PradyumSomebody/whisper-small-hi-custom3

Whisper Small EN - Pradyum Agarwal

This model is a fine-tuned version of openai/whisper-small on the Audio Medical Combined Dataset dataset. It achieves the following results on the evaluation set:

• Loss: 0.1223
• Wer: 4.8794

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: 10
• 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
• training_steps: 120
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.7764	0.25	10	0.8395	18.9812
0.796	0.5	20	0.8292	18.2306
0.7006	0.75	30	0.7916	17.1046
0.6618	1.0	40	0.7266	15.7105
0.621	1.25	50	0.6598	13.9946
0.4566	1.5	60	0.5669	12.3861
0.3225	1.75	70	0.4023	11.5282
0.2363	2.0	80	0.2502	10.9920
0.1248	2.25	90	0.2035	9.3298
0.1482	2.5	100	0.1727	7.9357
0.1016	2.75	110	0.1462	6.4879
0.1274	3.0	120	0.1223	4.8794

Framework versions

• Transformers 4.42.4
• Pytorch 2.3.0+cu121
• Datasets 2.20.0
• Tokenizers 0.19.1