wav2vec2-1b-Yfreq_pause

This model is a fine-tuned version of facebook/wav2vec2-xls-r-1b on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 1.2859
• Cer: 33.4880

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.0001
• train_batch_size: 2
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 8
• total_train_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 50
• num_epochs: 5
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer
12.7189	0.2580	200	4.6447	97.3156
3.4682	0.5160	400	2.9509	67.6574
1.7355	0.7741	600	2.5755	56.9666
1.3168	1.0321	800	1.8127	48.0557
1.0608	1.2901	1000	1.6916	45.5240
0.9486	1.5481	1200	1.5227	41.0009
0.8555	1.8062	1400	1.8282	48.3905
0.7964	2.0642	1600	2.1936	54.9988
0.6832	2.3222	1800	1.6408	40.3783
0.5939	2.5802	2000	1.4571	39.3327
0.5738	2.8383	2200	1.3655	37.2827
0.4881	3.0963	2400	1.3322	34.9977
0.4278	3.3543	2600	1.4489	36.8832
0.3979	3.6123	2800	1.4433	37.5940
0.36	3.8703	3000	1.3372	36.1490
0.3348	4.1284	3200	1.3533	35.6203
0.2949	4.3864	3400	1.3401	34.7568
0.2703	4.6444	3600	1.3240	34.6746
0.2519	4.9024	3800	1.2859	33.4880

Framework versions

• Transformers 4.45.2
• Pytorch 2.3.1.post100
• Datasets 2.19.1
• Tokenizers 0.20.1