wav2vec2-1b-E10_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: 0.7836
• Cer: 21.4932

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
10.4923	0.2580	200	4.9842	96.2289
4.7652	0.5160	400	5.0307	96.2230
4.0439	0.7741	600	3.6510	78.3365
1.695	1.0321	800	1.5850	43.7030
1.117	1.2901	1000	1.4627	38.2108
0.947	1.5481	1200	1.3708	35.0740
0.839	1.8062	1400	1.1110	32.1781
0.7141	2.0642	1600	1.2207	33.8463
0.5813	2.3222	1800	1.1949	32.5658
0.5561	2.5802	2000	1.0831	29.2411
0.4987	2.8383	2200	1.0766	29.0942
0.4244	3.0963	2400	0.9577	26.7328
0.3747	3.3543	2600	0.9733	26.6154
0.3351	3.6123	2800	0.8039	22.2686
0.3073	3.8703	3000	0.8293	23.2613
0.276	4.1284	3200	0.8784	23.6372
0.24	4.3864	3400	0.8575	23.6783
0.2281	4.6444	3600	0.8142	22.2509
0.2074	4.9024	3800	0.7836	21.4932

Framework versions

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