w2v-bert-cv-grain-lg_both_v2

This model is a fine-tuned version of facebook/w2v-bert-2.0 on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.0892
• Wer: 0.0443
• Cer: 0.0123

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: 8
• eval_batch_size: 4
• 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
• num_epochs: 80
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
0.2889	1.0	10812	0.1708	0.1703	0.0386
0.1849	2.0	21624	0.1342	0.1274	0.0285
0.1512	3.0	32436	0.1144	0.1044	0.0244
0.1313	4.0	43248	0.1033	0.0918	0.0217
0.117	5.0	54060	0.1034	0.0738	0.0191
0.1056	6.0	64872	0.0906	0.0738	0.0181
0.0962	7.0	75684	0.0959	0.0655	0.0168
0.0885	8.0	86496	0.0860	0.0592	0.0155
0.0807	9.0	97308	0.0844	0.0603	0.0154
0.0742	10.0	108120	0.0814	0.0573	0.0144
0.0683	11.0	118932	0.0858	0.0588	0.0154
0.0629	12.0	129744	0.0944	0.0538	0.0146
0.0581	13.0	140556	0.0842	0.0558	0.0151
0.0528	14.0	151368	0.0873	0.0503	0.0141
0.0479	15.0	162180	0.0820	0.0503	0.0138
0.0429	16.0	172992	0.0815	0.0427	0.0125
0.0392	17.0	183804	0.0864	0.0466	0.0128
0.035	18.0	194616	0.0899	0.0479	0.0128
0.0316	19.0	205428	0.0872	0.0430	0.0120
0.0286	20.0	216240	0.0821	0.0425	0.0114
0.0254	21.0	227052	0.0898	0.0466	0.0122
0.0229	22.0	237864	0.0864	0.0417	0.0120
0.021	23.0	248676	0.0893	0.0408	0.0122
0.0192	24.0	259488	0.0878	0.0430	0.0118
0.0171	25.0	270300	0.0994	0.0473	0.0128
0.0156	26.0	281112	0.0892	0.0443	0.0123

Framework versions

• Transformers 4.46.1
• Pytorch 2.1.0+cu118
• Datasets 3.1.0
• Tokenizers 0.20.1