w2v3

This model is a fine-tuned version of facebook/w2v-bert-2.0 on the common_voice_11_0 dataset. It achieves the following results on the evaluation set:

• Loss: 0.1747
• Wer: 0.1444
• Cer: 0.0349
• Bleu: 0.6254
• Bert Score F1: 0.9721

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: 1e-05
• train_batch_size: 4
• eval_batch_size: 8
• seed: 42
• optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• training_steps: 5000

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer	Bleu	Bert Score F1
0.3995	0.0357	250	0.3664	0.2725	0.0732	0.4414	0.9332
0.3065	0.0713	500	0.3399	0.2119	0.0593	0.5188	0.9465
0.2648	0.1070	750	0.3095	0.2327	0.0633	0.4970	0.9430
0.2393	0.1426	1000	0.2885	0.2134	0.0551	0.5156	0.9545
0.2756	0.1783	1250	0.2486	0.1817	0.0467	0.5670	0.9614
0.2005	0.2139	1500	0.2448	0.1935	0.0482	0.5485	0.9588
0.2112	0.2496	1750	0.2377	0.1823	0.0464	0.5617	0.9622
0.1934	0.2853	2000	0.2226	0.1674	0.0420	0.5888	0.9658
0.1631	0.3209	2250	0.2205	0.1660	0.0421	0.5888	0.9647
0.1905	0.3566	2500	0.2249	0.1679	0.0429	0.5879	0.9651
0.1639	0.3922	2750	0.2026	0.1625	0.0403	0.5975	0.9673
0.1567	0.4279	3000	0.1895	0.1516	0.0379	0.6150	0.9685
0.1641	0.4636	3250	0.1984	0.1555	0.0379	0.6076	0.9693
0.1404	0.4992	3500	0.1876	0.1528	0.0370	0.6124	0.9696
0.1475	0.5349	3750	0.1913	0.1568	0.0381	0.6055	0.9691
0.1586	0.5705	4000	0.1846	0.1510	0.0366	0.6151	0.9705
0.1322	0.6062	4250	0.1801	0.1475	0.0356	0.6208	0.9715
0.1396	0.6418	4500	0.1788	0.1454	0.0351	0.6242	0.9720
0.1287	0.6775	4750	0.1755	0.1455	0.0352	0.6233	0.9718
0.1376	0.7132	5000	0.1747	0.1444	0.0349	0.6254	0.9721

Framework versions

• Transformers 4.50.0
• Pytorch 2.5.1+cu121
• Datasets 3.3.1
• Tokenizers 0.21.0