miosipof/asr2_medium_v0.6

Whisper Medium

This model is a fine-tuned version of openai/whisper-medium on the b-brave-clean dataset. It achieves the following results on the evaluation set:

• Loss: 0.4423
• Wer: 37.1060
• Cer: 27.9222
• Lr: 0.0000

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: 0.0003
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• 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
• lr_scheduler_warmup_ratio: 0.3
• num_epochs: 16
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer	Lr
3.8497	1.0	168	2.1950	120.4871	80.3257	0.0001
1.0999	2.0	336	0.8517	88.6819	73.6538	0.0001
0.7773	3.0	504	0.7060	224.9284	212.4245	0.0002
0.5742	4.0	672	0.5562	50.1433	34.0163	0.0002
0.3652	5.0	840	0.5426	103.2951	88.8101	0.0003
0.2244	6.0	1008	0.5211	100.2865	66.5353	0.0003
0.1522	7.0	1176	0.4991	48.9971	35.5398	0.0002
0.0863	8.0	1344	0.4682	39.1117	28.3951	0.0002
0.0472	9.0	1512	0.4743	44.1261	30.9693	0.0002
0.021	10.0	1680	0.4590	40.5444	29.1305	0.0002
0.0106	11.0	1848	0.4460	37.3926	27.1868	0.0001
0.0053	12.0	2016	0.4420	36.6762	27.4494	0.0001
0.0034	13.0	2184	0.4395	37.9656	28.4476	0.0001
0.0021	14.0	2352	0.4398	36.8195	27.6596	0.0001
0.0023	15.0	2520	0.4422	37.1060	27.9222	0.0000
0.0021	15.9075	2672	0.4423	37.1060	27.9222	0.0000

Framework versions

• PEFT 0.14.0
• Transformers 4.48.3
• Pytorch 2.2.0
• Datasets 3.2.0
• Tokenizers 0.21.0