11:01:14,706 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.9929, 4.4712, 4.7786, 4.7463, 1.7526, 4.4687, 3.8049, 4.4901], device='cuda:1'), covar=tensor([0.1770, 0.0970, 0.0610, 0.0793, 0.6376, 0.1064, 0.0724, 0.1199], device='cuda:1'), in_proj_covar=tensor([0.0824, 0.0785, 0.0998, 0.0874, 0.0867, 0.0762, 0.0588, 0.0926], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-03 11:01:35,152 INFO [train.py:903] (1/4) Epoch 28, batch 1100, loss[loss=0.275, simple_loss=0.3349, pruned_loss=0.1075, over 13278.00 frames. ], tot_loss[loss=0.2015, simple_loss=0.2822, pruned_loss=0.06036, over 3815918.45 frames. ], batch size: 136, lr: 2.94e-03, grad_scale: 8.0 2023-04-03 11:01:57,233 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.219e+02 4.947e+02 6.329e+02 8.288e+02 1.903e+03, threshold=1.266e+03, percent-clipped=3.0 2023-04-03 11:02:13,020 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=185486.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:02:18,717 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=185491.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:02:31,470 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.7839, 1.4464, 1.6055, 1.5836, 3.3650, 1.0198, 2.2884, 3.8738], device='cuda:1'), covar=tensor([0.0488, 0.2908, 0.2942, 0.1926, 0.0705, 0.2772, 0.1476, 0.0215], device='cuda:1'), in_proj_covar=tensor([0.0422, 0.0377, 0.0396, 0.0351, 0.0381, 0.0357, 0.0393, 0.0415], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 11:02:35,892 INFO [train.py:903] (1/4) Epoch 28, batch 1150, loss[loss=0.1708, simple_loss=0.2495, pruned_loss=0.04603, over 19752.00 frames. ], tot_loss[loss=0.2025, simple_loss=0.2831, pruned_loss=0.06097, over 3822780.73 frames. ], batch size: 46, lr: 2.94e-03, grad_scale: 8.0 2023-04-03 11:03:15,647 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=185537.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:03:28,125 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.15 vs. limit=5.0 2023-04-03 11:03:40,781 INFO [train.py:903] (1/4) Epoch 28, batch 1200, loss[loss=0.1672, simple_loss=0.2515, pruned_loss=0.04143, over 19836.00 frames. ], tot_loss[loss=0.2013, simple_loss=0.2824, pruned_loss=0.0601, over 3817407.57 frames. ], batch size: 52, lr: 2.94e-03, grad_scale: 8.0 2023-04-03 11:04:01,639 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.374e+02 5.383e+02 6.799e+02 8.653e+02 1.626e+03, threshold=1.360e+03, percent-clipped=3.0 2023-04-03 11:04:11,775 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 2023-04-03 11:04:34,064 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=185599.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:04:41,623 INFO [train.py:903] (1/4) Epoch 28, batch 1250, loss[loss=0.2054, simple_loss=0.2837, pruned_loss=0.06356, over 19596.00 frames. ], tot_loss[loss=0.2026, simple_loss=0.2834, pruned_loss=0.06092, over 3816652.42 frames. ], batch size: 52, lr: 2.94e-03, grad_scale: 16.0 2023-04-03 11:04:42,042 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=185606.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:05:03,712 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=185624.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:05:44,893 INFO [train.py:903] (1/4) Epoch 28, batch 1300, loss[loss=0.2109, simple_loss=0.2991, pruned_loss=0.06131, over 17483.00 frames. ], tot_loss[loss=0.2028, simple_loss=0.2835, pruned_loss=0.06101, over 3805758.17 frames. ], batch size: 101, lr: 2.94e-03, grad_scale: 8.0 2023-04-03 11:05:46,475 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.9068, 1.4610, 1.6374, 1.5863, 4.4021, 1.1070, 2.6406, 4.8012], device='cuda:1'), covar=tensor([0.0499, 0.2875, 0.3153, 0.2147, 0.0767, 0.2796, 0.1475, 0.0186], device='cuda:1'), in_proj_covar=tensor([0.0421, 0.0376, 0.0396, 0.0351, 0.0381, 0.0357, 0.0392, 0.0415], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 11:06:04,624 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=185672.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:06:06,496 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.213e+02 4.542e+02 5.530e+02 7.592e+02 1.164e+03, threshold=1.106e+03, percent-clipped=0.0 2023-04-03 11:06:35,902 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=185697.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:06:39,506 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0260, 2.5358, 1.8111, 1.8807, 2.3659, 1.7425, 1.7506, 2.2248], device='cuda:1'), covar=tensor([0.0921, 0.0777, 0.0835, 0.0762, 0.0508, 0.1037, 0.0677, 0.0544], device='cuda:1'), in_proj_covar=tensor([0.0303, 0.0320, 0.0339, 0.0273, 0.0250, 0.0346, 0.0292, 0.0277], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 11:06:46,291 INFO [train.py:903] (1/4) Epoch 28, batch 1350, loss[loss=0.2003, simple_loss=0.2824, pruned_loss=0.0591, over 19679.00 frames. ], tot_loss[loss=0.2036, simple_loss=0.2845, pruned_loss=0.06135, over 3791589.34 frames. ], batch size: 59, lr: 2.94e-03, grad_scale: 8.0 2023-04-03 11:07:48,954 INFO [train.py:903] (1/4) Epoch 28, batch 1400, loss[loss=0.1889, simple_loss=0.2668, pruned_loss=0.05551, over 19791.00 frames. ], tot_loss[loss=0.2021, simple_loss=0.2834, pruned_loss=0.06041, over 3807766.20 frames. ], batch size: 49, lr: 2.94e-03, grad_scale: 8.0 2023-04-03 11:08:11,867 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.444e+02 4.899e+02 6.106e+02 7.699e+02 1.518e+03, threshold=1.221e+03, percent-clipped=6.0 2023-04-03 11:08:15,727 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=185777.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:08:35,111 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=185793.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:08:44,814 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.8095, 3.2410, 3.3102, 3.3316, 1.3299, 3.1855, 2.7638, 3.1107], device='cuda:1'), covar=tensor([0.1843, 0.1266, 0.0887, 0.1003, 0.6045, 0.1276, 0.0926, 0.1349], device='cuda:1'), in_proj_covar=tensor([0.0826, 0.0788, 0.1001, 0.0875, 0.0869, 0.0763, 0.0591, 0.0928], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-03 11:08:49,361 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_rvb from training. Duration: 25.85 2023-04-03 11:08:51,546 INFO [train.py:903] (1/4) Epoch 28, batch 1450, loss[loss=0.2312, simple_loss=0.3226, pruned_loss=0.06989, over 19588.00 frames. ], tot_loss[loss=0.2009, simple_loss=0.2827, pruned_loss=0.05953, over 3821214.63 frames. ], batch size: 61, lr: 2.94e-03, grad_scale: 8.0 2023-04-03 11:09:05,625 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=185817.0, num_to_drop=1, layers_to_drop={1} 2023-04-03 11:09:06,849 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=185818.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:09:20,217 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=185830.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:09:54,220 INFO [train.py:903] (1/4) Epoch 28, batch 1500, loss[loss=0.2044, simple_loss=0.2796, pruned_loss=0.06463, over 19385.00 frames. ], tot_loss[loss=0.2022, simple_loss=0.2836, pruned_loss=0.06041, over 3816994.43 frames. ], batch size: 48, lr: 2.94e-03, grad_scale: 8.0 2023-04-03 11:10:02,009 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=185862.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:10:15,979 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.811e+02 5.017e+02 6.227e+02 8.666e+02 1.816e+03, threshold=1.245e+03, percent-clipped=11.0 2023-04-03 11:10:33,553 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=185887.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:10:56,450 INFO [train.py:903] (1/4) Epoch 28, batch 1550, loss[loss=0.1497, simple_loss=0.232, pruned_loss=0.0337, over 19741.00 frames. ], tot_loss[loss=0.202, simple_loss=0.2835, pruned_loss=0.06029, over 3828222.69 frames. ], batch size: 47, lr: 2.94e-03, grad_scale: 8.0 2023-04-03 11:11:45,513 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=185945.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:11:58,297 INFO [train.py:903] (1/4) Epoch 28, batch 1600, loss[loss=0.217, simple_loss=0.3145, pruned_loss=0.05981, over 19550.00 frames. ], tot_loss[loss=0.2023, simple_loss=0.2835, pruned_loss=0.06053, over 3821311.03 frames. ], batch size: 56, lr: 2.94e-03, grad_scale: 8.0 2023-04-03 11:12:20,811 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9_rvb from training. Duration: 30.1555625 2023-04-03 11:12:23,177 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.991e+02 4.773e+02 5.899e+02 6.974e+02 1.687e+03, threshold=1.180e+03, percent-clipped=2.0 2023-04-03 11:12:28,397 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.75 vs. limit=5.0 2023-04-03 11:13:03,463 INFO [train.py:903] (1/4) Epoch 28, batch 1650, loss[loss=0.2349, simple_loss=0.3106, pruned_loss=0.07965, over 19099.00 frames. ], tot_loss[loss=0.2032, simple_loss=0.2845, pruned_loss=0.06098, over 3830584.79 frames. ], batch size: 69, lr: 2.94e-03, grad_scale: 8.0 2023-04-03 11:13:07,307 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6668, 1.2554, 1.3116, 1.5560, 1.0483, 1.4109, 1.2877, 1.4983], device='cuda:1'), covar=tensor([0.1224, 0.1321, 0.1737, 0.1061, 0.1518, 0.0652, 0.1759, 0.0919], device='cuda:1'), in_proj_covar=tensor([0.0280, 0.0362, 0.0320, 0.0259, 0.0309, 0.0259, 0.0323, 0.0267], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-04-03 11:13:26,194 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.3690, 3.0797, 2.3661, 2.7786, 0.6270, 3.0754, 2.9365, 3.0318], device='cuda:1'), covar=tensor([0.1104, 0.1431, 0.1981, 0.1092, 0.4055, 0.0958, 0.1215, 0.1322], device='cuda:1'), in_proj_covar=tensor([0.0526, 0.0427, 0.0515, 0.0360, 0.0410, 0.0457, 0.0451, 0.0479], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 11:14:07,007 INFO [train.py:903] (1/4) Epoch 28, batch 1700, loss[loss=0.1866, simple_loss=0.2638, pruned_loss=0.05465, over 18991.00 frames. ], tot_loss[loss=0.2013, simple_loss=0.2824, pruned_loss=0.06004, over 3838211.01 frames. ], batch size: 42, lr: 2.94e-03, grad_scale: 8.0 2023-04-03 11:14:29,606 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.897e+02 4.751e+02 5.670e+02 7.170e+02 1.723e+03, threshold=1.134e+03, percent-clipped=7.0 2023-04-03 11:14:44,418 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 2023-04-03 11:15:08,856 INFO [train.py:903] (1/4) Epoch 28, batch 1750, loss[loss=0.1907, simple_loss=0.2714, pruned_loss=0.05498, over 19584.00 frames. ], tot_loss[loss=0.2006, simple_loss=0.2818, pruned_loss=0.05966, over 3821531.63 frames. ], batch size: 52, lr: 2.94e-03, grad_scale: 8.0 2023-04-03 11:15:24,237 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=186118.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:15:28,460 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=186121.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:16:11,526 INFO [train.py:903] (1/4) Epoch 28, batch 1800, loss[loss=0.1829, simple_loss=0.2637, pruned_loss=0.05103, over 19675.00 frames. ], tot_loss[loss=0.2009, simple_loss=0.2822, pruned_loss=0.05983, over 3814122.58 frames. ], batch size: 53, lr: 2.94e-03, grad_scale: 8.0 2023-04-03 11:16:18,364 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=186161.0, num_to_drop=1, layers_to_drop={1} 2023-04-03 11:16:36,643 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.127e+02 5.184e+02 6.048e+02 8.514e+02 1.613e+03, threshold=1.210e+03, percent-clipped=4.0 2023-04-03 11:16:38,107 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=186176.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:16:42,875 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5184, 1.4929, 2.0227, 1.7194, 2.9899, 4.7608, 4.6559, 5.1792], device='cuda:1'), covar=tensor([0.1541, 0.3889, 0.3447, 0.2413, 0.0674, 0.0207, 0.0163, 0.0184], device='cuda:1'), in_proj_covar=tensor([0.0281, 0.0333, 0.0366, 0.0273, 0.0257, 0.0197, 0.0220, 0.0279], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-03 11:17:02,768 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.12 vs. limit=2.0 2023-04-03 11:17:08,561 WARNING [train.py:1073] (1/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9_rvb from training. Duration: 25.0944375 2023-04-03 11:17:09,023 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=186201.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:17:15,219 INFO [train.py:903] (1/4) Epoch 28, batch 1850, loss[loss=0.2096, simple_loss=0.296, pruned_loss=0.0616, over 19579.00 frames. ], tot_loss[loss=0.2015, simple_loss=0.2828, pruned_loss=0.06015, over 3825957.44 frames. ], batch size: 61, lr: 2.94e-03, grad_scale: 8.0 2023-04-03 11:17:39,893 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=186226.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:17:46,433 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9_rvb from training. Duration: 27.8166875 2023-04-03 11:17:52,379 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=186236.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:18:17,540 INFO [train.py:903] (1/4) Epoch 28, batch 1900, loss[loss=0.2031, simple_loss=0.2832, pruned_loss=0.0615, over 19762.00 frames. ], tot_loss[loss=0.2015, simple_loss=0.2826, pruned_loss=0.06015, over 3818826.28 frames. ], batch size: 54, lr: 2.94e-03, grad_scale: 8.0 2023-04-03 11:18:33,594 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 2023-04-03 11:18:38,425 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9_rvb from training. Duration: 27.02225 2023-04-03 11:18:39,534 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.824e+02 4.995e+02 5.845e+02 7.128e+02 1.193e+03, threshold=1.169e+03, percent-clipped=0.0 2023-04-03 11:18:41,962 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=186276.0, num_to_drop=1, layers_to_drop={1} 2023-04-03 11:18:42,901 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=186277.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:19:02,962 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 2023-04-03 11:19:19,221 INFO [train.py:903] (1/4) Epoch 28, batch 1950, loss[loss=0.1599, simple_loss=0.2455, pruned_loss=0.03716, over 19284.00 frames. ], tot_loss[loss=0.2025, simple_loss=0.2838, pruned_loss=0.06063, over 3824245.82 frames. ], batch size: 44, lr: 2.94e-03, grad_scale: 8.0 2023-04-03 11:20:20,372 INFO [train.py:903] (1/4) Epoch 28, batch 2000, loss[loss=0.2232, simple_loss=0.2999, pruned_loss=0.0733, over 18945.00 frames. ], tot_loss[loss=0.2022, simple_loss=0.283, pruned_loss=0.06073, over 3823414.73 frames. ], batch size: 74, lr: 2.94e-03, grad_scale: 8.0 2023-04-03 11:20:32,032 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.1591, 2.0867, 2.0086, 1.8323, 1.6566, 1.7489, 0.7161, 1.1279], device='cuda:1'), covar=tensor([0.0639, 0.0667, 0.0469, 0.0786, 0.1207, 0.0897, 0.1323, 0.1107], device='cuda:1'), in_proj_covar=tensor([0.0366, 0.0365, 0.0370, 0.0392, 0.0470, 0.0396, 0.0346, 0.0349], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 11:20:33,485 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 2023-04-03 11:20:45,110 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.192e+02 5.060e+02 6.550e+02 8.587e+02 3.446e+03, threshold=1.310e+03, percent-clipped=8.0 2023-04-03 11:21:10,085 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5199, 1.7315, 2.1271, 1.8011, 3.1772, 2.5303, 3.4586, 1.5914], device='cuda:1'), covar=tensor([0.2767, 0.4596, 0.2879, 0.2066, 0.1572, 0.2307, 0.1601, 0.4660], device='cuda:1'), in_proj_covar=tensor([0.0555, 0.0671, 0.0757, 0.0509, 0.0638, 0.0546, 0.0671, 0.0574], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-03 11:21:19,993 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_rvb from training. Duration: 26.8349375 2023-04-03 11:21:24,400 INFO [train.py:903] (1/4) Epoch 28, batch 2050, loss[loss=0.1807, simple_loss=0.267, pruned_loss=0.04721, over 19485.00 frames. ], tot_loss[loss=0.2012, simple_loss=0.2825, pruned_loss=0.06001, over 3817305.83 frames. ], batch size: 49, lr: 2.94e-03, grad_scale: 8.0 2023-04-03 11:21:40,390 WARNING [train.py:1073] (1/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 2023-04-03 11:21:42,556 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 2023-04-03 11:22:02,217 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 2023-04-03 11:22:04,680 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=186439.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:22:26,598 INFO [train.py:903] (1/4) Epoch 28, batch 2100, loss[loss=0.2015, simple_loss=0.2899, pruned_loss=0.05662, over 19501.00 frames. ], tot_loss[loss=0.1996, simple_loss=0.2808, pruned_loss=0.05919, over 3820235.11 frames. ], batch size: 64, lr: 2.94e-03, grad_scale: 8.0 2023-04-03 11:22:35,104 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=186462.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:22:49,639 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.238e+02 5.089e+02 6.089e+02 7.390e+02 1.324e+03, threshold=1.218e+03, percent-clipped=1.0 2023-04-03 11:22:58,694 WARNING [train.py:1073] (1/4) Exclude cut with ID 453-131332-0000-131866_sp0.9_rvb from training. Duration: 25.3333125 2023-04-03 11:23:11,858 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=186492.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:23:13,751 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=186493.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:23:20,654 WARNING [train.py:1073] (1/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9_rvb from training. Duration: 26.6166875 2023-04-03 11:23:28,689 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.6233, 4.2424, 2.7450, 3.7747, 1.0881, 4.2356, 4.0105, 4.1468], device='cuda:1'), covar=tensor([0.0610, 0.1003, 0.1941, 0.0828, 0.3896, 0.0650, 0.1019, 0.1111], device='cuda:1'), in_proj_covar=tensor([0.0528, 0.0430, 0.0516, 0.0360, 0.0410, 0.0458, 0.0452, 0.0481], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 11:23:29,683 INFO [train.py:903] (1/4) Epoch 28, batch 2150, loss[loss=0.1922, simple_loss=0.2787, pruned_loss=0.05287, over 19595.00 frames. ], tot_loss[loss=0.1999, simple_loss=0.2811, pruned_loss=0.05941, over 3820103.01 frames. ], batch size: 57, lr: 2.93e-03, grad_scale: 8.0 2023-04-03 11:23:42,638 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=186517.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:23:46,771 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=186520.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:24:02,599 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=186532.0, num_to_drop=1, layers_to_drop={1} 2023-04-03 11:24:06,039 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=186535.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:24:30,566 INFO [train.py:903] (1/4) Epoch 28, batch 2200, loss[loss=0.1991, simple_loss=0.2832, pruned_loss=0.05747, over 19510.00 frames. ], tot_loss[loss=0.2, simple_loss=0.2813, pruned_loss=0.05936, over 3823390.64 frames. ], batch size: 56, lr: 2.93e-03, grad_scale: 8.0 2023-04-03 11:24:32,076 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=186557.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:24:32,183 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=186557.0, num_to_drop=1, layers_to_drop={0} 2023-04-03 11:24:55,772 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.530e+02 4.864e+02 5.746e+02 7.619e+02 1.717e+03, threshold=1.149e+03, percent-clipped=3.0 2023-04-03 11:24:58,668 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=186577.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:25:09,387 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=186585.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:25:32,836 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=186604.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:25:34,941 INFO [train.py:903] (1/4) Epoch 28, batch 2250, loss[loss=0.1953, simple_loss=0.2838, pruned_loss=0.05338, over 19459.00 frames. ], tot_loss[loss=0.2006, simple_loss=0.282, pruned_loss=0.05962, over 3806124.40 frames. ], batch size: 64, lr: 2.93e-03, grad_scale: 8.0 2023-04-03 11:25:54,488 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=186621.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:26:09,358 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.12 vs. limit=2.0 2023-04-03 11:26:12,580 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=186635.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:26:39,407 INFO [train.py:903] (1/4) Epoch 28, batch 2300, loss[loss=0.2167, simple_loss=0.3011, pruned_loss=0.06618, over 17335.00 frames. ], tot_loss[loss=0.2004, simple_loss=0.282, pruned_loss=0.05939, over 3802227.79 frames. ], batch size: 101, lr: 2.93e-03, grad_scale: 8.0 2023-04-03 11:26:55,469 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_rvb from training. Duration: 26.205 2023-04-03 11:27:02,265 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.323e+02 4.896e+02 5.750e+02 7.152e+02 2.246e+03, threshold=1.150e+03, percent-clipped=6.0 2023-04-03 11:27:42,250 INFO [train.py:903] (1/4) Epoch 28, batch 2350, loss[loss=0.196, simple_loss=0.2821, pruned_loss=0.05501, over 19759.00 frames. ], tot_loss[loss=0.2006, simple_loss=0.282, pruned_loss=0.05956, over 3797071.67 frames. ], batch size: 54, lr: 2.93e-03, grad_scale: 8.0 2023-04-03 11:28:07,857 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=186727.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:28:20,530 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=186736.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:28:26,056 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_rvb from training. Duration: 25.775 2023-04-03 11:28:42,177 WARNING [train.py:1073] (1/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9_rvb from training. Duration: 25.45 2023-04-03 11:28:44,642 INFO [train.py:903] (1/4) Epoch 28, batch 2400, loss[loss=0.2, simple_loss=0.2785, pruned_loss=0.06082, over 19597.00 frames. ], tot_loss[loss=0.2004, simple_loss=0.2821, pruned_loss=0.05938, over 3800577.83 frames. ], batch size: 52, lr: 2.93e-03, grad_scale: 8.0 2023-04-03 11:29:08,814 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.283e+02 4.439e+02 5.729e+02 7.466e+02 1.887e+03, threshold=1.146e+03, percent-clipped=9.0 2023-04-03 11:29:20,260 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=186783.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:29:47,720 INFO [train.py:903] (1/4) Epoch 28, batch 2450, loss[loss=0.1733, simple_loss=0.2579, pruned_loss=0.04431, over 19581.00 frames. ], tot_loss[loss=0.2009, simple_loss=0.2829, pruned_loss=0.05945, over 3816064.12 frames. ], batch size: 52, lr: 2.93e-03, grad_scale: 8.0 2023-04-03 11:30:01,665 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 2023-04-03 11:30:02,241 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.5292, 2.5243, 2.2585, 2.6647, 2.3520, 2.1437, 2.1118, 2.4814], device='cuda:1'), covar=tensor([0.1017, 0.1517, 0.1384, 0.1161, 0.1422, 0.0564, 0.1455, 0.0716], device='cuda:1'), in_proj_covar=tensor([0.0281, 0.0362, 0.0321, 0.0259, 0.0308, 0.0258, 0.0323, 0.0265], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-04-03 11:30:22,236 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=186833.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:30:26,704 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=186837.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:30:50,779 INFO [train.py:903] (1/4) Epoch 28, batch 2500, loss[loss=0.195, simple_loss=0.2827, pruned_loss=0.05362, over 19522.00 frames. ], tot_loss[loss=0.2014, simple_loss=0.2837, pruned_loss=0.0596, over 3818548.90 frames. ], batch size: 54, lr: 2.93e-03, grad_scale: 8.0 2023-04-03 11:30:54,427 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=186858.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:30:54,537 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=186858.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:31:15,070 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.341e+02 4.860e+02 5.866e+02 7.110e+02 2.029e+03, threshold=1.173e+03, percent-clipped=7.0 2023-04-03 11:31:19,829 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=186879.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:31:35,086 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=186891.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:31:45,070 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=186898.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:31:48,258 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=186901.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:31:54,933 INFO [train.py:903] (1/4) Epoch 28, batch 2550, loss[loss=0.1818, simple_loss=0.2628, pruned_loss=0.05041, over 19598.00 frames. ], tot_loss[loss=0.2016, simple_loss=0.2836, pruned_loss=0.05979, over 3809668.34 frames. ], batch size: 50, lr: 2.93e-03, grad_scale: 8.0 2023-04-03 11:32:06,770 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=186916.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:32:22,502 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=186929.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:32:22,652 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=186929.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:32:47,702 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=186948.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:32:50,817 WARNING [train.py:1073] (1/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 2023-04-03 11:32:52,347 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=186952.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:32:56,633 INFO [train.py:903] (1/4) Epoch 28, batch 2600, loss[loss=0.2015, simple_loss=0.2853, pruned_loss=0.05889, over 19725.00 frames. ], tot_loss[loss=0.2018, simple_loss=0.2833, pruned_loss=0.06012, over 3819781.88 frames. ], batch size: 63, lr: 2.93e-03, grad_scale: 8.0 2023-04-03 11:33:07,516 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2697, 1.2739, 1.4182, 1.3935, 1.6721, 1.7325, 1.7157, 0.6959], device='cuda:1'), covar=tensor([0.2420, 0.4263, 0.2704, 0.1981, 0.1628, 0.2333, 0.1448, 0.5061], device='cuda:1'), in_proj_covar=tensor([0.0553, 0.0669, 0.0755, 0.0507, 0.0633, 0.0545, 0.0667, 0.0571], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 11:33:20,686 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.371e+02 5.127e+02 5.898e+02 7.788e+02 1.720e+03, threshold=1.180e+03, percent-clipped=7.0 2023-04-03 11:33:43,096 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=186992.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:33:45,238 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=186994.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:33:59,862 INFO [train.py:903] (1/4) Epoch 28, batch 2650, loss[loss=0.2381, simple_loss=0.3257, pruned_loss=0.07523, over 19729.00 frames. ], tot_loss[loss=0.202, simple_loss=0.2835, pruned_loss=0.06025, over 3819184.20 frames. ], batch size: 63, lr: 2.93e-03, grad_scale: 8.0 2023-04-03 11:34:12,839 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=187016.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:34:14,004 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=187017.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:34:22,394 WARNING [train.py:1073] (1/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9_rvb from training. Duration: 27.25 2023-04-03 11:34:42,920 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0001, 2.0898, 2.3079, 2.6376, 2.0264, 2.5752, 2.3250, 2.1322], device='cuda:1'), covar=tensor([0.4435, 0.4086, 0.2039, 0.2512, 0.4346, 0.2206, 0.5038, 0.3534], device='cuda:1'), in_proj_covar=tensor([0.0941, 0.1020, 0.0747, 0.0955, 0.0917, 0.0858, 0.0867, 0.0814], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-03 11:34:46,275 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=187043.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:34:47,536 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=187044.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:35:02,637 INFO [train.py:903] (1/4) Epoch 28, batch 2700, loss[loss=0.1866, simple_loss=0.2599, pruned_loss=0.05668, over 19775.00 frames. ], tot_loss[loss=0.2011, simple_loss=0.2827, pruned_loss=0.0598, over 3831668.76 frames. ], batch size: 46, lr: 2.93e-03, grad_scale: 8.0 2023-04-03 11:35:02,868 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=187056.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:35:13,307 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=187063.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:35:22,266 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=187071.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:35:26,786 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.432e+02 4.685e+02 5.845e+02 7.450e+02 1.306e+03, threshold=1.169e+03, percent-clipped=4.0 2023-04-03 11:36:06,488 INFO [train.py:903] (1/4) Epoch 28, batch 2750, loss[loss=0.1974, simple_loss=0.2856, pruned_loss=0.05467, over 19577.00 frames. ], tot_loss[loss=0.2002, simple_loss=0.282, pruned_loss=0.05926, over 3843693.55 frames. ], batch size: 61, lr: 2.93e-03, grad_scale: 8.0 2023-04-03 11:36:10,419 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6692, 2.2617, 1.6442, 1.5393, 2.0665, 1.3013, 1.4673, 2.0597], device='cuda:1'), covar=tensor([0.1072, 0.0821, 0.1133, 0.0911, 0.0578, 0.1407, 0.0823, 0.0517], device='cuda:1'), in_proj_covar=tensor([0.0303, 0.0320, 0.0341, 0.0273, 0.0252, 0.0345, 0.0293, 0.0277], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 11:36:13,203 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.14 vs. limit=2.0 2023-04-03 11:36:15,445 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.13 vs. limit=2.0 2023-04-03 11:36:18,795 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0633, 1.6554, 1.8400, 2.6729, 1.8048, 2.3679, 2.2623, 2.0196], device='cuda:1'), covar=tensor([0.0843, 0.1013, 0.1027, 0.0808, 0.0972, 0.0804, 0.0908, 0.0746], device='cuda:1'), in_proj_covar=tensor([0.0212, 0.0223, 0.0227, 0.0239, 0.0225, 0.0213, 0.0187, 0.0205], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-04-03 11:36:43,426 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([0.9002, 1.2825, 1.5957, 0.5935, 2.0250, 2.4626, 2.1378, 2.6308], device='cuda:1'), covar=tensor([0.1696, 0.3758, 0.3424, 0.2925, 0.0645, 0.0284, 0.0368, 0.0393], device='cuda:1'), in_proj_covar=tensor([0.0281, 0.0333, 0.0365, 0.0272, 0.0256, 0.0197, 0.0220, 0.0280], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-03 11:36:59,813 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.50 vs. limit=2.0 2023-04-03 11:37:06,480 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=187154.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:37:08,439 INFO [train.py:903] (1/4) Epoch 28, batch 2800, loss[loss=0.1701, simple_loss=0.2418, pruned_loss=0.04924, over 19296.00 frames. ], tot_loss[loss=0.2009, simple_loss=0.2827, pruned_loss=0.05952, over 3842223.46 frames. ], batch size: 44, lr: 2.93e-03, grad_scale: 8.0 2023-04-03 11:37:31,619 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.065e+02 4.906e+02 5.640e+02 7.444e+02 1.445e+03, threshold=1.128e+03, percent-clipped=1.0 2023-04-03 11:37:36,957 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=187179.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:37:46,965 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=187186.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:38:03,796 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9948, 2.0691, 1.7741, 2.1772, 1.9108, 1.7088, 1.7694, 1.9110], device='cuda:1'), covar=tensor([0.1171, 0.1433, 0.1481, 0.1000, 0.1384, 0.0613, 0.1457, 0.0806], device='cuda:1'), in_proj_covar=tensor([0.0280, 0.0360, 0.0319, 0.0257, 0.0305, 0.0257, 0.0322, 0.0264], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 11:38:05,766 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=187202.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:38:10,174 INFO [train.py:903] (1/4) Epoch 28, batch 2850, loss[loss=0.2131, simple_loss=0.2792, pruned_loss=0.0735, over 19627.00 frames. ], tot_loss[loss=0.2006, simple_loss=0.2826, pruned_loss=0.0593, over 3832375.47 frames. ], batch size: 50, lr: 2.93e-03, grad_scale: 8.0 2023-04-03 11:38:13,784 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=187208.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:38:44,959 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=187233.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:39:05,579 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4894, 1.2899, 1.3791, 1.9913, 1.4154, 1.6963, 1.6586, 1.4898], device='cuda:1'), covar=tensor([0.1041, 0.1197, 0.1194, 0.0728, 0.0950, 0.0990, 0.0964, 0.0890], device='cuda:1'), in_proj_covar=tensor([0.0212, 0.0223, 0.0227, 0.0239, 0.0226, 0.0213, 0.0187, 0.0206], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-04-03 11:39:05,636 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=187250.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:39:11,667 WARNING [train.py:1073] (1/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 2023-04-03 11:39:12,697 INFO [train.py:903] (1/4) Epoch 28, batch 2900, loss[loss=0.2234, simple_loss=0.309, pruned_loss=0.06892, over 19595.00 frames. ], tot_loss[loss=0.2008, simple_loss=0.2827, pruned_loss=0.05941, over 3821548.49 frames. ], batch size: 57, lr: 2.93e-03, grad_scale: 8.0 2023-04-03 11:39:13,295 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.13 vs. limit=2.0 2023-04-03 11:39:29,214 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6031, 1.6001, 1.7770, 1.8010, 2.4519, 2.2857, 2.5888, 1.0598], device='cuda:1'), covar=tensor([0.2464, 0.4422, 0.2870, 0.1969, 0.1561, 0.2282, 0.1466, 0.4857], device='cuda:1'), in_proj_covar=tensor([0.0553, 0.0670, 0.0755, 0.0508, 0.0635, 0.0546, 0.0668, 0.0573], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 11:39:32,576 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=187272.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:39:33,467 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=187273.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:39:35,537 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.364e+02 5.012e+02 6.172e+02 7.442e+02 2.226e+03, threshold=1.234e+03, percent-clipped=8.0 2023-04-03 11:39:35,965 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=187275.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:40:02,566 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=187297.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:40:06,079 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=187300.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:40:13,522 INFO [train.py:903] (1/4) Epoch 28, batch 2950, loss[loss=0.158, simple_loss=0.234, pruned_loss=0.04106, over 18580.00 frames. ], tot_loss[loss=0.1999, simple_loss=0.2817, pruned_loss=0.05908, over 3825729.36 frames. ], batch size: 41, lr: 2.93e-03, grad_scale: 8.0 2023-04-03 11:40:25,550 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3940, 2.0761, 1.5462, 1.4576, 1.9179, 1.2226, 1.3634, 1.8574], device='cuda:1'), covar=tensor([0.1103, 0.0892, 0.1261, 0.0892, 0.0629, 0.1426, 0.0782, 0.0530], device='cuda:1'), in_proj_covar=tensor([0.0303, 0.0320, 0.0342, 0.0273, 0.0253, 0.0346, 0.0294, 0.0277], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 11:40:26,515 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=187317.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:40:28,978 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=187319.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:40:35,700 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=187325.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:41:00,003 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=187344.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:41:13,833 INFO [train.py:903] (1/4) Epoch 28, batch 3000, loss[loss=0.1681, simple_loss=0.2443, pruned_loss=0.04589, over 19749.00 frames. ], tot_loss[loss=0.2005, simple_loss=0.2817, pruned_loss=0.05967, over 3831206.47 frames. ], batch size: 46, lr: 2.93e-03, grad_scale: 8.0 2023-04-03 11:41:13,834 INFO [train.py:928] (1/4) Computing validation loss 2023-04-03 11:41:26,712 INFO [train.py:937] (1/4) Epoch 28, validation: loss=0.1673, simple_loss=0.2667, pruned_loss=0.03394, over 944034.00 frames. 2023-04-03 11:41:26,713 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18821MB 2023-04-03 11:41:29,142 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_rvb from training. Duration: 29.735 2023-04-03 11:41:49,225 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.693e+02 4.898e+02 6.411e+02 7.995e+02 1.373e+03, threshold=1.282e+03, percent-clipped=5.0 2023-04-03 11:42:05,123 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=187387.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:42:06,465 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=187388.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:42:08,749 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=187390.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:42:18,223 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.56 vs. limit=2.0 2023-04-03 11:42:21,157 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=187400.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:42:27,790 INFO [train.py:903] (1/4) Epoch 28, batch 3050, loss[loss=0.2387, simple_loss=0.313, pruned_loss=0.08218, over 14003.00 frames. ], tot_loss[loss=0.2012, simple_loss=0.2826, pruned_loss=0.05992, over 3809525.85 frames. ], batch size: 135, lr: 2.93e-03, grad_scale: 8.0 2023-04-03 11:43:13,154 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=187442.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:43:29,484 INFO [train.py:903] (1/4) Epoch 28, batch 3100, loss[loss=0.1889, simple_loss=0.2634, pruned_loss=0.05722, over 19319.00 frames. ], tot_loss[loss=0.2009, simple_loss=0.2823, pruned_loss=0.05974, over 3808479.11 frames. ], batch size: 44, lr: 2.93e-03, grad_scale: 8.0 2023-04-03 11:43:43,949 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=187467.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:43:54,451 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.395e+02 4.792e+02 5.797e+02 7.675e+02 1.223e+03, threshold=1.159e+03, percent-clipped=0.0 2023-04-03 11:44:27,059 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=187502.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:44:32,169 INFO [train.py:903] (1/4) Epoch 28, batch 3150, loss[loss=0.1874, simple_loss=0.2681, pruned_loss=0.05331, over 19779.00 frames. ], tot_loss[loss=0.2002, simple_loss=0.2821, pruned_loss=0.05922, over 3814194.68 frames. ], batch size: 48, lr: 2.93e-03, grad_scale: 8.0 2023-04-03 11:44:40,711 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4266, 1.5162, 1.7582, 1.6566, 2.4088, 2.1112, 2.6140, 1.0388], device='cuda:1'), covar=tensor([0.2572, 0.4505, 0.2862, 0.2043, 0.1587, 0.2357, 0.1438, 0.5076], device='cuda:1'), in_proj_covar=tensor([0.0552, 0.0669, 0.0753, 0.0507, 0.0635, 0.0545, 0.0668, 0.0572], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 11:44:44,918 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=187515.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:44:54,642 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 2023-04-03 11:45:01,682 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=187530.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:45:10,624 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=187537.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:45:15,235 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=187541.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:45:35,285 INFO [train.py:903] (1/4) Epoch 28, batch 3200, loss[loss=0.212, simple_loss=0.2876, pruned_loss=0.06816, over 19676.00 frames. ], tot_loss[loss=0.2001, simple_loss=0.2818, pruned_loss=0.05921, over 3813327.00 frames. ], batch size: 53, lr: 2.93e-03, grad_scale: 8.0 2023-04-03 11:45:55,536 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=187573.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:45:57,397 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.995e+02 4.914e+02 6.228e+02 8.001e+02 2.182e+03, threshold=1.246e+03, percent-clipped=10.0 2023-04-03 11:46:08,978 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.32 vs. limit=5.0 2023-04-03 11:46:27,882 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=187598.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:46:36,786 INFO [train.py:903] (1/4) Epoch 28, batch 3250, loss[loss=0.1925, simple_loss=0.2818, pruned_loss=0.05155, over 19593.00 frames. ], tot_loss[loss=0.2011, simple_loss=0.2823, pruned_loss=0.06, over 3817377.34 frames. ], batch size: 57, lr: 2.93e-03, grad_scale: 8.0 2023-04-03 11:47:18,376 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.1384, 1.9323, 1.7343, 2.1389, 1.8340, 1.7506, 1.6938, 1.9508], device='cuda:1'), covar=tensor([0.1005, 0.1536, 0.1522, 0.1005, 0.1395, 0.0593, 0.1505, 0.0780], device='cuda:1'), in_proj_covar=tensor([0.0279, 0.0360, 0.0320, 0.0257, 0.0306, 0.0256, 0.0322, 0.0263], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 11:47:23,889 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=187644.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:47:37,437 INFO [train.py:903] (1/4) Epoch 28, batch 3300, loss[loss=0.2347, simple_loss=0.3064, pruned_loss=0.08149, over 19465.00 frames. ], tot_loss[loss=0.2024, simple_loss=0.2835, pruned_loss=0.06071, over 3823598.41 frames. ], batch size: 70, lr: 2.93e-03, grad_scale: 8.0 2023-04-03 11:47:37,459 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 2023-04-03 11:47:51,893 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.2924, 3.8172, 3.9250, 3.9382, 1.6720, 3.7326, 3.2334, 3.7000], device='cuda:1'), covar=tensor([0.1731, 0.0976, 0.0703, 0.0826, 0.5751, 0.1058, 0.0819, 0.1174], device='cuda:1'), in_proj_covar=tensor([0.0825, 0.0790, 0.1002, 0.0880, 0.0864, 0.0768, 0.0594, 0.0929], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-03 11:47:54,289 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=187669.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:48:01,989 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.234e+02 4.770e+02 6.093e+02 7.830e+02 1.620e+03, threshold=1.219e+03, percent-clipped=4.0 2023-04-03 11:48:40,831 INFO [train.py:903] (1/4) Epoch 28, batch 3350, loss[loss=0.1917, simple_loss=0.2769, pruned_loss=0.05322, over 19761.00 frames. ], tot_loss[loss=0.2026, simple_loss=0.2834, pruned_loss=0.0609, over 3811686.31 frames. ], batch size: 54, lr: 2.93e-03, grad_scale: 8.0 2023-04-03 11:49:14,754 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=187734.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:49:42,984 INFO [train.py:903] (1/4) Epoch 28, batch 3400, loss[loss=0.1804, simple_loss=0.2631, pruned_loss=0.04883, over 19413.00 frames. ], tot_loss[loss=0.2019, simple_loss=0.2828, pruned_loss=0.06044, over 3814904.84 frames. ], batch size: 48, lr: 2.92e-03, grad_scale: 4.0 2023-04-03 11:49:45,551 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=187758.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:50:02,005 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=187771.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:50:07,731 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.151e+02 4.801e+02 5.865e+02 7.575e+02 1.695e+03, threshold=1.173e+03, percent-clipped=1.0 2023-04-03 11:50:16,885 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=187783.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:50:29,633 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 2023-04-03 11:50:32,904 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=187796.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:50:46,602 INFO [train.py:903] (1/4) Epoch 28, batch 3450, loss[loss=0.1884, simple_loss=0.2761, pruned_loss=0.05033, over 19594.00 frames. ], tot_loss[loss=0.2018, simple_loss=0.2829, pruned_loss=0.0603, over 3805959.84 frames. ], batch size: 61, lr: 2.92e-03, grad_scale: 4.0 2023-04-03 11:50:47,816 WARNING [train.py:1073] (1/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9_rvb from training. Duration: 25.2444375 2023-04-03 11:51:41,378 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=187849.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:51:44,903 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.1177, 2.8891, 1.8497, 1.8447, 2.6209, 1.6749, 1.7424, 2.3557], device='cuda:1'), covar=tensor([0.1189, 0.0749, 0.1131, 0.0948, 0.0603, 0.1255, 0.0874, 0.0609], device='cuda:1'), in_proj_covar=tensor([0.0303, 0.0320, 0.0344, 0.0273, 0.0253, 0.0347, 0.0294, 0.0277], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 11:51:49,293 INFO [train.py:903] (1/4) Epoch 28, batch 3500, loss[loss=0.1814, simple_loss=0.2581, pruned_loss=0.05233, over 15107.00 frames. ], tot_loss[loss=0.2002, simple_loss=0.2815, pruned_loss=0.05949, over 3802671.57 frames. ], batch size: 33, lr: 2.92e-03, grad_scale: 4.0 2023-04-03 11:52:11,985 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=187874.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:52:14,011 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.602e+02 4.936e+02 6.004e+02 7.154e+02 1.224e+03, threshold=1.201e+03, percent-clipped=1.0 2023-04-03 11:52:20,884 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=187881.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:52:26,774 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=187885.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:52:50,774 INFO [train.py:903] (1/4) Epoch 28, batch 3550, loss[loss=0.208, simple_loss=0.2775, pruned_loss=0.06921, over 19747.00 frames. ], tot_loss[loss=0.2003, simple_loss=0.2817, pruned_loss=0.05943, over 3811462.98 frames. ], batch size: 46, lr: 2.92e-03, grad_scale: 4.0 2023-04-03 11:53:23,562 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5118, 1.7237, 2.0808, 1.8484, 2.9899, 2.5264, 3.3889, 1.6845], device='cuda:1'), covar=tensor([0.2717, 0.4539, 0.2816, 0.2040, 0.1698, 0.2296, 0.1673, 0.4592], device='cuda:1'), in_proj_covar=tensor([0.0553, 0.0667, 0.0753, 0.0507, 0.0634, 0.0545, 0.0669, 0.0572], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 11:53:52,895 INFO [train.py:903] (1/4) Epoch 28, batch 3600, loss[loss=0.1823, simple_loss=0.273, pruned_loss=0.04581, over 19535.00 frames. ], tot_loss[loss=0.1993, simple_loss=0.2809, pruned_loss=0.05887, over 3821916.17 frames. ], batch size: 56, lr: 2.92e-03, grad_scale: 8.0 2023-04-03 11:54:17,594 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.816e+02 4.904e+02 5.870e+02 7.567e+02 1.667e+03, threshold=1.174e+03, percent-clipped=3.0 2023-04-03 11:54:34,361 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=187989.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:54:42,627 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=187996.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:54:48,378 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=188000.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:54:56,060 INFO [train.py:903] (1/4) Epoch 28, batch 3650, loss[loss=0.2019, simple_loss=0.2906, pruned_loss=0.0566, over 19617.00 frames. ], tot_loss[loss=0.2008, simple_loss=0.2823, pruned_loss=0.05965, over 3835100.39 frames. ], batch size: 61, lr: 2.92e-03, grad_scale: 8.0 2023-04-03 11:55:36,435 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1913, 1.2354, 1.6681, 1.1186, 2.4554, 3.3514, 2.9666, 3.5350], device='cuda:1'), covar=tensor([0.1655, 0.4062, 0.3669, 0.2842, 0.0695, 0.0225, 0.0255, 0.0294], device='cuda:1'), in_proj_covar=tensor([0.0279, 0.0331, 0.0364, 0.0270, 0.0255, 0.0196, 0.0220, 0.0278], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-03 11:55:48,390 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.42 vs. limit=2.0 2023-04-03 11:55:57,939 INFO [train.py:903] (1/4) Epoch 28, batch 3700, loss[loss=0.1669, simple_loss=0.247, pruned_loss=0.04336, over 19784.00 frames. ], tot_loss[loss=0.2005, simple_loss=0.2822, pruned_loss=0.05935, over 3840013.11 frames. ], batch size: 47, lr: 2.92e-03, grad_scale: 8.0 2023-04-03 11:56:12,058 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.8914, 1.5680, 1.5194, 1.8424, 1.4577, 1.6190, 1.5101, 1.7410], device='cuda:1'), covar=tensor([0.1106, 0.1369, 0.1555, 0.1088, 0.1341, 0.0589, 0.1554, 0.0788], device='cuda:1'), in_proj_covar=tensor([0.0278, 0.0361, 0.0320, 0.0257, 0.0306, 0.0257, 0.0322, 0.0263], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 11:56:23,991 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.791e+02 5.013e+02 6.013e+02 7.793e+02 2.143e+03, threshold=1.203e+03, percent-clipped=3.0 2023-04-03 11:56:41,380 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9631, 1.7460, 1.6250, 1.9455, 1.6996, 1.6967, 1.5742, 1.8450], device='cuda:1'), covar=tensor([0.1097, 0.1531, 0.1576, 0.1116, 0.1377, 0.0599, 0.1567, 0.0793], device='cuda:1'), in_proj_covar=tensor([0.0279, 0.0361, 0.0320, 0.0258, 0.0307, 0.0257, 0.0322, 0.0263], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 11:56:59,855 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=188105.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:57:00,129 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.68 vs. limit=2.0 2023-04-03 11:57:00,595 INFO [train.py:903] (1/4) Epoch 28, batch 3750, loss[loss=0.1782, simple_loss=0.2535, pruned_loss=0.05147, over 19737.00 frames. ], tot_loss[loss=0.2014, simple_loss=0.2831, pruned_loss=0.05984, over 3829552.64 frames. ], batch size: 46, lr: 2.92e-03, grad_scale: 4.0 2023-04-03 11:57:32,238 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=188130.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 11:58:03,792 INFO [train.py:903] (1/4) Epoch 28, batch 3800, loss[loss=0.2132, simple_loss=0.2978, pruned_loss=0.06426, over 19274.00 frames. ], tot_loss[loss=0.2011, simple_loss=0.283, pruned_loss=0.05959, over 3841099.81 frames. ], batch size: 66, lr: 2.92e-03, grad_scale: 4.0 2023-04-03 11:58:29,767 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1061, 1.0823, 1.5755, 1.0934, 2.1665, 2.9659, 2.6958, 3.3062], device='cuda:1'), covar=tensor([0.1821, 0.5373, 0.4502, 0.2900, 0.0819, 0.0292, 0.0360, 0.0345], device='cuda:1'), in_proj_covar=tensor([0.0281, 0.0332, 0.0365, 0.0272, 0.0256, 0.0197, 0.0221, 0.0279], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-03 11:58:30,496 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.894e+02 5.177e+02 6.017e+02 8.386e+02 1.721e+03, threshold=1.203e+03, percent-clipped=7.0 2023-04-03 11:58:33,980 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9_rvb from training. Duration: 29.1166875 2023-04-03 11:59:06,271 INFO [train.py:903] (1/4) Epoch 28, batch 3850, loss[loss=0.2256, simple_loss=0.3072, pruned_loss=0.07197, over 18246.00 frames. ], tot_loss[loss=0.2017, simple_loss=0.2834, pruned_loss=0.05999, over 3840107.64 frames. ], batch size: 83, lr: 2.92e-03, grad_scale: 4.0 2023-04-03 11:59:56,095 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=188245.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:00:00,686 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.7862, 1.6314, 1.6643, 2.3093, 1.7148, 1.9778, 2.0567, 1.7910], device='cuda:1'), covar=tensor([0.0855, 0.0893, 0.0993, 0.0722, 0.0895, 0.0791, 0.0877, 0.0697], device='cuda:1'), in_proj_covar=tensor([0.0212, 0.0223, 0.0227, 0.0239, 0.0225, 0.0214, 0.0187, 0.0205], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-04-03 12:00:05,190 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=188252.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:00:09,178 INFO [train.py:903] (1/4) Epoch 28, batch 3900, loss[loss=0.1911, simple_loss=0.2815, pruned_loss=0.05037, over 19760.00 frames. ], tot_loss[loss=0.2002, simple_loss=0.2822, pruned_loss=0.05906, over 3844356.97 frames. ], batch size: 63, lr: 2.92e-03, grad_scale: 4.0 2023-04-03 12:00:09,599 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=188256.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:00:26,418 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=188270.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:00:34,023 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.127e+02 5.053e+02 6.264e+02 8.116e+02 1.975e+03, threshold=1.253e+03, percent-clipped=4.0 2023-04-03 12:00:34,452 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=188277.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:00:39,120 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=188281.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:00:50,587 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.44 vs. limit=5.0 2023-04-03 12:01:04,070 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=188301.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:01:09,682 INFO [train.py:903] (1/4) Epoch 28, batch 3950, loss[loss=0.1713, simple_loss=0.2499, pruned_loss=0.04641, over 19398.00 frames. ], tot_loss[loss=0.2004, simple_loss=0.2822, pruned_loss=0.05932, over 3823699.89 frames. ], batch size: 48, lr: 2.92e-03, grad_scale: 4.0 2023-04-03 12:01:15,479 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 2023-04-03 12:01:17,799 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([6.2185, 5.6618, 2.9719, 4.9186, 1.2481, 5.7505, 5.5792, 5.7908], device='cuda:1'), covar=tensor([0.0374, 0.0770, 0.1918, 0.0677, 0.3813, 0.0506, 0.0809, 0.0923], device='cuda:1'), in_proj_covar=tensor([0.0530, 0.0431, 0.0518, 0.0362, 0.0412, 0.0458, 0.0454, 0.0485], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 12:02:12,182 INFO [train.py:903] (1/4) Epoch 28, batch 4000, loss[loss=0.2239, simple_loss=0.303, pruned_loss=0.07239, over 19698.00 frames. ], tot_loss[loss=0.2002, simple_loss=0.2821, pruned_loss=0.05911, over 3826214.97 frames. ], batch size: 59, lr: 2.92e-03, grad_scale: 8.0 2023-04-03 12:02:38,285 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.378e+02 4.980e+02 6.272e+02 8.090e+02 1.579e+03, threshold=1.254e+03, percent-clipped=5.0 2023-04-03 12:02:59,869 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1_rvb from training. Duration: 0.7545625 2023-04-03 12:03:14,927 INFO [train.py:903] (1/4) Epoch 28, batch 4050, loss[loss=0.2297, simple_loss=0.3052, pruned_loss=0.07708, over 19863.00 frames. ], tot_loss[loss=0.2001, simple_loss=0.2817, pruned_loss=0.05924, over 3823062.26 frames. ], batch size: 52, lr: 2.92e-03, grad_scale: 8.0 2023-04-03 12:04:17,051 INFO [train.py:903] (1/4) Epoch 28, batch 4100, loss[loss=0.1749, simple_loss=0.2638, pruned_loss=0.04304, over 19657.00 frames. ], tot_loss[loss=0.2001, simple_loss=0.2816, pruned_loss=0.05931, over 3811851.69 frames. ], batch size: 55, lr: 2.92e-03, grad_scale: 8.0 2023-04-03 12:04:43,427 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.476e+02 5.116e+02 6.170e+02 8.694e+02 1.686e+03, threshold=1.234e+03, percent-clipped=5.0 2023-04-03 12:04:53,774 WARNING [train.py:1073] (1/4) Exclude cut with ID 497-129325-0061-9566_sp1.1_rvb from training. Duration: 0.97725 2023-04-03 12:05:11,282 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=188499.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:05:19,165 INFO [train.py:903] (1/4) Epoch 28, batch 4150, loss[loss=0.1807, simple_loss=0.2779, pruned_loss=0.0418, over 19675.00 frames. ], tot_loss[loss=0.198, simple_loss=0.2797, pruned_loss=0.05818, over 3826289.32 frames. ], batch size: 59, lr: 2.92e-03, grad_scale: 8.0 2023-04-03 12:05:33,298 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.70 vs. limit=5.0 2023-04-03 12:06:21,498 INFO [train.py:903] (1/4) Epoch 28, batch 4200, loss[loss=0.199, simple_loss=0.2771, pruned_loss=0.06043, over 19377.00 frames. ], tot_loss[loss=0.1994, simple_loss=0.2808, pruned_loss=0.05904, over 3807262.07 frames. ], batch size: 47, lr: 2.92e-03, grad_scale: 8.0 2023-04-03 12:06:24,956 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 2023-04-03 12:06:46,561 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.847e+02 4.775e+02 5.885e+02 7.540e+02 1.202e+03, threshold=1.177e+03, percent-clipped=0.0 2023-04-03 12:07:22,474 INFO [train.py:903] (1/4) Epoch 28, batch 4250, loss[loss=0.1942, simple_loss=0.2832, pruned_loss=0.05263, over 19537.00 frames. ], tot_loss[loss=0.2018, simple_loss=0.2825, pruned_loss=0.06051, over 3799863.51 frames. ], batch size: 56, lr: 2.92e-03, grad_scale: 4.0 2023-04-03 12:07:33,753 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9_rvb from training. Duration: 29.816625 2023-04-03 12:07:44,094 WARNING [train.py:1073] (1/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 2023-04-03 12:07:53,730 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0 2023-04-03 12:08:10,326 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=188645.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:08:24,251 INFO [train.py:903] (1/4) Epoch 28, batch 4300, loss[loss=0.1871, simple_loss=0.2787, pruned_loss=0.0478, over 18058.00 frames. ], tot_loss[loss=0.2006, simple_loss=0.2816, pruned_loss=0.05978, over 3798919.20 frames. ], batch size: 83, lr: 2.92e-03, grad_scale: 4.0 2023-04-03 12:08:35,241 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.5411, 1.2244, 1.5478, 1.6736, 2.9491, 1.3503, 2.5506, 3.4417], device='cuda:1'), covar=tensor([0.0687, 0.3549, 0.3098, 0.1982, 0.1074, 0.2627, 0.1290, 0.0386], device='cuda:1'), in_proj_covar=tensor([0.0423, 0.0379, 0.0396, 0.0353, 0.0381, 0.0357, 0.0397, 0.0418], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 12:08:35,283 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3784, 1.3083, 1.7735, 1.3631, 2.6532, 3.5607, 3.2404, 3.7380], device='cuda:1'), covar=tensor([0.1550, 0.3984, 0.3438, 0.2493, 0.0669, 0.0197, 0.0251, 0.0359], device='cuda:1'), in_proj_covar=tensor([0.0282, 0.0333, 0.0367, 0.0273, 0.0257, 0.0198, 0.0222, 0.0280], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-03 12:08:51,757 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.988e+02 4.883e+02 5.889e+02 8.258e+02 1.553e+03, threshold=1.178e+03, percent-clipped=6.0 2023-04-03 12:09:15,636 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 2023-04-03 12:09:26,307 INFO [train.py:903] (1/4) Epoch 28, batch 4350, loss[loss=0.162, simple_loss=0.2431, pruned_loss=0.04051, over 19026.00 frames. ], tot_loss[loss=0.2009, simple_loss=0.2823, pruned_loss=0.05977, over 3805028.42 frames. ], batch size: 42, lr: 2.92e-03, grad_scale: 4.0 2023-04-03 12:10:30,423 INFO [train.py:903] (1/4) Epoch 28, batch 4400, loss[loss=0.1726, simple_loss=0.2539, pruned_loss=0.04567, over 19824.00 frames. ], tot_loss[loss=0.2006, simple_loss=0.2819, pruned_loss=0.05961, over 3806949.38 frames. ], batch size: 49, lr: 2.92e-03, grad_scale: 8.0 2023-04-03 12:10:35,397 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=188760.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:10:52,225 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_rvb from training. Duration: 25.285 2023-04-03 12:10:57,774 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.415e+02 4.797e+02 5.749e+02 7.341e+02 1.454e+03, threshold=1.150e+03, percent-clipped=2.0 2023-04-03 12:11:02,388 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 2023-04-03 12:11:31,826 INFO [train.py:903] (1/4) Epoch 28, batch 4450, loss[loss=0.1945, simple_loss=0.2721, pruned_loss=0.05851, over 19763.00 frames. ], tot_loss[loss=0.2019, simple_loss=0.2831, pruned_loss=0.06033, over 3813172.83 frames. ], batch size: 48, lr: 2.92e-03, grad_scale: 8.0 2023-04-03 12:12:11,511 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.7553, 1.9994, 2.2424, 2.0568, 3.3247, 2.9306, 3.5269, 1.6038], device='cuda:1'), covar=tensor([0.2442, 0.4150, 0.2855, 0.1837, 0.1487, 0.1966, 0.1517, 0.4614], device='cuda:1'), in_proj_covar=tensor([0.0557, 0.0673, 0.0759, 0.0510, 0.0638, 0.0549, 0.0672, 0.0576], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-03 12:12:19,630 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=188843.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:12:32,891 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6837, 1.2540, 1.3249, 1.5658, 1.1441, 1.4238, 1.2792, 1.4982], device='cuda:1'), covar=tensor([0.1117, 0.1252, 0.1651, 0.1007, 0.1353, 0.0646, 0.1688, 0.0873], device='cuda:1'), in_proj_covar=tensor([0.0276, 0.0358, 0.0317, 0.0255, 0.0303, 0.0254, 0.0319, 0.0261], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 12:12:36,100 INFO [train.py:903] (1/4) Epoch 28, batch 4500, loss[loss=0.2516, simple_loss=0.32, pruned_loss=0.09161, over 13716.00 frames. ], tot_loss[loss=0.2014, simple_loss=0.2829, pruned_loss=0.05996, over 3808855.79 frames. ], batch size: 135, lr: 2.92e-03, grad_scale: 8.0 2023-04-03 12:12:44,874 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0202, 2.0888, 2.2978, 2.6284, 2.0378, 2.5039, 2.2429, 2.0996], device='cuda:1'), covar=tensor([0.4226, 0.4096, 0.2002, 0.2578, 0.4335, 0.2353, 0.5197, 0.3604], device='cuda:1'), in_proj_covar=tensor([0.0941, 0.1018, 0.0745, 0.0954, 0.0917, 0.0860, 0.0863, 0.0811], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-03 12:13:04,412 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.382e+02 4.536e+02 5.549e+02 7.562e+02 1.666e+03, threshold=1.110e+03, percent-clipped=5.0 2023-04-03 12:13:10,821 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.69 vs. limit=2.0 2023-04-03 12:13:38,229 INFO [train.py:903] (1/4) Epoch 28, batch 4550, loss[loss=0.198, simple_loss=0.2841, pruned_loss=0.05596, over 19135.00 frames. ], tot_loss[loss=0.2016, simple_loss=0.2832, pruned_loss=0.05997, over 3827578.86 frames. ], batch size: 69, lr: 2.92e-03, grad_scale: 8.0 2023-04-03 12:13:46,950 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9_rvb from training. Duration: 28.72225 2023-04-03 12:14:12,237 WARNING [train.py:1073] (1/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 2023-04-03 12:14:29,503 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=188947.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:14:31,816 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=188949.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:14:39,522 INFO [train.py:903] (1/4) Epoch 28, batch 4600, loss[loss=0.1919, simple_loss=0.2731, pruned_loss=0.0554, over 19632.00 frames. ], tot_loss[loss=0.2025, simple_loss=0.2842, pruned_loss=0.06036, over 3831094.93 frames. ], batch size: 50, lr: 2.92e-03, grad_scale: 8.0 2023-04-03 12:14:43,215 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=188958.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:15:07,935 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.099e+02 4.624e+02 5.779e+02 7.629e+02 1.899e+03, threshold=1.156e+03, percent-clipped=6.0 2023-04-03 12:15:08,213 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([6.2118, 5.6495, 3.2047, 4.9312, 1.3420, 5.8140, 5.6505, 5.8089], device='cuda:1'), covar=tensor([0.0332, 0.0731, 0.1681, 0.0731, 0.3627, 0.0445, 0.0735, 0.0752], device='cuda:1'), in_proj_covar=tensor([0.0525, 0.0428, 0.0512, 0.0359, 0.0407, 0.0454, 0.0450, 0.0480], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 12:15:13,473 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.46 vs. limit=5.0 2023-04-03 12:15:43,022 INFO [train.py:903] (1/4) Epoch 28, batch 4650, loss[loss=0.1633, simple_loss=0.2523, pruned_loss=0.03713, over 19657.00 frames. ], tot_loss[loss=0.2021, simple_loss=0.2838, pruned_loss=0.06018, over 3833764.91 frames. ], batch size: 53, lr: 2.92e-03, grad_scale: 8.0 2023-04-03 12:15:55,977 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=189016.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:16:01,294 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_rvb from training. Duration: 0.92 2023-04-03 12:16:12,621 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_rvb from training. Duration: 0.83 2023-04-03 12:16:26,483 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=189041.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:16:34,710 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=189048.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:16:34,913 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9595, 2.0368, 2.3049, 2.6205, 1.9825, 2.5050, 2.2745, 2.1194], device='cuda:1'), covar=tensor([0.4478, 0.4222, 0.2021, 0.2638, 0.4575, 0.2435, 0.5122, 0.3531], device='cuda:1'), in_proj_covar=tensor([0.0937, 0.1014, 0.0741, 0.0951, 0.0913, 0.0854, 0.0859, 0.0806], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-03 12:16:44,440 INFO [train.py:903] (1/4) Epoch 28, batch 4700, loss[loss=0.2804, simple_loss=0.3382, pruned_loss=0.1113, over 13942.00 frames. ], tot_loss[loss=0.2024, simple_loss=0.284, pruned_loss=0.06042, over 3834475.73 frames. ], batch size: 139, lr: 2.91e-03, grad_scale: 8.0 2023-04-03 12:17:07,716 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9_rvb from training. Duration: 0.92225 2023-04-03 12:17:10,864 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.010e+02 4.848e+02 5.842e+02 7.352e+02 2.015e+03, threshold=1.168e+03, percent-clipped=3.0 2023-04-03 12:17:17,565 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=189082.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:17:36,140 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=189097.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:17:46,801 INFO [train.py:903] (1/4) Epoch 28, batch 4750, loss[loss=0.1702, simple_loss=0.2578, pruned_loss=0.04129, over 19656.00 frames. ], tot_loss[loss=0.2026, simple_loss=0.2838, pruned_loss=0.06065, over 3823678.16 frames. ], batch size: 55, lr: 2.91e-03, grad_scale: 8.0 2023-04-03 12:18:47,827 INFO [train.py:903] (1/4) Epoch 28, batch 4800, loss[loss=0.1895, simple_loss=0.2733, pruned_loss=0.05285, over 19597.00 frames. ], tot_loss[loss=0.2022, simple_loss=0.2835, pruned_loss=0.06046, over 3822538.66 frames. ], batch size: 52, lr: 2.91e-03, grad_scale: 8.0 2023-04-03 12:19:16,002 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.214e+02 5.029e+02 6.156e+02 7.557e+02 1.439e+03, threshold=1.231e+03, percent-clipped=4.0 2023-04-03 12:19:43,396 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 2023-04-03 12:19:50,824 INFO [train.py:903] (1/4) Epoch 28, batch 4850, loss[loss=0.2027, simple_loss=0.2885, pruned_loss=0.05849, over 19493.00 frames. ], tot_loss[loss=0.2026, simple_loss=0.2838, pruned_loss=0.06069, over 3810321.22 frames. ], batch size: 64, lr: 2.91e-03, grad_scale: 8.0 2023-04-03 12:20:01,556 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=189214.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:20:10,566 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.3224, 1.9299, 2.1327, 3.0529, 2.1760, 2.4790, 2.6037, 2.2254], device='cuda:1'), covar=tensor([0.0784, 0.0936, 0.0954, 0.0738, 0.0830, 0.0763, 0.0848, 0.0682], device='cuda:1'), in_proj_covar=tensor([0.0213, 0.0224, 0.0228, 0.0242, 0.0226, 0.0214, 0.0189, 0.0207], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-04-03 12:20:12,593 WARNING [train.py:1073] (1/4) Exclude cut with ID 774-127930-0014-48411_sp1.1_rvb from training. Duration: 0.95 2023-04-03 12:20:32,917 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=189239.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:20:33,808 WARNING [train.py:1073] (1/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1_rvb from training. Duration: 0.9409375 2023-04-03 12:20:39,535 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 2023-04-03 12:20:40,697 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 2023-04-03 12:20:51,044 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1_rvb from training. Duration: 25.3818125 2023-04-03 12:20:53,318 INFO [train.py:903] (1/4) Epoch 28, batch 4900, loss[loss=0.1818, simple_loss=0.2604, pruned_loss=0.05163, over 19597.00 frames. ], tot_loss[loss=0.2021, simple_loss=0.2833, pruned_loss=0.06044, over 3796615.23 frames. ], batch size: 50, lr: 2.91e-03, grad_scale: 8.0 2023-04-03 12:21:05,669 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=189266.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:21:05,770 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9777, 1.8811, 1.6545, 2.0634, 1.7221, 1.6589, 1.6312, 1.8990], device='cuda:1'), covar=tensor([0.1142, 0.1399, 0.1653, 0.1109, 0.1453, 0.0630, 0.1607, 0.0810], device='cuda:1'), in_proj_covar=tensor([0.0278, 0.0360, 0.0319, 0.0257, 0.0305, 0.0255, 0.0321, 0.0263], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 12:21:10,805 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 2023-04-03 12:21:14,415 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.8201, 1.4272, 1.6964, 1.7438, 4.3088, 1.1677, 2.4438, 4.7247], device='cuda:1'), covar=tensor([0.0481, 0.3025, 0.2997, 0.1938, 0.0774, 0.2764, 0.1672, 0.0181], device='cuda:1'), in_proj_covar=tensor([0.0426, 0.0381, 0.0400, 0.0354, 0.0383, 0.0359, 0.0399, 0.0421], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 12:21:19,662 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.182e+02 4.742e+02 6.083e+02 7.635e+02 1.565e+03, threshold=1.217e+03, percent-clipped=2.0 2023-04-03 12:21:35,181 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=189291.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:21:38,179 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=189293.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:21:53,518 INFO [train.py:903] (1/4) Epoch 28, batch 4950, loss[loss=0.1988, simple_loss=0.2805, pruned_loss=0.05852, over 19390.00 frames. ], tot_loss[loss=0.203, simple_loss=0.2842, pruned_loss=0.06092, over 3801382.72 frames. ], batch size: 48, lr: 2.91e-03, grad_scale: 8.0 2023-04-03 12:22:10,592 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 2023-04-03 12:22:34,599 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 2023-04-03 12:22:41,890 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6406, 1.6993, 1.9200, 1.8188, 2.6045, 2.3394, 2.7062, 1.5266], device='cuda:1'), covar=tensor([0.2465, 0.4074, 0.2626, 0.1922, 0.1567, 0.2104, 0.1491, 0.4560], device='cuda:1'), in_proj_covar=tensor([0.0556, 0.0673, 0.0758, 0.0508, 0.0637, 0.0548, 0.0673, 0.0575], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-03 12:22:56,069 INFO [train.py:903] (1/4) Epoch 28, batch 5000, loss[loss=0.2079, simple_loss=0.2894, pruned_loss=0.0632, over 19589.00 frames. ], tot_loss[loss=0.2033, simple_loss=0.2845, pruned_loss=0.06103, over 3779720.49 frames. ], batch size: 52, lr: 2.91e-03, grad_scale: 8.0 2023-04-03 12:23:04,841 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_rvb from training. Duration: 27.14 2023-04-03 12:23:08,817 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=189365.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:23:15,678 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 2023-04-03 12:23:24,839 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.397e+02 5.003e+02 6.019e+02 8.012e+02 1.544e+03, threshold=1.204e+03, percent-clipped=7.0 2023-04-03 12:23:42,151 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=189392.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:23:59,717 INFO [train.py:903] (1/4) Epoch 28, batch 5050, loss[loss=0.2007, simple_loss=0.2839, pruned_loss=0.05881, over 19592.00 frames. ], tot_loss[loss=0.2028, simple_loss=0.2844, pruned_loss=0.06065, over 3795202.29 frames. ], batch size: 61, lr: 2.91e-03, grad_scale: 8.0 2023-04-03 12:24:00,151 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=189406.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:24:02,595 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=189408.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:24:24,214 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=189426.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:24:33,178 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9_rvb from training. Duration: 26.62775 2023-04-03 12:24:44,053 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=189441.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:25:01,213 INFO [train.py:903] (1/4) Epoch 28, batch 5100, loss[loss=0.1885, simple_loss=0.2664, pruned_loss=0.05524, over 19741.00 frames. ], tot_loss[loss=0.2025, simple_loss=0.2843, pruned_loss=0.06041, over 3808934.65 frames. ], batch size: 51, lr: 2.91e-03, grad_scale: 8.0 2023-04-03 12:25:10,101 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 2023-04-03 12:25:12,435 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_rvb from training. Duration: 25.035 2023-04-03 12:25:19,116 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_rvb from training. Duration: 27.92 2023-04-03 12:25:27,912 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.508e+02 5.191e+02 6.540e+02 8.236e+02 1.634e+03, threshold=1.308e+03, percent-clipped=9.0 2023-04-03 12:26:01,392 INFO [train.py:903] (1/4) Epoch 28, batch 5150, loss[loss=0.2519, simple_loss=0.3193, pruned_loss=0.09223, over 17385.00 frames. ], tot_loss[loss=0.2035, simple_loss=0.285, pruned_loss=0.061, over 3783376.92 frames. ], batch size: 101, lr: 2.91e-03, grad_scale: 8.0 2023-04-03 12:26:02,894 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=189507.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:26:11,409 WARNING [train.py:1073] (1/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 2023-04-03 12:26:18,140 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=189519.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:26:34,978 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.7531, 1.7726, 1.7306, 1.5259, 1.4195, 1.5286, 0.3000, 0.7308], device='cuda:1'), covar=tensor([0.0712, 0.0652, 0.0446, 0.0682, 0.1347, 0.0811, 0.1434, 0.1220], device='cuda:1'), in_proj_covar=tensor([0.0366, 0.0364, 0.0369, 0.0392, 0.0471, 0.0397, 0.0346, 0.0348], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 12:26:44,823 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=189541.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:26:46,849 WARNING [train.py:1073] (1/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1_rvb from training. Duration: 0.9681875 2023-04-03 12:27:02,655 INFO [train.py:903] (1/4) Epoch 28, batch 5200, loss[loss=0.227, simple_loss=0.3012, pruned_loss=0.0764, over 19689.00 frames. ], tot_loss[loss=0.2038, simple_loss=0.2853, pruned_loss=0.06117, over 3791908.45 frames. ], batch size: 59, lr: 2.91e-03, grad_scale: 8.0 2023-04-03 12:27:03,046 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=189556.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:27:04,961 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.7742, 4.3724, 2.6991, 3.9330, 0.9323, 4.4025, 4.2223, 4.3274], device='cuda:1'), covar=tensor([0.0612, 0.0957, 0.2017, 0.0781, 0.4154, 0.0595, 0.0912, 0.1110], device='cuda:1'), in_proj_covar=tensor([0.0529, 0.0429, 0.0517, 0.0360, 0.0410, 0.0456, 0.0451, 0.0483], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 12:27:17,238 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 2023-04-03 12:27:30,437 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.217e+02 4.850e+02 5.941e+02 7.550e+02 1.552e+03, threshold=1.188e+03, percent-clipped=2.0 2023-04-03 12:28:01,620 WARNING [train.py:1073] (1/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 2023-04-03 12:28:04,867 INFO [train.py:903] (1/4) Epoch 28, batch 5250, loss[loss=0.2152, simple_loss=0.2877, pruned_loss=0.07129, over 19732.00 frames. ], tot_loss[loss=0.2033, simple_loss=0.2849, pruned_loss=0.06081, over 3810341.87 frames. ], batch size: 51, lr: 2.91e-03, grad_scale: 8.0 2023-04-03 12:28:09,606 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=189610.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:29:04,828 INFO [train.py:903] (1/4) Epoch 28, batch 5300, loss[loss=0.1926, simple_loss=0.2856, pruned_loss=0.04977, over 19672.00 frames. ], tot_loss[loss=0.2029, simple_loss=0.2844, pruned_loss=0.06072, over 3814184.51 frames. ], batch size: 58, lr: 2.91e-03, grad_scale: 8.0 2023-04-03 12:29:08,388 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=189658.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:29:12,986 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=189662.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:29:14,942 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=189664.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:29:21,474 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 2023-04-03 12:29:31,784 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.136e+02 4.811e+02 6.188e+02 7.440e+02 1.460e+03, threshold=1.238e+03, percent-clipped=2.0 2023-04-03 12:29:44,190 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=189687.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:29:46,196 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=189689.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:30:05,519 INFO [train.py:903] (1/4) Epoch 28, batch 5350, loss[loss=0.2231, simple_loss=0.3037, pruned_loss=0.07122, over 19615.00 frames. ], tot_loss[loss=0.2024, simple_loss=0.2837, pruned_loss=0.06055, over 3794834.69 frames. ], batch size: 57, lr: 2.91e-03, grad_scale: 8.0 2023-04-03 12:30:09,135 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=189709.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:30:17,205 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.7706, 1.8921, 2.1644, 2.2529, 1.7377, 2.1614, 2.0941, 1.9619], device='cuda:1'), covar=tensor([0.4252, 0.3852, 0.2051, 0.2512, 0.4057, 0.2425, 0.5204, 0.3570], device='cuda:1'), in_proj_covar=tensor([0.0940, 0.1018, 0.0745, 0.0954, 0.0916, 0.0858, 0.0863, 0.0812], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-03 12:30:18,029 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.8298, 4.9137, 5.7977, 5.7695, 2.5643, 5.5637, 4.5307, 5.1856], device='cuda:1'), covar=tensor([0.1884, 0.1260, 0.0632, 0.0750, 0.6524, 0.1704, 0.0949, 0.1367], device='cuda:1'), in_proj_covar=tensor([0.0814, 0.0782, 0.0991, 0.0870, 0.0861, 0.0756, 0.0583, 0.0922], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-03 12:30:25,491 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=189722.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:30:29,855 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=189725.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:30:38,357 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9_rvb from training. Duration: 26.438875 2023-04-03 12:30:54,236 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=189746.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:31:04,019 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.8891, 4.4598, 2.7086, 3.9066, 1.0067, 4.4115, 4.3260, 4.3751], device='cuda:1'), covar=tensor([0.0481, 0.0876, 0.2049, 0.0845, 0.3829, 0.0596, 0.0843, 0.0996], device='cuda:1'), in_proj_covar=tensor([0.0530, 0.0430, 0.0518, 0.0361, 0.0411, 0.0457, 0.0452, 0.0483], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 12:31:04,989 INFO [train.py:903] (1/4) Epoch 28, batch 5400, loss[loss=0.1636, simple_loss=0.2387, pruned_loss=0.04429, over 19765.00 frames. ], tot_loss[loss=0.2025, simple_loss=0.2838, pruned_loss=0.06055, over 3810711.92 frames. ], batch size: 45, lr: 2.91e-03, grad_scale: 8.0 2023-04-03 12:31:15,129 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=189763.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:31:33,363 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.324e+02 4.839e+02 5.999e+02 7.745e+02 2.007e+03, threshold=1.200e+03, percent-clipped=4.0 2023-04-03 12:31:44,953 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=189788.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:31:56,108 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=189797.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:32:07,656 INFO [train.py:903] (1/4) Epoch 28, batch 5450, loss[loss=0.182, simple_loss=0.2586, pruned_loss=0.05265, over 19395.00 frames. ], tot_loss[loss=0.2016, simple_loss=0.2831, pruned_loss=0.06005, over 3819360.65 frames. ], batch size: 47, lr: 2.91e-03, grad_scale: 8.0 2023-04-03 12:32:15,153 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=189812.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:32:27,686 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=189822.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:32:30,068 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=189824.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:32:45,795 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=189837.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:33:08,050 INFO [train.py:903] (1/4) Epoch 28, batch 5500, loss[loss=0.1941, simple_loss=0.2902, pruned_loss=0.04897, over 19672.00 frames. ], tot_loss[loss=0.2013, simple_loss=0.2828, pruned_loss=0.05988, over 3831131.13 frames. ], batch size: 58, lr: 2.91e-03, grad_scale: 8.0 2023-04-03 12:33:17,138 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=189863.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:33:30,502 WARNING [train.py:1073] (1/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9_rvb from training. Duration: 27.47775 2023-04-03 12:33:35,032 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.121e+02 4.755e+02 5.876e+02 8.325e+02 1.733e+03, threshold=1.175e+03, percent-clipped=4.0 2023-04-03 12:34:07,023 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9980, 2.1126, 2.2952, 2.6202, 1.9685, 2.4671, 2.2405, 2.1174], device='cuda:1'), covar=tensor([0.4451, 0.4362, 0.2144, 0.2635, 0.4573, 0.2431, 0.5381, 0.3722], device='cuda:1'), in_proj_covar=tensor([0.0940, 0.1017, 0.0745, 0.0954, 0.0917, 0.0858, 0.0862, 0.0811], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-03 12:34:10,042 INFO [train.py:903] (1/4) Epoch 28, batch 5550, loss[loss=0.1679, simple_loss=0.2392, pruned_loss=0.04829, over 19769.00 frames. ], tot_loss[loss=0.2008, simple_loss=0.2823, pruned_loss=0.0596, over 3826938.12 frames. ], batch size: 45, lr: 2.91e-03, grad_scale: 8.0 2023-04-03 12:34:17,008 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 2023-04-03 12:35:06,749 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 2023-04-03 12:35:11,467 INFO [train.py:903] (1/4) Epoch 28, batch 5600, loss[loss=0.2043, simple_loss=0.2926, pruned_loss=0.05794, over 19436.00 frames. ], tot_loss[loss=0.1999, simple_loss=0.2818, pruned_loss=0.05904, over 3812597.17 frames. ], batch size: 70, lr: 2.91e-03, grad_scale: 8.0 2023-04-03 12:35:40,028 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.214e+02 4.597e+02 5.831e+02 7.652e+02 2.230e+03, threshold=1.166e+03, percent-clipped=4.0 2023-04-03 12:35:40,280 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=189978.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:35:44,668 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=189981.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:36:11,925 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=190002.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:36:17,389 INFO [train.py:903] (1/4) Epoch 28, batch 5650, loss[loss=0.1763, simple_loss=0.2599, pruned_loss=0.04632, over 19578.00 frames. ], tot_loss[loss=0.2005, simple_loss=0.2825, pruned_loss=0.05926, over 3814214.13 frames. ], batch size: 52, lr: 2.91e-03, grad_scale: 8.0 2023-04-03 12:36:17,743 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=190006.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:37:04,793 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9_rvb from training. Duration: 28.0944375 2023-04-03 12:37:18,834 INFO [train.py:903] (1/4) Epoch 28, batch 5700, loss[loss=0.1851, simple_loss=0.2674, pruned_loss=0.0514, over 19455.00 frames. ], tot_loss[loss=0.2022, simple_loss=0.2839, pruned_loss=0.06026, over 3820825.56 frames. ], batch size: 49, lr: 2.91e-03, grad_scale: 8.0 2023-04-03 12:37:26,617 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.42 vs. limit=2.0 2023-04-03 12:37:27,080 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.7715, 4.9014, 5.5013, 5.5348, 2.2826, 5.2249, 4.4865, 5.2145], device='cuda:1'), covar=tensor([0.1621, 0.1238, 0.0559, 0.0615, 0.6006, 0.0939, 0.0641, 0.1061], device='cuda:1'), in_proj_covar=tensor([0.0822, 0.0788, 0.0999, 0.0879, 0.0869, 0.0765, 0.0589, 0.0928], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-03 12:37:31,551 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=190066.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:37:45,568 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.971e+02 5.283e+02 6.430e+02 7.892e+02 1.365e+03, threshold=1.286e+03, percent-clipped=3.0 2023-04-03 12:37:49,398 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=190080.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:37:54,215 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.8133, 1.9319, 2.1972, 2.3206, 1.7840, 2.2824, 2.1514, 1.9716], device='cuda:1'), covar=tensor([0.4347, 0.3991, 0.2096, 0.2500, 0.4171, 0.2334, 0.5205, 0.3671], device='cuda:1'), in_proj_covar=tensor([0.0940, 0.1018, 0.0745, 0.0955, 0.0917, 0.0858, 0.0862, 0.0811], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-03 12:38:02,440 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=190090.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:38:21,031 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=190105.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:38:21,819 INFO [train.py:903] (1/4) Epoch 28, batch 5750, loss[loss=0.1759, simple_loss=0.2596, pruned_loss=0.04603, over 19787.00 frames. ], tot_loss[loss=0.201, simple_loss=0.2831, pruned_loss=0.0595, over 3820160.01 frames. ], batch size: 49, lr: 2.91e-03, grad_scale: 8.0 2023-04-03 12:38:24,215 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9_rvb from training. Duration: 33.038875 2023-04-03 12:38:32,477 WARNING [train.py:1073] (1/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 2023-04-03 12:38:35,173 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=190117.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:38:37,089 WARNING [train.py:1073] (1/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 2023-04-03 12:38:45,190 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6649, 1.7630, 2.0683, 1.8908, 2.9101, 2.4023, 3.0774, 1.4170], device='cuda:1'), covar=tensor([0.2442, 0.4293, 0.2685, 0.2015, 0.1365, 0.2272, 0.1320, 0.4497], device='cuda:1'), in_proj_covar=tensor([0.0558, 0.0675, 0.0760, 0.0512, 0.0640, 0.0551, 0.0674, 0.0579], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-03 12:39:23,777 INFO [train.py:903] (1/4) Epoch 28, batch 5800, loss[loss=0.177, simple_loss=0.2535, pruned_loss=0.05025, over 19743.00 frames. ], tot_loss[loss=0.2005, simple_loss=0.2825, pruned_loss=0.05922, over 3820969.28 frames. ], batch size: 46, lr: 2.91e-03, grad_scale: 8.0 2023-04-03 12:39:25,235 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=190157.0, num_to_drop=1, layers_to_drop={1} 2023-04-03 12:39:41,876 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=190169.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:39:52,005 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.261e+02 4.681e+02 5.738e+02 7.022e+02 1.341e+03, threshold=1.148e+03, percent-clipped=1.0 2023-04-03 12:39:56,735 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=190181.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:40:01,273 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.5723, 4.1746, 2.7500, 3.6525, 1.0293, 4.1430, 4.0086, 4.0996], device='cuda:1'), covar=tensor([0.0645, 0.0986, 0.1984, 0.0946, 0.4185, 0.0743, 0.0980, 0.1498], device='cuda:1'), in_proj_covar=tensor([0.0531, 0.0431, 0.0519, 0.0361, 0.0411, 0.0458, 0.0451, 0.0484], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 12:40:26,142 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=190205.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:40:27,625 INFO [train.py:903] (1/4) Epoch 28, batch 5850, loss[loss=0.1882, simple_loss=0.2677, pruned_loss=0.0544, over 19838.00 frames. ], tot_loss[loss=0.2003, simple_loss=0.2822, pruned_loss=0.05917, over 3819656.09 frames. ], batch size: 52, lr: 2.91e-03, grad_scale: 8.0 2023-04-03 12:41:02,546 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=190234.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:41:29,981 INFO [train.py:903] (1/4) Epoch 28, batch 5900, loss[loss=0.1885, simple_loss=0.2771, pruned_loss=0.04991, over 19679.00 frames. ], tot_loss[loss=0.1996, simple_loss=0.2817, pruned_loss=0.0588, over 3822038.62 frames. ], batch size: 59, lr: 2.91e-03, grad_scale: 8.0 2023-04-03 12:41:32,292 WARNING [train.py:1073] (1/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 2023-04-03 12:41:32,687 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1913, 1.2421, 1.1751, 1.0148, 1.0715, 1.0417, 0.0962, 0.3442], device='cuda:1'), covar=tensor([0.0725, 0.0740, 0.0533, 0.0715, 0.1361, 0.0751, 0.1519, 0.1315], device='cuda:1'), in_proj_covar=tensor([0.0367, 0.0366, 0.0367, 0.0393, 0.0471, 0.0398, 0.0346, 0.0349], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 12:41:33,923 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=190259.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:41:54,171 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9_rvb from training. Duration: 27.511125 2023-04-03 12:41:56,485 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.705e+02 4.583e+02 5.674e+02 7.520e+02 1.844e+03, threshold=1.135e+03, percent-clipped=9.0 2023-04-03 12:42:24,247 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.45 vs. limit=2.0 2023-04-03 12:42:31,996 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5972, 1.7039, 1.9533, 1.9438, 1.5127, 1.8795, 1.9340, 1.8122], device='cuda:1'), covar=tensor([0.4232, 0.3955, 0.2155, 0.2558, 0.4049, 0.2424, 0.5406, 0.3660], device='cuda:1'), in_proj_covar=tensor([0.0938, 0.1017, 0.0746, 0.0955, 0.0916, 0.0857, 0.0862, 0.0810], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-03 12:42:32,649 INFO [train.py:903] (1/4) Epoch 28, batch 5950, loss[loss=0.2167, simple_loss=0.3, pruned_loss=0.06666, over 19772.00 frames. ], tot_loss[loss=0.2009, simple_loss=0.2827, pruned_loss=0.05954, over 3808689.00 frames. ], batch size: 63, lr: 2.91e-03, grad_scale: 8.0 2023-04-03 12:43:07,967 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=190334.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:43:13,914 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.4035, 1.5370, 1.6137, 1.5680, 3.0101, 1.1447, 2.3688, 3.4421], device='cuda:1'), covar=tensor([0.0558, 0.2624, 0.2802, 0.1821, 0.0683, 0.2526, 0.1277, 0.0264], device='cuda:1'), in_proj_covar=tensor([0.0425, 0.0379, 0.0399, 0.0354, 0.0383, 0.0359, 0.0399, 0.0420], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 12:43:34,404 INFO [train.py:903] (1/4) Epoch 28, batch 6000, loss[loss=0.1837, simple_loss=0.2752, pruned_loss=0.04613, over 19653.00 frames. ], tot_loss[loss=0.2013, simple_loss=0.2833, pruned_loss=0.05968, over 3804253.80 frames. ], batch size: 55, lr: 2.90e-03, grad_scale: 8.0 2023-04-03 12:43:34,405 INFO [train.py:928] (1/4) Computing validation loss 2023-04-03 12:43:48,493 INFO [train.py:937] (1/4) Epoch 28, validation: loss=0.1668, simple_loss=0.2663, pruned_loss=0.03368, over 944034.00 frames. 2023-04-03 12:43:48,494 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18821MB 2023-04-03 12:44:08,735 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=190373.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:44:15,160 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.216e+02 4.800e+02 5.909e+02 8.119e+02 1.607e+03, threshold=1.182e+03, percent-clipped=4.0 2023-04-03 12:44:41,367 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=190398.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:44:50,436 INFO [train.py:903] (1/4) Epoch 28, batch 6050, loss[loss=0.2289, simple_loss=0.2974, pruned_loss=0.08022, over 19751.00 frames. ], tot_loss[loss=0.2014, simple_loss=0.2835, pruned_loss=0.05964, over 3802846.12 frames. ], batch size: 47, lr: 2.90e-03, grad_scale: 8.0 2023-04-03 12:45:04,562 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=190417.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:45:11,915 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.8437, 4.4552, 2.9493, 3.9098, 1.0623, 4.4380, 4.2590, 4.3504], device='cuda:1'), covar=tensor([0.0579, 0.0812, 0.1808, 0.0866, 0.3854, 0.0600, 0.0903, 0.1186], device='cuda:1'), in_proj_covar=tensor([0.0531, 0.0431, 0.0519, 0.0361, 0.0411, 0.0458, 0.0450, 0.0483], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 12:45:28,511 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=190437.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:45:50,732 INFO [train.py:903] (1/4) Epoch 28, batch 6100, loss[loss=0.2077, simple_loss=0.2687, pruned_loss=0.0734, over 19370.00 frames. ], tot_loss[loss=0.2024, simple_loss=0.284, pruned_loss=0.06045, over 3802584.34 frames. ], batch size: 47, lr: 2.90e-03, grad_scale: 8.0 2023-04-03 12:45:57,946 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=190461.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:45:59,867 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=190462.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:46:17,942 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.255e+02 4.836e+02 5.950e+02 7.277e+02 1.439e+03, threshold=1.190e+03, percent-clipped=1.0 2023-04-03 12:46:21,259 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.57 vs. limit=5.0 2023-04-03 12:46:28,813 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=190486.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:46:46,944 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=190501.0, num_to_drop=1, layers_to_drop={0} 2023-04-03 12:46:53,524 INFO [train.py:903] (1/4) Epoch 28, batch 6150, loss[loss=0.1848, simple_loss=0.2641, pruned_loss=0.05272, over 19528.00 frames. ], tot_loss[loss=0.201, simple_loss=0.2823, pruned_loss=0.05986, over 3812874.21 frames. ], batch size: 54, lr: 2.90e-03, grad_scale: 8.0 2023-04-03 12:47:02,052 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=190513.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:47:23,094 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9_rvb from training. Duration: 31.02225 2023-04-03 12:47:56,364 INFO [train.py:903] (1/4) Epoch 28, batch 6200, loss[loss=0.1906, simple_loss=0.2785, pruned_loss=0.05132, over 19779.00 frames. ], tot_loss[loss=0.1994, simple_loss=0.2813, pruned_loss=0.05879, over 3823473.33 frames. ], batch size: 56, lr: 2.90e-03, grad_scale: 8.0 2023-04-03 12:48:23,170 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.156e+02 4.897e+02 5.800e+02 7.277e+02 1.783e+03, threshold=1.160e+03, percent-clipped=5.0 2023-04-03 12:48:59,510 INFO [train.py:903] (1/4) Epoch 28, batch 6250, loss[loss=0.1997, simple_loss=0.2895, pruned_loss=0.05497, over 19116.00 frames. ], tot_loss[loss=0.199, simple_loss=0.2808, pruned_loss=0.05861, over 3822529.52 frames. ], batch size: 69, lr: 2.90e-03, grad_scale: 16.0 2023-04-03 12:49:03,309 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=190609.0, num_to_drop=1, layers_to_drop={0} 2023-04-03 12:49:11,048 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=190616.0, num_to_drop=1, layers_to_drop={0} 2023-04-03 12:49:26,319 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=190627.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:49:27,628 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=190628.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:49:31,959 WARNING [train.py:1073] (1/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9_rvb from training. Duration: 25.988875 2023-04-03 12:49:58,813 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.13 vs. limit=2.0 2023-04-03 12:50:00,482 INFO [train.py:903] (1/4) Epoch 28, batch 6300, loss[loss=0.1842, simple_loss=0.2718, pruned_loss=0.04834, over 19590.00 frames. ], tot_loss[loss=0.1974, simple_loss=0.2793, pruned_loss=0.05776, over 3828703.89 frames. ], batch size: 52, lr: 2.90e-03, grad_scale: 8.0 2023-04-03 12:50:28,879 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=190678.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:50:29,804 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.019e+02 4.803e+02 6.276e+02 7.910e+02 2.564e+03, threshold=1.255e+03, percent-clipped=8.0 2023-04-03 12:50:47,605 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=190693.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:51:04,099 INFO [train.py:903] (1/4) Epoch 28, batch 6350, loss[loss=0.1767, simple_loss=0.2705, pruned_loss=0.0414, over 19633.00 frames. ], tot_loss[loss=0.198, simple_loss=0.2798, pruned_loss=0.0581, over 3822943.17 frames. ], batch size: 55, lr: 2.90e-03, grad_scale: 8.0 2023-04-03 12:52:06,522 INFO [train.py:903] (1/4) Epoch 28, batch 6400, loss[loss=0.1998, simple_loss=0.29, pruned_loss=0.05483, over 18796.00 frames. ], tot_loss[loss=0.1981, simple_loss=0.28, pruned_loss=0.05808, over 3820160.94 frames. ], batch size: 74, lr: 2.90e-03, grad_scale: 8.0 2023-04-03 12:52:13,581 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=190761.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:52:36,095 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.096e+02 4.755e+02 6.032e+02 8.092e+02 1.400e+03, threshold=1.206e+03, percent-clipped=2.0 2023-04-03 12:52:52,702 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=190793.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:53:08,684 INFO [train.py:903] (1/4) Epoch 28, batch 6450, loss[loss=0.2076, simple_loss=0.2955, pruned_loss=0.05979, over 19609.00 frames. ], tot_loss[loss=0.1985, simple_loss=0.2806, pruned_loss=0.05819, over 3825535.82 frames. ], batch size: 61, lr: 2.90e-03, grad_scale: 4.0 2023-04-03 12:53:09,391 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 2023-04-03 12:53:55,127 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9_rvb from training. Duration: 28.638875 2023-04-03 12:54:10,281 INFO [train.py:903] (1/4) Epoch 28, batch 6500, loss[loss=0.2014, simple_loss=0.286, pruned_loss=0.05842, over 19604.00 frames. ], tot_loss[loss=0.1982, simple_loss=0.2805, pruned_loss=0.05795, over 3837679.53 frames. ], batch size: 61, lr: 2.90e-03, grad_scale: 4.0 2023-04-03 12:54:17,913 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1_rvb from training. Duration: 0.836375 2023-04-03 12:54:32,004 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=190872.0, num_to_drop=1, layers_to_drop={0} 2023-04-03 12:54:36,601 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=190876.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:54:36,729 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6031, 1.7038, 1.9821, 1.8970, 1.5046, 1.8852, 1.9454, 1.8181], device='cuda:1'), covar=tensor([0.4335, 0.3921, 0.2018, 0.2619, 0.4095, 0.2430, 0.5273, 0.3621], device='cuda:1'), in_proj_covar=tensor([0.0939, 0.1018, 0.0746, 0.0954, 0.0915, 0.0857, 0.0861, 0.0809], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-03 12:54:40,950 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.047e+02 4.758e+02 5.991e+02 7.680e+02 1.999e+03, threshold=1.198e+03, percent-clipped=5.0 2023-04-03 12:54:47,140 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=190884.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:54:49,293 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.4035, 2.3437, 2.1925, 2.6624, 2.2626, 2.0527, 2.0354, 2.5114], device='cuda:1'), covar=tensor([0.1084, 0.1705, 0.1531, 0.1061, 0.1459, 0.0585, 0.1588, 0.0688], device='cuda:1'), in_proj_covar=tensor([0.0276, 0.0358, 0.0316, 0.0256, 0.0304, 0.0253, 0.0319, 0.0260], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 12:55:01,693 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=190897.0, num_to_drop=1, layers_to_drop={0} 2023-04-03 12:55:13,651 INFO [train.py:903] (1/4) Epoch 28, batch 6550, loss[loss=0.1706, simple_loss=0.2519, pruned_loss=0.04465, over 19774.00 frames. ], tot_loss[loss=0.199, simple_loss=0.2812, pruned_loss=0.05842, over 3833104.25 frames. ], batch size: 46, lr: 2.90e-03, grad_scale: 4.0 2023-04-03 12:55:18,347 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=190909.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:55:45,225 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.3988, 4.0292, 2.7485, 3.5743, 1.0655, 3.9629, 3.8569, 3.9535], device='cuda:1'), covar=tensor([0.0681, 0.1014, 0.1919, 0.0950, 0.3874, 0.0760, 0.0921, 0.1132], device='cuda:1'), in_proj_covar=tensor([0.0534, 0.0433, 0.0521, 0.0362, 0.0411, 0.0459, 0.0452, 0.0485], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 12:56:14,194 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=190953.0, num_to_drop=1, layers_to_drop={0} 2023-04-03 12:56:17,102 INFO [train.py:903] (1/4) Epoch 28, batch 6600, loss[loss=0.1832, simple_loss=0.2603, pruned_loss=0.05307, over 19771.00 frames. ], tot_loss[loss=0.1988, simple_loss=0.2807, pruned_loss=0.05846, over 3836416.84 frames. ], batch size: 46, lr: 2.90e-03, grad_scale: 4.0 2023-04-03 12:56:35,866 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=190971.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:56:48,225 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.060e+02 4.870e+02 6.050e+02 8.150e+02 2.542e+03, threshold=1.210e+03, percent-clipped=13.0 2023-04-03 12:56:50,190 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 2023-04-03 12:56:53,376 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.7629, 1.6756, 1.8631, 1.6775, 4.2756, 1.2089, 2.7167, 4.6533], device='cuda:1'), covar=tensor([0.0441, 0.2904, 0.2911, 0.2111, 0.0738, 0.2930, 0.1556, 0.0178], device='cuda:1'), in_proj_covar=tensor([0.0427, 0.0382, 0.0400, 0.0356, 0.0386, 0.0361, 0.0401, 0.0421], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 12:57:19,915 INFO [train.py:903] (1/4) Epoch 28, batch 6650, loss[loss=0.2117, simple_loss=0.2985, pruned_loss=0.06243, over 19702.00 frames. ], tot_loss[loss=0.1999, simple_loss=0.2817, pruned_loss=0.05906, over 3830733.50 frames. ], batch size: 59, lr: 2.90e-03, grad_scale: 2.0 2023-04-03 12:57:54,086 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.88 vs. limit=5.0 2023-04-03 12:57:59,360 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=191037.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:58:14,554 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=191049.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:58:22,332 INFO [train.py:903] (1/4) Epoch 28, batch 6700, loss[loss=0.209, simple_loss=0.2829, pruned_loss=0.06759, over 19563.00 frames. ], tot_loss[loss=0.1991, simple_loss=0.281, pruned_loss=0.05854, over 3838807.05 frames. ], batch size: 52, lr: 2.90e-03, grad_scale: 2.0 2023-04-03 12:58:38,460 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=191068.0, num_to_drop=1, layers_to_drop={0} 2023-04-03 12:58:45,528 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=191074.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:58:52,989 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.114e+02 5.245e+02 6.198e+02 8.145e+02 2.088e+03, threshold=1.240e+03, percent-clipped=7.0 2023-04-03 12:58:59,002 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=191086.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:58:59,067 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=191086.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 12:59:21,819 INFO [train.py:903] (1/4) Epoch 28, batch 6750, loss[loss=0.1897, simple_loss=0.2711, pruned_loss=0.05416, over 19579.00 frames. ], tot_loss[loss=0.2008, simple_loss=0.2822, pruned_loss=0.05969, over 3811546.49 frames. ], batch size: 52, lr: 2.90e-03, grad_scale: 2.0 2023-04-03 12:59:51,532 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=191132.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:00:11,310 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=191150.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:00:13,780 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=191152.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:00:17,982 INFO [train.py:903] (1/4) Epoch 28, batch 6800, loss[loss=0.2517, simple_loss=0.3185, pruned_loss=0.09244, over 13536.00 frames. ], tot_loss[loss=0.2018, simple_loss=0.2828, pruned_loss=0.0604, over 3789208.81 frames. ], batch size: 136, lr: 2.90e-03, grad_scale: 4.0 2023-04-03 13:00:19,447 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=191157.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:00:20,961 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.08 vs. limit=2.0 2023-04-03 13:00:45,258 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.070e+02 5.099e+02 6.292e+02 8.518e+02 1.501e+03, threshold=1.258e+03, percent-clipped=3.0 2023-04-03 13:01:04,676 WARNING [train.py:1073] (1/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 2023-04-03 13:01:05,146 WARNING [train.py:1073] (1/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 2023-04-03 13:01:08,829 INFO [train.py:903] (1/4) Epoch 29, batch 0, loss[loss=0.1662, simple_loss=0.2426, pruned_loss=0.0449, over 19484.00 frames. ], tot_loss[loss=0.1662, simple_loss=0.2426, pruned_loss=0.0449, over 19484.00 frames. ], batch size: 49, lr: 2.85e-03, grad_scale: 8.0 2023-04-03 13:01:08,829 INFO [train.py:928] (1/4) Computing validation loss 2023-04-03 13:01:20,493 INFO [train.py:937] (1/4) Epoch 29, validation: loss=0.1669, simple_loss=0.2669, pruned_loss=0.03339, over 944034.00 frames. 2023-04-03 13:01:20,494 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18821MB 2023-04-03 13:01:31,766 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 2023-04-03 13:01:43,679 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=191203.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:01:46,163 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6475, 1.7165, 2.0091, 2.0344, 1.5570, 2.0039, 1.9898, 1.8382], device='cuda:1'), covar=tensor([0.4229, 0.3731, 0.2019, 0.2253, 0.3850, 0.2173, 0.5293, 0.3591], device='cuda:1'), in_proj_covar=tensor([0.0946, 0.1023, 0.0750, 0.0960, 0.0921, 0.0863, 0.0868, 0.0815], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-03 13:02:19,662 INFO [train.py:903] (1/4) Epoch 29, batch 50, loss[loss=0.1955, simple_loss=0.2657, pruned_loss=0.06267, over 19764.00 frames. ], tot_loss[loss=0.1988, simple_loss=0.2798, pruned_loss=0.05887, over 871960.96 frames. ], batch size: 46, lr: 2.85e-03, grad_scale: 8.0 2023-04-03 13:02:55,022 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1_rvb from training. Duration: 27.0318125 2023-04-03 13:03:15,128 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.166e+02 5.210e+02 6.215e+02 8.429e+02 1.754e+03, threshold=1.243e+03, percent-clipped=6.0 2023-04-03 13:03:18,588 INFO [train.py:903] (1/4) Epoch 29, batch 100, loss[loss=0.1926, simple_loss=0.279, pruned_loss=0.05307, over 19655.00 frames. ], tot_loss[loss=0.1987, simple_loss=0.2806, pruned_loss=0.05841, over 1523997.28 frames. ], batch size: 55, lr: 2.85e-03, grad_scale: 8.0 2023-04-03 13:03:33,065 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 2023-04-03 13:04:07,317 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=191324.0, num_to_drop=1, layers_to_drop={1} 2023-04-03 13:04:19,665 INFO [train.py:903] (1/4) Epoch 29, batch 150, loss[loss=0.2008, simple_loss=0.2879, pruned_loss=0.05687, over 18093.00 frames. ], tot_loss[loss=0.1972, simple_loss=0.2791, pruned_loss=0.05766, over 2042649.02 frames. ], batch size: 83, lr: 2.85e-03, grad_scale: 8.0 2023-04-03 13:04:29,101 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=191342.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:04:36,957 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=191349.0, num_to_drop=1, layers_to_drop={0} 2023-04-03 13:04:59,842 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=191367.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:05:15,367 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.397e+02 5.272e+02 6.303e+02 7.742e+02 1.475e+03, threshold=1.261e+03, percent-clipped=3.0 2023-04-03 13:05:15,429 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 2023-04-03 13:05:18,047 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=191383.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:05:18,900 INFO [train.py:903] (1/4) Epoch 29, batch 200, loss[loss=0.2314, simple_loss=0.3061, pruned_loss=0.07833, over 17462.00 frames. ], tot_loss[loss=0.2011, simple_loss=0.2826, pruned_loss=0.05981, over 2421080.09 frames. ], batch size: 101, lr: 2.85e-03, grad_scale: 8.0 2023-04-03 13:05:37,010 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.2899, 2.1437, 2.0553, 1.9065, 1.7327, 1.8148, 0.7591, 1.2330], device='cuda:1'), covar=tensor([0.0700, 0.0699, 0.0523, 0.0862, 0.1223, 0.1031, 0.1379, 0.1155], device='cuda:1'), in_proj_covar=tensor([0.0368, 0.0366, 0.0367, 0.0393, 0.0471, 0.0398, 0.0346, 0.0349], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 13:05:48,183 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=191408.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:06:13,706 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=191430.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:06:17,601 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=191433.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:06:20,064 INFO [train.py:903] (1/4) Epoch 29, batch 250, loss[loss=0.1867, simple_loss=0.2785, pruned_loss=0.04746, over 19348.00 frames. ], tot_loss[loss=0.1994, simple_loss=0.281, pruned_loss=0.05892, over 2736289.48 frames. ], batch size: 66, lr: 2.85e-03, grad_scale: 8.0 2023-04-03 13:06:41,368 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0 2023-04-03 13:07:16,620 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.382e+02 5.311e+02 6.503e+02 8.385e+02 2.446e+03, threshold=1.301e+03, percent-clipped=7.0 2023-04-03 13:07:20,116 INFO [train.py:903] (1/4) Epoch 29, batch 300, loss[loss=0.1608, simple_loss=0.2366, pruned_loss=0.04247, over 19119.00 frames. ], tot_loss[loss=0.2004, simple_loss=0.282, pruned_loss=0.05935, over 2980967.65 frames. ], batch size: 42, lr: 2.85e-03, grad_scale: 8.0 2023-04-03 13:07:33,140 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=191494.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:08:19,683 INFO [train.py:903] (1/4) Epoch 29, batch 350, loss[loss=0.2197, simple_loss=0.3084, pruned_loss=0.06554, over 19529.00 frames. ], tot_loss[loss=0.2011, simple_loss=0.2827, pruned_loss=0.05978, over 3179790.32 frames. ], batch size: 56, lr: 2.84e-03, grad_scale: 8.0 2023-04-03 13:08:27,199 WARNING [train.py:1073] (1/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1_rvb from training. Duration: 0.9681875 2023-04-03 13:08:33,092 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=191545.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:08:35,157 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=191547.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:09:16,623 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.400e+02 4.772e+02 5.697e+02 6.938e+02 1.378e+03, threshold=1.139e+03, percent-clipped=1.0 2023-04-03 13:09:19,970 INFO [train.py:903] (1/4) Epoch 29, batch 400, loss[loss=0.1858, simple_loss=0.2764, pruned_loss=0.04763, over 19546.00 frames. ], tot_loss[loss=0.2011, simple_loss=0.2828, pruned_loss=0.05974, over 3322433.00 frames. ], batch size: 56, lr: 2.84e-03, grad_scale: 8.0 2023-04-03 13:09:50,420 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=191609.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:10:11,891 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.53 vs. limit=5.0 2023-04-03 13:10:19,939 INFO [train.py:903] (1/4) Epoch 29, batch 450, loss[loss=0.2344, simple_loss=0.3035, pruned_loss=0.0826, over 12950.00 frames. ], tot_loss[loss=0.2007, simple_loss=0.2824, pruned_loss=0.05948, over 3444718.73 frames. ], batch size: 137, lr: 2.84e-03, grad_scale: 8.0 2023-04-03 13:10:55,268 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=191662.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:10:57,256 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 2023-04-03 13:10:58,390 WARNING [train.py:1073] (1/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 2023-04-03 13:11:17,385 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.061e+02 4.702e+02 5.973e+02 7.083e+02 1.398e+03, threshold=1.195e+03, percent-clipped=2.0 2023-04-03 13:11:20,347 INFO [train.py:903] (1/4) Epoch 29, batch 500, loss[loss=0.2097, simple_loss=0.293, pruned_loss=0.0632, over 19716.00 frames. ], tot_loss[loss=0.2008, simple_loss=0.2826, pruned_loss=0.05953, over 3520280.66 frames. ], batch size: 63, lr: 2.84e-03, grad_scale: 8.0 2023-04-03 13:12:13,382 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=191727.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:12:21,067 INFO [train.py:903] (1/4) Epoch 29, batch 550, loss[loss=0.2198, simple_loss=0.3001, pruned_loss=0.06976, over 17690.00 frames. ], tot_loss[loss=0.2014, simple_loss=0.2832, pruned_loss=0.0598, over 3565320.13 frames. ], batch size: 101, lr: 2.84e-03, grad_scale: 4.0 2023-04-03 13:12:45,844 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0821, 1.8101, 1.7002, 2.0001, 1.6503, 1.7291, 1.6057, 1.9446], device='cuda:1'), covar=tensor([0.1040, 0.1323, 0.1458, 0.1091, 0.1390, 0.0549, 0.1587, 0.0732], device='cuda:1'), in_proj_covar=tensor([0.0279, 0.0362, 0.0320, 0.0259, 0.0307, 0.0256, 0.0323, 0.0262], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-04-03 13:13:04,092 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([0.9850, 1.2779, 1.7197, 0.8847, 2.3163, 3.0551, 2.7115, 3.2321], device='cuda:1'), covar=tensor([0.1725, 0.4068, 0.3382, 0.2937, 0.0652, 0.0231, 0.0279, 0.0322], device='cuda:1'), in_proj_covar=tensor([0.0281, 0.0334, 0.0366, 0.0272, 0.0256, 0.0198, 0.0221, 0.0279], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-03 13:13:04,201 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4049, 1.4497, 1.7196, 1.6218, 2.3106, 2.0752, 2.3950, 0.9958], device='cuda:1'), covar=tensor([0.2551, 0.4438, 0.2698, 0.2014, 0.1511, 0.2340, 0.1379, 0.4786], device='cuda:1'), in_proj_covar=tensor([0.0555, 0.0672, 0.0759, 0.0509, 0.0635, 0.0546, 0.0671, 0.0575], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-03 13:13:06,445 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5305, 1.6244, 1.9438, 1.8063, 2.8633, 2.4021, 2.9817, 1.4658], device='cuda:1'), covar=tensor([0.2623, 0.4432, 0.2870, 0.2026, 0.1486, 0.2300, 0.1448, 0.4586], device='cuda:1'), in_proj_covar=tensor([0.0555, 0.0672, 0.0759, 0.0509, 0.0635, 0.0546, 0.0671, 0.0575], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-03 13:13:19,226 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.607e+02 5.419e+02 6.471e+02 8.968e+02 1.855e+03, threshold=1.294e+03, percent-clipped=10.0 2023-04-03 13:13:22,383 INFO [train.py:903] (1/4) Epoch 29, batch 600, loss[loss=0.1857, simple_loss=0.2599, pruned_loss=0.05576, over 19079.00 frames. ], tot_loss[loss=0.202, simple_loss=0.2837, pruned_loss=0.06009, over 3624961.13 frames. ], batch size: 42, lr: 2.84e-03, grad_scale: 4.0 2023-04-03 13:13:41,913 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=191801.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:14:02,330 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 2023-04-03 13:14:13,338 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=191826.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:14:22,139 INFO [train.py:903] (1/4) Epoch 29, batch 650, loss[loss=0.1932, simple_loss=0.2863, pruned_loss=0.05004, over 19736.00 frames. ], tot_loss[loss=0.2015, simple_loss=0.2832, pruned_loss=0.05989, over 3666017.06 frames. ], batch size: 63, lr: 2.84e-03, grad_scale: 4.0 2023-04-03 13:14:31,462 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=191842.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:14:42,693 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.53 vs. limit=2.0 2023-04-03 13:15:01,023 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=191865.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:15:18,380 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.813e+02 4.883e+02 6.286e+02 7.850e+02 2.365e+03, threshold=1.257e+03, percent-clipped=4.0 2023-04-03 13:15:21,594 INFO [train.py:903] (1/4) Epoch 29, batch 700, loss[loss=0.2119, simple_loss=0.2935, pruned_loss=0.06519, over 19743.00 frames. ], tot_loss[loss=0.2023, simple_loss=0.2844, pruned_loss=0.06008, over 3704310.43 frames. ], batch size: 63, lr: 2.84e-03, grad_scale: 4.0 2023-04-03 13:15:29,659 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=191890.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:16:03,792 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=191918.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:16:23,390 INFO [train.py:903] (1/4) Epoch 29, batch 750, loss[loss=0.1938, simple_loss=0.2772, pruned_loss=0.05524, over 19710.00 frames. ], tot_loss[loss=0.2015, simple_loss=0.2836, pruned_loss=0.05973, over 3717825.89 frames. ], batch size: 59, lr: 2.84e-03, grad_scale: 4.0 2023-04-03 13:16:34,317 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=191943.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:17:21,354 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.970e+02 4.754e+02 5.591e+02 7.301e+02 1.189e+03, threshold=1.118e+03, percent-clipped=0.0 2023-04-03 13:17:23,700 INFO [train.py:903] (1/4) Epoch 29, batch 800, loss[loss=0.2228, simple_loss=0.3083, pruned_loss=0.06869, over 18249.00 frames. ], tot_loss[loss=0.202, simple_loss=0.2841, pruned_loss=0.05993, over 3747217.81 frames. ], batch size: 83, lr: 2.84e-03, grad_scale: 8.0 2023-04-03 13:17:23,992 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.8184, 3.3035, 3.3566, 3.3660, 1.3434, 3.2404, 2.8220, 3.1433], device='cuda:1'), covar=tensor([0.1859, 0.1077, 0.0828, 0.1026, 0.5996, 0.1055, 0.0890, 0.1361], device='cuda:1'), in_proj_covar=tensor([0.0821, 0.0786, 0.0994, 0.0877, 0.0864, 0.0762, 0.0588, 0.0927], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-03 13:17:36,670 WARNING [train.py:1073] (1/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 2023-04-03 13:18:16,905 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.6354, 2.5367, 2.3354, 2.6216, 2.4226, 2.1319, 2.1858, 2.4865], device='cuda:1'), covar=tensor([0.0992, 0.1506, 0.1427, 0.1106, 0.1363, 0.0583, 0.1484, 0.0708], device='cuda:1'), in_proj_covar=tensor([0.0279, 0.0362, 0.0321, 0.0260, 0.0307, 0.0257, 0.0323, 0.0263], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-04-03 13:18:26,428 INFO [train.py:903] (1/4) Epoch 29, batch 850, loss[loss=0.1698, simple_loss=0.2477, pruned_loss=0.046, over 19746.00 frames. ], tot_loss[loss=0.2017, simple_loss=0.284, pruned_loss=0.0597, over 3772989.34 frames. ], batch size: 46, lr: 2.84e-03, grad_scale: 8.0 2023-04-03 13:19:17,572 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9_rvb from training. Duration: 25.061125 2023-04-03 13:19:23,133 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.899e+02 4.771e+02 5.573e+02 7.809e+02 2.111e+03, threshold=1.115e+03, percent-clipped=7.0 2023-04-03 13:19:25,393 INFO [train.py:903] (1/4) Epoch 29, batch 900, loss[loss=0.2267, simple_loss=0.303, pruned_loss=0.07518, over 19504.00 frames. ], tot_loss[loss=0.2022, simple_loss=0.2841, pruned_loss=0.06016, over 3786407.05 frames. ], batch size: 64, lr: 2.84e-03, grad_scale: 8.0 2023-04-03 13:19:43,262 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=192098.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:20:07,631 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.82 vs. limit=2.0 2023-04-03 13:20:09,365 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=192120.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:20:12,966 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=192123.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:20:26,582 INFO [train.py:903] (1/4) Epoch 29, batch 950, loss[loss=0.2124, simple_loss=0.2975, pruned_loss=0.06366, over 19694.00 frames. ], tot_loss[loss=0.2016, simple_loss=0.2832, pruned_loss=0.05998, over 3806967.50 frames. ], batch size: 59, lr: 2.84e-03, grad_scale: 8.0 2023-04-03 13:20:30,182 WARNING [train.py:1073] (1/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9_rvb from training. Duration: 26.32775 2023-04-03 13:21:13,704 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=192173.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:21:24,654 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.322e+02 5.272e+02 6.130e+02 7.632e+02 1.213e+03, threshold=1.226e+03, percent-clipped=2.0 2023-04-03 13:21:27,713 INFO [train.py:903] (1/4) Epoch 29, batch 1000, loss[loss=0.2028, simple_loss=0.2852, pruned_loss=0.0602, over 19513.00 frames. ], tot_loss[loss=0.2001, simple_loss=0.2815, pruned_loss=0.05936, over 3817643.07 frames. ], batch size: 64, lr: 2.84e-03, grad_scale: 8.0 2023-04-03 13:22:23,179 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 2023-04-03 13:22:27,655 INFO [train.py:903] (1/4) Epoch 29, batch 1050, loss[loss=0.2322, simple_loss=0.3113, pruned_loss=0.07654, over 17407.00 frames. ], tot_loss[loss=0.2007, simple_loss=0.282, pruned_loss=0.05969, over 3815665.84 frames. ], batch size: 101, lr: 2.84e-03, grad_scale: 8.0 2023-04-03 13:22:31,123 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=192237.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:22:34,316 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=192239.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:23:02,374 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 2023-04-03 13:23:25,657 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.003e+02 4.755e+02 5.709e+02 7.263e+02 1.610e+03, threshold=1.142e+03, percent-clipped=3.0 2023-04-03 13:23:28,074 INFO [train.py:903] (1/4) Epoch 29, batch 1100, loss[loss=0.1998, simple_loss=0.2947, pruned_loss=0.05239, over 19776.00 frames. ], tot_loss[loss=0.2012, simple_loss=0.2826, pruned_loss=0.05993, over 3816287.97 frames. ], batch size: 56, lr: 2.84e-03, grad_scale: 8.0 2023-04-03 13:24:29,303 INFO [train.py:903] (1/4) Epoch 29, batch 1150, loss[loss=0.2135, simple_loss=0.2968, pruned_loss=0.06513, over 19741.00 frames. ], tot_loss[loss=0.2019, simple_loss=0.2833, pruned_loss=0.06023, over 3803659.37 frames. ], batch size: 63, lr: 2.84e-03, grad_scale: 8.0 2023-04-03 13:24:53,861 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5708, 1.5847, 1.9300, 1.8282, 2.6899, 2.3935, 2.8808, 1.3124], device='cuda:1'), covar=tensor([0.2586, 0.4613, 0.2857, 0.1992, 0.1538, 0.2183, 0.1408, 0.4830], device='cuda:1'), in_proj_covar=tensor([0.0555, 0.0672, 0.0760, 0.0511, 0.0635, 0.0545, 0.0670, 0.0574], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 13:25:27,744 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.011e+02 4.839e+02 5.803e+02 7.271e+02 1.538e+03, threshold=1.161e+03, percent-clipped=1.0 2023-04-03 13:25:30,847 INFO [train.py:903] (1/4) Epoch 29, batch 1200, loss[loss=0.1989, simple_loss=0.2885, pruned_loss=0.05465, over 19663.00 frames. ], tot_loss[loss=0.2013, simple_loss=0.2828, pruned_loss=0.05996, over 3802126.80 frames. ], batch size: 58, lr: 2.84e-03, grad_scale: 8.0 2023-04-03 13:26:01,131 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=192410.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:26:01,920 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 2023-04-03 13:26:10,304 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=192417.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:26:30,211 INFO [train.py:903] (1/4) Epoch 29, batch 1250, loss[loss=0.1797, simple_loss=0.2616, pruned_loss=0.04888, over 19580.00 frames. ], tot_loss[loss=0.2007, simple_loss=0.2824, pruned_loss=0.05947, over 3818166.91 frames. ], batch size: 52, lr: 2.84e-03, grad_scale: 8.0 2023-04-03 13:26:50,089 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=192451.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:27:07,153 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=192464.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:27:28,114 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.043e+02 5.075e+02 6.202e+02 7.494e+02 1.680e+03, threshold=1.240e+03, percent-clipped=4.0 2023-04-03 13:27:30,147 INFO [train.py:903] (1/4) Epoch 29, batch 1300, loss[loss=0.2544, simple_loss=0.3242, pruned_loss=0.09225, over 17378.00 frames. ], tot_loss[loss=0.2004, simple_loss=0.2818, pruned_loss=0.05948, over 3821628.90 frames. ], batch size: 101, lr: 2.84e-03, grad_scale: 8.0 2023-04-03 13:28:10,137 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=192517.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:28:14,819 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2442, 1.2910, 1.2537, 1.0898, 1.1244, 1.0506, 0.0971, 0.4249], device='cuda:1'), covar=tensor([0.0813, 0.0770, 0.0559, 0.0718, 0.1575, 0.0798, 0.1514, 0.1310], device='cuda:1'), in_proj_covar=tensor([0.0368, 0.0365, 0.0367, 0.0391, 0.0471, 0.0398, 0.0345, 0.0347], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 13:28:15,739 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.7265, 3.4190, 2.5801, 3.0645, 1.5390, 3.3546, 3.2680, 3.3473], device='cuda:1'), covar=tensor([0.0953, 0.1132, 0.2052, 0.0990, 0.2834, 0.0872, 0.1155, 0.1411], device='cuda:1'), in_proj_covar=tensor([0.0531, 0.0431, 0.0519, 0.0361, 0.0410, 0.0457, 0.0452, 0.0482], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 13:28:30,905 INFO [train.py:903] (1/4) Epoch 29, batch 1350, loss[loss=0.1994, simple_loss=0.2781, pruned_loss=0.06032, over 19493.00 frames. ], tot_loss[loss=0.2, simple_loss=0.2813, pruned_loss=0.05933, over 3831825.63 frames. ], batch size: 49, lr: 2.84e-03, grad_scale: 8.0 2023-04-03 13:29:25,921 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=192579.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:29:27,883 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=192581.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:29:28,924 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.238e+02 5.533e+02 7.774e+02 1.028e+03 2.542e+03, threshold=1.555e+03, percent-clipped=13.0 2023-04-03 13:29:31,159 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=192583.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:29:32,191 INFO [train.py:903] (1/4) Epoch 29, batch 1400, loss[loss=0.2234, simple_loss=0.3055, pruned_loss=0.07069, over 19384.00 frames. ], tot_loss[loss=0.1995, simple_loss=0.2813, pruned_loss=0.05887, over 3823371.60 frames. ], batch size: 70, lr: 2.84e-03, grad_scale: 8.0 2023-04-03 13:29:42,125 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.74 vs. limit=2.0 2023-04-03 13:29:44,071 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4676, 1.4847, 1.7049, 1.7126, 2.3821, 2.2030, 2.5409, 1.0523], device='cuda:1'), covar=tensor([0.2619, 0.4675, 0.2997, 0.2051, 0.1709, 0.2254, 0.1549, 0.5032], device='cuda:1'), in_proj_covar=tensor([0.0555, 0.0671, 0.0760, 0.0510, 0.0633, 0.0545, 0.0668, 0.0574], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 13:30:30,367 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=192632.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:30:32,441 INFO [train.py:903] (1/4) Epoch 29, batch 1450, loss[loss=0.1812, simple_loss=0.257, pruned_loss=0.05271, over 19726.00 frames. ], tot_loss[loss=0.1995, simple_loss=0.2813, pruned_loss=0.05889, over 3826493.23 frames. ], batch size: 46, lr: 2.84e-03, grad_scale: 8.0 2023-04-03 13:30:32,470 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_rvb from training. Duration: 25.85 2023-04-03 13:30:32,732 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=192634.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:31:29,879 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.888e+02 4.834e+02 5.734e+02 7.377e+02 1.406e+03, threshold=1.147e+03, percent-clipped=0.0 2023-04-03 13:31:32,941 INFO [train.py:903] (1/4) Epoch 29, batch 1500, loss[loss=0.1776, simple_loss=0.2653, pruned_loss=0.04494, over 19585.00 frames. ], tot_loss[loss=0.1991, simple_loss=0.2809, pruned_loss=0.05866, over 3821689.99 frames. ], batch size: 52, lr: 2.84e-03, grad_scale: 8.0 2023-04-03 13:31:41,353 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=192691.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:31:47,403 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=192696.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:31:50,556 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=192698.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:32:32,341 INFO [train.py:903] (1/4) Epoch 29, batch 1550, loss[loss=0.2244, simple_loss=0.3006, pruned_loss=0.07409, over 19524.00 frames. ], tot_loss[loss=0.2012, simple_loss=0.2828, pruned_loss=0.05982, over 3813945.24 frames. ], batch size: 54, lr: 2.84e-03, grad_scale: 8.0 2023-04-03 13:32:56,579 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=192753.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:32:57,536 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=192754.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:33:05,626 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=192761.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:33:31,827 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.013e+02 5.130e+02 6.287e+02 7.695e+02 1.691e+03, threshold=1.257e+03, percent-clipped=3.0 2023-04-03 13:33:34,855 INFO [train.py:903] (1/4) Epoch 29, batch 1600, loss[loss=0.1823, simple_loss=0.2725, pruned_loss=0.04599, over 19660.00 frames. ], tot_loss[loss=0.2005, simple_loss=0.2823, pruned_loss=0.05933, over 3810482.84 frames. ], batch size: 53, lr: 2.84e-03, grad_scale: 8.0 2023-04-03 13:33:47,550 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=192795.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:33:51,041 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3733, 1.3647, 1.4942, 1.5413, 1.7503, 1.8843, 1.7912, 0.6160], device='cuda:1'), covar=tensor([0.2448, 0.4291, 0.2888, 0.2021, 0.1733, 0.2340, 0.1525, 0.5210], device='cuda:1'), in_proj_covar=tensor([0.0554, 0.0669, 0.0758, 0.0510, 0.0632, 0.0545, 0.0666, 0.0574], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 13:33:57,595 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9_rvb from training. Duration: 30.1555625 2023-04-03 13:34:34,144 INFO [train.py:903] (1/4) Epoch 29, batch 1650, loss[loss=0.2404, simple_loss=0.3173, pruned_loss=0.08174, over 19324.00 frames. ], tot_loss[loss=0.2008, simple_loss=0.2825, pruned_loss=0.05953, over 3823680.51 frames. ], batch size: 66, lr: 2.84e-03, grad_scale: 8.0 2023-04-03 13:34:35,675 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=192835.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:34:51,822 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=192849.0, num_to_drop=1, layers_to_drop={0} 2023-04-03 13:35:04,026 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=192860.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:35:15,785 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=192869.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:35:23,787 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=192876.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:35:30,386 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.635e+02 4.944e+02 5.872e+02 7.824e+02 1.796e+03, threshold=1.174e+03, percent-clipped=2.0 2023-04-03 13:35:32,716 INFO [train.py:903] (1/4) Epoch 29, batch 1700, loss[loss=0.2085, simple_loss=0.2832, pruned_loss=0.06688, over 19618.00 frames. ], tot_loss[loss=0.2012, simple_loss=0.2827, pruned_loss=0.05982, over 3812518.11 frames. ], batch size: 50, lr: 2.83e-03, grad_scale: 8.0 2023-04-03 13:35:38,395 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=192888.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:35:40,583 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3722, 1.4488, 1.6645, 1.5739, 2.4832, 2.1020, 2.5909, 1.2486], device='cuda:1'), covar=tensor([0.2547, 0.4481, 0.2761, 0.2125, 0.1508, 0.2365, 0.1383, 0.4659], device='cuda:1'), in_proj_covar=tensor([0.0554, 0.0671, 0.0760, 0.0510, 0.0634, 0.0546, 0.0667, 0.0575], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 13:36:04,808 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=192910.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:36:08,234 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=192913.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:36:13,404 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 2023-04-03 13:36:32,294 INFO [train.py:903] (1/4) Epoch 29, batch 1750, loss[loss=0.1783, simple_loss=0.2695, pruned_loss=0.04354, over 19657.00 frames. ], tot_loss[loss=0.201, simple_loss=0.2827, pruned_loss=0.05963, over 3804956.79 frames. ], batch size: 60, lr: 2.83e-03, grad_scale: 8.0 2023-04-03 13:36:55,825 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=192952.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:36:58,009 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=192954.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:37:24,864 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=192977.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:37:25,785 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=192978.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:37:27,139 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=192979.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:37:31,715 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.864e+02 4.741e+02 6.117e+02 7.797e+02 1.758e+03, threshold=1.223e+03, percent-clipped=7.0 2023-04-03 13:37:34,997 INFO [train.py:903] (1/4) Epoch 29, batch 1800, loss[loss=0.1682, simple_loss=0.2467, pruned_loss=0.04486, over 19741.00 frames. ], tot_loss[loss=0.2013, simple_loss=0.2831, pruned_loss=0.05978, over 3805044.86 frames. ], batch size: 46, lr: 2.83e-03, grad_scale: 8.0 2023-04-03 13:37:46,814 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0785, 2.0692, 1.7606, 2.1179, 1.8329, 1.7355, 1.6683, 1.9229], device='cuda:1'), covar=tensor([0.1059, 0.1321, 0.1597, 0.1100, 0.1438, 0.0600, 0.1603, 0.0786], device='cuda:1'), in_proj_covar=tensor([0.0279, 0.0362, 0.0321, 0.0259, 0.0309, 0.0258, 0.0324, 0.0264], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-04-03 13:37:47,939 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6971, 1.5733, 1.6524, 2.1085, 1.7831, 1.8168, 1.8260, 1.7082], device='cuda:1'), covar=tensor([0.0676, 0.0771, 0.0803, 0.0539, 0.0848, 0.0698, 0.0859, 0.0626], device='cuda:1'), in_proj_covar=tensor([0.0213, 0.0222, 0.0229, 0.0239, 0.0226, 0.0214, 0.0188, 0.0207], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-04-03 13:38:29,366 WARNING [train.py:1073] (1/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9_rvb from training. Duration: 25.0944375 2023-04-03 13:38:33,981 INFO [train.py:903] (1/4) Epoch 29, batch 1850, loss[loss=0.1816, simple_loss=0.2754, pruned_loss=0.04396, over 19668.00 frames. ], tot_loss[loss=0.2019, simple_loss=0.2836, pruned_loss=0.06009, over 3809616.09 frames. ], batch size: 58, lr: 2.83e-03, grad_scale: 8.0 2023-04-03 13:38:35,276 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=193035.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:38:35,441 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.7503, 1.9244, 2.2609, 1.9812, 2.9494, 3.3683, 3.2473, 3.5248], device='cuda:1'), covar=tensor([0.1428, 0.3129, 0.2781, 0.2342, 0.0920, 0.0306, 0.0203, 0.0370], device='cuda:1'), in_proj_covar=tensor([0.0282, 0.0334, 0.0367, 0.0274, 0.0257, 0.0198, 0.0221, 0.0280], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-03 13:39:04,658 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=193059.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:39:05,538 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9_rvb from training. Duration: 27.8166875 2023-04-03 13:39:21,240 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4959, 1.6749, 2.1980, 1.8450, 3.0981, 2.2978, 3.2076, 1.8280], device='cuda:1'), covar=tensor([0.2891, 0.4891, 0.3011, 0.2233, 0.1579, 0.2748, 0.1937, 0.4704], device='cuda:1'), in_proj_covar=tensor([0.0556, 0.0674, 0.0763, 0.0512, 0.0637, 0.0549, 0.0669, 0.0576], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 13:39:28,424 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 2023-04-03 13:39:32,458 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.498e+02 4.794e+02 5.817e+02 7.567e+02 2.266e+03, threshold=1.163e+03, percent-clipped=4.0 2023-04-03 13:39:34,763 INFO [train.py:903] (1/4) Epoch 29, batch 1900, loss[loss=0.1918, simple_loss=0.2881, pruned_loss=0.04776, over 19694.00 frames. ], tot_loss[loss=0.2022, simple_loss=0.2843, pruned_loss=0.0601, over 3807499.77 frames. ], batch size: 59, lr: 2.83e-03, grad_scale: 8.0 2023-04-03 13:39:45,981 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=193093.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:39:49,044 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 2023-04-03 13:39:50,327 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=193097.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:39:55,324 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9_rvb from training. Duration: 27.02225 2023-04-03 13:40:18,684 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 2023-04-03 13:40:25,443 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=193125.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:40:32,279 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4541, 1.4346, 1.6528, 1.6223, 2.2093, 2.1354, 2.3194, 0.9194], device='cuda:1'), covar=tensor([0.2557, 0.4687, 0.3008, 0.2099, 0.1675, 0.2315, 0.1481, 0.5100], device='cuda:1'), in_proj_covar=tensor([0.0557, 0.0676, 0.0765, 0.0513, 0.0639, 0.0551, 0.0671, 0.0578], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-03 13:40:33,560 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=193132.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:40:33,928 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 2023-04-03 13:40:35,431 INFO [train.py:903] (1/4) Epoch 29, batch 1950, loss[loss=0.228, simple_loss=0.309, pruned_loss=0.07348, over 19569.00 frames. ], tot_loss[loss=0.2006, simple_loss=0.2829, pruned_loss=0.05918, over 3817998.95 frames. ], batch size: 61, lr: 2.83e-03, grad_scale: 8.0 2023-04-03 13:40:56,425 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=193150.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:40:56,463 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=193150.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:41:04,201 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=193157.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:41:15,373 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=193166.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:41:34,710 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.104e+02 4.788e+02 5.811e+02 7.006e+02 1.429e+03, threshold=1.162e+03, percent-clipped=3.0 2023-04-03 13:41:37,846 INFO [train.py:903] (1/4) Epoch 29, batch 2000, loss[loss=0.1777, simple_loss=0.2688, pruned_loss=0.04334, over 19766.00 frames. ], tot_loss[loss=0.1993, simple_loss=0.2815, pruned_loss=0.05851, over 3818905.20 frames. ], batch size: 54, lr: 2.83e-03, grad_scale: 8.0 2023-04-03 13:41:46,094 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=193191.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:41:48,027 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=193193.0, num_to_drop=1, layers_to_drop={0} 2023-04-03 13:42:10,521 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=193212.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:42:31,348 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_rvb from training. Duration: 26.8349375 2023-04-03 13:42:31,630 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=193229.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:42:37,085 INFO [train.py:903] (1/4) Epoch 29, batch 2050, loss[loss=0.2067, simple_loss=0.2945, pruned_loss=0.05952, over 19667.00 frames. ], tot_loss[loss=0.2007, simple_loss=0.2826, pruned_loss=0.05942, over 3815833.07 frames. ], batch size: 58, lr: 2.83e-03, grad_scale: 8.0 2023-04-03 13:42:50,736 WARNING [train.py:1073] (1/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 2023-04-03 13:42:50,761 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 2023-04-03 13:43:00,864 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.6548, 0.8970, 1.3134, 1.4907, 3.0013, 1.1822, 2.5139, 3.5433], device='cuda:1'), covar=tensor([0.0742, 0.4217, 0.3764, 0.2424, 0.1190, 0.3127, 0.1497, 0.0410], device='cuda:1'), in_proj_covar=tensor([0.0428, 0.0382, 0.0402, 0.0357, 0.0386, 0.0362, 0.0401, 0.0422], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 13:43:13,503 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 2023-04-03 13:43:34,576 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.211e+02 4.912e+02 6.403e+02 8.614e+02 2.001e+03, threshold=1.281e+03, percent-clipped=8.0 2023-04-03 13:43:36,896 INFO [train.py:903] (1/4) Epoch 29, batch 2100, loss[loss=0.1982, simple_loss=0.2847, pruned_loss=0.05584, over 19661.00 frames. ], tot_loss[loss=0.2011, simple_loss=0.2827, pruned_loss=0.05973, over 3822370.41 frames. ], batch size: 55, lr: 2.83e-03, grad_scale: 8.0 2023-04-03 13:43:40,750 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([0.8898, 1.3590, 1.0875, 1.0647, 1.1926, 1.0301, 0.9432, 1.2839], device='cuda:1'), covar=tensor([0.0786, 0.0965, 0.1151, 0.0875, 0.0662, 0.1463, 0.0659, 0.0548], device='cuda:1'), in_proj_covar=tensor([0.0303, 0.0318, 0.0342, 0.0272, 0.0253, 0.0346, 0.0293, 0.0274], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 13:43:59,250 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=193302.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:44:07,582 WARNING [train.py:1073] (1/4) Exclude cut with ID 453-131332-0000-131866_sp0.9_rvb from training. Duration: 25.3333125 2023-04-03 13:44:07,929 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=193308.0, num_to_drop=1, layers_to_drop={0} 2023-04-03 13:44:25,614 WARNING [train.py:1073] (1/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9_rvb from training. Duration: 26.6166875 2023-04-03 13:44:37,260 INFO [train.py:903] (1/4) Epoch 29, batch 2150, loss[loss=0.1926, simple_loss=0.266, pruned_loss=0.05961, over 19356.00 frames. ], tot_loss[loss=0.2014, simple_loss=0.2825, pruned_loss=0.06015, over 3836632.83 frames. ], batch size: 47, lr: 2.83e-03, grad_scale: 8.0 2023-04-03 13:44:56,808 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=193349.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:45:25,877 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=193374.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:45:35,169 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.248e+02 5.242e+02 6.391e+02 8.913e+02 2.261e+03, threshold=1.278e+03, percent-clipped=9.0 2023-04-03 13:45:37,407 INFO [train.py:903] (1/4) Epoch 29, batch 2200, loss[loss=0.2089, simple_loss=0.2902, pruned_loss=0.06385, over 19627.00 frames. ], tot_loss[loss=0.2013, simple_loss=0.2824, pruned_loss=0.06011, over 3835600.97 frames. ], batch size: 61, lr: 2.83e-03, grad_scale: 8.0 2023-04-03 13:46:00,710 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=193403.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:46:04,394 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=193406.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:46:29,042 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 2023-04-03 13:46:36,073 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=193431.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:46:39,103 INFO [train.py:903] (1/4) Epoch 29, batch 2250, loss[loss=0.1898, simple_loss=0.2698, pruned_loss=0.05486, over 19843.00 frames. ], tot_loss[loss=0.2013, simple_loss=0.2824, pruned_loss=0.06011, over 3847623.40 frames. ], batch size: 52, lr: 2.83e-03, grad_scale: 8.0 2023-04-03 13:46:40,737 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.3431, 1.9317, 1.9710, 2.9479, 2.0298, 2.5747, 2.6246, 2.2863], device='cuda:1'), covar=tensor([0.0761, 0.0897, 0.1002, 0.0731, 0.0848, 0.0708, 0.0813, 0.0646], device='cuda:1'), in_proj_covar=tensor([0.0213, 0.0223, 0.0229, 0.0240, 0.0227, 0.0214, 0.0188, 0.0208], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-04-03 13:46:45,248 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=193439.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:47:15,470 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 2023-04-03 13:47:20,667 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=193468.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:47:25,760 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=193472.0, num_to_drop=1, layers_to_drop={1} 2023-04-03 13:47:36,426 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.451e+02 4.884e+02 5.894e+02 7.303e+02 1.661e+03, threshold=1.179e+03, percent-clipped=4.0 2023-04-03 13:47:38,736 INFO [train.py:903] (1/4) Epoch 29, batch 2300, loss[loss=0.1865, simple_loss=0.2739, pruned_loss=0.0496, over 19664.00 frames. ], tot_loss[loss=0.2019, simple_loss=0.2827, pruned_loss=0.06054, over 3847958.61 frames. ], batch size: 53, lr: 2.83e-03, grad_scale: 8.0 2023-04-03 13:47:39,134 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.2761, 2.1594, 1.9872, 1.8780, 1.6406, 1.8462, 0.6940, 1.2098], device='cuda:1'), covar=tensor([0.0650, 0.0715, 0.0563, 0.0961, 0.1227, 0.1080, 0.1468, 0.1243], device='cuda:1'), in_proj_covar=tensor([0.0370, 0.0367, 0.0371, 0.0395, 0.0475, 0.0401, 0.0349, 0.0351], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 13:47:50,425 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=193493.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:47:53,665 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_rvb from training. Duration: 26.205 2023-04-03 13:48:12,804 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.41 vs. limit=2.0 2023-04-03 13:48:21,010 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=193518.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:48:39,106 INFO [train.py:903] (1/4) Epoch 29, batch 2350, loss[loss=0.227, simple_loss=0.3132, pruned_loss=0.07047, over 19641.00 frames. ], tot_loss[loss=0.2015, simple_loss=0.2824, pruned_loss=0.06032, over 3850050.84 frames. ], batch size: 57, lr: 2.83e-03, grad_scale: 8.0 2023-04-03 13:49:16,316 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=193564.0, num_to_drop=1, layers_to_drop={1} 2023-04-03 13:49:20,451 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_rvb from training. Duration: 25.775 2023-04-03 13:49:27,314 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=193573.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:49:36,484 WARNING [train.py:1073] (1/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9_rvb from training. Duration: 25.45 2023-04-03 13:49:37,593 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.068e+02 4.546e+02 5.901e+02 7.828e+02 1.716e+03, threshold=1.180e+03, percent-clipped=11.0 2023-04-03 13:49:40,757 INFO [train.py:903] (1/4) Epoch 29, batch 2400, loss[loss=0.1895, simple_loss=0.2824, pruned_loss=0.04825, over 19676.00 frames. ], tot_loss[loss=0.2016, simple_loss=0.2831, pruned_loss=0.0601, over 3836967.20 frames. ], batch size: 53, lr: 2.83e-03, grad_scale: 8.0 2023-04-03 13:49:47,421 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=193589.0, num_to_drop=1, layers_to_drop={0} 2023-04-03 13:49:52,738 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.1449, 2.2180, 2.4809, 2.7668, 2.1821, 2.6599, 2.4792, 2.3199], device='cuda:1'), covar=tensor([0.4094, 0.3900, 0.1866, 0.2427, 0.3962, 0.2189, 0.4785, 0.3301], device='cuda:1'), in_proj_covar=tensor([0.0950, 0.1031, 0.0753, 0.0962, 0.0927, 0.0866, 0.0870, 0.0817], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-03 13:50:14,632 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6897, 1.2595, 1.3232, 1.5817, 1.1339, 1.4445, 1.2917, 1.5129], device='cuda:1'), covar=tensor([0.1144, 0.1231, 0.1638, 0.0958, 0.1359, 0.0620, 0.1652, 0.0879], device='cuda:1'), in_proj_covar=tensor([0.0279, 0.0362, 0.0322, 0.0259, 0.0309, 0.0259, 0.0324, 0.0264], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-04-03 13:50:40,470 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=193633.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:50:41,249 INFO [train.py:903] (1/4) Epoch 29, batch 2450, loss[loss=0.1926, simple_loss=0.268, pruned_loss=0.05857, over 19733.00 frames. ], tot_loss[loss=0.2017, simple_loss=0.2829, pruned_loss=0.06022, over 3846165.61 frames. ], batch size: 51, lr: 2.83e-03, grad_scale: 8.0 2023-04-03 13:50:54,654 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=193646.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:51:39,133 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.050e+02 4.882e+02 5.957e+02 8.043e+02 1.393e+03, threshold=1.191e+03, percent-clipped=5.0 2023-04-03 13:51:41,319 INFO [train.py:903] (1/4) Epoch 29, batch 2500, loss[loss=0.3255, simple_loss=0.3728, pruned_loss=0.1391, over 13323.00 frames. ], tot_loss[loss=0.2026, simple_loss=0.2836, pruned_loss=0.0608, over 3829593.53 frames. ], batch size: 135, lr: 2.83e-03, grad_scale: 8.0 2023-04-03 13:51:45,922 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=193688.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:51:50,505 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.0694, 1.2422, 1.7740, 1.0515, 2.5111, 3.3766, 3.0869, 3.5742], device='cuda:1'), covar=tensor([0.1723, 0.3978, 0.3429, 0.2839, 0.0708, 0.0201, 0.0230, 0.0290], device='cuda:1'), in_proj_covar=tensor([0.0280, 0.0334, 0.0366, 0.0273, 0.0257, 0.0198, 0.0220, 0.0278], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-03 13:52:40,697 INFO [train.py:903] (1/4) Epoch 29, batch 2550, loss[loss=0.204, simple_loss=0.2869, pruned_loss=0.06052, over 19324.00 frames. ], tot_loss[loss=0.2026, simple_loss=0.2834, pruned_loss=0.06088, over 3822648.27 frames. ], batch size: 66, lr: 2.83e-03, grad_scale: 8.0 2023-04-03 13:52:46,734 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9776, 1.8719, 1.7185, 2.0125, 1.7762, 1.7256, 1.6813, 1.8950], device='cuda:1'), covar=tensor([0.1025, 0.1298, 0.1409, 0.1031, 0.1286, 0.0543, 0.1416, 0.0726], device='cuda:1'), in_proj_covar=tensor([0.0279, 0.0361, 0.0320, 0.0259, 0.0308, 0.0258, 0.0323, 0.0263], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-04-03 13:52:54,644 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.88 vs. limit=5.0 2023-04-03 13:53:09,924 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 2023-04-03 13:53:15,297 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=193761.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:53:30,092 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=193774.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:53:35,068 WARNING [train.py:1073] (1/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 2023-04-03 13:53:40,999 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.098e+02 5.186e+02 6.141e+02 8.140e+02 1.584e+03, threshold=1.228e+03, percent-clipped=10.0 2023-04-03 13:53:41,155 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=193783.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:53:42,120 INFO [train.py:903] (1/4) Epoch 29, batch 2600, loss[loss=0.2172, simple_loss=0.3008, pruned_loss=0.06681, over 18710.00 frames. ], tot_loss[loss=0.2031, simple_loss=0.284, pruned_loss=0.06109, over 3804360.17 frames. ], batch size: 74, lr: 2.83e-03, grad_scale: 8.0 2023-04-03 13:54:02,270 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=193799.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:54:22,059 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=193816.0, num_to_drop=1, layers_to_drop={0} 2023-04-03 13:54:44,186 INFO [train.py:903] (1/4) Epoch 29, batch 2650, loss[loss=0.2079, simple_loss=0.2952, pruned_loss=0.06025, over 19523.00 frames. ], tot_loss[loss=0.2035, simple_loss=0.2844, pruned_loss=0.0613, over 3783645.40 frames. ], batch size: 54, lr: 2.83e-03, grad_scale: 8.0 2023-04-03 13:54:59,844 WARNING [train.py:1073] (1/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9_rvb from training. Duration: 27.25 2023-04-03 13:55:38,465 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.12 vs. limit=2.0 2023-04-03 13:55:43,538 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.739e+02 4.600e+02 5.538e+02 7.187e+02 1.315e+03, threshold=1.108e+03, percent-clipped=1.0 2023-04-03 13:55:44,725 INFO [train.py:903] (1/4) Epoch 29, batch 2700, loss[loss=0.1515, simple_loss=0.2379, pruned_loss=0.03253, over 19739.00 frames. ], tot_loss[loss=0.2018, simple_loss=0.2831, pruned_loss=0.06025, over 3796290.84 frames. ], batch size: 46, lr: 2.83e-03, grad_scale: 8.0 2023-04-03 13:55:55,015 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4828, 1.5524, 1.7902, 1.7252, 2.5487, 2.2642, 2.7788, 1.2299], device='cuda:1'), covar=tensor([0.2630, 0.4541, 0.2895, 0.2053, 0.1594, 0.2256, 0.1410, 0.4782], device='cuda:1'), in_proj_covar=tensor([0.0557, 0.0676, 0.0764, 0.0512, 0.0640, 0.0550, 0.0672, 0.0577], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-03 13:56:00,526 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=193898.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:56:05,882 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=193902.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:56:42,116 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=193931.0, num_to_drop=1, layers_to_drop={1} 2023-04-03 13:56:45,192 INFO [train.py:903] (1/4) Epoch 29, batch 2750, loss[loss=0.2243, simple_loss=0.2977, pruned_loss=0.0754, over 13371.00 frames. ], tot_loss[loss=0.2015, simple_loss=0.2826, pruned_loss=0.06018, over 3798585.75 frames. ], batch size: 135, lr: 2.83e-03, grad_scale: 8.0 2023-04-03 13:56:58,028 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=193944.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:57:27,795 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=193969.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:57:29,114 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4353, 1.5798, 1.8289, 1.7181, 2.4551, 2.1370, 2.6409, 0.9680], device='cuda:1'), covar=tensor([0.2654, 0.4409, 0.2899, 0.2047, 0.1586, 0.2417, 0.1437, 0.5172], device='cuda:1'), in_proj_covar=tensor([0.0556, 0.0674, 0.0762, 0.0511, 0.0638, 0.0549, 0.0671, 0.0576], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-03 13:57:36,659 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=193977.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:57:44,120 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.356e+02 4.680e+02 5.858e+02 7.532e+02 1.982e+03, threshold=1.172e+03, percent-clipped=4.0 2023-04-03 13:57:45,309 INFO [train.py:903] (1/4) Epoch 29, batch 2800, loss[loss=0.1794, simple_loss=0.2728, pruned_loss=0.04302, over 19648.00 frames. ], tot_loss[loss=0.2019, simple_loss=0.2834, pruned_loss=0.06016, over 3813123.99 frames. ], batch size: 55, lr: 2.83e-03, grad_scale: 8.0 2023-04-03 13:58:07,523 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3209, 1.4402, 1.8918, 1.4298, 2.7192, 3.8008, 3.5338, 3.9811], device='cuda:1'), covar=tensor([0.1524, 0.3737, 0.3188, 0.2493, 0.0696, 0.0184, 0.0194, 0.0271], device='cuda:1'), in_proj_covar=tensor([0.0280, 0.0334, 0.0366, 0.0273, 0.0256, 0.0198, 0.0221, 0.0279], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-03 13:58:26,686 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=194017.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:58:48,030 INFO [train.py:903] (1/4) Epoch 29, batch 2850, loss[loss=0.2044, simple_loss=0.2807, pruned_loss=0.06409, over 19595.00 frames. ], tot_loss[loss=0.1995, simple_loss=0.2813, pruned_loss=0.05886, over 3833168.97 frames. ], batch size: 50, lr: 2.83e-03, grad_scale: 8.0 2023-04-03 13:58:58,403 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=194042.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:59:45,132 WARNING [train.py:1073] (1/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 2023-04-03 13:59:47,409 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.156e+02 4.652e+02 5.661e+02 7.187e+02 2.087e+03, threshold=1.132e+03, percent-clipped=4.0 2023-04-03 13:59:48,549 INFO [train.py:903] (1/4) Epoch 29, batch 2900, loss[loss=0.1937, simple_loss=0.2826, pruned_loss=0.05237, over 19660.00 frames. ], tot_loss[loss=0.2003, simple_loss=0.2823, pruned_loss=0.05914, over 3825958.81 frames. ], batch size: 55, lr: 2.83e-03, grad_scale: 8.0 2023-04-03 13:59:55,297 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=194090.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 13:59:57,596 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=194092.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 14:00:47,076 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.1272, 3.0899, 1.8522, 1.9465, 2.7837, 1.7456, 1.6623, 2.3004], device='cuda:1'), covar=tensor([0.1378, 0.0693, 0.1163, 0.0906, 0.0549, 0.1332, 0.0977, 0.0720], device='cuda:1'), in_proj_covar=tensor([0.0304, 0.0320, 0.0344, 0.0274, 0.0253, 0.0347, 0.0292, 0.0277], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 14:00:47,879 INFO [train.py:903] (1/4) Epoch 29, batch 2950, loss[loss=0.1934, simple_loss=0.2837, pruned_loss=0.05156, over 18279.00 frames. ], tot_loss[loss=0.2005, simple_loss=0.2824, pruned_loss=0.05937, over 3821341.18 frames. ], batch size: 83, lr: 2.83e-03, grad_scale: 8.0 2023-04-03 14:01:13,340 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=194154.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 14:01:37,556 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 2023-04-03 14:01:42,871 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=194179.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 14:01:46,908 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.793e+02 4.745e+02 5.839e+02 7.587e+02 1.918e+03, threshold=1.168e+03, percent-clipped=4.0 2023-04-03 14:01:48,092 INFO [train.py:903] (1/4) Epoch 29, batch 3000, loss[loss=0.1894, simple_loss=0.269, pruned_loss=0.05492, over 19703.00 frames. ], tot_loss[loss=0.2004, simple_loss=0.2821, pruned_loss=0.05938, over 3821191.13 frames. ], batch size: 51, lr: 2.83e-03, grad_scale: 8.0 2023-04-03 14:01:48,092 INFO [train.py:928] (1/4) Computing validation loss 2023-04-03 14:02:01,075 INFO [train.py:937] (1/4) Epoch 29, validation: loss=0.1668, simple_loss=0.2661, pruned_loss=0.03375, over 944034.00 frames. 2023-04-03 14:02:01,077 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18821MB 2023-04-03 14:02:02,339 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_rvb from training. Duration: 29.735 2023-04-03 14:02:05,060 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=194187.0, num_to_drop=1, layers_to_drop={1} 2023-04-03 14:02:36,788 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=194212.0, num_to_drop=1, layers_to_drop={1} 2023-04-03 14:03:01,745 INFO [train.py:903] (1/4) Epoch 29, batch 3050, loss[loss=0.1676, simple_loss=0.2428, pruned_loss=0.04625, over 19362.00 frames. ], tot_loss[loss=0.1999, simple_loss=0.2818, pruned_loss=0.05902, over 3819588.42 frames. ], batch size: 47, lr: 2.82e-03, grad_scale: 8.0 2023-04-03 14:03:17,757 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=194246.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 14:04:02,892 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.301e+02 5.546e+02 6.449e+02 7.988e+02 2.570e+03, threshold=1.290e+03, percent-clipped=8.0 2023-04-03 14:04:03,996 INFO [train.py:903] (1/4) Epoch 29, batch 3100, loss[loss=0.1906, simple_loss=0.2809, pruned_loss=0.0501, over 19667.00 frames. ], tot_loss[loss=0.2, simple_loss=0.2818, pruned_loss=0.0591, over 3828090.83 frames. ], batch size: 60, lr: 2.82e-03, grad_scale: 8.0 2023-04-03 14:05:04,146 INFO [train.py:903] (1/4) Epoch 29, batch 3150, loss[loss=0.212, simple_loss=0.2831, pruned_loss=0.07047, over 19756.00 frames. ], tot_loss[loss=0.2003, simple_loss=0.2821, pruned_loss=0.05929, over 3835237.94 frames. ], batch size: 45, lr: 2.82e-03, grad_scale: 8.0 2023-04-03 14:05:06,679 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=194336.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 14:05:19,948 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=194348.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 14:05:27,479 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 2023-04-03 14:05:36,313 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=194361.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 14:05:49,268 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=194371.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 14:05:51,742 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=194373.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 14:05:59,650 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=194380.0, num_to_drop=1, layers_to_drop={0} 2023-04-03 14:06:02,722 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.204e+02 4.902e+02 5.790e+02 7.151e+02 3.080e+03, threshold=1.158e+03, percent-clipped=4.0 2023-04-03 14:06:03,882 INFO [train.py:903] (1/4) Epoch 29, batch 3200, loss[loss=0.1922, simple_loss=0.2671, pruned_loss=0.05862, over 19336.00 frames. ], tot_loss[loss=0.2006, simple_loss=0.2822, pruned_loss=0.05944, over 3829062.13 frames. ], batch size: 47, lr: 2.82e-03, grad_scale: 8.0 2023-04-03 14:07:04,910 INFO [train.py:903] (1/4) Epoch 29, batch 3250, loss[loss=0.1656, simple_loss=0.2447, pruned_loss=0.04321, over 19772.00 frames. ], tot_loss[loss=0.2001, simple_loss=0.2817, pruned_loss=0.05929, over 3823040.42 frames. ], batch size: 47, lr: 2.82e-03, grad_scale: 8.0 2023-04-03 14:07:05,064 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=194434.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 14:07:23,391 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.9948, 3.6647, 2.5259, 3.2707, 0.7321, 3.6468, 3.4750, 3.6014], device='cuda:1'), covar=tensor([0.0815, 0.1128, 0.2002, 0.0965, 0.4111, 0.0769, 0.1096, 0.1288], device='cuda:1'), in_proj_covar=tensor([0.0532, 0.0429, 0.0519, 0.0357, 0.0410, 0.0458, 0.0452, 0.0483], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 14:08:04,290 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.869e+02 4.916e+02 6.594e+02 9.076e+02 2.390e+03, threshold=1.319e+03, percent-clipped=9.0 2023-04-03 14:08:05,462 INFO [train.py:903] (1/4) Epoch 29, batch 3300, loss[loss=0.1803, simple_loss=0.2509, pruned_loss=0.05491, over 19399.00 frames. ], tot_loss[loss=0.2007, simple_loss=0.2824, pruned_loss=0.05954, over 3825223.89 frames. ], batch size: 47, lr: 2.82e-03, grad_scale: 8.0 2023-04-03 14:08:09,431 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 2023-04-03 14:09:07,386 INFO [train.py:903] (1/4) Epoch 29, batch 3350, loss[loss=0.2109, simple_loss=0.2811, pruned_loss=0.07031, over 19743.00 frames. ], tot_loss[loss=0.2007, simple_loss=0.282, pruned_loss=0.05973, over 3822349.70 frames. ], batch size: 47, lr: 2.82e-03, grad_scale: 8.0 2023-04-03 14:09:24,594 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=194549.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 14:10:05,401 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6990, 1.7093, 1.6021, 1.4068, 1.3607, 1.4077, 0.2900, 0.6949], device='cuda:1'), covar=tensor([0.0776, 0.0702, 0.0477, 0.0771, 0.1313, 0.0877, 0.1465, 0.1266], device='cuda:1'), in_proj_covar=tensor([0.0371, 0.0368, 0.0373, 0.0396, 0.0477, 0.0402, 0.0350, 0.0352], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 14:10:06,126 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.313e+02 5.166e+02 6.518e+02 8.363e+02 1.379e+03, threshold=1.304e+03, percent-clipped=1.0 2023-04-03 14:10:07,289 INFO [train.py:903] (1/4) Epoch 29, batch 3400, loss[loss=0.2259, simple_loss=0.3049, pruned_loss=0.07346, over 17342.00 frames. ], tot_loss[loss=0.2006, simple_loss=0.2816, pruned_loss=0.05985, over 3813868.46 frames. ], batch size: 101, lr: 2.82e-03, grad_scale: 8.0 2023-04-03 14:10:48,233 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=194617.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 14:11:06,749 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 2023-04-03 14:11:07,040 INFO [train.py:903] (1/4) Epoch 29, batch 3450, loss[loss=0.196, simple_loss=0.2722, pruned_loss=0.05993, over 19726.00 frames. ], tot_loss[loss=0.2006, simple_loss=0.2818, pruned_loss=0.0597, over 3823730.21 frames. ], batch size: 51, lr: 2.82e-03, grad_scale: 8.0 2023-04-03 14:11:09,269 WARNING [train.py:1073] (1/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9_rvb from training. Duration: 25.2444375 2023-04-03 14:11:17,003 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=194642.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 14:11:29,965 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=194652.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 14:12:01,641 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=194680.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 14:12:05,596 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.890e+02 4.627e+02 5.712e+02 7.363e+02 1.612e+03, threshold=1.142e+03, percent-clipped=3.0 2023-04-03 14:12:06,670 INFO [train.py:903] (1/4) Epoch 29, batch 3500, loss[loss=0.199, simple_loss=0.2733, pruned_loss=0.0624, over 19470.00 frames. ], tot_loss[loss=0.2016, simple_loss=0.2827, pruned_loss=0.06027, over 3824303.89 frames. ], batch size: 49, lr: 2.82e-03, grad_scale: 8.0 2023-04-03 14:12:43,647 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=194715.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 14:12:54,403 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=194724.0, num_to_drop=1, layers_to_drop={0} 2023-04-03 14:12:55,741 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5192, 1.5715, 1.8519, 1.7879, 2.7479, 2.3619, 2.9589, 1.2208], device='cuda:1'), covar=tensor([0.2612, 0.4543, 0.3019, 0.2015, 0.1448, 0.2188, 0.1366, 0.4835], device='cuda:1'), in_proj_covar=tensor([0.0556, 0.0672, 0.0760, 0.0510, 0.0635, 0.0547, 0.0670, 0.0574], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 14:13:06,874 INFO [train.py:903] (1/4) Epoch 29, batch 3550, loss[loss=0.2733, simple_loss=0.3433, pruned_loss=0.1017, over 19754.00 frames. ], tot_loss[loss=0.2023, simple_loss=0.2834, pruned_loss=0.06057, over 3822475.30 frames. ], batch size: 63, lr: 2.82e-03, grad_scale: 8.0 2023-04-03 14:14:06,359 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.972e+02 4.806e+02 5.826e+02 6.989e+02 1.690e+03, threshold=1.165e+03, percent-clipped=2.0 2023-04-03 14:14:07,571 INFO [train.py:903] (1/4) Epoch 29, batch 3600, loss[loss=0.1909, simple_loss=0.276, pruned_loss=0.05291, over 19680.00 frames. ], tot_loss[loss=0.2015, simple_loss=0.2829, pruned_loss=0.0601, over 3822471.27 frames. ], batch size: 59, lr: 2.82e-03, grad_scale: 8.0 2023-04-03 14:14:20,128 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=194795.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 14:14:33,714 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=194805.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 14:14:37,434 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.56 vs. limit=2.0 2023-04-03 14:15:02,712 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=194830.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 14:15:02,752 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=194830.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 14:15:06,686 INFO [train.py:903] (1/4) Epoch 29, batch 3650, loss[loss=0.2211, simple_loss=0.3037, pruned_loss=0.06919, over 19669.00 frames. ], tot_loss[loss=0.2013, simple_loss=0.2824, pruned_loss=0.06012, over 3803328.41 frames. ], batch size: 60, lr: 2.82e-03, grad_scale: 8.0 2023-04-03 14:15:12,771 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=194839.0, num_to_drop=1, layers_to_drop={1} 2023-04-03 14:15:45,377 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=194865.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 14:16:05,571 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.663e+02 5.135e+02 6.261e+02 7.520e+02 2.080e+03, threshold=1.252e+03, percent-clipped=4.0 2023-04-03 14:16:06,488 INFO [train.py:903] (1/4) Epoch 29, batch 3700, loss[loss=0.275, simple_loss=0.3438, pruned_loss=0.1031, over 19756.00 frames. ], tot_loss[loss=0.2023, simple_loss=0.2835, pruned_loss=0.06051, over 3805676.93 frames. ], batch size: 63, lr: 2.82e-03, grad_scale: 8.0 2023-04-03 14:17:08,165 INFO [train.py:903] (1/4) Epoch 29, batch 3750, loss[loss=0.1772, simple_loss=0.2683, pruned_loss=0.04303, over 19613.00 frames. ], tot_loss[loss=0.2025, simple_loss=0.2836, pruned_loss=0.06071, over 3811819.62 frames. ], batch size: 57, lr: 2.82e-03, grad_scale: 8.0 2023-04-03 14:18:06,324 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.993e+02 4.741e+02 6.225e+02 7.602e+02 2.236e+03, threshold=1.245e+03, percent-clipped=10.0 2023-04-03 14:18:07,440 INFO [train.py:903] (1/4) Epoch 29, batch 3800, loss[loss=0.2062, simple_loss=0.2919, pruned_loss=0.06025, over 19339.00 frames. ], tot_loss[loss=0.203, simple_loss=0.2843, pruned_loss=0.06088, over 3802667.90 frames. ], batch size: 66, lr: 2.82e-03, grad_scale: 8.0 2023-04-03 14:18:21,207 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=194996.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 14:18:34,663 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9_rvb from training. Duration: 29.1166875 2023-04-03 14:19:01,231 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=195029.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 14:19:06,512 INFO [train.py:903] (1/4) Epoch 29, batch 3850, loss[loss=0.2676, simple_loss=0.3411, pruned_loss=0.097, over 19523.00 frames. ], tot_loss[loss=0.2027, simple_loss=0.2841, pruned_loss=0.0607, over 3813242.34 frames. ], batch size: 64, lr: 2.82e-03, grad_scale: 4.0 2023-04-03 14:19:27,245 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=195051.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 14:19:41,483 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=195062.0, num_to_drop=1, layers_to_drop={0} 2023-04-03 14:19:43,970 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.3041, 2.3130, 2.6312, 2.9652, 2.3558, 2.8688, 2.5825, 2.3771], device='cuda:1'), covar=tensor([0.4297, 0.4162, 0.1830, 0.2655, 0.4364, 0.2210, 0.4802, 0.3285], device='cuda:1'), in_proj_covar=tensor([0.0951, 0.1030, 0.0754, 0.0964, 0.0928, 0.0866, 0.0869, 0.0818], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-03 14:19:56,971 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=195076.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 14:20:06,400 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.076e+02 4.743e+02 5.822e+02 7.676e+02 1.767e+03, threshold=1.164e+03, percent-clipped=5.0 2023-04-03 14:20:06,418 INFO [train.py:903] (1/4) Epoch 29, batch 3900, loss[loss=0.2285, simple_loss=0.3058, pruned_loss=0.07562, over 19670.00 frames. ], tot_loss[loss=0.2031, simple_loss=0.2845, pruned_loss=0.06085, over 3820802.94 frames. ], batch size: 58, lr: 2.82e-03, grad_scale: 4.0 2023-04-03 14:20:08,952 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=195086.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 14:20:13,189 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=195089.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 14:20:21,686 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=195095.0, num_to_drop=1, layers_to_drop={1} 2023-04-03 14:20:23,804 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.7929, 4.9548, 5.5827, 5.5939, 2.0918, 5.2661, 4.4864, 5.3185], device='cuda:1'), covar=tensor([0.1819, 0.0994, 0.0570, 0.0686, 0.6126, 0.0924, 0.0685, 0.1139], device='cuda:1'), in_proj_covar=tensor([0.0826, 0.0794, 0.1004, 0.0880, 0.0871, 0.0769, 0.0594, 0.0935], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-03 14:20:39,895 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=195111.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 14:20:39,935 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=195111.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 14:20:49,223 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.73 vs. limit=5.0 2023-04-03 14:20:50,182 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=195120.0, num_to_drop=1, layers_to_drop={0} 2023-04-03 14:21:07,592 INFO [train.py:903] (1/4) Epoch 29, batch 3950, loss[loss=0.2051, simple_loss=0.2876, pruned_loss=0.06131, over 19785.00 frames. ], tot_loss[loss=0.2015, simple_loss=0.2829, pruned_loss=0.06004, over 3821583.96 frames. ], batch size: 56, lr: 2.82e-03, grad_scale: 4.0 2023-04-03 14:21:12,684 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 2023-04-03 14:21:43,433 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2585, 1.5215, 1.9499, 1.6123, 3.0567, 4.6268, 4.5471, 5.0534], device='cuda:1'), covar=tensor([0.1706, 0.3826, 0.3492, 0.2461, 0.0656, 0.0195, 0.0184, 0.0217], device='cuda:1'), in_proj_covar=tensor([0.0281, 0.0334, 0.0367, 0.0273, 0.0258, 0.0198, 0.0221, 0.0278], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-03 14:22:08,284 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.119e+02 4.632e+02 5.828e+02 7.591e+02 1.503e+03, threshold=1.166e+03, percent-clipped=4.0 2023-04-03 14:22:08,302 INFO [train.py:903] (1/4) Epoch 29, batch 4000, loss[loss=0.1948, simple_loss=0.2843, pruned_loss=0.05264, over 19482.00 frames. ], tot_loss[loss=0.2005, simple_loss=0.2821, pruned_loss=0.05939, over 3817884.49 frames. ], batch size: 64, lr: 2.82e-03, grad_scale: 8.0 2023-04-03 14:22:38,703 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=195209.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 14:22:52,711 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1_rvb from training. Duration: 0.7545625 2023-04-03 14:23:07,863 INFO [train.py:903] (1/4) Epoch 29, batch 4050, loss[loss=0.2037, simple_loss=0.2886, pruned_loss=0.05939, over 19348.00 frames. ], tot_loss[loss=0.2, simple_loss=0.2818, pruned_loss=0.05914, over 3819770.98 frames. ], batch size: 70, lr: 2.82e-03, grad_scale: 8.0 2023-04-03 14:23:19,843 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.13 vs. limit=2.0 2023-04-03 14:23:24,411 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.31 vs. limit=2.0 2023-04-03 14:24:08,395 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.197e+02 5.118e+02 6.147e+02 8.377e+02 1.783e+03, threshold=1.229e+03, percent-clipped=5.0 2023-04-03 14:24:08,413 INFO [train.py:903] (1/4) Epoch 29, batch 4100, loss[loss=0.2131, simple_loss=0.2929, pruned_loss=0.06662, over 19487.00 frames. ], tot_loss[loss=0.2, simple_loss=0.2819, pruned_loss=0.05904, over 3807374.65 frames. ], batch size: 64, lr: 2.82e-03, grad_scale: 8.0 2023-04-03 14:24:41,888 WARNING [train.py:1073] (1/4) Exclude cut with ID 497-129325-0061-9566_sp1.1_rvb from training. Duration: 0.97725 2023-04-03 14:24:56,223 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=195324.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 14:25:08,838 INFO [train.py:903] (1/4) Epoch 29, batch 4150, loss[loss=0.1995, simple_loss=0.2776, pruned_loss=0.06066, over 19682.00 frames. ], tot_loss[loss=0.1997, simple_loss=0.2817, pruned_loss=0.05881, over 3822570.60 frames. ], batch size: 60, lr: 2.82e-03, grad_scale: 8.0 2023-04-03 14:25:49,116 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=195367.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 14:25:56,297 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=195373.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 14:26:10,052 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.652e+02 4.925e+02 6.547e+02 8.865e+02 3.200e+03, threshold=1.309e+03, percent-clipped=9.0 2023-04-03 14:26:10,070 INFO [train.py:903] (1/4) Epoch 29, batch 4200, loss[loss=0.1754, simple_loss=0.2544, pruned_loss=0.04821, over 18751.00 frames. ], tot_loss[loss=0.1987, simple_loss=0.2808, pruned_loss=0.05826, over 3828601.01 frames. ], batch size: 41, lr: 2.82e-03, grad_scale: 8.0 2023-04-03 14:26:13,440 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 2023-04-03 14:26:19,416 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=195392.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 14:26:35,640 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=195406.0, num_to_drop=1, layers_to_drop={0} 2023-04-03 14:26:40,122 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=195410.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 14:27:00,193 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.48 vs. limit=2.0 2023-04-03 14:27:07,660 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=195433.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 14:27:08,634 INFO [train.py:903] (1/4) Epoch 29, batch 4250, loss[loss=0.2063, simple_loss=0.2717, pruned_loss=0.07047, over 19753.00 frames. ], tot_loss[loss=0.1991, simple_loss=0.2808, pruned_loss=0.05876, over 3844092.85 frames. ], batch size: 47, lr: 2.82e-03, grad_scale: 8.0 2023-04-03 14:27:22,556 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9_rvb from training. Duration: 29.816625 2023-04-03 14:27:32,649 WARNING [train.py:1073] (1/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 2023-04-03 14:28:08,329 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.783e+02 4.660e+02 5.852e+02 7.750e+02 1.648e+03, threshold=1.170e+03, percent-clipped=3.0 2023-04-03 14:28:08,347 INFO [train.py:903] (1/4) Epoch 29, batch 4300, loss[loss=0.1669, simple_loss=0.2489, pruned_loss=0.04243, over 19746.00 frames. ], tot_loss[loss=0.1988, simple_loss=0.2806, pruned_loss=0.05852, over 3843657.94 frames. ], batch size: 45, lr: 2.82e-03, grad_scale: 8.0 2023-04-03 14:28:14,115 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=195488.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 14:28:54,535 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=195521.0, num_to_drop=1, layers_to_drop={0} 2023-04-03 14:29:01,485 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 2023-04-03 14:29:09,900 INFO [train.py:903] (1/4) Epoch 29, batch 4350, loss[loss=0.1924, simple_loss=0.2704, pruned_loss=0.0572, over 19611.00 frames. ], tot_loss[loss=0.1997, simple_loss=0.2811, pruned_loss=0.05913, over 3830473.00 frames. ], batch size: 50, lr: 2.82e-03, grad_scale: 8.0 2023-04-03 14:29:28,095 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=195548.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 14:29:58,861 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4689, 1.5612, 2.2129, 1.8419, 3.2423, 4.8956, 4.6739, 5.1844], device='cuda:1'), covar=tensor([0.1610, 0.3835, 0.3230, 0.2239, 0.0574, 0.0178, 0.0166, 0.0196], device='cuda:1'), in_proj_covar=tensor([0.0280, 0.0334, 0.0366, 0.0272, 0.0257, 0.0198, 0.0220, 0.0279], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-03 14:30:05,583 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=195580.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 14:30:10,384 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.562e+02 4.994e+02 6.103e+02 7.630e+02 1.747e+03, threshold=1.221e+03, percent-clipped=1.0 2023-04-03 14:30:10,402 INFO [train.py:903] (1/4) Epoch 29, batch 4400, loss[loss=0.229, simple_loss=0.3124, pruned_loss=0.07278, over 19559.00 frames. ], tot_loss[loss=0.201, simple_loss=0.2826, pruned_loss=0.05971, over 3837914.88 frames. ], batch size: 61, lr: 2.82e-03, grad_scale: 8.0 2023-04-03 14:30:29,901 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_rvb from training. Duration: 25.285 2023-04-03 14:30:35,591 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=195605.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 14:30:38,555 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 2023-04-03 14:30:40,905 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.6005, 1.1779, 1.3967, 1.2995, 2.2677, 1.1848, 2.1830, 2.4993], device='cuda:1'), covar=tensor([0.0673, 0.2835, 0.2851, 0.1698, 0.0813, 0.1986, 0.1006, 0.0441], device='cuda:1'), in_proj_covar=tensor([0.0427, 0.0383, 0.0403, 0.0356, 0.0386, 0.0360, 0.0400, 0.0420], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 14:30:45,375 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.13 vs. limit=2.0 2023-04-03 14:31:09,645 INFO [train.py:903] (1/4) Epoch 29, batch 4450, loss[loss=0.1789, simple_loss=0.2559, pruned_loss=0.05101, over 18247.00 frames. ], tot_loss[loss=0.202, simple_loss=0.2832, pruned_loss=0.06044, over 3818080.34 frames. ], batch size: 40, lr: 2.81e-03, grad_scale: 8.0 2023-04-03 14:32:08,771 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.510e+02 5.083e+02 5.988e+02 7.824e+02 1.664e+03, threshold=1.198e+03, percent-clipped=3.0 2023-04-03 14:32:08,790 INFO [train.py:903] (1/4) Epoch 29, batch 4500, loss[loss=0.1841, simple_loss=0.2627, pruned_loss=0.05273, over 19584.00 frames. ], tot_loss[loss=0.2023, simple_loss=0.2836, pruned_loss=0.06052, over 3822020.92 frames. ], batch size: 52, lr: 2.81e-03, grad_scale: 8.0 2023-04-03 14:33:10,063 INFO [train.py:903] (1/4) Epoch 29, batch 4550, loss[loss=0.2414, simple_loss=0.3206, pruned_loss=0.08108, over 19728.00 frames. ], tot_loss[loss=0.2033, simple_loss=0.2844, pruned_loss=0.06115, over 3808279.78 frames. ], batch size: 63, lr: 2.81e-03, grad_scale: 8.0 2023-04-03 14:33:15,627 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9_rvb from training. Duration: 28.72225 2023-04-03 14:33:21,588 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=195744.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 14:33:33,983 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=195754.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 14:33:39,337 WARNING [train.py:1073] (1/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 2023-04-03 14:33:48,577 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.7845, 4.2758, 4.4733, 4.4971, 1.6595, 4.2066, 3.7130, 4.2160], device='cuda:1'), covar=tensor([0.1668, 0.0892, 0.0621, 0.0682, 0.6102, 0.0963, 0.0684, 0.1129], device='cuda:1'), in_proj_covar=tensor([0.0826, 0.0796, 0.1005, 0.0880, 0.0872, 0.0771, 0.0593, 0.0937], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-03 14:33:51,771 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=195769.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 14:34:01,454 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=195777.0, num_to_drop=1, layers_to_drop={0} 2023-04-03 14:34:08,709 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.333e+02 5.047e+02 5.905e+02 8.200e+02 1.793e+03, threshold=1.181e+03, percent-clipped=7.0 2023-04-03 14:34:08,728 INFO [train.py:903] (1/4) Epoch 29, batch 4600, loss[loss=0.2004, simple_loss=0.2981, pruned_loss=0.0513, over 19669.00 frames. ], tot_loss[loss=0.2026, simple_loss=0.2839, pruned_loss=0.06062, over 3814381.82 frames. ], batch size: 58, lr: 2.81e-03, grad_scale: 8.0 2023-04-03 14:34:31,273 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=195802.0, num_to_drop=1, layers_to_drop={0} 2023-04-03 14:34:33,629 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=195804.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 14:35:05,198 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=195829.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 14:35:10,613 INFO [train.py:903] (1/4) Epoch 29, batch 4650, loss[loss=0.162, simple_loss=0.2456, pruned_loss=0.03919, over 19354.00 frames. ], tot_loss[loss=0.2025, simple_loss=0.2837, pruned_loss=0.06066, over 3811187.21 frames. ], batch size: 48, lr: 2.81e-03, grad_scale: 8.0 2023-04-03 14:35:22,166 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_rvb from training. Duration: 0.92 2023-04-03 14:35:33,357 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_rvb from training. Duration: 0.83 2023-04-03 14:35:54,034 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=195869.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 14:36:10,278 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.073e+02 5.115e+02 6.253e+02 8.443e+02 2.371e+03, threshold=1.251e+03, percent-clipped=7.0 2023-04-03 14:36:10,296 INFO [train.py:903] (1/4) Epoch 29, batch 4700, loss[loss=0.2372, simple_loss=0.3078, pruned_loss=0.08333, over 13123.00 frames. ], tot_loss[loss=0.2026, simple_loss=0.2838, pruned_loss=0.06069, over 3809743.24 frames. ], batch size: 135, lr: 2.81e-03, grad_scale: 8.0 2023-04-03 14:36:29,173 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9_rvb from training. Duration: 0.92225 2023-04-03 14:36:29,929 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.40 vs. limit=2.0 2023-04-03 14:37:11,407 INFO [train.py:903] (1/4) Epoch 29, batch 4750, loss[loss=0.2449, simple_loss=0.3185, pruned_loss=0.08566, over 18184.00 frames. ], tot_loss[loss=0.2021, simple_loss=0.2835, pruned_loss=0.06036, over 3818001.12 frames. ], batch size: 83, lr: 2.81e-03, grad_scale: 8.0 2023-04-03 14:38:11,675 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.651e+02 4.980e+02 6.350e+02 8.555e+02 2.103e+03, threshold=1.270e+03, percent-clipped=3.0 2023-04-03 14:38:11,694 INFO [train.py:903] (1/4) Epoch 29, batch 4800, loss[loss=0.1702, simple_loss=0.2584, pruned_loss=0.041, over 19864.00 frames. ], tot_loss[loss=0.202, simple_loss=0.2837, pruned_loss=0.06016, over 3812993.58 frames. ], batch size: 52, lr: 2.81e-03, grad_scale: 8.0 2023-04-03 14:38:28,761 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 2023-04-03 14:39:14,512 INFO [train.py:903] (1/4) Epoch 29, batch 4850, loss[loss=0.1691, simple_loss=0.2517, pruned_loss=0.04324, over 19378.00 frames. ], tot_loss[loss=0.2013, simple_loss=0.2831, pruned_loss=0.05969, over 3824312.12 frames. ], batch size: 47, lr: 2.81e-03, grad_scale: 8.0 2023-04-03 14:39:31,311 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0 2023-04-03 14:39:37,082 WARNING [train.py:1073] (1/4) Exclude cut with ID 774-127930-0014-48411_sp1.1_rvb from training. Duration: 0.95 2023-04-03 14:39:55,388 WARNING [train.py:1073] (1/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1_rvb from training. Duration: 0.9409375 2023-04-03 14:40:01,747 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 2023-04-03 14:40:02,668 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 2023-04-03 14:40:11,417 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1_rvb from training. Duration: 25.3818125 2023-04-03 14:40:13,827 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.943e+02 4.641e+02 5.767e+02 7.663e+02 1.512e+03, threshold=1.153e+03, percent-clipped=2.0 2023-04-03 14:40:13,845 INFO [train.py:903] (1/4) Epoch 29, batch 4900, loss[loss=0.1909, simple_loss=0.2618, pruned_loss=0.06001, over 15283.00 frames. ], tot_loss[loss=0.2016, simple_loss=0.2833, pruned_loss=0.05991, over 3821335.99 frames. ], batch size: 33, lr: 2.81e-03, grad_scale: 8.0 2023-04-03 14:40:32,811 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 2023-04-03 14:40:44,459 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5304, 2.2424, 1.7044, 1.5353, 2.0300, 1.3411, 1.4523, 1.9347], device='cuda:1'), covar=tensor([0.1130, 0.0820, 0.1159, 0.0951, 0.0672, 0.1379, 0.0791, 0.0540], device='cuda:1'), in_proj_covar=tensor([0.0305, 0.0321, 0.0344, 0.0274, 0.0255, 0.0348, 0.0292, 0.0276], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 14:41:03,530 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=196125.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 14:41:14,060 INFO [train.py:903] (1/4) Epoch 29, batch 4950, loss[loss=0.243, simple_loss=0.3187, pruned_loss=0.08363, over 19669.00 frames. ], tot_loss[loss=0.2017, simple_loss=0.2833, pruned_loss=0.06007, over 3816914.38 frames. ], batch size: 58, lr: 2.81e-03, grad_scale: 8.0 2023-04-03 14:41:14,800 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.08 vs. limit=2.0 2023-04-03 14:41:16,697 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.5450, 1.5546, 1.8090, 1.6972, 3.2251, 1.3434, 2.5667, 3.5549], device='cuda:1'), covar=tensor([0.0492, 0.2715, 0.2609, 0.1870, 0.0597, 0.2420, 0.1374, 0.0268], device='cuda:1'), in_proj_covar=tensor([0.0427, 0.0384, 0.0404, 0.0358, 0.0387, 0.0361, 0.0402, 0.0422], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 14:41:31,268 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 2023-04-03 14:41:33,828 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=196150.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 14:41:54,193 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 2023-04-03 14:41:58,582 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.8843, 1.4473, 1.7019, 1.6826, 3.4788, 1.3018, 2.4309, 3.9841], device='cuda:1'), covar=tensor([0.0462, 0.2881, 0.2717, 0.1888, 0.0665, 0.2492, 0.1414, 0.0182], device='cuda:1'), in_proj_covar=tensor([0.0428, 0.0385, 0.0404, 0.0358, 0.0387, 0.0362, 0.0403, 0.0423], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 14:42:14,317 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.427e+02 4.864e+02 6.197e+02 7.725e+02 1.739e+03, threshold=1.239e+03, percent-clipped=10.0 2023-04-03 14:42:14,335 INFO [train.py:903] (1/4) Epoch 29, batch 5000, loss[loss=0.2113, simple_loss=0.3016, pruned_loss=0.06047, over 18331.00 frames. ], tot_loss[loss=0.2013, simple_loss=0.2828, pruned_loss=0.05994, over 3829666.99 frames. ], batch size: 84, lr: 2.81e-03, grad_scale: 8.0 2023-04-03 14:42:14,674 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3483, 1.3342, 1.7827, 1.4138, 2.8728, 3.6788, 3.3834, 3.8287], device='cuda:1'), covar=tensor([0.1682, 0.4061, 0.3653, 0.2768, 0.0658, 0.0229, 0.0228, 0.0313], device='cuda:1'), in_proj_covar=tensor([0.0281, 0.0335, 0.0367, 0.0274, 0.0258, 0.0199, 0.0221, 0.0279], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-03 14:42:16,057 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.37 vs. limit=5.0 2023-04-03 14:42:23,515 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_rvb from training. Duration: 27.14 2023-04-03 14:42:35,047 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 2023-04-03 14:42:54,571 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.84 vs. limit=5.0 2023-04-03 14:43:14,837 INFO [train.py:903] (1/4) Epoch 29, batch 5050, loss[loss=0.1739, simple_loss=0.2505, pruned_loss=0.04865, over 19062.00 frames. ], tot_loss[loss=0.2002, simple_loss=0.2818, pruned_loss=0.05923, over 3825744.29 frames. ], batch size: 42, lr: 2.81e-03, grad_scale: 8.0 2023-04-03 14:43:49,699 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9_rvb from training. Duration: 26.62775 2023-04-03 14:44:15,610 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.659e+02 4.774e+02 5.782e+02 7.176e+02 1.455e+03, threshold=1.156e+03, percent-clipped=5.0 2023-04-03 14:44:15,628 INFO [train.py:903] (1/4) Epoch 29, batch 5100, loss[loss=0.2549, simple_loss=0.3238, pruned_loss=0.09305, over 12476.00 frames. ], tot_loss[loss=0.201, simple_loss=0.2826, pruned_loss=0.05976, over 3805282.61 frames. ], batch size: 136, lr: 2.81e-03, grad_scale: 8.0 2023-04-03 14:44:24,662 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 2023-04-03 14:44:27,971 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_rvb from training. Duration: 25.035 2023-04-03 14:44:32,505 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_rvb from training. Duration: 27.92 2023-04-03 14:45:16,433 INFO [train.py:903] (1/4) Epoch 29, batch 5150, loss[loss=0.1836, simple_loss=0.2618, pruned_loss=0.05266, over 19750.00 frames. ], tot_loss[loss=0.2007, simple_loss=0.2821, pruned_loss=0.05958, over 3805523.48 frames. ], batch size: 45, lr: 2.81e-03, grad_scale: 4.0 2023-04-03 14:45:17,953 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.2498, 2.8758, 2.1877, 2.2401, 2.1782, 2.5203, 1.0867, 2.0444], device='cuda:1'), covar=tensor([0.0710, 0.0671, 0.0784, 0.1331, 0.1105, 0.1239, 0.1526, 0.1206], device='cuda:1'), in_proj_covar=tensor([0.0366, 0.0363, 0.0369, 0.0393, 0.0473, 0.0398, 0.0347, 0.0349], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 14:45:23,032 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=196339.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 14:45:26,216 WARNING [train.py:1073] (1/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 2023-04-03 14:45:59,984 WARNING [train.py:1073] (1/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1_rvb from training. Duration: 0.9681875 2023-04-03 14:46:10,215 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=196378.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 14:46:15,862 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.3892, 2.4327, 2.6484, 3.1401, 2.4468, 3.0620, 2.6041, 2.4433], device='cuda:1'), covar=tensor([0.4430, 0.3931, 0.1971, 0.2630, 0.4500, 0.2230, 0.5377, 0.3502], device='cuda:1'), in_proj_covar=tensor([0.0945, 0.1023, 0.0748, 0.0958, 0.0923, 0.0862, 0.0863, 0.0812], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-03 14:46:17,530 INFO [train.py:903] (1/4) Epoch 29, batch 5200, loss[loss=0.2021, simple_loss=0.2846, pruned_loss=0.05984, over 19402.00 frames. ], tot_loss[loss=0.2014, simple_loss=0.2829, pruned_loss=0.05992, over 3799622.89 frames. ], batch size: 48, lr: 2.81e-03, grad_scale: 8.0 2023-04-03 14:46:18,467 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.574e+02 5.187e+02 6.337e+02 8.295e+02 1.863e+03, threshold=1.267e+03, percent-clipped=5.0 2023-04-03 14:46:28,775 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 2023-04-03 14:47:12,666 WARNING [train.py:1073] (1/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 2023-04-03 14:47:19,216 INFO [train.py:903] (1/4) Epoch 29, batch 5250, loss[loss=0.187, simple_loss=0.2691, pruned_loss=0.05251, over 19757.00 frames. ], tot_loss[loss=0.2014, simple_loss=0.283, pruned_loss=0.05988, over 3793087.65 frames. ], batch size: 54, lr: 2.81e-03, grad_scale: 8.0 2023-04-03 14:47:43,271 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.8076, 1.9083, 2.2180, 2.2954, 1.7973, 2.2593, 2.1444, 1.9965], device='cuda:1'), covar=tensor([0.4384, 0.3971, 0.2009, 0.2465, 0.4167, 0.2320, 0.5242, 0.3585], device='cuda:1'), in_proj_covar=tensor([0.0949, 0.1026, 0.0751, 0.0961, 0.0926, 0.0864, 0.0866, 0.0815], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-03 14:48:07,755 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.7968, 1.9077, 2.1942, 2.2645, 1.7712, 2.2344, 2.1582, 1.9747], device='cuda:1'), covar=tensor([0.4136, 0.3722, 0.1933, 0.2482, 0.3952, 0.2216, 0.4906, 0.3463], device='cuda:1'), in_proj_covar=tensor([0.0948, 0.1025, 0.0750, 0.0960, 0.0925, 0.0864, 0.0865, 0.0814], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-03 14:48:20,444 INFO [train.py:903] (1/4) Epoch 29, batch 5300, loss[loss=0.2106, simple_loss=0.2926, pruned_loss=0.06435, over 19580.00 frames. ], tot_loss[loss=0.2006, simple_loss=0.2824, pruned_loss=0.05942, over 3801135.80 frames. ], batch size: 52, lr: 2.81e-03, grad_scale: 8.0 2023-04-03 14:48:21,521 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.666e+02 4.792e+02 5.709e+02 7.130e+02 1.478e+03, threshold=1.142e+03, percent-clipped=2.0 2023-04-03 14:48:35,845 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 2023-04-03 14:49:20,274 INFO [train.py:903] (1/4) Epoch 29, batch 5350, loss[loss=0.1707, simple_loss=0.2586, pruned_loss=0.04139, over 19535.00 frames. ], tot_loss[loss=0.2008, simple_loss=0.2824, pruned_loss=0.05956, over 3794906.95 frames. ], batch size: 54, lr: 2.81e-03, grad_scale: 8.0 2023-04-03 14:49:51,985 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9_rvb from training. Duration: 26.438875 2023-04-03 14:50:09,932 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.3807, 3.1022, 2.2288, 2.8041, 0.7625, 3.1118, 2.9972, 3.0539], device='cuda:1'), covar=tensor([0.1117, 0.1354, 0.2122, 0.1160, 0.3877, 0.0965, 0.1137, 0.1594], device='cuda:1'), in_proj_covar=tensor([0.0534, 0.0432, 0.0518, 0.0357, 0.0409, 0.0457, 0.0452, 0.0487], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 14:50:21,630 INFO [train.py:903] (1/4) Epoch 29, batch 5400, loss[loss=0.1836, simple_loss=0.2551, pruned_loss=0.05601, over 19783.00 frames. ], tot_loss[loss=0.2015, simple_loss=0.283, pruned_loss=0.06001, over 3801795.90 frames. ], batch size: 47, lr: 2.81e-03, grad_scale: 8.0 2023-04-03 14:50:22,761 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.031e+02 4.834e+02 5.946e+02 7.840e+02 1.782e+03, threshold=1.189e+03, percent-clipped=5.0 2023-04-03 14:50:36,247 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=196596.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 14:50:59,748 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3279, 2.0523, 1.6802, 1.4256, 1.8207, 1.3453, 1.3219, 1.8490], device='cuda:1'), covar=tensor([0.0973, 0.0873, 0.1064, 0.0927, 0.0657, 0.1370, 0.0731, 0.0476], device='cuda:1'), in_proj_covar=tensor([0.0307, 0.0324, 0.0346, 0.0277, 0.0255, 0.0350, 0.0294, 0.0278], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 14:51:22,494 INFO [train.py:903] (1/4) Epoch 29, batch 5450, loss[loss=0.2226, simple_loss=0.3068, pruned_loss=0.0692, over 19583.00 frames. ], tot_loss[loss=0.2023, simple_loss=0.2839, pruned_loss=0.06039, over 3802163.99 frames. ], batch size: 61, lr: 2.81e-03, grad_scale: 8.0 2023-04-03 14:51:34,866 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.5689, 4.1533, 4.3054, 4.2944, 1.7243, 4.0526, 3.5194, 4.0579], device='cuda:1'), covar=tensor([0.1813, 0.0846, 0.0666, 0.0805, 0.6195, 0.0996, 0.0753, 0.1135], device='cuda:1'), in_proj_covar=tensor([0.0829, 0.0802, 0.1013, 0.0886, 0.0879, 0.0774, 0.0596, 0.0938], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-03 14:51:37,064 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.1893, 2.8531, 2.3704, 2.4379, 2.3038, 2.5628, 1.1953, 2.1242], device='cuda:1'), covar=tensor([0.0748, 0.0714, 0.0725, 0.1262, 0.1092, 0.1300, 0.1535, 0.1164], device='cuda:1'), in_proj_covar=tensor([0.0368, 0.0365, 0.0370, 0.0394, 0.0474, 0.0400, 0.0348, 0.0350], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 14:51:41,431 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.3088, 3.0379, 2.2375, 2.7376, 0.9729, 3.0214, 2.8768, 2.9444], device='cuda:1'), covar=tensor([0.1237, 0.1396, 0.2096, 0.1134, 0.3663, 0.0977, 0.1229, 0.1788], device='cuda:1'), in_proj_covar=tensor([0.0537, 0.0435, 0.0522, 0.0360, 0.0413, 0.0460, 0.0456, 0.0491], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 14:52:21,546 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=196683.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 14:52:22,558 INFO [train.py:903] (1/4) Epoch 29, batch 5500, loss[loss=0.1895, simple_loss=0.2618, pruned_loss=0.05857, over 19607.00 frames. ], tot_loss[loss=0.2015, simple_loss=0.2831, pruned_loss=0.05992, over 3801495.09 frames. ], batch size: 50, lr: 2.81e-03, grad_scale: 8.0 2023-04-03 14:52:23,694 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.565e+02 4.767e+02 5.823e+02 7.066e+02 1.237e+03, threshold=1.165e+03, percent-clipped=1.0 2023-04-03 14:52:46,224 WARNING [train.py:1073] (1/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9_rvb from training. Duration: 27.47775 2023-04-03 14:53:08,302 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=196722.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 14:53:22,918 INFO [train.py:903] (1/4) Epoch 29, batch 5550, loss[loss=0.1694, simple_loss=0.2584, pruned_loss=0.04018, over 19578.00 frames. ], tot_loss[loss=0.2018, simple_loss=0.2835, pruned_loss=0.06006, over 3801075.96 frames. ], batch size: 52, lr: 2.81e-03, grad_scale: 8.0 2023-04-03 14:53:30,039 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 2023-04-03 14:53:40,087 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.38 vs. limit=5.0 2023-04-03 14:54:18,739 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 2023-04-03 14:54:22,906 INFO [train.py:903] (1/4) Epoch 29, batch 5600, loss[loss=0.1884, simple_loss=0.2693, pruned_loss=0.05375, over 19637.00 frames. ], tot_loss[loss=0.2018, simple_loss=0.2833, pruned_loss=0.06011, over 3806458.24 frames. ], batch size: 50, lr: 2.81e-03, grad_scale: 8.0 2023-04-03 14:54:24,067 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.131e+02 4.937e+02 5.977e+02 7.468e+02 1.263e+03, threshold=1.195e+03, percent-clipped=3.0 2023-04-03 14:54:39,818 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=196797.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 14:54:40,986 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=196798.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 14:55:24,435 INFO [train.py:903] (1/4) Epoch 29, batch 5650, loss[loss=0.2036, simple_loss=0.291, pruned_loss=0.05817, over 19528.00 frames. ], tot_loss[loss=0.2012, simple_loss=0.2829, pruned_loss=0.0597, over 3797651.04 frames. ], batch size: 56, lr: 2.81e-03, grad_scale: 8.0 2023-04-03 14:55:28,150 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=196837.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 14:56:10,100 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=196872.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 14:56:11,809 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9_rvb from training. Duration: 28.0944375 2023-04-03 14:56:24,235 INFO [train.py:903] (1/4) Epoch 29, batch 5700, loss[loss=0.1749, simple_loss=0.249, pruned_loss=0.05045, over 19797.00 frames. ], tot_loss[loss=0.2004, simple_loss=0.2822, pruned_loss=0.0593, over 3797652.02 frames. ], batch size: 48, lr: 2.81e-03, grad_scale: 8.0 2023-04-03 14:56:25,388 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.192e+02 5.159e+02 6.169e+02 7.777e+02 1.148e+03, threshold=1.234e+03, percent-clipped=0.0 2023-04-03 14:56:41,823 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=196899.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 14:56:46,521 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.72 vs. limit=2.0 2023-04-03 14:57:24,057 INFO [train.py:903] (1/4) Epoch 29, batch 5750, loss[loss=0.2162, simple_loss=0.294, pruned_loss=0.06917, over 19663.00 frames. ], tot_loss[loss=0.2003, simple_loss=0.2818, pruned_loss=0.05938, over 3803028.89 frames. ], batch size: 55, lr: 2.81e-03, grad_scale: 8.0 2023-04-03 14:57:25,244 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9_rvb from training. Duration: 33.038875 2023-04-03 14:57:30,934 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=196940.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 14:57:31,938 WARNING [train.py:1073] (1/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 2023-04-03 14:57:38,015 WARNING [train.py:1073] (1/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 2023-04-03 14:58:24,228 INFO [train.py:903] (1/4) Epoch 29, batch 5800, loss[loss=0.2197, simple_loss=0.302, pruned_loss=0.06866, over 19748.00 frames. ], tot_loss[loss=0.2011, simple_loss=0.2827, pruned_loss=0.05971, over 3821939.26 frames. ], batch size: 54, lr: 2.81e-03, grad_scale: 8.0 2023-04-03 14:58:25,406 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.058e+02 4.987e+02 6.141e+02 8.368e+02 1.662e+03, threshold=1.228e+03, percent-clipped=5.0 2023-04-03 14:59:24,882 INFO [train.py:903] (1/4) Epoch 29, batch 5850, loss[loss=0.1946, simple_loss=0.2741, pruned_loss=0.05759, over 19635.00 frames. ], tot_loss[loss=0.2002, simple_loss=0.282, pruned_loss=0.05917, over 3826514.97 frames. ], batch size: 50, lr: 2.80e-03, grad_scale: 8.0 2023-04-03 14:59:27,409 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.13 vs. limit=2.0 2023-04-03 14:59:49,585 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=197054.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 14:59:50,524 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=197055.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 15:00:19,100 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=197079.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 15:00:24,408 INFO [train.py:903] (1/4) Epoch 29, batch 5900, loss[loss=0.2088, simple_loss=0.2952, pruned_loss=0.06126, over 18765.00 frames. ], tot_loss[loss=0.2011, simple_loss=0.2828, pruned_loss=0.05966, over 3822706.97 frames. ], batch size: 74, lr: 2.80e-03, grad_scale: 8.0 2023-04-03 15:00:25,160 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.74 vs. limit=5.0 2023-04-03 15:00:25,528 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.208e+02 4.889e+02 5.905e+02 7.501e+02 2.168e+03, threshold=1.181e+03, percent-clipped=6.0 2023-04-03 15:00:26,680 WARNING [train.py:1073] (1/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 2023-04-03 15:00:35,898 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=197093.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 15:00:46,559 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9_rvb from training. Duration: 27.511125 2023-04-03 15:00:47,965 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.4424, 4.0697, 2.5282, 3.6121, 1.0854, 4.0367, 3.9327, 3.9644], device='cuda:1'), covar=tensor([0.0650, 0.0940, 0.2084, 0.0837, 0.3669, 0.0680, 0.0899, 0.1128], device='cuda:1'), in_proj_covar=tensor([0.0534, 0.0433, 0.0520, 0.0358, 0.0409, 0.0459, 0.0454, 0.0487], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 15:01:06,458 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=197118.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 15:01:24,661 INFO [train.py:903] (1/4) Epoch 29, batch 5950, loss[loss=0.2034, simple_loss=0.288, pruned_loss=0.05941, over 19756.00 frames. ], tot_loss[loss=0.2011, simple_loss=0.2828, pruned_loss=0.05967, over 3816746.85 frames. ], batch size: 54, lr: 2.80e-03, grad_scale: 8.0 2023-04-03 15:01:32,742 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=197141.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 15:01:59,496 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.6016, 1.2196, 1.4686, 1.2982, 2.2639, 1.1450, 2.1318, 2.5733], device='cuda:1'), covar=tensor([0.0712, 0.2936, 0.2865, 0.1755, 0.0865, 0.2060, 0.1039, 0.0437], device='cuda:1'), in_proj_covar=tensor([0.0427, 0.0383, 0.0402, 0.0357, 0.0385, 0.0360, 0.0400, 0.0421], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 15:02:18,305 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1888, 1.2746, 1.2496, 1.0218, 1.1326, 1.0330, 0.0702, 0.3285], device='cuda:1'), covar=tensor([0.0816, 0.0747, 0.0502, 0.0698, 0.1334, 0.0799, 0.1572, 0.1403], device='cuda:1'), in_proj_covar=tensor([0.0366, 0.0363, 0.0369, 0.0393, 0.0471, 0.0400, 0.0347, 0.0348], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 15:02:24,686 INFO [train.py:903] (1/4) Epoch 29, batch 6000, loss[loss=0.2398, simple_loss=0.3073, pruned_loss=0.08614, over 13693.00 frames. ], tot_loss[loss=0.2014, simple_loss=0.2828, pruned_loss=0.06003, over 3817734.15 frames. ], batch size: 136, lr: 2.80e-03, grad_scale: 8.0 2023-04-03 15:02:24,686 INFO [train.py:928] (1/4) Computing validation loss 2023-04-03 15:02:43,145 INFO [train.py:937] (1/4) Epoch 29, validation: loss=0.167, simple_loss=0.2662, pruned_loss=0.03392, over 944034.00 frames. 2023-04-03 15:02:43,146 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18821MB 2023-04-03 15:02:44,363 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.479e+02 4.704e+02 5.833e+02 7.372e+02 1.660e+03, threshold=1.167e+03, percent-clipped=5.0 2023-04-03 15:02:48,713 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.76 vs. limit=5.0 2023-04-03 15:03:23,053 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=197216.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 15:03:44,117 INFO [train.py:903] (1/4) Epoch 29, batch 6050, loss[loss=0.1883, simple_loss=0.2638, pruned_loss=0.05643, over 19584.00 frames. ], tot_loss[loss=0.2006, simple_loss=0.2819, pruned_loss=0.05967, over 3826213.97 frames. ], batch size: 52, lr: 2.80e-03, grad_scale: 8.0 2023-04-03 15:03:55,189 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=197243.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 15:04:11,945 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=197256.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 15:04:45,073 INFO [train.py:903] (1/4) Epoch 29, batch 6100, loss[loss=0.1601, simple_loss=0.2518, pruned_loss=0.03422, over 19851.00 frames. ], tot_loss[loss=0.1996, simple_loss=0.2805, pruned_loss=0.05934, over 3826349.91 frames. ], batch size: 52, lr: 2.80e-03, grad_scale: 8.0 2023-04-03 15:04:46,195 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.008e+02 5.045e+02 6.568e+02 8.070e+02 1.422e+03, threshold=1.314e+03, percent-clipped=5.0 2023-04-03 15:05:16,415 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.3579, 2.4176, 2.6883, 3.1006, 2.3659, 2.9477, 2.6713, 2.5700], device='cuda:1'), covar=tensor([0.4398, 0.4269, 0.2021, 0.2735, 0.4547, 0.2359, 0.4972, 0.3376], device='cuda:1'), in_proj_covar=tensor([0.0947, 0.1027, 0.0750, 0.0960, 0.0926, 0.0864, 0.0868, 0.0815], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-03 15:05:17,434 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=197311.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 15:05:42,448 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=197331.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 15:05:45,466 INFO [train.py:903] (1/4) Epoch 29, batch 6150, loss[loss=0.2016, simple_loss=0.2916, pruned_loss=0.05583, over 19616.00 frames. ], tot_loss[loss=0.1988, simple_loss=0.2802, pruned_loss=0.05869, over 3839772.89 frames. ], batch size: 61, lr: 2.80e-03, grad_scale: 8.0 2023-04-03 15:05:48,275 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=197336.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 15:06:13,339 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9_rvb from training. Duration: 31.02225 2023-04-03 15:06:14,663 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=197358.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 15:06:46,115 INFO [train.py:903] (1/4) Epoch 29, batch 6200, loss[loss=0.1684, simple_loss=0.2532, pruned_loss=0.04176, over 19621.00 frames. ], tot_loss[loss=0.1996, simple_loss=0.2809, pruned_loss=0.05911, over 3833982.06 frames. ], batch size: 50, lr: 2.80e-03, grad_scale: 8.0 2023-04-03 15:06:47,111 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.451e+02 5.150e+02 5.975e+02 7.505e+02 2.002e+03, threshold=1.195e+03, percent-clipped=6.0 2023-04-03 15:07:11,161 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6720, 1.5118, 1.5100, 2.0310, 1.5579, 1.8587, 1.8434, 1.6947], device='cuda:1'), covar=tensor([0.0820, 0.0920, 0.1015, 0.0652, 0.0825, 0.0751, 0.0802, 0.0703], device='cuda:1'), in_proj_covar=tensor([0.0214, 0.0226, 0.0230, 0.0242, 0.0228, 0.0215, 0.0189, 0.0209], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-04-03 15:07:31,626 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.45 vs. limit=5.0 2023-04-03 15:07:45,774 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=197433.0, num_to_drop=1, layers_to_drop={0} 2023-04-03 15:07:46,599 INFO [train.py:903] (1/4) Epoch 29, batch 6250, loss[loss=0.1932, simple_loss=0.2629, pruned_loss=0.06175, over 18230.00 frames. ], tot_loss[loss=0.2004, simple_loss=0.2815, pruned_loss=0.0596, over 3839268.71 frames. ], batch size: 40, lr: 2.80e-03, grad_scale: 8.0 2023-04-03 15:08:18,032 WARNING [train.py:1073] (1/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9_rvb from training. Duration: 25.988875 2023-04-03 15:08:48,228 INFO [train.py:903] (1/4) Epoch 29, batch 6300, loss[loss=0.1667, simple_loss=0.2419, pruned_loss=0.04575, over 19099.00 frames. ], tot_loss[loss=0.199, simple_loss=0.2804, pruned_loss=0.05884, over 3838529.23 frames. ], batch size: 42, lr: 2.80e-03, grad_scale: 4.0 2023-04-03 15:08:50,595 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.231e+02 4.620e+02 5.575e+02 6.491e+02 1.508e+03, threshold=1.115e+03, percent-clipped=2.0 2023-04-03 15:09:21,954 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=197512.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 15:09:49,202 INFO [train.py:903] (1/4) Epoch 29, batch 6350, loss[loss=0.1872, simple_loss=0.2692, pruned_loss=0.05258, over 19472.00 frames. ], tot_loss[loss=0.1985, simple_loss=0.28, pruned_loss=0.05846, over 3837899.01 frames. ], batch size: 49, lr: 2.80e-03, grad_scale: 4.0 2023-04-03 15:09:49,506 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=197534.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 15:09:52,928 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=197537.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 15:10:23,557 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=197563.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 15:10:49,098 INFO [train.py:903] (1/4) Epoch 29, batch 6400, loss[loss=0.1931, simple_loss=0.2688, pruned_loss=0.05872, over 19377.00 frames. ], tot_loss[loss=0.1985, simple_loss=0.2805, pruned_loss=0.05826, over 3851678.02 frames. ], batch size: 47, lr: 2.80e-03, grad_scale: 8.0 2023-04-03 15:10:52,229 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.915e+02 4.287e+02 5.708e+02 7.339e+02 1.464e+03, threshold=1.142e+03, percent-clipped=5.0 2023-04-03 15:10:53,775 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=197587.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 15:10:53,985 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 2023-04-03 15:11:23,600 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=197612.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 15:11:25,873 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=197614.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 15:11:49,896 INFO [train.py:903] (1/4) Epoch 29, batch 6450, loss[loss=0.2176, simple_loss=0.2998, pruned_loss=0.06767, over 19575.00 frames. ], tot_loss[loss=0.1982, simple_loss=0.2802, pruned_loss=0.05809, over 3860847.29 frames. ], batch size: 61, lr: 2.80e-03, grad_scale: 8.0 2023-04-03 15:11:56,895 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=197639.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 15:12:24,333 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([0.8132, 1.1413, 0.9282, 0.8966, 1.0380, 0.9002, 0.8654, 1.0928], device='cuda:1'), covar=tensor([0.0600, 0.0863, 0.1013, 0.0712, 0.0558, 0.1178, 0.0555, 0.0485], device='cuda:1'), in_proj_covar=tensor([0.0307, 0.0322, 0.0345, 0.0277, 0.0255, 0.0347, 0.0293, 0.0278], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 15:12:37,427 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9_rvb from training. Duration: 28.638875 2023-04-03 15:12:49,912 INFO [train.py:903] (1/4) Epoch 29, batch 6500, loss[loss=0.2025, simple_loss=0.2881, pruned_loss=0.0585, over 19475.00 frames. ], tot_loss[loss=0.2, simple_loss=0.2819, pruned_loss=0.05907, over 3855848.32 frames. ], batch size: 64, lr: 2.80e-03, grad_scale: 8.0 2023-04-03 15:12:52,112 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.446e+02 4.970e+02 6.097e+02 8.194e+02 1.467e+03, threshold=1.219e+03, percent-clipped=8.0 2023-04-03 15:12:58,634 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1_rvb from training. Duration: 0.836375 2023-04-03 15:13:50,299 INFO [train.py:903] (1/4) Epoch 29, batch 6550, loss[loss=0.2214, simple_loss=0.305, pruned_loss=0.06891, over 19669.00 frames. ], tot_loss[loss=0.2005, simple_loss=0.2822, pruned_loss=0.0594, over 3845427.50 frames. ], batch size: 53, lr: 2.80e-03, grad_scale: 8.0 2023-04-03 15:14:42,386 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=197777.0, num_to_drop=1, layers_to_drop={1} 2023-04-03 15:14:49,852 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.7757, 1.9095, 2.2458, 2.0406, 2.8694, 3.2649, 3.1539, 3.4399], device='cuda:1'), covar=tensor([0.1390, 0.3112, 0.2849, 0.2347, 0.1036, 0.0263, 0.0217, 0.0390], device='cuda:1'), in_proj_covar=tensor([0.0281, 0.0334, 0.0367, 0.0275, 0.0256, 0.0199, 0.0221, 0.0279], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-03 15:14:50,648 INFO [train.py:903] (1/4) Epoch 29, batch 6600, loss[loss=0.2399, simple_loss=0.3053, pruned_loss=0.08722, over 13152.00 frames. ], tot_loss[loss=0.1991, simple_loss=0.2809, pruned_loss=0.0586, over 3836925.13 frames. ], batch size: 137, lr: 2.80e-03, grad_scale: 4.0 2023-04-03 15:14:54,077 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.141e+02 4.758e+02 5.997e+02 7.159e+02 2.116e+03, threshold=1.199e+03, percent-clipped=4.0 2023-04-03 15:15:42,977 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2630, 0.9797, 1.1954, 1.9081, 1.2915, 1.1432, 1.2843, 1.1483], device='cuda:1'), covar=tensor([0.1196, 0.1851, 0.1416, 0.0793, 0.1152, 0.1554, 0.1285, 0.1230], device='cuda:1'), in_proj_covar=tensor([0.0214, 0.0226, 0.0230, 0.0244, 0.0228, 0.0216, 0.0190, 0.0210], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-04-03 15:15:51,170 INFO [train.py:903] (1/4) Epoch 29, batch 6650, loss[loss=0.2132, simple_loss=0.2997, pruned_loss=0.06342, over 19477.00 frames. ], tot_loss[loss=0.1984, simple_loss=0.2806, pruned_loss=0.05813, over 3831511.18 frames. ], batch size: 64, lr: 2.80e-03, grad_scale: 4.0 2023-04-03 15:16:45,535 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=197878.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 15:16:52,899 INFO [train.py:903] (1/4) Epoch 29, batch 6700, loss[loss=0.2571, simple_loss=0.3225, pruned_loss=0.0959, over 13261.00 frames. ], tot_loss[loss=0.1983, simple_loss=0.2802, pruned_loss=0.05817, over 3814977.66 frames. ], batch size: 139, lr: 2.80e-03, grad_scale: 4.0 2023-04-03 15:16:56,373 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.106e+02 5.093e+02 5.944e+02 7.634e+02 1.783e+03, threshold=1.189e+03, percent-clipped=4.0 2023-04-03 15:17:02,450 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=197892.0, num_to_drop=1, layers_to_drop={1} 2023-04-03 15:17:19,974 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=197907.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 15:17:23,564 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=197910.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 15:17:44,097 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=4.15 vs. limit=5.0 2023-04-03 15:17:50,660 INFO [train.py:903] (1/4) Epoch 29, batch 6750, loss[loss=0.1961, simple_loss=0.2851, pruned_loss=0.05355, over 19141.00 frames. ], tot_loss[loss=0.199, simple_loss=0.2809, pruned_loss=0.05861, over 3809826.51 frames. ], batch size: 69, lr: 2.80e-03, grad_scale: 4.0 2023-04-03 15:18:29,601 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4750, 1.5398, 1.7910, 1.7113, 2.5387, 2.2520, 2.7245, 1.1563], device='cuda:1'), covar=tensor([0.2587, 0.4445, 0.2825, 0.2027, 0.1654, 0.2268, 0.1528, 0.4906], device='cuda:1'), in_proj_covar=tensor([0.0558, 0.0680, 0.0768, 0.0515, 0.0642, 0.0551, 0.0675, 0.0579], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-03 15:18:45,947 INFO [train.py:903] (1/4) Epoch 29, batch 6800, loss[loss=0.2309, simple_loss=0.3126, pruned_loss=0.07456, over 18859.00 frames. ], tot_loss[loss=0.1997, simple_loss=0.2811, pruned_loss=0.05912, over 3802495.08 frames. ], batch size: 74, lr: 2.80e-03, grad_scale: 8.0 2023-04-03 15:18:49,168 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.543e+02 4.853e+02 6.089e+02 7.661e+02 1.249e+03, threshold=1.218e+03, percent-clipped=1.0 2023-04-03 15:18:56,334 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=197993.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 15:18:59,789 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.8160, 1.9271, 2.1670, 2.2795, 1.7920, 2.2280, 2.1269, 1.9957], device='cuda:1'), covar=tensor([0.4379, 0.3916, 0.2082, 0.2599, 0.4077, 0.2323, 0.5300, 0.3647], device='cuda:1'), in_proj_covar=tensor([0.0946, 0.1027, 0.0749, 0.0959, 0.0926, 0.0863, 0.0863, 0.0817], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-03 15:19:09,598 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.1515, 3.3719, 3.6533, 3.6749, 1.9910, 3.4022, 3.0582, 3.4656], device='cuda:1'), covar=tensor([0.1599, 0.3706, 0.0715, 0.0831, 0.5141, 0.1617, 0.0721, 0.1104], device='cuda:1'), in_proj_covar=tensor([0.0824, 0.0797, 0.1004, 0.0881, 0.0873, 0.0771, 0.0592, 0.0937], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-03 15:19:33,727 WARNING [train.py:1073] (1/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 2023-04-03 15:19:34,154 WARNING [train.py:1073] (1/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 2023-04-03 15:19:37,016 INFO [train.py:903] (1/4) Epoch 30, batch 0, loss[loss=0.2303, simple_loss=0.3067, pruned_loss=0.07698, over 18204.00 frames. ], tot_loss[loss=0.2303, simple_loss=0.3067, pruned_loss=0.07698, over 18204.00 frames. ], batch size: 83, lr: 2.75e-03, grad_scale: 8.0 2023-04-03 15:19:37,016 INFO [train.py:928] (1/4) Computing validation loss 2023-04-03 15:19:49,307 INFO [train.py:937] (1/4) Epoch 30, validation: loss=0.167, simple_loss=0.2667, pruned_loss=0.03362, over 944034.00 frames. 2023-04-03 15:19:49,308 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18821MB 2023-04-03 15:20:02,490 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=198022.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 15:20:03,299 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 2023-04-03 15:20:25,395 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.09 vs. limit=2.0 2023-04-03 15:20:51,416 INFO [train.py:903] (1/4) Epoch 30, batch 50, loss[loss=0.1981, simple_loss=0.2913, pruned_loss=0.05246, over 19624.00 frames. ], tot_loss[loss=0.1996, simple_loss=0.2812, pruned_loss=0.05902, over 853155.80 frames. ], batch size: 57, lr: 2.75e-03, grad_scale: 8.0 2023-04-03 15:21:20,382 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.592e+02 4.959e+02 5.838e+02 7.711e+02 1.808e+03, threshold=1.168e+03, percent-clipped=5.0 2023-04-03 15:21:26,252 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1_rvb from training. Duration: 27.0318125 2023-04-03 15:21:36,615 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=198099.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 15:21:52,085 INFO [train.py:903] (1/4) Epoch 30, batch 100, loss[loss=0.1916, simple_loss=0.2803, pruned_loss=0.05144, over 19665.00 frames. ], tot_loss[loss=0.2015, simple_loss=0.2833, pruned_loss=0.05979, over 1519029.05 frames. ], batch size: 55, lr: 2.75e-03, grad_scale: 8.0 2023-04-03 15:22:04,483 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 2023-04-03 15:22:37,707 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=198148.0, num_to_drop=1, layers_to_drop={0} 2023-04-03 15:22:54,436 INFO [train.py:903] (1/4) Epoch 30, batch 150, loss[loss=0.1987, simple_loss=0.2723, pruned_loss=0.06258, over 19403.00 frames. ], tot_loss[loss=0.2021, simple_loss=0.2838, pruned_loss=0.06017, over 2023172.64 frames. ], batch size: 48, lr: 2.75e-03, grad_scale: 8.0 2023-04-03 15:23:07,432 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=198173.0, num_to_drop=1, layers_to_drop={1} 2023-04-03 15:23:25,375 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.519e+02 4.863e+02 5.858e+02 7.546e+02 1.579e+03, threshold=1.172e+03, percent-clipped=2.0 2023-04-03 15:23:52,146 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 2023-04-03 15:23:55,535 INFO [train.py:903] (1/4) Epoch 30, batch 200, loss[loss=0.1965, simple_loss=0.278, pruned_loss=0.05755, over 19678.00 frames. ], tot_loss[loss=0.2015, simple_loss=0.2834, pruned_loss=0.05976, over 2422954.71 frames. ], batch size: 53, lr: 2.75e-03, grad_scale: 8.0 2023-04-03 15:23:58,310 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6323, 1.7216, 1.8915, 1.8504, 2.5540, 2.3809, 2.6658, 1.1714], device='cuda:1'), covar=tensor([0.2553, 0.4504, 0.2922, 0.1991, 0.1510, 0.2200, 0.1395, 0.4785], device='cuda:1'), in_proj_covar=tensor([0.0561, 0.0681, 0.0770, 0.0516, 0.0642, 0.0552, 0.0676, 0.0579], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-03 15:24:42,400 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=198249.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 15:24:47,693 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=198254.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 15:24:57,534 INFO [train.py:903] (1/4) Epoch 30, batch 250, loss[loss=0.2171, simple_loss=0.3097, pruned_loss=0.06222, over 19540.00 frames. ], tot_loss[loss=0.2011, simple_loss=0.2829, pruned_loss=0.05961, over 2718994.11 frames. ], batch size: 54, lr: 2.75e-03, grad_scale: 8.0 2023-04-03 15:25:13,977 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=198274.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 15:25:19,533 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=198278.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 15:25:29,213 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.056e+02 4.793e+02 5.720e+02 7.294e+02 1.524e+03, threshold=1.144e+03, percent-clipped=6.0 2023-04-03 15:25:49,178 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=198303.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 15:26:01,275 INFO [train.py:903] (1/4) Epoch 30, batch 300, loss[loss=0.2064, simple_loss=0.2976, pruned_loss=0.05761, over 19330.00 frames. ], tot_loss[loss=0.2013, simple_loss=0.2831, pruned_loss=0.05974, over 2965442.98 frames. ], batch size: 66, lr: 2.75e-03, grad_scale: 8.0 2023-04-03 15:27:03,943 INFO [train.py:903] (1/4) Epoch 30, batch 350, loss[loss=0.1552, simple_loss=0.2366, pruned_loss=0.03693, over 19769.00 frames. ], tot_loss[loss=0.2017, simple_loss=0.2833, pruned_loss=0.06003, over 3146913.35 frames. ], batch size: 47, lr: 2.75e-03, grad_scale: 8.0 2023-04-03 15:27:06,269 WARNING [train.py:1073] (1/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1_rvb from training. Duration: 0.9681875 2023-04-03 15:27:12,156 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=198369.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 15:27:33,418 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.869e+02 4.876e+02 6.987e+02 8.997e+02 2.429e+03, threshold=1.397e+03, percent-clipped=9.0 2023-04-03 15:27:48,286 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6402, 1.4694, 1.8294, 1.4751, 2.8214, 3.8694, 3.5908, 4.0303], device='cuda:1'), covar=tensor([0.1403, 0.3782, 0.3402, 0.2486, 0.0622, 0.0176, 0.0207, 0.0276], device='cuda:1'), in_proj_covar=tensor([0.0279, 0.0331, 0.0364, 0.0272, 0.0254, 0.0198, 0.0219, 0.0277], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-03 15:28:04,750 INFO [train.py:903] (1/4) Epoch 30, batch 400, loss[loss=0.2171, simple_loss=0.3046, pruned_loss=0.06479, over 19321.00 frames. ], tot_loss[loss=0.2018, simple_loss=0.2834, pruned_loss=0.06009, over 3295692.59 frames. ], batch size: 70, lr: 2.75e-03, grad_scale: 8.0 2023-04-03 15:28:08,526 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.0444, 3.7142, 2.7131, 3.3276, 0.9847, 3.6986, 3.5418, 3.5873], device='cuda:1'), covar=tensor([0.0752, 0.1069, 0.1894, 0.0897, 0.3895, 0.0736, 0.0992, 0.1209], device='cuda:1'), in_proj_covar=tensor([0.0535, 0.0435, 0.0523, 0.0359, 0.0412, 0.0460, 0.0454, 0.0488], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 15:28:44,695 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=198443.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 15:28:56,374 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.0620, 3.7477, 2.6412, 3.3184, 0.8250, 3.7206, 3.5483, 3.6191], device='cuda:1'), covar=tensor([0.0770, 0.1143, 0.1933, 0.0923, 0.3844, 0.0743, 0.1027, 0.1194], device='cuda:1'), in_proj_covar=tensor([0.0533, 0.0434, 0.0522, 0.0358, 0.0411, 0.0459, 0.0453, 0.0487], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 15:29:06,535 INFO [train.py:903] (1/4) Epoch 30, batch 450, loss[loss=0.198, simple_loss=0.2846, pruned_loss=0.05567, over 19625.00 frames. ], tot_loss[loss=0.1999, simple_loss=0.2818, pruned_loss=0.05896, over 3430314.02 frames. ], batch size: 57, lr: 2.75e-03, grad_scale: 8.0 2023-04-03 15:29:38,785 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.061e+02 5.102e+02 6.085e+02 7.779e+02 1.785e+03, threshold=1.217e+03, percent-clipped=4.0 2023-04-03 15:29:40,856 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 2023-04-03 15:29:42,045 WARNING [train.py:1073] (1/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 2023-04-03 15:30:08,496 INFO [train.py:903] (1/4) Epoch 30, batch 500, loss[loss=0.1941, simple_loss=0.2604, pruned_loss=0.06393, over 19385.00 frames. ], tot_loss[loss=0.1996, simple_loss=0.2815, pruned_loss=0.05882, over 3506092.69 frames. ], batch size: 47, lr: 2.75e-03, grad_scale: 8.0 2023-04-03 15:30:15,662 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=198517.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 15:31:06,284 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=198558.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 15:31:11,508 INFO [train.py:903] (1/4) Epoch 30, batch 550, loss[loss=0.2004, simple_loss=0.2883, pruned_loss=0.05622, over 19284.00 frames. ], tot_loss[loss=0.1999, simple_loss=0.2821, pruned_loss=0.0589, over 3588307.71 frames. ], batch size: 66, lr: 2.75e-03, grad_scale: 8.0 2023-04-03 15:31:40,932 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.425e+02 4.928e+02 6.417e+02 8.667e+02 2.852e+03, threshold=1.283e+03, percent-clipped=10.0 2023-04-03 15:32:13,286 INFO [train.py:903] (1/4) Epoch 30, batch 600, loss[loss=0.2219, simple_loss=0.3034, pruned_loss=0.07016, over 17413.00 frames. ], tot_loss[loss=0.2002, simple_loss=0.2822, pruned_loss=0.0591, over 3640161.76 frames. ], batch size: 101, lr: 2.75e-03, grad_scale: 8.0 2023-04-03 15:32:28,413 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=198625.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 15:32:38,378 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=198633.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 15:32:51,217 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.43 vs. limit=5.0 2023-04-03 15:32:52,895 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 2023-04-03 15:33:01,368 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=198650.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 15:33:14,980 INFO [train.py:903] (1/4) Epoch 30, batch 650, loss[loss=0.1628, simple_loss=0.246, pruned_loss=0.03984, over 19786.00 frames. ], tot_loss[loss=0.1984, simple_loss=0.2805, pruned_loss=0.0582, over 3696815.19 frames. ], batch size: 46, lr: 2.75e-03, grad_scale: 8.0 2023-04-03 15:33:46,600 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.134e+02 5.000e+02 5.780e+02 7.067e+02 2.104e+03, threshold=1.156e+03, percent-clipped=2.0 2023-04-03 15:34:16,540 INFO [train.py:903] (1/4) Epoch 30, batch 700, loss[loss=0.18, simple_loss=0.2561, pruned_loss=0.05197, over 19592.00 frames. ], tot_loss[loss=0.1985, simple_loss=0.2801, pruned_loss=0.05839, over 3723815.13 frames. ], batch size: 52, lr: 2.74e-03, grad_scale: 8.0 2023-04-03 15:35:18,682 INFO [train.py:903] (1/4) Epoch 30, batch 750, loss[loss=0.2166, simple_loss=0.2809, pruned_loss=0.07617, over 19769.00 frames. ], tot_loss[loss=0.1986, simple_loss=0.2802, pruned_loss=0.05847, over 3742555.90 frames. ], batch size: 47, lr: 2.74e-03, grad_scale: 8.0 2023-04-03 15:35:29,204 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.3717, 1.3813, 1.5328, 1.5368, 1.6731, 1.8703, 1.7691, 0.4786], device='cuda:1'), covar=tensor([0.2634, 0.4634, 0.2835, 0.2086, 0.1917, 0.2523, 0.1573, 0.5431], device='cuda:1'), in_proj_covar=tensor([0.0559, 0.0679, 0.0766, 0.0515, 0.0640, 0.0551, 0.0674, 0.0578], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-03 15:35:51,149 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.284e+02 4.908e+02 5.891e+02 7.920e+02 1.978e+03, threshold=1.178e+03, percent-clipped=7.0 2023-04-03 15:36:04,794 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.81 vs. limit=2.0 2023-04-03 15:36:05,380 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6028, 1.6947, 1.9372, 1.9098, 1.4808, 1.8899, 1.9125, 1.7931], device='cuda:1'), covar=tensor([0.4247, 0.3808, 0.2097, 0.2528, 0.4039, 0.2380, 0.5385, 0.3560], device='cuda:1'), in_proj_covar=tensor([0.0946, 0.1029, 0.0750, 0.0960, 0.0926, 0.0866, 0.0866, 0.0815], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-03 15:36:23,529 INFO [train.py:903] (1/4) Epoch 30, batch 800, loss[loss=0.2128, simple_loss=0.2945, pruned_loss=0.06552, over 19616.00 frames. ], tot_loss[loss=0.1989, simple_loss=0.2807, pruned_loss=0.05852, over 3765009.85 frames. ], batch size: 61, lr: 2.74e-03, grad_scale: 8.0 2023-04-03 15:36:27,272 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=198814.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 15:36:39,846 WARNING [train.py:1073] (1/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 2023-04-03 15:36:56,795 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=198839.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 15:37:25,167 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=198861.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 15:37:26,181 INFO [train.py:903] (1/4) Epoch 30, batch 850, loss[loss=0.2074, simple_loss=0.2967, pruned_loss=0.05907, over 19660.00 frames. ], tot_loss[loss=0.1995, simple_loss=0.2814, pruned_loss=0.05884, over 3773753.71 frames. ], batch size: 60, lr: 2.74e-03, grad_scale: 8.0 2023-04-03 15:37:35,466 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=198870.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 15:37:55,569 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.084e+02 4.755e+02 5.781e+02 7.125e+02 1.514e+03, threshold=1.156e+03, percent-clipped=6.0 2023-04-03 15:37:59,387 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9876, 1.6747, 1.5643, 1.8649, 1.5880, 1.6512, 1.5110, 1.7998], device='cuda:1'), covar=tensor([0.1011, 0.1201, 0.1587, 0.1026, 0.1257, 0.0610, 0.1554, 0.0780], device='cuda:1'), in_proj_covar=tensor([0.0281, 0.0360, 0.0321, 0.0259, 0.0311, 0.0260, 0.0324, 0.0265], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-04-03 15:38:13,693 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=198901.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 15:38:20,325 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9_rvb from training. Duration: 25.061125 2023-04-03 15:38:26,175 INFO [train.py:903] (1/4) Epoch 30, batch 900, loss[loss=0.1809, simple_loss=0.2705, pruned_loss=0.04566, over 19770.00 frames. ], tot_loss[loss=0.2002, simple_loss=0.2821, pruned_loss=0.05916, over 3787384.12 frames. ], batch size: 54, lr: 2.74e-03, grad_scale: 8.0 2023-04-03 15:39:07,728 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=198945.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 15:39:27,824 INFO [train.py:903] (1/4) Epoch 30, batch 950, loss[loss=0.206, simple_loss=0.2968, pruned_loss=0.0576, over 19488.00 frames. ], tot_loss[loss=0.2009, simple_loss=0.283, pruned_loss=0.05946, over 3809474.44 frames. ], batch size: 64, lr: 2.74e-03, grad_scale: 8.0 2023-04-03 15:39:32,353 WARNING [train.py:1073] (1/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9_rvb from training. Duration: 26.32775 2023-04-03 15:39:47,014 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=198976.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 15:39:48,041 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=198977.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 15:39:59,676 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.142e+02 5.052e+02 6.305e+02 7.870e+02 1.588e+03, threshold=1.261e+03, percent-clipped=4.0 2023-04-03 15:40:30,253 INFO [train.py:903] (1/4) Epoch 30, batch 1000, loss[loss=0.2046, simple_loss=0.2932, pruned_loss=0.05805, over 19606.00 frames. ], tot_loss[loss=0.2005, simple_loss=0.2829, pruned_loss=0.05907, over 3824851.00 frames. ], batch size: 57, lr: 2.74e-03, grad_scale: 8.0 2023-04-03 15:40:46,863 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.81 vs. limit=2.0 2023-04-03 15:41:23,995 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 2023-04-03 15:41:33,183 INFO [train.py:903] (1/4) Epoch 30, batch 1050, loss[loss=0.1701, simple_loss=0.2571, pruned_loss=0.0415, over 19845.00 frames. ], tot_loss[loss=0.2007, simple_loss=0.2826, pruned_loss=0.05939, over 3818809.85 frames. ], batch size: 52, lr: 2.74e-03, grad_scale: 8.0 2023-04-03 15:42:03,436 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.270e+02 4.631e+02 5.961e+02 7.864e+02 1.953e+03, threshold=1.192e+03, percent-clipped=2.0 2023-04-03 15:42:04,594 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 2023-04-03 15:42:07,343 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.7880, 2.0643, 2.3733, 2.1433, 3.1844, 3.6655, 3.5323, 3.9514], device='cuda:1'), covar=tensor([0.1458, 0.3070, 0.2833, 0.2197, 0.1000, 0.0387, 0.0217, 0.0356], device='cuda:1'), in_proj_covar=tensor([0.0281, 0.0333, 0.0365, 0.0273, 0.0256, 0.0199, 0.0220, 0.0279], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-03 15:42:11,044 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=199092.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 15:42:35,185 INFO [train.py:903] (1/4) Epoch 30, batch 1100, loss[loss=0.164, simple_loss=0.2416, pruned_loss=0.04323, over 19724.00 frames. ], tot_loss[loss=0.1994, simple_loss=0.2813, pruned_loss=0.05872, over 3828901.27 frames. ], batch size: 46, lr: 2.74e-03, grad_scale: 8.0 2023-04-03 15:43:38,421 INFO [train.py:903] (1/4) Epoch 30, batch 1150, loss[loss=0.1876, simple_loss=0.2704, pruned_loss=0.05238, over 19589.00 frames. ], tot_loss[loss=0.1999, simple_loss=0.2817, pruned_loss=0.05906, over 3816040.99 frames. ], batch size: 52, lr: 2.74e-03, grad_scale: 8.0 2023-04-03 15:44:03,860 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.1651, 1.2976, 1.5192, 1.4588, 2.8157, 1.2441, 2.2932, 3.1629], device='cuda:1'), covar=tensor([0.0543, 0.2898, 0.2852, 0.1766, 0.0671, 0.2256, 0.1184, 0.0301], device='cuda:1'), in_proj_covar=tensor([0.0426, 0.0381, 0.0400, 0.0355, 0.0385, 0.0359, 0.0401, 0.0422], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 15:44:10,291 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.218e+02 5.110e+02 5.915e+02 7.639e+02 1.372e+03, threshold=1.183e+03, percent-clipped=6.0 2023-04-03 15:44:18,752 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=199194.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 15:44:41,421 INFO [train.py:903] (1/4) Epoch 30, batch 1200, loss[loss=0.1702, simple_loss=0.2546, pruned_loss=0.04287, over 19842.00 frames. ], tot_loss[loss=0.2009, simple_loss=0.2828, pruned_loss=0.05945, over 3820071.15 frames. ], batch size: 52, lr: 2.74e-03, grad_scale: 8.0 2023-04-03 15:44:43,950 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=199214.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 15:45:06,839 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=199232.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 15:45:14,390 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 2023-04-03 15:45:22,181 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=199245.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 15:45:38,774 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=199257.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 15:45:44,960 INFO [train.py:903] (1/4) Epoch 30, batch 1250, loss[loss=0.2234, simple_loss=0.306, pruned_loss=0.07044, over 19371.00 frames. ], tot_loss[loss=0.1998, simple_loss=0.2819, pruned_loss=0.05881, over 3822855.63 frames. ], batch size: 66, lr: 2.74e-03, grad_scale: 8.0 2023-04-03 15:46:14,637 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.608e+02 5.004e+02 5.991e+02 7.632e+02 1.398e+03, threshold=1.198e+03, percent-clipped=2.0 2023-04-03 15:46:17,137 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=199289.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 15:46:46,509 INFO [train.py:903] (1/4) Epoch 30, batch 1300, loss[loss=0.2025, simple_loss=0.2827, pruned_loss=0.06113, over 18342.00 frames. ], tot_loss[loss=0.1994, simple_loss=0.2816, pruned_loss=0.0586, over 3821683.26 frames. ], batch size: 84, lr: 2.74e-03, grad_scale: 8.0 2023-04-03 15:47:07,591 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=199329.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 15:47:22,756 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4685, 2.1454, 1.6153, 1.4547, 2.0335, 1.3214, 1.3321, 1.9581], device='cuda:1'), covar=tensor([0.1201, 0.0932, 0.1140, 0.0936, 0.0629, 0.1378, 0.0825, 0.0585], device='cuda:1'), in_proj_covar=tensor([0.0305, 0.0320, 0.0344, 0.0275, 0.0253, 0.0348, 0.0290, 0.0275], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 15:47:32,044 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=199348.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 15:47:46,099 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=199360.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 15:47:47,888 INFO [train.py:903] (1/4) Epoch 30, batch 1350, loss[loss=0.1668, simple_loss=0.2471, pruned_loss=0.04319, over 19407.00 frames. ], tot_loss[loss=0.1986, simple_loss=0.2811, pruned_loss=0.0581, over 3839538.92 frames. ], batch size: 47, lr: 2.74e-03, grad_scale: 4.0 2023-04-03 15:48:03,198 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=199373.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 15:48:13,453 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.2891, 2.2258, 2.1153, 2.3696, 2.1232, 1.9960, 2.1007, 2.2563], device='cuda:1'), covar=tensor([0.0817, 0.1112, 0.1144, 0.0777, 0.1068, 0.0525, 0.1124, 0.0608], device='cuda:1'), in_proj_covar=tensor([0.0281, 0.0360, 0.0321, 0.0259, 0.0310, 0.0260, 0.0323, 0.0264], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-04-03 15:48:20,886 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.443e+02 4.672e+02 5.739e+02 7.239e+02 1.592e+03, threshold=1.148e+03, percent-clipped=7.0 2023-04-03 15:48:40,961 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=199404.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 15:48:50,924 INFO [train.py:903] (1/4) Epoch 30, batch 1400, loss[loss=0.1813, simple_loss=0.2691, pruned_loss=0.04673, over 19678.00 frames. ], tot_loss[loss=0.1985, simple_loss=0.2811, pruned_loss=0.05797, over 3831576.18 frames. ], batch size: 53, lr: 2.74e-03, grad_scale: 4.0 2023-04-03 15:48:59,010 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=199418.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 15:49:55,027 INFO [train.py:903] (1/4) Epoch 30, batch 1450, loss[loss=0.2166, simple_loss=0.2922, pruned_loss=0.07051, over 18373.00 frames. ], tot_loss[loss=0.1988, simple_loss=0.2812, pruned_loss=0.05815, over 3823417.71 frames. ], batch size: 84, lr: 2.74e-03, grad_scale: 4.0 2023-04-03 15:49:56,191 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_rvb from training. Duration: 25.85 2023-04-03 15:50:11,578 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=199476.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 15:50:25,539 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.487e+02 5.040e+02 6.121e+02 7.927e+02 1.972e+03, threshold=1.224e+03, percent-clipped=6.0 2023-04-03 15:50:56,161 INFO [train.py:903] (1/4) Epoch 30, batch 1500, loss[loss=0.1739, simple_loss=0.2563, pruned_loss=0.04572, over 19389.00 frames. ], tot_loss[loss=0.1996, simple_loss=0.2817, pruned_loss=0.05872, over 3825067.30 frames. ], batch size: 47, lr: 2.74e-03, grad_scale: 4.0 2023-04-03 15:50:56,495 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=199512.0, num_to_drop=1, layers_to_drop={1} 2023-04-03 15:51:28,278 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=199538.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 15:51:46,261 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.6897, 1.3063, 1.4881, 1.5690, 3.3129, 1.1805, 2.4298, 3.6054], device='cuda:1'), covar=tensor([0.0473, 0.2936, 0.2980, 0.1853, 0.0634, 0.2592, 0.1393, 0.0275], device='cuda:1'), in_proj_covar=tensor([0.0427, 0.0383, 0.0403, 0.0356, 0.0388, 0.0362, 0.0403, 0.0424], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 15:51:47,401 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=199553.0, num_to_drop=1, layers_to_drop={1} 2023-04-03 15:51:58,248 INFO [train.py:903] (1/4) Epoch 30, batch 1550, loss[loss=0.2147, simple_loss=0.2932, pruned_loss=0.06813, over 17808.00 frames. ], tot_loss[loss=0.199, simple_loss=0.2808, pruned_loss=0.05859, over 3823817.54 frames. ], batch size: 101, lr: 2.74e-03, grad_scale: 4.0 2023-04-03 15:52:28,212 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=199585.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 15:52:31,298 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.284e+02 4.717e+02 5.804e+02 6.903e+02 1.639e+03, threshold=1.161e+03, percent-clipped=1.0 2023-04-03 15:52:45,773 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6867, 1.6703, 1.6451, 1.4343, 1.3667, 1.3687, 0.3387, 0.7122], device='cuda:1'), covar=tensor([0.0708, 0.0701, 0.0480, 0.0723, 0.1318, 0.0867, 0.1471, 0.1202], device='cuda:1'), in_proj_covar=tensor([0.0368, 0.0364, 0.0370, 0.0395, 0.0472, 0.0398, 0.0347, 0.0348], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 15:52:59,355 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=199610.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 15:53:01,356 INFO [train.py:903] (1/4) Epoch 30, batch 1600, loss[loss=0.219, simple_loss=0.2871, pruned_loss=0.07544, over 19625.00 frames. ], tot_loss[loss=0.1985, simple_loss=0.2806, pruned_loss=0.05817, over 3831083.03 frames. ], batch size: 50, lr: 2.74e-03, grad_scale: 8.0 2023-04-03 15:53:06,439 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=199616.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 15:53:26,395 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9_rvb from training. Duration: 30.1555625 2023-04-03 15:53:37,720 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=199641.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 15:53:53,130 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=199653.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 15:54:01,264 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=199660.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 15:54:03,841 INFO [train.py:903] (1/4) Epoch 30, batch 1650, loss[loss=0.2453, simple_loss=0.3224, pruned_loss=0.08413, over 17294.00 frames. ], tot_loss[loss=0.1986, simple_loss=0.2806, pruned_loss=0.05832, over 3819171.37 frames. ], batch size: 101, lr: 2.74e-03, grad_scale: 8.0 2023-04-03 15:54:32,753 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=199685.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 15:54:35,819 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.122e+02 4.724e+02 5.858e+02 7.730e+02 3.208e+03, threshold=1.172e+03, percent-clipped=5.0 2023-04-03 15:54:36,276 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9370, 1.8857, 1.8124, 1.6425, 1.4889, 1.5685, 0.5869, 0.8959], device='cuda:1'), covar=tensor([0.0769, 0.0694, 0.0522, 0.0800, 0.1302, 0.0943, 0.1390, 0.1281], device='cuda:1'), in_proj_covar=tensor([0.0370, 0.0365, 0.0372, 0.0397, 0.0475, 0.0401, 0.0348, 0.0350], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 15:54:50,264 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.0755, 1.2587, 1.7367, 0.9820, 2.3736, 3.0577, 2.7332, 3.2280], device='cuda:1'), covar=tensor([0.1764, 0.4150, 0.3495, 0.2968, 0.0693, 0.0270, 0.0280, 0.0340], device='cuda:1'), in_proj_covar=tensor([0.0283, 0.0336, 0.0369, 0.0276, 0.0258, 0.0201, 0.0222, 0.0282], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-03 15:55:06,071 INFO [train.py:903] (1/4) Epoch 30, batch 1700, loss[loss=0.1763, simple_loss=0.2593, pruned_loss=0.0466, over 19674.00 frames. ], tot_loss[loss=0.1973, simple_loss=0.2794, pruned_loss=0.05762, over 3830096.76 frames. ], batch size: 53, lr: 2.74e-03, grad_scale: 8.0 2023-04-03 15:55:24,010 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0822, 1.9510, 1.8295, 1.7171, 1.4779, 1.6597, 0.6486, 1.0433], device='cuda:1'), covar=tensor([0.0751, 0.0700, 0.0617, 0.0982, 0.1463, 0.1007, 0.1476, 0.1314], device='cuda:1'), in_proj_covar=tensor([0.0371, 0.0366, 0.0373, 0.0398, 0.0476, 0.0402, 0.0350, 0.0351], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 15:55:40,547 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.71 vs. limit=2.0 2023-04-03 15:55:46,466 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 2023-04-03 15:56:09,056 INFO [train.py:903] (1/4) Epoch 30, batch 1750, loss[loss=0.1834, simple_loss=0.2563, pruned_loss=0.05526, over 19085.00 frames. ], tot_loss[loss=0.1978, simple_loss=0.2801, pruned_loss=0.0578, over 3816919.27 frames. ], batch size: 42, lr: 2.74e-03, grad_scale: 4.0 2023-04-03 15:56:09,206 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=199762.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 15:56:42,923 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.076e+02 4.893e+02 6.270e+02 7.375e+02 1.627e+03, threshold=1.254e+03, percent-clipped=1.0 2023-04-03 15:57:00,414 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([5.1156, 5.1657, 5.9233, 5.9452, 1.9456, 5.5837, 4.7128, 5.6206], device='cuda:1'), covar=tensor([0.1837, 0.0858, 0.0593, 0.0680, 0.6596, 0.0962, 0.0666, 0.1172], device='cuda:1'), in_proj_covar=tensor([0.0829, 0.0803, 0.1015, 0.0891, 0.0877, 0.0776, 0.0600, 0.0943], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-03 15:57:11,410 INFO [train.py:903] (1/4) Epoch 30, batch 1800, loss[loss=0.192, simple_loss=0.2774, pruned_loss=0.0533, over 19781.00 frames. ], tot_loss[loss=0.1984, simple_loss=0.2806, pruned_loss=0.05807, over 3815363.82 frames. ], batch size: 56, lr: 2.74e-03, grad_scale: 4.0 2023-04-03 15:57:21,123 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=199820.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 15:58:06,475 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=199856.0, num_to_drop=1, layers_to_drop={1} 2023-04-03 15:58:08,426 WARNING [train.py:1073] (1/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9_rvb from training. Duration: 25.0944375 2023-04-03 15:58:13,034 INFO [train.py:903] (1/4) Epoch 30, batch 1850, loss[loss=0.1704, simple_loss=0.2516, pruned_loss=0.0446, over 19404.00 frames. ], tot_loss[loss=0.1973, simple_loss=0.2797, pruned_loss=0.0575, over 3820649.14 frames. ], batch size: 48, lr: 2.74e-03, grad_scale: 4.0 2023-04-03 15:58:15,397 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.3363, 3.8425, 3.9621, 3.9641, 1.5640, 3.7871, 3.2847, 3.7388], device='cuda:1'), covar=tensor([0.1708, 0.0962, 0.0684, 0.0757, 0.5912, 0.0951, 0.0797, 0.1112], device='cuda:1'), in_proj_covar=tensor([0.0831, 0.0806, 0.1020, 0.0894, 0.0881, 0.0779, 0.0603, 0.0948], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-03 15:58:32,552 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=199877.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 15:58:46,819 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.213e+02 5.043e+02 6.338e+02 8.325e+02 2.069e+03, threshold=1.268e+03, percent-clipped=7.0 2023-04-03 15:58:46,854 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9_rvb from training. Duration: 27.8166875 2023-04-03 15:58:57,067 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=199897.0, num_to_drop=1, layers_to_drop={0} 2023-04-03 15:59:12,866 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=199909.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 15:59:15,994 INFO [train.py:903] (1/4) Epoch 30, batch 1900, loss[loss=0.1825, simple_loss=0.2731, pruned_loss=0.04598, over 19697.00 frames. ], tot_loss[loss=0.1988, simple_loss=0.2809, pruned_loss=0.0584, over 3811036.62 frames. ], batch size: 59, lr: 2.74e-03, grad_scale: 4.0 2023-04-03 15:59:33,278 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 2023-04-03 15:59:33,521 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=199927.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 15:59:38,032 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9_rvb from training. Duration: 27.02225 2023-04-03 15:59:42,751 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=199934.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 15:59:43,926 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=199935.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 16:00:03,960 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 2023-04-03 16:00:17,076 INFO [train.py:903] (1/4) Epoch 30, batch 1950, loss[loss=0.2318, simple_loss=0.3139, pruned_loss=0.07481, over 19340.00 frames. ], tot_loss[loss=0.199, simple_loss=0.281, pruned_loss=0.05851, over 3802666.44 frames. ], batch size: 66, lr: 2.74e-03, grad_scale: 4.0 2023-04-03 16:00:28,470 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=199971.0, num_to_drop=1, layers_to_drop={0} 2023-04-03 16:00:51,089 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.452e+02 5.211e+02 6.525e+02 7.685e+02 1.771e+03, threshold=1.305e+03, percent-clipped=2.0 2023-04-03 16:01:21,185 INFO [train.py:903] (1/4) Epoch 30, batch 2000, loss[loss=0.2156, simple_loss=0.295, pruned_loss=0.06816, over 19281.00 frames. ], tot_loss[loss=0.1987, simple_loss=0.2808, pruned_loss=0.05827, over 3807091.47 frames. ], batch size: 66, lr: 2.74e-03, grad_scale: 8.0 2023-04-03 16:01:21,557 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=200012.0, num_to_drop=1, layers_to_drop={0} 2023-04-03 16:02:21,484 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_rvb from training. Duration: 26.8349375 2023-04-03 16:02:23,760 INFO [train.py:903] (1/4) Epoch 30, batch 2050, loss[loss=0.2196, simple_loss=0.2981, pruned_loss=0.07056, over 17470.00 frames. ], tot_loss[loss=0.1983, simple_loss=0.2805, pruned_loss=0.05806, over 3814231.86 frames. ], batch size: 101, lr: 2.74e-03, grad_scale: 8.0 2023-04-03 16:02:43,234 WARNING [train.py:1073] (1/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 2023-04-03 16:02:43,266 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 2023-04-03 16:02:50,419 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.3251, 2.3371, 2.5311, 3.0086, 2.3530, 2.7847, 2.5275, 2.3337], device='cuda:1'), covar=tensor([0.4165, 0.3959, 0.1933, 0.2525, 0.4269, 0.2290, 0.4751, 0.3295], device='cuda:1'), in_proj_covar=tensor([0.0947, 0.1032, 0.0750, 0.0959, 0.0926, 0.0866, 0.0867, 0.0815], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-03 16:02:57,788 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.348e+02 4.986e+02 6.252e+02 8.206e+02 1.738e+03, threshold=1.250e+03, percent-clipped=4.0 2023-04-03 16:03:03,565 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 2023-04-03 16:03:11,190 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.35 vs. limit=5.0 2023-04-03 16:03:26,457 INFO [train.py:903] (1/4) Epoch 30, batch 2100, loss[loss=0.1832, simple_loss=0.2783, pruned_loss=0.04408, over 19533.00 frames. ], tot_loss[loss=0.1977, simple_loss=0.2799, pruned_loss=0.05779, over 3813445.77 frames. ], batch size: 56, lr: 2.74e-03, grad_scale: 8.0 2023-04-03 16:03:52,970 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=200133.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 16:03:57,147 WARNING [train.py:1073] (1/4) Exclude cut with ID 453-131332-0000-131866_sp0.9_rvb from training. Duration: 25.3333125 2023-04-03 16:04:18,959 WARNING [train.py:1073] (1/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9_rvb from training. Duration: 26.6166875 2023-04-03 16:04:24,950 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=200158.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 16:04:29,063 INFO [train.py:903] (1/4) Epoch 30, batch 2150, loss[loss=0.1955, simple_loss=0.2836, pruned_loss=0.0537, over 19683.00 frames. ], tot_loss[loss=0.1972, simple_loss=0.2794, pruned_loss=0.05748, over 3812405.61 frames. ], batch size: 60, lr: 2.74e-03, grad_scale: 8.0 2023-04-03 16:05:02,778 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.654e+02 4.819e+02 5.957e+02 8.119e+02 2.108e+03, threshold=1.191e+03, percent-clipped=2.0 2023-04-03 16:05:05,537 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=200191.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 16:05:10,671 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=200195.0, num_to_drop=1, layers_to_drop={1} 2023-04-03 16:05:30,902 INFO [train.py:903] (1/4) Epoch 30, batch 2200, loss[loss=0.2313, simple_loss=0.31, pruned_loss=0.07626, over 19518.00 frames. ], tot_loss[loss=0.197, simple_loss=0.2794, pruned_loss=0.05729, over 3831234.93 frames. ], batch size: 64, lr: 2.73e-03, grad_scale: 8.0 2023-04-03 16:05:37,176 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=200216.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 16:05:39,436 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.9043, 4.3901, 4.6498, 4.6573, 1.7233, 4.3638, 3.7963, 4.3509], device='cuda:1'), covar=tensor([0.1795, 0.0948, 0.0651, 0.0688, 0.6489, 0.1062, 0.0744, 0.1256], device='cuda:1'), in_proj_covar=tensor([0.0824, 0.0798, 0.1011, 0.0885, 0.0872, 0.0772, 0.0597, 0.0936], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-03 16:05:50,054 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=200227.0, num_to_drop=1, layers_to_drop={0} 2023-04-03 16:06:20,820 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=200252.0, num_to_drop=1, layers_to_drop={1} 2023-04-03 16:06:33,031 INFO [train.py:903] (1/4) Epoch 30, batch 2250, loss[loss=0.2293, simple_loss=0.3041, pruned_loss=0.07728, over 13369.00 frames. ], tot_loss[loss=0.199, simple_loss=0.2811, pruned_loss=0.05845, over 3808965.81 frames. ], batch size: 135, lr: 2.73e-03, grad_scale: 4.0 2023-04-03 16:06:41,174 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=200268.0, num_to_drop=1, layers_to_drop={0} 2023-04-03 16:06:44,363 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=200271.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 16:07:08,302 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.532e+02 5.298e+02 6.530e+02 8.599e+02 1.543e+03, threshold=1.306e+03, percent-clipped=6.0 2023-04-03 16:07:12,069 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=200293.0, num_to_drop=1, layers_to_drop={1} 2023-04-03 16:07:17,571 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=200297.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 16:07:35,303 INFO [train.py:903] (1/4) Epoch 30, batch 2300, loss[loss=0.1691, simple_loss=0.2485, pruned_loss=0.04491, over 19753.00 frames. ], tot_loss[loss=0.1996, simple_loss=0.2819, pruned_loss=0.05862, over 3824458.14 frames. ], batch size: 47, lr: 2.73e-03, grad_scale: 4.0 2023-04-03 16:07:51,099 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_rvb from training. Duration: 26.205 2023-04-03 16:08:18,932 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.41 vs. limit=2.0 2023-04-03 16:08:37,753 INFO [train.py:903] (1/4) Epoch 30, batch 2350, loss[loss=0.2081, simple_loss=0.2815, pruned_loss=0.06741, over 19582.00 frames. ], tot_loss[loss=0.1997, simple_loss=0.2819, pruned_loss=0.05878, over 3822576.36 frames. ], batch size: 52, lr: 2.73e-03, grad_scale: 4.0 2023-04-03 16:08:47,329 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.41 vs. limit=5.0 2023-04-03 16:09:07,632 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=200386.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 16:09:12,872 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.379e+02 4.836e+02 5.833e+02 7.163e+02 1.475e+03, threshold=1.167e+03, percent-clipped=1.0 2023-04-03 16:09:21,097 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_rvb from training. Duration: 25.775 2023-04-03 16:09:34,074 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.9953, 1.4104, 1.6285, 1.5209, 3.5879, 1.2328, 2.5042, 4.0033], device='cuda:1'), covar=tensor([0.0488, 0.2849, 0.2908, 0.2025, 0.0665, 0.2601, 0.1362, 0.0221], device='cuda:1'), in_proj_covar=tensor([0.0426, 0.0383, 0.0404, 0.0356, 0.0386, 0.0361, 0.0402, 0.0423], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 16:09:38,134 WARNING [train.py:1073] (1/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9_rvb from training. Duration: 25.45 2023-04-03 16:09:40,554 INFO [train.py:903] (1/4) Epoch 30, batch 2400, loss[loss=0.1947, simple_loss=0.2783, pruned_loss=0.05557, over 19598.00 frames. ], tot_loss[loss=0.2002, simple_loss=0.2821, pruned_loss=0.05918, over 3809325.05 frames. ], batch size: 57, lr: 2.73e-03, grad_scale: 8.0 2023-04-03 16:10:24,210 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.6760, 1.2860, 1.5308, 1.5380, 3.2531, 1.3356, 2.5306, 3.6291], device='cuda:1'), covar=tensor([0.0553, 0.3049, 0.3137, 0.2034, 0.0737, 0.2512, 0.1302, 0.0294], device='cuda:1'), in_proj_covar=tensor([0.0427, 0.0383, 0.0403, 0.0356, 0.0386, 0.0361, 0.0402, 0.0423], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 16:10:43,351 INFO [train.py:903] (1/4) Epoch 30, batch 2450, loss[loss=0.1945, simple_loss=0.2795, pruned_loss=0.05472, over 19700.00 frames. ], tot_loss[loss=0.1998, simple_loss=0.282, pruned_loss=0.05878, over 3822445.14 frames. ], batch size: 60, lr: 2.73e-03, grad_scale: 8.0 2023-04-03 16:11:19,105 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.119e+02 5.121e+02 5.946e+02 7.814e+02 2.121e+03, threshold=1.189e+03, percent-clipped=7.0 2023-04-03 16:11:26,427 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.8249, 3.2958, 3.3466, 3.3537, 1.3420, 3.2249, 2.8223, 3.1408], device='cuda:1'), covar=tensor([0.1759, 0.1028, 0.0844, 0.0997, 0.5805, 0.1177, 0.0910, 0.1323], device='cuda:1'), in_proj_covar=tensor([0.0819, 0.0795, 0.1006, 0.0883, 0.0868, 0.0770, 0.0593, 0.0935], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-03 16:11:46,465 INFO [train.py:903] (1/4) Epoch 30, batch 2500, loss[loss=0.1709, simple_loss=0.2498, pruned_loss=0.04605, over 19402.00 frames. ], tot_loss[loss=0.199, simple_loss=0.281, pruned_loss=0.05846, over 3822825.20 frames. ], batch size: 48, lr: 2.73e-03, grad_scale: 8.0 2023-04-03 16:12:20,429 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=200539.0, num_to_drop=1, layers_to_drop={1} 2023-04-03 16:12:48,792 INFO [train.py:903] (1/4) Epoch 30, batch 2550, loss[loss=0.1904, simple_loss=0.2703, pruned_loss=0.05526, over 19479.00 frames. ], tot_loss[loss=0.1985, simple_loss=0.2806, pruned_loss=0.0582, over 3815430.67 frames. ], batch size: 49, lr: 2.73e-03, grad_scale: 8.0 2023-04-03 16:13:23,054 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.619e+02 4.923e+02 6.163e+02 7.973e+02 2.573e+03, threshold=1.233e+03, percent-clipped=12.0 2023-04-03 16:13:41,801 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.0392, 3.6848, 2.7221, 3.3076, 0.8158, 3.6835, 3.5376, 3.6340], device='cuda:1'), covar=tensor([0.0786, 0.1197, 0.1808, 0.0975, 0.4058, 0.0764, 0.1016, 0.1213], device='cuda:1'), in_proj_covar=tensor([0.0533, 0.0431, 0.0517, 0.0358, 0.0409, 0.0458, 0.0449, 0.0487], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 16:13:47,180 WARNING [train.py:1073] (1/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 2023-04-03 16:13:51,615 INFO [train.py:903] (1/4) Epoch 30, batch 2600, loss[loss=0.1757, simple_loss=0.2537, pruned_loss=0.0488, over 19296.00 frames. ], tot_loss[loss=0.1986, simple_loss=0.2805, pruned_loss=0.05833, over 3809953.69 frames. ], batch size: 44, lr: 2.73e-03, grad_scale: 8.0 2023-04-03 16:14:28,347 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=200641.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 16:14:29,696 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=200642.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 16:14:44,533 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=200654.0, num_to_drop=1, layers_to_drop={1} 2023-04-03 16:14:54,163 INFO [train.py:903] (1/4) Epoch 30, batch 2650, loss[loss=0.1894, simple_loss=0.2794, pruned_loss=0.04975, over 19657.00 frames. ], tot_loss[loss=0.1991, simple_loss=0.2809, pruned_loss=0.05864, over 3811967.07 frames. ], batch size: 58, lr: 2.73e-03, grad_scale: 8.0 2023-04-03 16:14:55,567 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=200663.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 16:15:00,393 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=200667.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 16:15:12,706 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.8549, 1.3882, 1.5612, 1.6767, 3.4394, 1.3130, 2.4792, 3.8878], device='cuda:1'), covar=tensor([0.0493, 0.2857, 0.2947, 0.1793, 0.0673, 0.2498, 0.1329, 0.0224], device='cuda:1'), in_proj_covar=tensor([0.0425, 0.0382, 0.0402, 0.0355, 0.0386, 0.0361, 0.0402, 0.0422], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 16:15:15,798 WARNING [train.py:1073] (1/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9_rvb from training. Duration: 27.25 2023-04-03 16:15:28,264 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.276e+02 4.881e+02 6.412e+02 8.116e+02 1.737e+03, threshold=1.282e+03, percent-clipped=7.0 2023-04-03 16:15:55,161 INFO [train.py:903] (1/4) Epoch 30, batch 2700, loss[loss=0.1927, simple_loss=0.284, pruned_loss=0.05076, over 18059.00 frames. ], tot_loss[loss=0.1992, simple_loss=0.2812, pruned_loss=0.05857, over 3817929.66 frames. ], batch size: 83, lr: 2.73e-03, grad_scale: 8.0 2023-04-03 16:16:34,930 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 2023-04-03 16:16:51,799 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=200756.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 16:16:59,613 INFO [train.py:903] (1/4) Epoch 30, batch 2750, loss[loss=0.2297, simple_loss=0.3133, pruned_loss=0.07302, over 18241.00 frames. ], tot_loss[loss=0.199, simple_loss=0.281, pruned_loss=0.05843, over 3814749.50 frames. ], batch size: 84, lr: 2.73e-03, grad_scale: 8.0 2023-04-03 16:17:34,085 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.266e+02 4.594e+02 5.767e+02 6.844e+02 1.269e+03, threshold=1.153e+03, percent-clipped=0.0 2023-04-03 16:18:02,566 INFO [train.py:903] (1/4) Epoch 30, batch 2800, loss[loss=0.1823, simple_loss=0.2715, pruned_loss=0.0465, over 18740.00 frames. ], tot_loss[loss=0.1989, simple_loss=0.281, pruned_loss=0.05838, over 3809912.87 frames. ], batch size: 74, lr: 2.73e-03, grad_scale: 8.0 2023-04-03 16:19:05,167 INFO [train.py:903] (1/4) Epoch 30, batch 2850, loss[loss=0.1553, simple_loss=0.2441, pruned_loss=0.03324, over 19737.00 frames. ], tot_loss[loss=0.1988, simple_loss=0.2808, pruned_loss=0.05837, over 3805170.21 frames. ], batch size: 51, lr: 2.73e-03, grad_scale: 8.0 2023-04-03 16:19:39,283 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.089e+02 4.865e+02 5.683e+02 7.765e+02 1.857e+03, threshold=1.137e+03, percent-clipped=6.0 2023-04-03 16:20:04,681 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=200910.0, num_to_drop=1, layers_to_drop={0} 2023-04-03 16:20:06,533 INFO [train.py:903] (1/4) Epoch 30, batch 2900, loss[loss=0.2271, simple_loss=0.3114, pruned_loss=0.07135, over 19620.00 frames. ], tot_loss[loss=0.2002, simple_loss=0.2825, pruned_loss=0.05901, over 3803485.66 frames. ], batch size: 57, lr: 2.73e-03, grad_scale: 8.0 2023-04-03 16:20:08,647 WARNING [train.py:1073] (1/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 2023-04-03 16:20:36,311 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=200935.0, num_to_drop=1, layers_to_drop={0} 2023-04-03 16:21:00,030 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([0.9057, 1.3322, 1.0930, 0.9099, 1.1582, 0.9569, 0.9798, 1.2613], device='cuda:1'), covar=tensor([0.0682, 0.1002, 0.1235, 0.0917, 0.0672, 0.1496, 0.0690, 0.0566], device='cuda:1'), in_proj_covar=tensor([0.0305, 0.0321, 0.0345, 0.0277, 0.0255, 0.0350, 0.0291, 0.0276], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 16:21:08,764 INFO [train.py:903] (1/4) Epoch 30, batch 2950, loss[loss=0.2055, simple_loss=0.2961, pruned_loss=0.05745, over 19657.00 frames. ], tot_loss[loss=0.2006, simple_loss=0.283, pruned_loss=0.05913, over 3814483.69 frames. ], batch size: 58, lr: 2.73e-03, grad_scale: 8.0 2023-04-03 16:21:44,099 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.449e+02 4.844e+02 6.179e+02 7.399e+02 1.416e+03, threshold=1.236e+03, percent-clipped=4.0 2023-04-03 16:22:06,363 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=201007.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 16:22:11,911 INFO [train.py:903] (1/4) Epoch 30, batch 3000, loss[loss=0.1678, simple_loss=0.2436, pruned_loss=0.04594, over 19744.00 frames. ], tot_loss[loss=0.1998, simple_loss=0.2822, pruned_loss=0.05868, over 3825596.32 frames. ], batch size: 46, lr: 2.73e-03, grad_scale: 8.0 2023-04-03 16:22:11,912 INFO [train.py:928] (1/4) Computing validation loss 2023-04-03 16:22:26,179 INFO [train.py:937] (1/4) Epoch 30, validation: loss=0.1666, simple_loss=0.266, pruned_loss=0.03357, over 944034.00 frames. 2023-04-03 16:22:26,180 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18821MB 2023-04-03 16:22:26,628 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=201012.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 16:22:32,376 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_rvb from training. Duration: 29.735 2023-04-03 16:22:57,381 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=201037.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 16:23:27,828 INFO [train.py:903] (1/4) Epoch 30, batch 3050, loss[loss=0.2026, simple_loss=0.2769, pruned_loss=0.06417, over 19408.00 frames. ], tot_loss[loss=0.2001, simple_loss=0.2823, pruned_loss=0.05892, over 3825700.55 frames. ], batch size: 48, lr: 2.73e-03, grad_scale: 8.0 2023-04-03 16:24:02,286 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=201089.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 16:24:03,235 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.915e+02 4.896e+02 5.871e+02 7.483e+02 2.064e+03, threshold=1.174e+03, percent-clipped=5.0 2023-04-03 16:24:17,335 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.6041, 2.0287, 2.2271, 2.1377, 3.3564, 1.7307, 2.8854, 3.5530], device='cuda:1'), covar=tensor([0.0464, 0.2311, 0.2198, 0.1552, 0.0507, 0.2110, 0.1356, 0.0252], device='cuda:1'), in_proj_covar=tensor([0.0428, 0.0385, 0.0404, 0.0356, 0.0388, 0.0363, 0.0404, 0.0425], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 16:24:32,128 INFO [train.py:903] (1/4) Epoch 30, batch 3100, loss[loss=0.178, simple_loss=0.2698, pruned_loss=0.0431, over 19792.00 frames. ], tot_loss[loss=0.2007, simple_loss=0.283, pruned_loss=0.05918, over 3818232.19 frames. ], batch size: 56, lr: 2.73e-03, grad_scale: 4.0 2023-04-03 16:24:38,535 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.84 vs. limit=2.0 2023-04-03 16:24:43,602 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=201122.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 16:25:33,506 INFO [train.py:903] (1/4) Epoch 30, batch 3150, loss[loss=0.1617, simple_loss=0.2518, pruned_loss=0.03586, over 19721.00 frames. ], tot_loss[loss=0.2011, simple_loss=0.2831, pruned_loss=0.05952, over 3809427.13 frames. ], batch size: 51, lr: 2.73e-03, grad_scale: 4.0 2023-04-03 16:25:36,007 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.9343, 4.4539, 2.7175, 3.9440, 0.8253, 4.5093, 4.3319, 4.4822], device='cuda:1'), covar=tensor([0.0568, 0.0958, 0.2132, 0.0876, 0.4390, 0.0638, 0.0990, 0.1240], device='cuda:1'), in_proj_covar=tensor([0.0537, 0.0433, 0.0521, 0.0360, 0.0411, 0.0460, 0.0453, 0.0490], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 16:26:02,508 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 2023-04-03 16:26:10,328 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.152e+02 4.625e+02 5.752e+02 7.402e+02 1.953e+03, threshold=1.150e+03, percent-clipped=5.0 2023-04-03 16:26:35,924 INFO [train.py:903] (1/4) Epoch 30, batch 3200, loss[loss=0.1501, simple_loss=0.2322, pruned_loss=0.03397, over 18499.00 frames. ], tot_loss[loss=0.2006, simple_loss=0.2826, pruned_loss=0.05934, over 3815049.62 frames. ], batch size: 41, lr: 2.73e-03, grad_scale: 8.0 2023-04-03 16:27:32,993 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.51 vs. limit=2.0 2023-04-03 16:27:40,086 INFO [train.py:903] (1/4) Epoch 30, batch 3250, loss[loss=0.1982, simple_loss=0.2831, pruned_loss=0.05667, over 19599.00 frames. ], tot_loss[loss=0.1993, simple_loss=0.2811, pruned_loss=0.05878, over 3820617.64 frames. ], batch size: 57, lr: 2.73e-03, grad_scale: 8.0 2023-04-03 16:27:51,496 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=201271.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 16:28:14,669 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.43 vs. limit=5.0 2023-04-03 16:28:14,872 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.466e+02 4.729e+02 5.856e+02 7.119e+02 1.424e+03, threshold=1.171e+03, percent-clipped=4.0 2023-04-03 16:28:16,380 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=201292.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 16:28:42,755 INFO [train.py:903] (1/4) Epoch 30, batch 3300, loss[loss=0.1918, simple_loss=0.2654, pruned_loss=0.05906, over 19620.00 frames. ], tot_loss[loss=0.199, simple_loss=0.2806, pruned_loss=0.05871, over 3826685.93 frames. ], batch size: 50, lr: 2.73e-03, grad_scale: 8.0 2023-04-03 16:28:49,687 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 2023-04-03 16:28:53,526 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.1571, 2.0854, 1.9411, 1.7804, 1.7309, 1.7889, 0.5864, 1.0924], device='cuda:1'), covar=tensor([0.0711, 0.0639, 0.0514, 0.0820, 0.1215, 0.0953, 0.1408, 0.1204], device='cuda:1'), in_proj_covar=tensor([0.0368, 0.0364, 0.0370, 0.0393, 0.0472, 0.0399, 0.0347, 0.0350], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 16:29:03,926 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.1357, 3.2547, 1.7814, 1.9508, 2.9385, 1.6191, 1.4847, 2.3348], device='cuda:1'), covar=tensor([0.1414, 0.0641, 0.1242, 0.0964, 0.0525, 0.1403, 0.1049, 0.0703], device='cuda:1'), in_proj_covar=tensor([0.0303, 0.0320, 0.0344, 0.0276, 0.0254, 0.0348, 0.0291, 0.0276], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 16:29:44,706 INFO [train.py:903] (1/4) Epoch 30, batch 3350, loss[loss=0.1842, simple_loss=0.2684, pruned_loss=0.04999, over 19597.00 frames. ], tot_loss[loss=0.1983, simple_loss=0.28, pruned_loss=0.05831, over 3824200.41 frames. ], batch size: 61, lr: 2.73e-03, grad_scale: 8.0 2023-04-03 16:30:05,887 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=201378.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 16:30:07,531 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 2023-04-03 16:30:21,792 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.184e+02 4.772e+02 5.759e+02 7.012e+02 1.305e+03, threshold=1.152e+03, percent-clipped=2.0 2023-04-03 16:30:37,302 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=201403.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 16:30:48,352 INFO [train.py:903] (1/4) Epoch 30, batch 3400, loss[loss=0.2363, simple_loss=0.3184, pruned_loss=0.07714, over 19603.00 frames. ], tot_loss[loss=0.1985, simple_loss=0.2802, pruned_loss=0.05834, over 3824484.35 frames. ], batch size: 57, lr: 2.73e-03, grad_scale: 8.0 2023-04-03 16:31:16,063 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=201433.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 16:31:51,817 INFO [train.py:903] (1/4) Epoch 30, batch 3450, loss[loss=0.221, simple_loss=0.3023, pruned_loss=0.06981, over 19684.00 frames. ], tot_loss[loss=0.1981, simple_loss=0.2803, pruned_loss=0.05796, over 3841480.54 frames. ], batch size: 55, lr: 2.73e-03, grad_scale: 8.0 2023-04-03 16:31:57,365 WARNING [train.py:1073] (1/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9_rvb from training. Duration: 25.2444375 2023-04-03 16:32:27,319 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.371e+02 5.269e+02 6.567e+02 8.372e+02 2.121e+03, threshold=1.313e+03, percent-clipped=9.0 2023-04-03 16:32:55,652 INFO [train.py:903] (1/4) Epoch 30, batch 3500, loss[loss=0.2041, simple_loss=0.293, pruned_loss=0.0576, over 17414.00 frames. ], tot_loss[loss=0.1979, simple_loss=0.2799, pruned_loss=0.05795, over 3834687.78 frames. ], batch size: 101, lr: 2.73e-03, grad_scale: 8.0 2023-04-03 16:33:18,879 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.13 vs. limit=2.0 2023-04-03 16:33:25,393 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.6673, 4.2734, 2.9099, 3.8031, 1.0373, 4.2416, 4.0842, 4.1758], device='cuda:1'), covar=tensor([0.0605, 0.0898, 0.1709, 0.0831, 0.3801, 0.0632, 0.0956, 0.0965], device='cuda:1'), in_proj_covar=tensor([0.0539, 0.0436, 0.0522, 0.0361, 0.0412, 0.0462, 0.0455, 0.0493], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 16:33:41,239 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=201548.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 16:33:58,202 INFO [train.py:903] (1/4) Epoch 30, batch 3550, loss[loss=0.2392, simple_loss=0.3067, pruned_loss=0.08587, over 14359.00 frames. ], tot_loss[loss=0.1988, simple_loss=0.281, pruned_loss=0.05827, over 3822661.64 frames. ], batch size: 141, lr: 2.73e-03, grad_scale: 8.0 2023-04-03 16:34:35,157 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.339e+02 4.645e+02 6.006e+02 7.208e+02 1.141e+03, threshold=1.201e+03, percent-clipped=0.0 2023-04-03 16:35:01,950 INFO [train.py:903] (1/4) Epoch 30, batch 3600, loss[loss=0.1857, simple_loss=0.2738, pruned_loss=0.0488, over 19516.00 frames. ], tot_loss[loss=0.1996, simple_loss=0.2818, pruned_loss=0.05864, over 3822406.45 frames. ], batch size: 54, lr: 2.73e-03, grad_scale: 8.0 2023-04-03 16:35:06,948 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=201615.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 16:35:32,742 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=201636.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 16:36:04,973 INFO [train.py:903] (1/4) Epoch 30, batch 3650, loss[loss=0.1884, simple_loss=0.2813, pruned_loss=0.04777, over 19769.00 frames. ], tot_loss[loss=0.1991, simple_loss=0.2816, pruned_loss=0.05832, over 3831383.29 frames. ], batch size: 54, lr: 2.72e-03, grad_scale: 8.0 2023-04-03 16:36:05,292 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=201662.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 16:36:27,551 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=201679.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 16:36:42,038 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.871e+02 4.844e+02 5.974e+02 7.556e+02 1.962e+03, threshold=1.195e+03, percent-clipped=4.0 2023-04-03 16:37:09,552 INFO [train.py:903] (1/4) Epoch 30, batch 3700, loss[loss=0.1769, simple_loss=0.2544, pruned_loss=0.04969, over 19759.00 frames. ], tot_loss[loss=0.1978, simple_loss=0.2801, pruned_loss=0.05776, over 3830990.76 frames. ], batch size: 47, lr: 2.72e-03, grad_scale: 8.0 2023-04-03 16:37:15,816 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9404, 2.0248, 1.9110, 3.0698, 2.2435, 2.9132, 1.9804, 1.6942], device='cuda:1'), covar=tensor([0.5377, 0.5372, 0.3194, 0.3325, 0.5366, 0.2695, 0.7169, 0.5601], device='cuda:1'), in_proj_covar=tensor([0.0953, 0.1036, 0.0755, 0.0962, 0.0929, 0.0870, 0.0869, 0.0820], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-03 16:37:19,156 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5834, 1.6363, 1.9401, 1.8297, 2.6938, 2.4034, 2.9419, 1.2099], device='cuda:1'), covar=tensor([0.2547, 0.4409, 0.2759, 0.1954, 0.1551, 0.2190, 0.1473, 0.4971], device='cuda:1'), in_proj_covar=tensor([0.0557, 0.0677, 0.0764, 0.0515, 0.0637, 0.0551, 0.0670, 0.0579], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 16:37:23,545 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=201724.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 16:37:31,452 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=201730.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 16:37:57,731 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=201751.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 16:37:57,783 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9095, 1.8772, 1.7160, 1.5068, 1.3714, 1.4331, 0.5566, 0.8205], device='cuda:1'), covar=tensor([0.0987, 0.0924, 0.0559, 0.1001, 0.1748, 0.1320, 0.1646, 0.1603], device='cuda:1'), in_proj_covar=tensor([0.0369, 0.0366, 0.0371, 0.0395, 0.0473, 0.0401, 0.0348, 0.0351], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 16:38:11,302 INFO [train.py:903] (1/4) Epoch 30, batch 3750, loss[loss=0.2422, simple_loss=0.3136, pruned_loss=0.08543, over 13825.00 frames. ], tot_loss[loss=0.1991, simple_loss=0.2813, pruned_loss=0.05843, over 3806972.83 frames. ], batch size: 136, lr: 2.72e-03, grad_scale: 8.0 2023-04-03 16:38:47,453 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.241e+02 4.921e+02 6.534e+02 8.381e+02 2.079e+03, threshold=1.307e+03, percent-clipped=7.0 2023-04-03 16:39:04,313 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=201804.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 16:39:14,010 INFO [train.py:903] (1/4) Epoch 30, batch 3800, loss[loss=0.2193, simple_loss=0.3064, pruned_loss=0.0661, over 19782.00 frames. ], tot_loss[loss=0.2, simple_loss=0.2822, pruned_loss=0.05891, over 3791862.44 frames. ], batch size: 56, lr: 2.72e-03, grad_scale: 8.0 2023-04-03 16:39:22,284 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.3996, 2.4618, 2.2598, 2.5738, 2.3767, 2.0950, 2.2387, 2.4015], device='cuda:1'), covar=tensor([0.0882, 0.1170, 0.1125, 0.0819, 0.1095, 0.0493, 0.1092, 0.0615], device='cuda:1'), in_proj_covar=tensor([0.0280, 0.0358, 0.0319, 0.0259, 0.0309, 0.0258, 0.0323, 0.0264], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-04-03 16:39:35,640 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=201829.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 16:39:41,267 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9_rvb from training. Duration: 29.1166875 2023-04-03 16:40:13,005 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=2.25 vs. limit=5.0 2023-04-03 16:40:15,944 INFO [train.py:903] (1/4) Epoch 30, batch 3850, loss[loss=0.1775, simple_loss=0.2543, pruned_loss=0.05031, over 19325.00 frames. ], tot_loss[loss=0.1997, simple_loss=0.2818, pruned_loss=0.05883, over 3798388.70 frames. ], batch size: 44, lr: 2.72e-03, grad_scale: 8.0 2023-04-03 16:40:51,661 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.891e+02 4.977e+02 6.719e+02 8.916e+02 2.147e+03, threshold=1.344e+03, percent-clipped=8.0 2023-04-03 16:41:12,702 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.4729, 1.4084, 1.3920, 1.7756, 1.3617, 1.6244, 1.6767, 1.4806], device='cuda:1'), covar=tensor([0.0869, 0.0921, 0.1047, 0.0660, 0.0846, 0.0760, 0.0817, 0.0752], device='cuda:1'), in_proj_covar=tensor([0.0212, 0.0222, 0.0227, 0.0239, 0.0227, 0.0215, 0.0187, 0.0207], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-04-03 16:41:18,294 INFO [train.py:903] (1/4) Epoch 30, batch 3900, loss[loss=0.1772, simple_loss=0.2494, pruned_loss=0.05249, over 16827.00 frames. ], tot_loss[loss=0.1989, simple_loss=0.2811, pruned_loss=0.05835, over 3813778.98 frames. ], batch size: 37, lr: 2.72e-03, grad_scale: 8.0 2023-04-03 16:42:20,922 INFO [train.py:903] (1/4) Epoch 30, batch 3950, loss[loss=0.1767, simple_loss=0.2558, pruned_loss=0.0488, over 19301.00 frames. ], tot_loss[loss=0.1993, simple_loss=0.2812, pruned_loss=0.05868, over 3799718.99 frames. ], batch size: 44, lr: 2.72e-03, grad_scale: 8.0 2023-04-03 16:42:20,947 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 2023-04-03 16:42:50,970 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.9385, 1.3445, 1.5779, 1.6062, 3.5297, 1.2455, 2.6019, 4.0176], device='cuda:1'), covar=tensor([0.0472, 0.3044, 0.3106, 0.1927, 0.0660, 0.2708, 0.1411, 0.0207], device='cuda:1'), in_proj_covar=tensor([0.0431, 0.0386, 0.0405, 0.0358, 0.0390, 0.0365, 0.0405, 0.0428], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 16:42:51,087 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=201986.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 16:42:56,903 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.990e+02 4.814e+02 5.725e+02 7.208e+02 1.816e+03, threshold=1.145e+03, percent-clipped=2.0 2023-04-03 16:43:16,592 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=202006.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 16:43:18,076 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=202007.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 16:43:19,112 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=202008.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 16:43:23,654 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=202011.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 16:43:24,451 INFO [train.py:903] (1/4) Epoch 30, batch 4000, loss[loss=0.2254, simple_loss=0.3033, pruned_loss=0.07374, over 19510.00 frames. ], tot_loss[loss=0.1991, simple_loss=0.2812, pruned_loss=0.05856, over 3808961.23 frames. ], batch size: 64, lr: 2.72e-03, grad_scale: 8.0 2023-04-03 16:43:38,594 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=202023.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 16:43:49,621 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=202032.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 16:44:08,204 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1_rvb from training. Duration: 0.7545625 2023-04-03 16:44:27,670 INFO [train.py:903] (1/4) Epoch 30, batch 4050, loss[loss=0.1797, simple_loss=0.2655, pruned_loss=0.04694, over 19847.00 frames. ], tot_loss[loss=0.1983, simple_loss=0.2803, pruned_loss=0.05813, over 3819593.17 frames. ], batch size: 52, lr: 2.72e-03, grad_scale: 8.0 2023-04-03 16:44:34,770 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=202068.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 16:45:02,977 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.018e+02 4.674e+02 6.115e+02 7.260e+02 2.667e+03, threshold=1.223e+03, percent-clipped=7.0 2023-04-03 16:45:25,459 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.9718, 2.0778, 2.3537, 2.6202, 2.0520, 2.5178, 2.3005, 2.0816], device='cuda:1'), covar=tensor([0.4293, 0.4048, 0.2007, 0.2547, 0.4213, 0.2326, 0.5048, 0.3532], device='cuda:1'), in_proj_covar=tensor([0.0952, 0.1035, 0.0752, 0.0962, 0.0929, 0.0870, 0.0868, 0.0818], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-03 16:45:30,403 INFO [train.py:903] (1/4) Epoch 30, batch 4100, loss[loss=0.2055, simple_loss=0.2915, pruned_loss=0.05973, over 19736.00 frames. ], tot_loss[loss=0.1989, simple_loss=0.2807, pruned_loss=0.05855, over 3814529.44 frames. ], batch size: 63, lr: 2.72e-03, grad_scale: 8.0 2023-04-03 16:45:40,905 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=202121.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 16:46:03,060 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=202138.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 16:46:04,985 WARNING [train.py:1073] (1/4) Exclude cut with ID 497-129325-0061-9566_sp1.1_rvb from training. Duration: 0.97725 2023-04-03 16:46:32,899 INFO [train.py:903] (1/4) Epoch 30, batch 4150, loss[loss=0.1788, simple_loss=0.26, pruned_loss=0.04878, over 19836.00 frames. ], tot_loss[loss=0.1983, simple_loss=0.2803, pruned_loss=0.05819, over 3820647.29 frames. ], batch size: 52, lr: 2.72e-03, grad_scale: 8.0 2023-04-03 16:46:59,287 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=202183.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 16:47:08,854 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.767e+02 4.823e+02 5.653e+02 7.013e+02 1.196e+03, threshold=1.131e+03, percent-clipped=0.0 2023-04-03 16:47:34,526 INFO [train.py:903] (1/4) Epoch 30, batch 4200, loss[loss=0.1936, simple_loss=0.2677, pruned_loss=0.05968, over 19838.00 frames. ], tot_loss[loss=0.1972, simple_loss=0.2795, pruned_loss=0.05744, over 3833710.03 frames. ], batch size: 52, lr: 2.72e-03, grad_scale: 8.0 2023-04-03 16:47:37,782 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 2023-04-03 16:47:59,111 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 2023-04-03 16:48:35,838 INFO [train.py:903] (1/4) Epoch 30, batch 4250, loss[loss=0.1951, simple_loss=0.284, pruned_loss=0.05307, over 19727.00 frames. ], tot_loss[loss=0.198, simple_loss=0.2803, pruned_loss=0.05788, over 3832094.03 frames. ], batch size: 63, lr: 2.72e-03, grad_scale: 8.0 2023-04-03 16:48:52,002 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9_rvb from training. Duration: 29.816625 2023-04-03 16:49:03,275 WARNING [train.py:1073] (1/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 2023-04-03 16:49:12,500 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.375e+02 5.334e+02 6.771e+02 9.277e+02 2.113e+03, threshold=1.354e+03, percent-clipped=7.0 2023-04-03 16:49:31,143 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 2023-04-03 16:49:38,444 INFO [train.py:903] (1/4) Epoch 30, batch 4300, loss[loss=0.2133, simple_loss=0.2939, pruned_loss=0.06631, over 19683.00 frames. ], tot_loss[loss=0.1994, simple_loss=0.2814, pruned_loss=0.05866, over 3831981.12 frames. ], batch size: 58, lr: 2.72e-03, grad_scale: 8.0 2023-04-03 16:50:16,160 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([4.0348, 3.6480, 2.5303, 3.2206, 0.7669, 3.6505, 3.5378, 3.5849], device='cuda:1'), covar=tensor([0.0792, 0.1158, 0.2028, 0.1063, 0.4010, 0.0781, 0.1054, 0.1262], device='cuda:1'), in_proj_covar=tensor([0.0540, 0.0439, 0.0524, 0.0364, 0.0416, 0.0463, 0.0458, 0.0494], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 16:50:28,703 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=202352.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 16:50:34,342 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 2023-04-03 16:50:41,233 INFO [train.py:903] (1/4) Epoch 30, batch 4350, loss[loss=0.1805, simple_loss=0.259, pruned_loss=0.05098, over 19735.00 frames. ], tot_loss[loss=0.2001, simple_loss=0.2823, pruned_loss=0.05898, over 3811280.92 frames. ], batch size: 46, lr: 2.72e-03, grad_scale: 4.0 2023-04-03 16:50:59,661 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=202377.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 16:51:03,967 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.7031, 1.6416, 1.4767, 1.7004, 1.5786, 1.4244, 1.4081, 1.6089], device='cuda:1'), covar=tensor([0.1031, 0.1342, 0.1377, 0.0928, 0.1277, 0.0682, 0.1579, 0.0759], device='cuda:1'), in_proj_covar=tensor([0.0283, 0.0361, 0.0322, 0.0262, 0.0312, 0.0260, 0.0326, 0.0266], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-04-03 16:51:18,785 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.299e+02 4.712e+02 5.570e+02 7.585e+02 1.545e+03, threshold=1.114e+03, percent-clipped=3.0 2023-04-03 16:51:21,605 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=202394.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 16:51:31,129 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=202402.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 16:51:43,046 INFO [train.py:903] (1/4) Epoch 30, batch 4400, loss[loss=0.1769, simple_loss=0.2567, pruned_loss=0.04855, over 19822.00 frames. ], tot_loss[loss=0.2004, simple_loss=0.2821, pruned_loss=0.05933, over 3802804.52 frames. ], batch size: 52, lr: 2.72e-03, grad_scale: 8.0 2023-04-03 16:51:52,553 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=202419.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 16:52:08,203 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_rvb from training. Duration: 25.285 2023-04-03 16:52:16,996 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 2023-04-03 16:52:17,423 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=202439.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 16:52:46,168 INFO [train.py:903] (1/4) Epoch 30, batch 4450, loss[loss=0.1973, simple_loss=0.2865, pruned_loss=0.05411, over 19606.00 frames. ], tot_loss[loss=0.1998, simple_loss=0.2819, pruned_loss=0.05884, over 3813565.97 frames. ], batch size: 57, lr: 2.72e-03, grad_scale: 8.0 2023-04-03 16:52:46,657 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.8612, 1.9956, 2.2392, 2.4606, 1.8779, 2.3225, 2.2164, 2.0396], device='cuda:1'), covar=tensor([0.4369, 0.4191, 0.2088, 0.2603, 0.4365, 0.2481, 0.5270, 0.3685], device='cuda:1'), in_proj_covar=tensor([0.0953, 0.1035, 0.0753, 0.0963, 0.0929, 0.0871, 0.0867, 0.0818], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-03 16:52:48,917 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=202464.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 16:52:52,458 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=202467.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 16:53:23,818 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.801e+02 4.882e+02 5.921e+02 7.342e+02 1.295e+03, threshold=1.184e+03, percent-clipped=2.0 2023-04-03 16:53:48,733 INFO [train.py:903] (1/4) Epoch 30, batch 4500, loss[loss=0.195, simple_loss=0.2797, pruned_loss=0.05516, over 19523.00 frames. ], tot_loss[loss=0.2, simple_loss=0.282, pruned_loss=0.059, over 3815630.97 frames. ], batch size: 56, lr: 2.72e-03, grad_scale: 8.0 2023-04-03 16:54:32,461 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.7900, 1.9072, 2.2290, 2.3724, 1.8067, 2.2709, 2.1813, 2.0112], device='cuda:1'), covar=tensor([0.4480, 0.4036, 0.2110, 0.2450, 0.4223, 0.2266, 0.5498, 0.3783], device='cuda:1'), in_proj_covar=tensor([0.0953, 0.1036, 0.0753, 0.0964, 0.0929, 0.0871, 0.0868, 0.0820], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-03 16:54:50,682 INFO [train.py:903] (1/4) Epoch 30, batch 4550, loss[loss=0.2351, simple_loss=0.32, pruned_loss=0.07507, over 18231.00 frames. ], tot_loss[loss=0.1994, simple_loss=0.2814, pruned_loss=0.0587, over 3827007.39 frames. ], batch size: 83, lr: 2.72e-03, grad_scale: 8.0 2023-04-03 16:55:01,218 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9_rvb from training. Duration: 28.72225 2023-04-03 16:55:25,942 WARNING [train.py:1073] (1/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 2023-04-03 16:55:27,102 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.113e+02 5.006e+02 6.254e+02 8.173e+02 1.576e+03, threshold=1.251e+03, percent-clipped=3.0 2023-04-03 16:55:52,912 INFO [train.py:903] (1/4) Epoch 30, batch 4600, loss[loss=0.2147, simple_loss=0.2968, pruned_loss=0.06627, over 18708.00 frames. ], tot_loss[loss=0.201, simple_loss=0.2826, pruned_loss=0.05972, over 3810815.26 frames. ], batch size: 74, lr: 2.72e-03, grad_scale: 8.0 2023-04-03 16:56:22,227 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.59 vs. limit=5.0 2023-04-03 16:56:51,357 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([5.6896, 5.2634, 3.1725, 4.5791, 1.5515, 5.2349, 5.1255, 5.2546], device='cuda:1'), covar=tensor([0.0402, 0.0752, 0.1768, 0.0725, 0.3426, 0.0556, 0.0877, 0.1081], device='cuda:1'), in_proj_covar=tensor([0.0537, 0.0436, 0.0520, 0.0361, 0.0412, 0.0460, 0.0454, 0.0491], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 16:56:54,623 INFO [train.py:903] (1/4) Epoch 30, batch 4650, loss[loss=0.1955, simple_loss=0.2822, pruned_loss=0.05436, over 19681.00 frames. ], tot_loss[loss=0.2004, simple_loss=0.2821, pruned_loss=0.05933, over 3815606.67 frames. ], batch size: 60, lr: 2.72e-03, grad_scale: 8.0 2023-04-03 16:57:12,732 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_rvb from training. Duration: 0.92 2023-04-03 16:57:23,001 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_rvb from training. Duration: 0.83 2023-04-03 16:57:31,775 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.354e+02 5.029e+02 5.947e+02 7.618e+02 1.686e+03, threshold=1.189e+03, percent-clipped=3.0 2023-04-03 16:57:55,569 INFO [train.py:903] (1/4) Epoch 30, batch 4700, loss[loss=0.1546, simple_loss=0.234, pruned_loss=0.03767, over 19306.00 frames. ], tot_loss[loss=0.2008, simple_loss=0.2823, pruned_loss=0.05964, over 3812433.50 frames. ], batch size: 44, lr: 2.72e-03, grad_scale: 8.0 2023-04-03 16:58:10,506 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=202723.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 16:58:18,348 WARNING [train.py:1073] (1/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9_rvb from training. Duration: 0.92225 2023-04-03 16:58:41,363 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=202748.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 16:58:59,214 INFO [train.py:903] (1/4) Epoch 30, batch 4750, loss[loss=0.2089, simple_loss=0.291, pruned_loss=0.06342, over 19683.00 frames. ], tot_loss[loss=0.1989, simple_loss=0.2806, pruned_loss=0.0586, over 3828170.81 frames. ], batch size: 60, lr: 2.72e-03, grad_scale: 8.0 2023-04-03 16:59:35,969 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.278e+02 5.055e+02 5.959e+02 7.636e+02 1.705e+03, threshold=1.192e+03, percent-clipped=6.0 2023-04-03 17:00:01,991 INFO [train.py:903] (1/4) Epoch 30, batch 4800, loss[loss=0.2202, simple_loss=0.3027, pruned_loss=0.06885, over 19608.00 frames. ], tot_loss[loss=0.1987, simple_loss=0.2807, pruned_loss=0.05839, over 3825474.47 frames. ], batch size: 57, lr: 2.72e-03, grad_scale: 8.0 2023-04-03 17:00:38,738 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.14 vs. limit=2.0 2023-04-03 17:00:47,792 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=202848.0, num_to_drop=1, layers_to_drop={1} 2023-04-03 17:01:03,238 INFO [train.py:903] (1/4) Epoch 30, batch 4850, loss[loss=0.1858, simple_loss=0.2725, pruned_loss=0.04956, over 19694.00 frames. ], tot_loss[loss=0.1988, simple_loss=0.2808, pruned_loss=0.05844, over 3829357.00 frames. ], batch size: 59, lr: 2.72e-03, grad_scale: 8.0 2023-04-03 17:01:27,935 WARNING [train.py:1073] (1/4) Exclude cut with ID 774-127930-0014-48411_sp1.1_rvb from training. Duration: 0.95 2023-04-03 17:01:41,380 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.446e+02 5.041e+02 6.050e+02 7.335e+02 2.031e+03, threshold=1.210e+03, percent-clipped=4.0 2023-04-03 17:01:47,266 WARNING [train.py:1073] (1/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1_rvb from training. Duration: 0.9409375 2023-04-03 17:01:54,131 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 2023-04-03 17:01:55,281 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 2023-04-03 17:02:03,482 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1_rvb from training. Duration: 25.3818125 2023-04-03 17:02:05,916 INFO [train.py:903] (1/4) Epoch 30, batch 4900, loss[loss=0.1942, simple_loss=0.2842, pruned_loss=0.05209, over 19667.00 frames. ], tot_loss[loss=0.1995, simple_loss=0.2814, pruned_loss=0.05879, over 3816657.53 frames. ], batch size: 58, lr: 2.72e-03, grad_scale: 8.0 2023-04-03 17:02:12,231 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.5979, 2.9979, 3.0447, 3.0633, 1.3750, 2.9110, 2.5843, 2.8682], device='cuda:1'), covar=tensor([0.1717, 0.1588, 0.0871, 0.1002, 0.5533, 0.1385, 0.0839, 0.1271], device='cuda:1'), in_proj_covar=tensor([0.0825, 0.0805, 0.1013, 0.0890, 0.0876, 0.0778, 0.0601, 0.0943], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-03 17:02:25,011 WARNING [train.py:1073] (1/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 2023-04-03 17:02:54,167 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.88 vs. limit=5.0 2023-04-03 17:03:07,819 INFO [train.py:903] (1/4) Epoch 30, batch 4950, loss[loss=0.1969, simple_loss=0.2864, pruned_loss=0.0537, over 19583.00 frames. ], tot_loss[loss=0.2016, simple_loss=0.2833, pruned_loss=0.05998, over 3814272.04 frames. ], batch size: 61, lr: 2.72e-03, grad_scale: 8.0 2023-04-03 17:03:22,384 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 2023-04-03 17:03:43,838 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.897e+02 4.771e+02 6.104e+02 8.250e+02 2.222e+03, threshold=1.221e+03, percent-clipped=2.0 2023-04-03 17:03:46,196 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 2023-04-03 17:03:53,451 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.50 vs. limit=2.0 2023-04-03 17:04:09,794 INFO [train.py:903] (1/4) Epoch 30, batch 5000, loss[loss=0.1934, simple_loss=0.283, pruned_loss=0.05189, over 19606.00 frames. ], tot_loss[loss=0.2015, simple_loss=0.2834, pruned_loss=0.05983, over 3817068.26 frames. ], batch size: 57, lr: 2.72e-03, grad_scale: 8.0 2023-04-03 17:04:16,543 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_rvb from training. Duration: 27.14 2023-04-03 17:04:18,986 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.9712, 1.4708, 1.5957, 1.5427, 3.5751, 1.1087, 2.7164, 4.0089], device='cuda:1'), covar=tensor([0.0410, 0.2795, 0.2902, 0.1897, 0.0635, 0.2600, 0.1158, 0.0215], device='cuda:1'), in_proj_covar=tensor([0.0429, 0.0385, 0.0404, 0.0358, 0.0389, 0.0363, 0.0403, 0.0426], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 17:04:28,733 WARNING [train.py:1073] (1/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 2023-04-03 17:05:02,498 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=203056.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 17:05:02,579 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.2045, 1.8763, 2.0552, 2.8518, 1.9751, 2.4791, 2.2657, 2.3326], device='cuda:1'), covar=tensor([0.0794, 0.0898, 0.0915, 0.0770, 0.0877, 0.0747, 0.0910, 0.0627], device='cuda:1'), in_proj_covar=tensor([0.0213, 0.0224, 0.0228, 0.0239, 0.0227, 0.0216, 0.0188, 0.0208], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-04-03 17:05:09,007 INFO [train.py:903] (1/4) Epoch 30, batch 5050, loss[loss=0.19, simple_loss=0.263, pruned_loss=0.05855, over 19732.00 frames. ], tot_loss[loss=0.2011, simple_loss=0.2829, pruned_loss=0.0597, over 3828311.68 frames. ], batch size: 45, lr: 2.72e-03, grad_scale: 8.0 2023-04-03 17:05:17,041 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.8792, 1.4304, 1.5613, 1.6918, 3.4676, 1.2536, 2.4727, 3.9469], device='cuda:1'), covar=tensor([0.0478, 0.2911, 0.2973, 0.1911, 0.0683, 0.2533, 0.1426, 0.0211], device='cuda:1'), in_proj_covar=tensor([0.0428, 0.0384, 0.0403, 0.0357, 0.0388, 0.0362, 0.0403, 0.0425], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 17:05:45,249 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9_rvb from training. Duration: 26.62775 2023-04-03 17:05:46,417 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.400e+02 4.511e+02 5.411e+02 6.959e+02 2.884e+03, threshold=1.082e+03, percent-clipped=1.0 2023-04-03 17:05:59,231 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=203103.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 17:06:10,216 INFO [train.py:903] (1/4) Epoch 30, batch 5100, loss[loss=0.2082, simple_loss=0.3007, pruned_loss=0.05789, over 19523.00 frames. ], tot_loss[loss=0.2004, simple_loss=0.2822, pruned_loss=0.05933, over 3829876.45 frames. ], batch size: 56, lr: 2.72e-03, grad_scale: 8.0 2023-04-03 17:06:23,802 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 2023-04-03 17:06:27,355 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0005-9467_rvb from training. Duration: 25.035 2023-04-03 17:06:30,779 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_rvb from training. Duration: 27.92 2023-04-03 17:07:12,616 INFO [train.py:903] (1/4) Epoch 30, batch 5150, loss[loss=0.1938, simple_loss=0.2735, pruned_loss=0.05708, over 19858.00 frames. ], tot_loss[loss=0.1992, simple_loss=0.2813, pruned_loss=0.05857, over 3839578.65 frames. ], batch size: 52, lr: 2.71e-03, grad_scale: 8.0 2023-04-03 17:07:27,923 WARNING [train.py:1073] (1/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 2023-04-03 17:07:49,456 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.244e+02 5.115e+02 6.435e+02 8.085e+02 2.061e+03, threshold=1.287e+03, percent-clipped=7.0 2023-04-03 17:07:49,623 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=203192.0, num_to_drop=1, layers_to_drop={1} 2023-04-03 17:08:01,701 WARNING [train.py:1073] (1/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1_rvb from training. Duration: 0.9681875 2023-04-03 17:08:14,988 INFO [train.py:903] (1/4) Epoch 30, batch 5200, loss[loss=0.2409, simple_loss=0.3177, pruned_loss=0.08208, over 19679.00 frames. ], tot_loss[loss=0.1994, simple_loss=0.2809, pruned_loss=0.05901, over 3815796.77 frames. ], batch size: 59, lr: 2.71e-03, grad_scale: 8.0 2023-04-03 17:08:30,097 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 2023-04-03 17:09:14,142 WARNING [train.py:1073] (1/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 2023-04-03 17:09:16,378 INFO [train.py:903] (1/4) Epoch 30, batch 5250, loss[loss=0.2007, simple_loss=0.2754, pruned_loss=0.06303, over 19403.00 frames. ], tot_loss[loss=0.1997, simple_loss=0.2813, pruned_loss=0.05905, over 3821414.07 frames. ], batch size: 48, lr: 2.71e-03, grad_scale: 8.0 2023-04-03 17:09:53,734 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.978e+02 4.929e+02 6.276e+02 8.119e+02 2.486e+03, threshold=1.255e+03, percent-clipped=4.0 2023-04-03 17:10:05,401 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.1369, 3.6405, 3.7684, 3.7769, 1.8221, 3.5331, 3.1622, 3.5551], device='cuda:1'), covar=tensor([0.1871, 0.1623, 0.0751, 0.0852, 0.5482, 0.1337, 0.0791, 0.1182], device='cuda:1'), in_proj_covar=tensor([0.0832, 0.0807, 0.1018, 0.0891, 0.0880, 0.0780, 0.0600, 0.0946], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-03 17:10:12,191 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=203307.0, num_to_drop=1, layers_to_drop={0} 2023-04-03 17:10:17,556 INFO [train.py:903] (1/4) Epoch 30, batch 5300, loss[loss=0.2403, simple_loss=0.3282, pruned_loss=0.07623, over 19708.00 frames. ], tot_loss[loss=0.1998, simple_loss=0.2816, pruned_loss=0.05903, over 3822760.26 frames. ], batch size: 59, lr: 2.71e-03, grad_scale: 8.0 2023-04-03 17:10:36,825 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 2023-04-03 17:11:08,662 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.1717, 2.0399, 2.0227, 1.8876, 1.5912, 1.7959, 0.6251, 1.1556], device='cuda:1'), covar=tensor([0.0739, 0.0732, 0.0501, 0.0827, 0.1370, 0.0939, 0.1476, 0.1193], device='cuda:1'), in_proj_covar=tensor([0.0367, 0.0366, 0.0371, 0.0393, 0.0471, 0.0399, 0.0346, 0.0350], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 17:11:18,765 INFO [train.py:903] (1/4) Epoch 30, batch 5350, loss[loss=0.2149, simple_loss=0.301, pruned_loss=0.06438, over 19526.00 frames. ], tot_loss[loss=0.2007, simple_loss=0.2826, pruned_loss=0.05938, over 3826526.69 frames. ], batch size: 56, lr: 2.71e-03, grad_scale: 8.0 2023-04-03 17:11:38,863 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 2023-04-03 17:11:53,867 WARNING [train.py:1073] (1/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9_rvb from training. Duration: 26.438875 2023-04-03 17:11:56,057 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.690e+02 5.524e+02 6.450e+02 8.325e+02 1.910e+03, threshold=1.290e+03, percent-clipped=5.0 2023-04-03 17:12:05,342 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=203400.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 17:12:08,779 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([0.8464, 1.1936, 0.9589, 0.8066, 1.0401, 0.8064, 0.9436, 1.1002], device='cuda:1'), covar=tensor([0.0711, 0.0862, 0.1083, 0.0890, 0.0621, 0.1428, 0.0571, 0.0547], device='cuda:1'), in_proj_covar=tensor([0.0308, 0.0324, 0.0348, 0.0280, 0.0257, 0.0352, 0.0294, 0.0280], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 17:12:20,976 INFO [train.py:903] (1/4) Epoch 30, batch 5400, loss[loss=0.2452, simple_loss=0.312, pruned_loss=0.08924, over 19696.00 frames. ], tot_loss[loss=0.2019, simple_loss=0.2835, pruned_loss=0.06022, over 3836202.66 frames. ], batch size: 63, lr: 2.71e-03, grad_scale: 8.0 2023-04-03 17:13:02,927 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=203447.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 17:13:20,914 INFO [train.py:903] (1/4) Epoch 30, batch 5450, loss[loss=0.2221, simple_loss=0.3116, pruned_loss=0.06631, over 18062.00 frames. ], tot_loss[loss=0.2019, simple_loss=0.2835, pruned_loss=0.06012, over 3836652.70 frames. ], batch size: 83, lr: 2.71e-03, grad_scale: 8.0 2023-04-03 17:13:21,329 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.7122, 1.2495, 1.3569, 1.5808, 1.1086, 1.4464, 1.3486, 1.5256], device='cuda:1'), covar=tensor([0.1137, 0.1300, 0.1624, 0.1046, 0.1486, 0.0655, 0.1653, 0.0863], device='cuda:1'), in_proj_covar=tensor([0.0283, 0.0364, 0.0323, 0.0262, 0.0312, 0.0261, 0.0327, 0.0266], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-04-03 17:13:50,568 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=203485.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 17:13:59,454 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.086e+02 4.910e+02 6.531e+02 8.117e+02 1.984e+03, threshold=1.306e+03, percent-clipped=3.0 2023-04-03 17:14:23,411 INFO [train.py:903] (1/4) Epoch 30, batch 5500, loss[loss=0.1873, simple_loss=0.2766, pruned_loss=0.04902, over 19726.00 frames. ], tot_loss[loss=0.2002, simple_loss=0.2821, pruned_loss=0.05916, over 3845605.51 frames. ], batch size: 63, lr: 2.71e-03, grad_scale: 8.0 2023-04-03 17:14:26,944 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=203515.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 17:14:48,311 WARNING [train.py:1073] (1/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9_rvb from training. Duration: 27.47775 2023-04-03 17:15:24,336 INFO [train.py:903] (1/4) Epoch 30, batch 5550, loss[loss=0.1968, simple_loss=0.2814, pruned_loss=0.05612, over 19592.00 frames. ], tot_loss[loss=0.1998, simple_loss=0.2817, pruned_loss=0.05892, over 3848966.44 frames. ], batch size: 61, lr: 2.71e-03, grad_scale: 8.0 2023-04-03 17:15:24,639 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=203562.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 17:15:25,780 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=203563.0, num_to_drop=1, layers_to_drop={0} 2023-04-03 17:15:31,867 WARNING [train.py:1073] (1/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 2023-04-03 17:15:56,787 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=203588.0, num_to_drop=1, layers_to_drop={0} 2023-04-03 17:16:01,700 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.118e+02 4.952e+02 6.311e+02 7.819e+02 2.120e+03, threshold=1.262e+03, percent-clipped=2.0 2023-04-03 17:16:22,144 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 2023-04-03 17:16:27,754 INFO [train.py:903] (1/4) Epoch 30, batch 5600, loss[loss=0.2281, simple_loss=0.3045, pruned_loss=0.07587, over 13307.00 frames. ], tot_loss[loss=0.2009, simple_loss=0.2826, pruned_loss=0.05958, over 3844655.06 frames. ], batch size: 136, lr: 2.71e-03, grad_scale: 8.0 2023-04-03 17:16:38,804 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.63 vs. limit=2.0 2023-04-03 17:16:59,740 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.84 vs. limit=2.0 2023-04-03 17:17:29,529 INFO [train.py:903] (1/4) Epoch 30, batch 5650, loss[loss=0.1938, simple_loss=0.2763, pruned_loss=0.05561, over 19606.00 frames. ], tot_loss[loss=0.2018, simple_loss=0.2836, pruned_loss=0.05999, over 3857858.26 frames. ], batch size: 50, lr: 2.71e-03, grad_scale: 8.0 2023-04-03 17:17:50,967 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([3.3359, 3.8425, 3.9511, 3.9627, 1.6544, 3.7647, 3.2848, 3.7287], device='cuda:1'), covar=tensor([0.1751, 0.0963, 0.0725, 0.0847, 0.5823, 0.1088, 0.0765, 0.1190], device='cuda:1'), in_proj_covar=tensor([0.0827, 0.0804, 0.1013, 0.0889, 0.0875, 0.0776, 0.0598, 0.0945], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-03 17:18:06,340 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.203e+02 4.913e+02 5.869e+02 7.597e+02 1.607e+03, threshold=1.174e+03, percent-clipped=2.0 2023-04-03 17:18:18,649 WARNING [train.py:1073] (1/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9_rvb from training. Duration: 28.0944375 2023-04-03 17:18:29,970 INFO [train.py:903] (1/4) Epoch 30, batch 5700, loss[loss=0.1977, simple_loss=0.2725, pruned_loss=0.06145, over 19332.00 frames. ], tot_loss[loss=0.2003, simple_loss=0.2822, pruned_loss=0.05924, over 3854801.45 frames. ], batch size: 47, lr: 2.71e-03, grad_scale: 8.0 2023-04-03 17:18:31,155 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([6.2355, 5.6717, 3.4117, 5.0192, 1.1307, 5.8802, 5.6862, 5.8957], device='cuda:1'), covar=tensor([0.0354, 0.0763, 0.1645, 0.0760, 0.3986, 0.0456, 0.0741, 0.0915], device='cuda:1'), in_proj_covar=tensor([0.0540, 0.0439, 0.0522, 0.0363, 0.0414, 0.0462, 0.0456, 0.0492], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 17:19:10,860 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.5376, 1.6222, 1.7839, 1.9253, 1.5429, 1.8443, 1.8052, 1.7397], device='cuda:1'), covar=tensor([0.3365, 0.2956, 0.1531, 0.1898, 0.3003, 0.1765, 0.3788, 0.2538], device='cuda:1'), in_proj_covar=tensor([0.0953, 0.1036, 0.0753, 0.0963, 0.0930, 0.0870, 0.0866, 0.0818], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-03 17:19:31,944 INFO [train.py:903] (1/4) Epoch 30, batch 5750, loss[loss=0.1932, simple_loss=0.2824, pruned_loss=0.05195, over 19776.00 frames. ], tot_loss[loss=0.2013, simple_loss=0.2832, pruned_loss=0.05968, over 3839857.31 frames. ], batch size: 56, lr: 2.71e-03, grad_scale: 8.0 2023-04-03 17:19:35,374 WARNING [train.py:1073] (1/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9_rvb from training. Duration: 33.038875 2023-04-03 17:19:35,644 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=203765.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 17:19:44,345 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=203771.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 17:19:45,186 WARNING [train.py:1073] (1/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 2023-04-03 17:19:51,014 WARNING [train.py:1073] (1/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 2023-04-03 17:20:00,646 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=203784.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 17:20:09,174 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.085e+02 5.021e+02 5.985e+02 7.897e+02 1.857e+03, threshold=1.197e+03, percent-clipped=7.0 2023-04-03 17:20:15,310 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=203796.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 17:20:35,335 INFO [train.py:903] (1/4) Epoch 30, batch 5800, loss[loss=0.2024, simple_loss=0.2717, pruned_loss=0.06658, over 19740.00 frames. ], tot_loss[loss=0.2012, simple_loss=0.283, pruned_loss=0.05974, over 3842589.38 frames. ], batch size: 46, lr: 2.71e-03, grad_scale: 8.0 2023-04-03 17:20:41,479 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.1941, 1.8237, 1.4953, 1.3102, 1.6087, 1.2195, 1.1767, 1.6263], device='cuda:1'), covar=tensor([0.0958, 0.0910, 0.1103, 0.0925, 0.0622, 0.1440, 0.0711, 0.0504], device='cuda:1'), in_proj_covar=tensor([0.0308, 0.0324, 0.0347, 0.0279, 0.0257, 0.0351, 0.0293, 0.0281], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 17:20:43,823 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=203818.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 17:20:56,418 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=203829.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 17:21:13,715 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=203843.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 17:21:37,905 INFO [train.py:903] (1/4) Epoch 30, batch 5850, loss[loss=0.2095, simple_loss=0.2977, pruned_loss=0.06068, over 19474.00 frames. ], tot_loss[loss=0.201, simple_loss=0.283, pruned_loss=0.05952, over 3832074.33 frames. ], batch size: 64, lr: 2.71e-03, grad_scale: 8.0 2023-04-03 17:21:41,664 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=203865.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 17:22:15,805 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.367e+02 4.919e+02 5.824e+02 7.977e+02 1.569e+03, threshold=1.165e+03, percent-clipped=4.0 2023-04-03 17:22:39,475 INFO [train.py:903] (1/4) Epoch 30, batch 5900, loss[loss=0.1961, simple_loss=0.2869, pruned_loss=0.05266, over 19537.00 frames. ], tot_loss[loss=0.2012, simple_loss=0.2831, pruned_loss=0.05962, over 3823197.12 frames. ], batch size: 56, lr: 2.71e-03, grad_scale: 8.0 2023-04-03 17:22:42,664 WARNING [train.py:1073] (1/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 2023-04-03 17:22:47,246 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=203918.0, num_to_drop=1, layers_to_drop={0} 2023-04-03 17:23:06,104 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9_rvb from training. Duration: 27.511125 2023-04-03 17:23:07,743 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2854, 1.9482, 1.5021, 1.2717, 1.7879, 1.1082, 1.2030, 1.7927], device='cuda:1'), covar=tensor([0.1130, 0.0876, 0.1245, 0.0994, 0.0697, 0.1569, 0.0878, 0.0525], device='cuda:1'), in_proj_covar=tensor([0.0306, 0.0323, 0.0346, 0.0278, 0.0256, 0.0350, 0.0293, 0.0280], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 17:23:19,592 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=203944.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 17:23:31,982 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.8212, 3.3119, 3.3389, 3.3572, 1.4464, 3.2101, 2.7657, 3.1339], device='cuda:1'), covar=tensor([0.1824, 0.1104, 0.0823, 0.0987, 0.5591, 0.1159, 0.0917, 0.1294], device='cuda:1'), in_proj_covar=tensor([0.0827, 0.0805, 0.1013, 0.0889, 0.0876, 0.0776, 0.0599, 0.0945], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-03 17:23:40,674 INFO [train.py:903] (1/4) Epoch 30, batch 5950, loss[loss=0.1986, simple_loss=0.2813, pruned_loss=0.05792, over 19668.00 frames. ], tot_loss[loss=0.2015, simple_loss=0.2832, pruned_loss=0.0599, over 3824207.55 frames. ], batch size: 60, lr: 2.71e-03, grad_scale: 8.0 2023-04-03 17:24:18,145 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.982e+02 5.092e+02 6.183e+02 7.281e+02 1.501e+03, threshold=1.237e+03, percent-clipped=4.0 2023-04-03 17:24:45,144 INFO [train.py:903] (1/4) Epoch 30, batch 6000, loss[loss=0.2046, simple_loss=0.2908, pruned_loss=0.05919, over 19697.00 frames. ], tot_loss[loss=0.201, simple_loss=0.2826, pruned_loss=0.05969, over 3817099.48 frames. ], batch size: 59, lr: 2.71e-03, grad_scale: 8.0 2023-04-03 17:24:45,144 INFO [train.py:928] (1/4) Computing validation loss 2023-04-03 17:24:58,727 INFO [train.py:937] (1/4) Epoch 30, validation: loss=0.167, simple_loss=0.2658, pruned_loss=0.03407, over 944034.00 frames. 2023-04-03 17:24:58,728 INFO [train.py:938] (1/4) Maximum memory allocated so far is 18821MB 2023-04-03 17:26:02,073 INFO [train.py:903] (1/4) Epoch 30, batch 6050, loss[loss=0.1774, simple_loss=0.2708, pruned_loss=0.04197, over 19664.00 frames. ], tot_loss[loss=0.2005, simple_loss=0.2819, pruned_loss=0.0595, over 3822094.02 frames. ], batch size: 58, lr: 2.71e-03, grad_scale: 8.0 2023-04-03 17:26:28,633 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0535, 1.8305, 1.6917, 1.9824, 1.6750, 1.7191, 1.6544, 1.9166], device='cuda:1'), covar=tensor([0.0999, 0.1376, 0.1458, 0.1048, 0.1398, 0.0600, 0.1478, 0.0771], device='cuda:1'), in_proj_covar=tensor([0.0281, 0.0362, 0.0323, 0.0261, 0.0312, 0.0260, 0.0325, 0.0265], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-04-03 17:26:38,218 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.289e+02 5.054e+02 6.352e+02 7.854e+02 1.582e+03, threshold=1.270e+03, percent-clipped=1.0 2023-04-03 17:26:42,118 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=204095.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 17:26:54,293 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0604, 1.6878, 1.8419, 2.6492, 1.9858, 2.1901, 2.3326, 2.0019], device='cuda:1'), covar=tensor([0.0832, 0.0960, 0.1031, 0.0772, 0.0882, 0.0840, 0.0881, 0.0746], device='cuda:1'), in_proj_covar=tensor([0.0214, 0.0223, 0.0229, 0.0240, 0.0227, 0.0217, 0.0188, 0.0209], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-04-03 17:27:00,734 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=204109.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 17:27:04,111 INFO [train.py:903] (1/4) Epoch 30, batch 6100, loss[loss=0.2123, simple_loss=0.2868, pruned_loss=0.06891, over 19768.00 frames. ], tot_loss[loss=0.2011, simple_loss=0.2828, pruned_loss=0.05971, over 3824350.00 frames. ], batch size: 54, lr: 2.71e-03, grad_scale: 8.0 2023-04-03 17:27:23,319 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=204128.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 17:28:04,387 INFO [train.py:903] (1/4) Epoch 30, batch 6150, loss[loss=0.22, simple_loss=0.3089, pruned_loss=0.06555, over 19740.00 frames. ], tot_loss[loss=0.2013, simple_loss=0.283, pruned_loss=0.05983, over 3815516.31 frames. ], batch size: 63, lr: 2.71e-03, grad_scale: 8.0 2023-04-03 17:28:29,836 INFO [zipformer.py:1188] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=204181.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 17:28:34,969 WARNING [train.py:1073] (1/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9_rvb from training. Duration: 31.02225 2023-04-03 17:28:42,856 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.206e+02 5.061e+02 6.485e+02 7.945e+02 1.594e+03, threshold=1.297e+03, percent-clipped=2.0 2023-04-03 17:28:52,805 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=204200.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 17:29:03,786 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=204209.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 17:29:07,242 INFO [train.py:903] (1/4) Epoch 30, batch 6200, loss[loss=0.1937, simple_loss=0.2681, pruned_loss=0.05969, over 19404.00 frames. ], tot_loss[loss=0.2016, simple_loss=0.283, pruned_loss=0.06008, over 3808727.84 frames. ], batch size: 48, lr: 2.71e-03, grad_scale: 8.0 2023-04-03 17:29:24,447 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=204224.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 17:29:25,533 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=204225.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 17:29:41,913 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=256, metric=3.73 vs. limit=5.0 2023-04-03 17:29:45,869 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=204243.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 17:29:50,462 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.6657, 1.4869, 1.4939, 2.0682, 1.5369, 1.9376, 1.8652, 1.7504], device='cuda:1'), covar=tensor([0.0808, 0.0891, 0.0987, 0.0653, 0.0789, 0.0737, 0.0781, 0.0669], device='cuda:1'), in_proj_covar=tensor([0.0212, 0.0222, 0.0227, 0.0238, 0.0225, 0.0215, 0.0186, 0.0207], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-04-03 17:30:09,325 INFO [train.py:903] (1/4) Epoch 30, batch 6250, loss[loss=0.1822, simple_loss=0.2657, pruned_loss=0.04934, over 19754.00 frames. ], tot_loss[loss=0.2003, simple_loss=0.282, pruned_loss=0.05929, over 3825506.79 frames. ], batch size: 54, lr: 2.71e-03, grad_scale: 8.0 2023-04-03 17:30:10,640 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=204262.0, num_to_drop=1, layers_to_drop={1} 2023-04-03 17:30:38,580 WARNING [train.py:1073] (1/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9_rvb from training. Duration: 25.988875 2023-04-03 17:30:46,052 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.091e+02 5.004e+02 5.938e+02 7.424e+02 1.100e+03, threshold=1.188e+03, percent-clipped=0.0 2023-04-03 17:31:10,304 INFO [train.py:903] (1/4) Epoch 30, batch 6300, loss[loss=0.2522, simple_loss=0.3277, pruned_loss=0.08834, over 19744.00 frames. ], tot_loss[loss=0.2005, simple_loss=0.2821, pruned_loss=0.05948, over 3828889.53 frames. ], batch size: 63, lr: 2.71e-03, grad_scale: 4.0 2023-04-03 17:31:24,153 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=204324.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 17:31:56,763 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.0929, 2.0589, 1.9529, 1.6930, 1.3817, 1.6137, 0.6685, 1.1267], device='cuda:1'), covar=tensor([0.0934, 0.0959, 0.0608, 0.1248, 0.1756, 0.1584, 0.1682, 0.1518], device='cuda:1'), in_proj_covar=tensor([0.0369, 0.0368, 0.0371, 0.0396, 0.0474, 0.0400, 0.0348, 0.0351], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-03 17:32:11,321 INFO [train.py:903] (1/4) Epoch 30, batch 6350, loss[loss=0.1696, simple_loss=0.2474, pruned_loss=0.04585, over 18509.00 frames. ], tot_loss[loss=0.2001, simple_loss=0.2819, pruned_loss=0.05915, over 3831962.32 frames. ], batch size: 41, lr: 2.71e-03, grad_scale: 4.0 2023-04-03 17:32:30,534 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=204377.0, num_to_drop=1, layers_to_drop={1} 2023-04-03 17:32:50,298 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.055e+02 4.728e+02 6.049e+02 7.442e+02 1.533e+03, threshold=1.210e+03, percent-clipped=4.0 2023-04-03 17:33:12,659 INFO [train.py:903] (1/4) Epoch 30, batch 6400, loss[loss=0.1957, simple_loss=0.2806, pruned_loss=0.05542, over 18021.00 frames. ], tot_loss[loss=0.2015, simple_loss=0.2832, pruned_loss=0.05985, over 3835072.73 frames. ], batch size: 83, lr: 2.71e-03, grad_scale: 8.0 2023-04-03 17:33:45,677 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=204439.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 17:34:13,526 INFO [train.py:903] (1/4) Epoch 30, batch 6450, loss[loss=0.2093, simple_loss=0.2881, pruned_loss=0.0652, over 18176.00 frames. ], tot_loss[loss=0.2025, simple_loss=0.2841, pruned_loss=0.06046, over 3816687.08 frames. ], batch size: 83, lr: 2.71e-03, grad_scale: 8.0 2023-04-03 17:34:23,500 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.2186, 2.2001, 2.4269, 2.9897, 2.3260, 2.8162, 2.4577, 2.2562], device='cuda:1'), covar=tensor([0.4425, 0.4518, 0.2123, 0.2796, 0.4689, 0.2569, 0.5362, 0.3674], device='cuda:1'), in_proj_covar=tensor([0.0954, 0.1037, 0.0755, 0.0965, 0.0932, 0.0871, 0.0868, 0.0819], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-03 17:34:35,898 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=204480.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 17:34:51,302 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.580e+02 4.981e+02 6.276e+02 8.010e+02 2.155e+03, threshold=1.255e+03, percent-clipped=8.0 2023-04-03 17:34:54,942 WARNING [train.py:1073] (1/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9_rvb from training. Duration: 28.638875 2023-04-03 17:34:59,488 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=204499.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 17:35:07,995 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=204505.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 17:35:16,831 INFO [train.py:903] (1/4) Epoch 30, batch 6500, loss[loss=0.2079, simple_loss=0.293, pruned_loss=0.06138, over 19460.00 frames. ], tot_loss[loss=0.2014, simple_loss=0.2834, pruned_loss=0.05973, over 3817301.69 frames. ], batch size: 70, lr: 2.71e-03, grad_scale: 8.0 2023-04-03 17:35:19,115 WARNING [train.py:1073] (1/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1_rvb from training. Duration: 0.836375 2023-04-03 17:35:30,901 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=204524.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 17:35:31,787 INFO [zipformer.py:1188] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=204525.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 17:36:09,107 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=204554.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 17:36:18,053 INFO [train.py:903] (1/4) Epoch 30, batch 6550, loss[loss=0.2773, simple_loss=0.3382, pruned_loss=0.1082, over 13755.00 frames. ], tot_loss[loss=0.2004, simple_loss=0.2825, pruned_loss=0.05911, over 3823032.99 frames. ], batch size: 135, lr: 2.71e-03, grad_scale: 8.0 2023-04-03 17:36:40,526 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=204580.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 17:36:56,884 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.390e+02 5.288e+02 6.916e+02 9.470e+02 2.608e+03, threshold=1.383e+03, percent-clipped=11.0 2023-04-03 17:37:11,422 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=204605.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 17:37:20,110 INFO [train.py:903] (1/4) Epoch 30, batch 6600, loss[loss=0.1719, simple_loss=0.2532, pruned_loss=0.04534, over 19381.00 frames. ], tot_loss[loss=0.2003, simple_loss=0.2823, pruned_loss=0.05919, over 3801405.46 frames. ], batch size: 47, lr: 2.71e-03, grad_scale: 8.0 2023-04-03 17:37:46,951 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=204633.0, num_to_drop=1, layers_to_drop={0} 2023-04-03 17:37:54,527 INFO [zipformer.py:1188] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=204640.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 17:38:02,586 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.2209, 1.7921, 1.4182, 1.2541, 1.5700, 1.1923, 1.2451, 1.6483], device='cuda:1'), covar=tensor([0.0838, 0.0861, 0.1121, 0.0891, 0.0626, 0.1298, 0.0626, 0.0452], device='cuda:1'), in_proj_covar=tensor([0.0306, 0.0323, 0.0346, 0.0278, 0.0256, 0.0350, 0.0292, 0.0281], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 17:38:17,129 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=204658.0, num_to_drop=1, layers_to_drop={1} 2023-04-03 17:38:21,418 INFO [train.py:903] (1/4) Epoch 30, batch 6650, loss[loss=0.1971, simple_loss=0.2845, pruned_loss=0.05487, over 19724.00 frames. ], tot_loss[loss=0.2003, simple_loss=0.2824, pruned_loss=0.05913, over 3796516.18 frames. ], batch size: 51, lr: 2.70e-03, grad_scale: 8.0 2023-04-03 17:38:53,458 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([2.1049, 2.0636, 1.9725, 2.2050, 2.0158, 1.8735, 1.9638, 2.0881], device='cuda:1'), covar=tensor([0.0931, 0.1150, 0.1160, 0.0873, 0.1126, 0.0527, 0.1180, 0.0617], device='cuda:1'), in_proj_covar=tensor([0.0282, 0.0361, 0.0321, 0.0261, 0.0311, 0.0261, 0.0325, 0.0266], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-04-03 17:38:59,515 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.417e+02 4.923e+02 6.497e+02 8.594e+02 3.232e+03, threshold=1.299e+03, percent-clipped=5.0 2023-04-03 17:39:24,423 INFO [train.py:903] (1/4) Epoch 30, batch 6700, loss[loss=0.2242, simple_loss=0.3034, pruned_loss=0.07249, over 18313.00 frames. ], tot_loss[loss=0.2002, simple_loss=0.2824, pruned_loss=0.05894, over 3799875.30 frames. ], batch size: 83, lr: 2.70e-03, grad_scale: 4.0 2023-04-03 17:40:21,754 INFO [train.py:903] (1/4) Epoch 30, batch 6750, loss[loss=0.2399, simple_loss=0.3239, pruned_loss=0.07801, over 19614.00 frames. ], tot_loss[loss=0.2005, simple_loss=0.2827, pruned_loss=0.05916, over 3807904.05 frames. ], batch size: 57, lr: 2.70e-03, grad_scale: 4.0 2023-04-03 17:40:58,539 INFO [optim.py:369] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.758e+02 4.849e+02 6.262e+02 9.905e+02 2.863e+03, threshold=1.252e+03, percent-clipped=11.0 2023-04-03 17:41:17,887 INFO [zipformer.py:1188] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=204810.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 17:41:19,747 INFO [train.py:903] (1/4) Epoch 30, batch 6800, loss[loss=0.2082, simple_loss=0.2954, pruned_loss=0.06046, over 19614.00 frames. ], tot_loss[loss=0.1993, simple_loss=0.2812, pruned_loss=0.05866, over 3817292.09 frames. ], batch size: 57, lr: 2.70e-03, grad_scale: 8.0 2023-04-03 17:41:35,116 INFO [zipformer.py:2441] (1/4) attn_weights_entropy = tensor([1.8597, 0.8989, 1.0898, 1.0451, 1.6443, 0.8528, 1.5504, 1.8726], device='cuda:1'), covar=tensor([0.0603, 0.2270, 0.2199, 0.1316, 0.0681, 0.1665, 0.1421, 0.0466], device='cuda:1'), in_proj_covar=tensor([0.0427, 0.0382, 0.0403, 0.0357, 0.0388, 0.0362, 0.0402, 0.0423], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-03 17:41:45,196 INFO [zipformer.py:1188] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=204835.0, num_to_drop=0, layers_to_drop=set() 2023-04-03 17:41:49,972 INFO [train.py:1171] (1/4) Done!