wav2vecvanilla_ctc_zero_infinity_longertrain

This model is a fine-tuned version of facebook/wav2vec2-base-960h on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 1.1396
• Wer: 0.2973

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: 4
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 15

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.4721	0.43	100	1.0565	0.4014
1.2574	0.85	200	0.9707	0.3704
1.1397	1.28	300	0.9644	0.3609
1.0939	1.71	400	0.9610	0.3637
1.0874	2.14	500	0.9508	0.3581
1.0573	2.56	600	0.8865	0.3518
1.0386	2.99	700	1.0304	0.3493
0.9792	3.42	800	0.8235	0.3523
0.9789	3.85	900	0.8404	0.3388
0.9095	4.27	1000	1.0925	0.3588
0.8947	4.7	1100	1.0126	0.3357
0.8571	5.13	1200	1.1404	0.3550
0.8276	5.56	1300	0.8135	0.3294
0.8631	5.98	1400	0.8342	0.3279
0.8134	6.41	1500	0.8524	0.3177
0.8027	6.84	1600	0.8182	0.3207
0.7556	7.26	1700	0.8445	0.3185
0.737	7.69	1800	0.8919	0.3197
0.7398	8.12	1900	0.8115	0.3167
0.7069	8.55	2000	0.8346	0.3174
0.7206	8.97	2100	0.9714	0.3147
0.6946	9.4	2200	0.8138	0.3124
0.6752	9.83	2300	0.8366	0.3086
0.7256	10.26	2400	0.8482	0.3044
0.7063	10.68	2500	0.8997	0.3041
0.6399	11.11	2600	0.8614	0.3045
0.6268	11.54	2700	0.8564	0.3018
0.6665	11.97	2800	0.8531	0.3006
0.622	12.39	2900	0.8759	0.3007
0.6568	12.82	3000	1.3093	0.3023
0.6296	13.25	3100	1.1312	0.3002
0.6448	13.68	3200	1.1779	0.2994
0.6188	14.1	3300	1.1203	0.2989
0.6216	14.53	3400	1.1421	0.2978
0.6238	14.96	3500	1.1396	0.2973

Framework versions

• Transformers 4.38.2
• Pytorch 2.2.1+cu121
• Datasets 2.18.0
• Tokenizers 0.15.2