[hops] 2024-09-24 15:33:02.854 | INFO | Initializing a parser from /workspace/configs/exp_camembertv2/camembertv2_base_p2_17k_last_layer.yaml [hops] 2024-09-24 15:33:02.888 | INFO | Generating a FastText model from the treebank [hops] 2024-09-24 15:33:02.897 | INFO | Training fasttext model [hops] 2024-09-24 15:33:04.173 | WARNING | Some weights of RobertaModel were not initialized from the model checkpoint at /scratch/camembertv2/runs/models/camembertv2-base-bf16/post/ckpt-p2-17000/pt/ and are newly initialized: ['roberta.pooler.dense.bias', 'roberta.pooler.dense.weight'] You should probably TRAIN this model on a down-stream task to be able to use it for predictions and inference. [hops] 2024-09-24 15:33:09.889 | INFO | Start training on cuda:0 [hops] 2024-09-24 15:33:09.892 | WARNING | You're using a RobertaTokenizerFast tokenizer. Please note that with a fast tokenizer, using the `__call__` method is faster than using a method to encode the text followed by a call to the `pad` method to get a padded encoding. [hops] 2024-09-24 15:33:21.621 | INFO | Epoch 0: train loss 3.2439 dev loss 2.8669 dev tag acc 00.26% dev head acc 23.97% dev deprel acc 32.17% [hops] 2024-09-24 15:33:21.625 | INFO | New best model: head accuracy 23.97% > 0.00% [hops] 2024-09-24 15:33:35.614 | INFO | Epoch 1: train loss 2.5748 dev loss 2.1781 dev tag acc 00.55% dev head acc 44.69% dev deprel acc 64.55% [hops] 2024-09-24 15:33:35.614 | INFO | New best model: head accuracy 44.69% > 23.97% [hops] 2024-09-24 15:33:49.487 | INFO | Epoch 2: train loss 2.0676 dev loss 1.7605 dev tag acc 37.83% dev head acc 53.20% dev deprel acc 72.15% [hops] 2024-09-24 15:33:49.488 | INFO | New best model: head accuracy 53.20% > 44.69% [hops] 2024-09-24 15:34:03.400 | INFO | Epoch 3: train loss 1.6440 dev loss 1.3460 dev tag acc 59.15% dev head acc 62.36% dev deprel acc 76.75% [hops] 2024-09-24 15:34:03.401 | INFO | New best model: head accuracy 62.36% > 53.20% [hops] 2024-09-24 15:34:17.979 | INFO | Epoch 4: train loss 1.2439 dev loss 1.0561 dev tag acc 68.45% dev head acc 69.04% dev deprel acc 80.28% [hops] 2024-09-24 15:34:17.980 | INFO | New best model: head accuracy 69.04% > 62.36% [hops] 2024-09-24 15:34:32.155 | INFO | Epoch 5: train loss 0.9863 dev loss 0.8724 dev tag acc 72.19% dev head acc 73.82% dev deprel acc 83.38% [hops] 2024-09-24 15:34:32.156 | INFO | New best model: head accuracy 73.82% > 69.04% [hops] 2024-09-24 15:34:46.008 | INFO | Epoch 6: train loss 0.8104 dev loss 0.7492 dev tag acc 82.46% dev head acc 76.78% dev deprel acc 84.17% [hops] 2024-09-24 15:34:46.009 | INFO | New best model: head accuracy 76.78% > 73.82% [hops] 2024-09-24 15:35:00.123 | INFO | Epoch 7: train loss 0.6813 dev loss 0.6608 dev tag acc 86.37% dev head acc 79.33% dev deprel acc 85.77% [hops] 2024-09-24 15:35:00.124 | INFO | New best model: head accuracy 79.33% > 76.78% [hops] 2024-09-24 15:35:14.106 | INFO | Epoch 8: train loss 0.5863 dev loss 0.6169 dev tag acc 87.87% dev head acc 79.97% dev deprel acc 86.28% [hops] 2024-09-24 15:35:14.107 | INFO | New best model: head accuracy 79.97% > 79.33% [hops] 2024-09-24 15:35:28.022 | INFO | Epoch 9: train loss 0.5135 dev loss 0.6030 dev tag acc 89.73% dev head acc 80.82% dev deprel acc 87.10% [hops] 2024-09-24 15:35:28.023 | INFO | New best model: head accuracy 80.82% > 79.97% [hops] 2024-09-24 15:35:41.556 | INFO | Epoch 10: train loss 0.4537 dev loss 0.5871 dev tag acc 91.19% dev head acc 82.06% dev deprel acc 88.05% [hops] 2024-09-24 15:35:41.557 | INFO | New best model: head accuracy 82.06% > 80.82% [hops] 2024-09-24 15:35:56.168 | INFO | Epoch 11: train loss 0.4092 dev loss 0.5826 dev tag acc 91.57% dev head acc 83.66% dev deprel acc 88.62% [hops] 2024-09-24 15:35:56.169 | INFO | New best model: head accuracy 83.66% > 82.06% [hops] 2024-09-24 15:36:10.303 | INFO | Epoch 12: train loss 0.3746 dev loss 0.5677 dev tag acc 92.30% dev head acc 83.92% dev deprel acc 88.77% [hops] 2024-09-24 15:36:10.304 | INFO | New best model: head accuracy 83.92% > 83.66% [hops] 2024-09-24 15:36:24.290 | INFO | Epoch 13: train loss 0.3392 dev loss 0.5757 dev tag acc 92.46% dev head acc 83.37% dev deprel acc 89.09% [hops] 2024-09-24 15:36:36.551 | INFO | Epoch 14: train loss 0.3124 dev loss 0.5494 dev tag acc 92.84% dev head acc 83.96% dev deprel acc 89.71% [hops] 2024-09-24 15:36:36.552 | INFO | New best model: head accuracy 83.96% > 83.92% [hops] 2024-09-24 15:36:50.657 | INFO | Epoch 15: train loss 0.2848 dev loss 0.5712 dev tag acc 93.29% dev head acc 84.60% dev deprel acc 89.87% [hops] 2024-09-24 15:36:50.658 | INFO | New best model: head accuracy 84.60% > 83.96% [hops] 2024-09-24 15:37:04.712 | INFO | Epoch 16: train loss 0.2680 dev loss 0.5854 dev tag acc 93.47% dev head acc 85.18% dev deprel acc 90.01% [hops] 2024-09-24 15:37:04.713 | INFO | New best model: head accuracy 85.18% > 84.60% [hops] 2024-09-24 15:37:18.922 | INFO | Epoch 17: train loss 0.2485 dev loss 0.5999 dev tag acc 93.55% dev head acc 84.95% dev deprel acc 90.24% [hops] 2024-09-24 15:37:30.981 | INFO | Epoch 18: train loss 0.2326 dev loss 0.6017 dev tag acc 93.86% dev head acc 85.64% dev deprel acc 89.91% [hops] 2024-09-24 15:37:30.982 | INFO | New best model: head accuracy 85.64% > 85.18% [hops] 2024-09-24 15:37:45.204 | INFO | Epoch 19: train loss 0.2163 dev loss 0.6003 dev tag acc 93.93% dev head acc 85.48% dev deprel acc 90.38% [hops] 2024-09-24 15:37:57.324 | INFO | Epoch 20: train loss 0.2016 dev loss 0.6100 dev tag acc 94.17% dev head acc 85.77% dev deprel acc 90.26% [hops] 2024-09-24 15:37:57.324 | INFO | New best model: head accuracy 85.77% > 85.64% [hops] 2024-09-24 15:38:11.713 | INFO | Epoch 21: train loss 0.1925 dev loss 0.6233 dev tag acc 94.24% dev head acc 85.53% dev deprel acc 90.73% [hops] 2024-09-24 15:38:23.640 | INFO | Epoch 22: train loss 0.1829 dev loss 0.6013 dev tag acc 94.30% dev head acc 85.94% dev deprel acc 90.72% [hops] 2024-09-24 15:38:23.641 | INFO | New best model: head accuracy 85.94% > 85.77% [hops] 2024-09-24 15:38:38.142 | INFO | Epoch 23: train loss 0.1698 dev loss 0.6447 dev tag acc 94.43% dev head acc 86.10% dev deprel acc 90.83% [hops] 2024-09-24 15:38:38.143 | INFO | New best model: head accuracy 86.10% > 85.94% [hops] 2024-09-24 15:38:51.582 | INFO | Epoch 24: train loss 0.1593 dev loss 0.6905 dev tag acc 94.60% dev head acc 85.94% dev deprel acc 91.06% [hops] 2024-09-24 15:39:03.061 | INFO | Epoch 25: train loss 0.1530 dev loss 0.6778 dev tag acc 94.61% dev head acc 85.93% dev deprel acc 90.78% [hops] 2024-09-24 15:39:14.938 | INFO | Epoch 26: train loss 0.1431 dev loss 0.6909 dev tag acc 94.61% dev head acc 86.17% dev deprel acc 91.01% [hops] 2024-09-24 15:39:14.939 | INFO | New best model: head accuracy 86.17% > 86.10% [hops] 2024-09-24 15:39:29.019 | INFO | Epoch 27: train loss 0.1379 dev loss 0.7448 dev tag acc 94.57% dev head acc 86.15% dev deprel acc 90.87% [hops] 2024-09-24 15:39:40.887 | INFO | Epoch 28: train loss 0.1292 dev loss 0.7414 dev tag acc 94.66% dev head acc 86.44% dev deprel acc 91.14% [hops] 2024-09-24 15:39:40.888 | INFO | New best model: head accuracy 86.44% > 86.17% [hops] 2024-09-24 15:39:55.098 | INFO | Epoch 29: train loss 0.1242 dev loss 0.7502 dev tag acc 94.61% dev head acc 86.45% dev deprel acc 91.12% [hops] 2024-09-24 15:39:55.099 | INFO | New best model: head accuracy 86.45% > 86.44% [hops] 2024-09-24 15:40:09.314 | INFO | Epoch 30: train loss 0.1201 dev loss 0.7615 dev tag acc 94.66% dev head acc 86.15% dev deprel acc 91.13% [hops] 2024-09-24 15:40:21.157 | INFO | Epoch 31: train loss 0.1145 dev loss 0.7454 dev tag acc 94.88% dev head acc 86.35% dev deprel acc 91.13% [hops] 2024-09-24 15:40:33.212 | INFO | Epoch 32: train loss 0.1104 dev loss 0.7493 dev tag acc 94.80% dev head acc 86.59% dev deprel acc 91.18% [hops] 2024-09-24 15:40:33.213 | INFO | New best model: head accuracy 86.59% > 86.45% [hops] 2024-09-24 15:40:46.985 | INFO | Epoch 33: train loss 0.1045 dev loss 0.7510 dev tag acc 94.86% dev head acc 86.49% dev deprel acc 91.26% [hops] 2024-09-24 15:40:58.689 | INFO | Epoch 34: train loss 0.1013 dev loss 0.8365 dev tag acc 94.84% dev head acc 86.13% dev deprel acc 91.15% [hops] 2024-09-24 15:41:10.761 | INFO | Epoch 35: train loss 0.0963 dev loss 0.8366 dev tag acc 94.94% dev head acc 86.39% dev deprel acc 91.15% [hops] 2024-09-24 15:41:22.642 | INFO | Epoch 36: train loss 0.0921 dev loss 0.8437 dev tag acc 94.94% dev head acc 85.95% dev deprel acc 91.17% [hops] 2024-09-24 15:41:34.395 | INFO | Epoch 37: train loss 0.0900 dev loss 0.8631 dev tag acc 94.90% dev head acc 86.14% dev deprel acc 91.31% [hops] 2024-09-24 15:41:46.080 | INFO | Epoch 38: train loss 0.0857 dev loss 0.8898 dev tag acc 95.02% dev head acc 86.29% dev deprel acc 91.26% [hops] 2024-09-24 15:41:58.024 | INFO | Epoch 39: train loss 0.0836 dev loss 0.8344 dev tag acc 95.00% dev head acc 86.53% dev deprel acc 91.41% [hops] 2024-09-24 15:42:10.046 | INFO | Epoch 40: train loss 0.0808 dev loss 0.9104 dev tag acc 94.94% dev head acc 86.55% dev deprel acc 91.31% [hops] 2024-09-24 15:42:22.012 | INFO | Epoch 41: train loss 0.0767 dev loss 0.8839 dev tag acc 95.01% dev head acc 86.53% dev deprel acc 91.37% [hops] 2024-09-24 15:42:33.859 | INFO | Epoch 42: train loss 0.0740 dev loss 0.9266 dev tag acc 94.99% dev head acc 86.36% dev deprel acc 91.30% [hops] 2024-09-24 15:42:45.857 | INFO | Epoch 43: train loss 0.0708 dev loss 0.9715 dev tag acc 95.08% dev head acc 86.20% dev deprel acc 91.31% [hops] 2024-09-24 15:42:57.382 | INFO | Epoch 44: train loss 0.0700 dev loss 0.9246 dev tag acc 95.11% dev head acc 86.53% dev deprel acc 91.38% [hops] 2024-09-24 15:43:09.412 | INFO | Epoch 45: train loss 0.0665 dev loss 0.9343 dev tag acc 95.07% dev head acc 86.45% dev deprel acc 91.56% [hops] 2024-09-24 15:43:21.084 | INFO | Epoch 46: train loss 0.0655 dev loss 0.9534 dev tag acc 95.21% dev head acc 86.46% dev deprel acc 91.45% [hops] 2024-09-24 15:43:33.081 | INFO | Epoch 47: train loss 0.0645 dev loss 0.9809 dev tag acc 95.12% dev head acc 86.47% dev deprel acc 91.23% [hops] 2024-09-24 15:43:45.664 | INFO | Epoch 48: train loss 0.0612 dev loss 0.9552 dev tag acc 95.12% dev head acc 86.59% dev deprel acc 91.48% [hops] 2024-09-24 15:43:58.409 | INFO | Epoch 49: train loss 0.0588 dev loss 0.9978 dev tag acc 95.12% dev head acc 86.55% dev deprel acc 91.45% [hops] 2024-09-24 15:44:10.971 | INFO | Epoch 50: train loss 0.0588 dev loss 1.0168 dev tag acc 95.20% dev head acc 86.90% dev deprel acc 91.61% [hops] 2024-09-24 15:44:10.972 | INFO | New best model: head accuracy 86.90% > 86.59% [hops] 2024-09-24 15:44:24.670 | INFO | Epoch 51: train loss 0.0553 dev loss 1.0430 dev tag acc 95.14% dev head acc 86.59% dev deprel acc 91.41% [hops] 2024-09-24 15:44:36.476 | INFO | Epoch 52: train loss 0.0546 dev loss 0.9933 dev tag acc 95.21% dev head acc 86.65% dev deprel acc 91.47% [hops] 2024-09-24 15:44:48.691 | INFO | Epoch 53: train loss 0.0531 dev loss 1.0337 dev tag acc 95.18% dev head acc 86.64% dev deprel acc 91.53% [hops] 2024-09-24 15:45:01.074 | INFO | Epoch 54: train loss 0.0496 dev loss 1.0584 dev tag acc 95.12% dev head acc 86.66% dev deprel acc 91.50% [hops] 2024-09-24 15:45:12.150 | INFO | Epoch 55: train loss 0.0518 dev loss 1.0425 dev tag acc 95.21% dev head acc 86.74% dev deprel acc 91.54% [hops] 2024-09-24 15:45:23.844 | INFO | Epoch 56: train loss 0.0506 dev loss 1.0478 dev tag acc 95.16% dev head acc 86.64% dev deprel acc 91.55% [hops] 2024-09-24 15:45:35.747 | INFO | Epoch 57: train loss 0.0479 dev loss 1.0688 dev tag acc 95.18% dev head acc 86.69% dev deprel acc 91.44% [hops] 2024-09-24 15:45:47.637 | INFO | Epoch 58: train loss 0.0484 dev loss 1.0632 dev tag acc 95.21% dev head acc 86.74% dev deprel acc 91.50% [hops] 2024-09-24 15:45:59.297 | INFO | Epoch 59: train loss 0.0460 dev loss 1.0741 dev tag acc 95.21% dev head acc 86.74% dev deprel acc 91.61% [hops] 2024-09-24 15:46:11.546 | INFO | Epoch 60: train loss 0.0462 dev loss 1.0848 dev tag acc 95.18% dev head acc 86.79% dev deprel acc 91.61% [hops] 2024-09-24 15:46:23.613 | INFO | Epoch 61: train loss 0.0466 dev loss 1.0859 dev tag acc 95.21% dev head acc 86.83% dev deprel acc 91.56% [hops] 2024-09-24 15:46:35.434 | INFO | Epoch 62: train loss 0.0440 dev loss 1.0883 dev tag acc 95.23% dev head acc 86.74% dev deprel acc 91.55% [hops] 2024-09-24 15:46:47.207 | INFO | Epoch 63: train loss 0.0452 dev loss 1.0858 dev tag acc 95.25% dev head acc 86.70% dev deprel acc 91.57% [hops] 2024-09-24 15:46:52.244 | WARNING | You're using a RobertaTokenizerFast tokenizer. Please note that with a fast tokenizer, using the `__call__` method is faster than using a method to encode the text followed by a call to the `pad` method to get a padded encoding. [hops] 2024-09-24 15:46:59.091 | WARNING | You're using a RobertaTokenizerFast tokenizer. Please note that with a fast tokenizer, using the `__call__` method is faster than using a method to encode the text followed by a call to the `pad` method to get a padded encoding. [hops] 2024-09-24 15:47:01.944 | INFO | Metrics for FSMB-camembertv2_base_p2_17k_last_layer+rand_seed=25 ─────────────────────────────── Split UPOS UAS LAS ─────────────────────────────── Dev 95.10 87.05 81.42 Test 95.18 87.05 81.88 ───────────────────────────────