whisper-a-nomimose-trial

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: 0.0481
• Wer: 132.6667

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: 20
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.6555	1.0	109	0.8617	56.6667
0.154	2.0	218	0.2293	52.3333
0.0778	3.0	327	0.3132	63.3333
0.0633	4.0	436	0.1483	80.0
0.0503	5.0	545	0.3422	54.6667
0.0533	6.0	654	0.1111	186.0
0.0332	7.0	763	0.1978	101.0
0.0242	8.0	872	0.1586	62.0
0.0204	9.0	981	0.0867	59.3333
0.0191	10.0	1090	0.1203	179.0
0.0139	11.0	1199	0.0813	40.6667
0.0083	12.0	1308	0.0827	60.0
0.0052	13.0	1417	0.0945	131.3333
0.0055	14.0	1526	0.0923	182.6667
0.0055	15.0	1635	0.0937	152.6667
0.0045	16.0	1744	0.0532	117.6667
0.0024	17.0	1853	0.0575	126.3333
0.002	18.0	1962	0.0505	136.0
0.0022	19.0	2071	0.0491	136.3333
0.0013	19.8203	2160	0.0481	132.6667

Framework versions

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