whisper-a-no-ag

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: 0.0365
• Wer: 26.0523

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: 30
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.7935	1.4317	100	0.1805	33.4471
0.164	2.8633	200	0.0531	18.6576
0.0556	4.2878	300	0.1691	21.3879
0.0531	5.7194	400	0.0423	26.6212
0.023	7.1439	500	0.1101	70.6485
0.027	8.5755	600	0.0749	25.3697
0.0092	10.0	700	0.0406	18.9989
0.0046	11.4317	800	0.0673	36.9738
0.0063	12.8633	900	0.0371	24.6871
0.0032	14.2878	1000	0.0428	27.1900
0.001	15.7194	1100	0.0536	27.3038
0.0001	17.1439	1200	0.0453	25.8248
0.0	18.5755	1300	0.0434	25.7110
0.0008	20.0	1400	0.0368	26.1661
0.0	21.4317	1500	0.0365	26.1661
0.0	22.8633	1600	0.0365	26.1661
0.0	24.2878	1700	0.0365	26.0523
0.0	25.7194	1800	0.0365	26.0523
0.0	27.1439	1900	0.0365	26.0523
0.0	28.5755	2000	0.0365	26.0523

Framework versions

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