whisper-a-normal-ls

This model is a fine-tuned version of openai/whisper-small on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.0292
• Wer: 3.1854

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: 0.0004
• 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_steps: 132
• num_epochs: 11
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
No log	1.0	70	0.2501	18.9989
0.8254	2.0	140	0.1896	25.1422
0.1344	3.0	210	0.0928	29.1240
0.1344	4.0	280	0.0523	17.8612
0.054	5.0	350	0.0681	33.4471
0.0174	6.0	420	0.0446	11.9454
0.0174	7.0	490	0.0394	7.0535
0.0121	8.0	560	0.0314	5.6883
0.0004	9.0	630	0.0292	3.2992
0.0	10.0	700	0.0293	3.1854
0.0	10.8489	759	0.0292	3.1854

Framework versions

• Transformers 4.47.0.dev0
• Pytorch 2.4.0
• Datasets 3.0.1
• Tokenizers 0.20.0