whisper-a-nomimo-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: 0.0304
• Wer: 18.6667

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
1.139	1.0	104	0.2078	37.6667
0.1123	2.0	208	0.1434	40.0
0.0676	3.0	312	0.0929	32.3333
0.1027	4.0	416	0.1140	45.3333
0.0304	5.0	520	0.1683	44.0
0.0309	6.0	624	0.1440	41.0
0.0231	7.0	728	0.0904	36.3333
0.0184	8.0	832	0.0931	35.3333
0.0181	9.0	936	0.0715	30.6667
0.0118	10.0	1040	0.0890	35.6667
0.0095	11.0	1144	0.0414	21.0
0.0092	12.0	1248	0.0721	37.3333
0.0065	13.0	1352	0.0940	28.3333
0.0061	14.0	1456	0.0560	28.3333
0.0058	15.0	1560	0.0534	29.0000
0.004	16.0	1664	0.0375	20.0
0.0029	17.0	1768	0.0436	22.3333
0.0026	18.0	1872	0.0318	16.0
0.0021	19.0	1976	0.0319	19.6667
0.0015	19.8116	2060	0.0304	18.6667

Framework versions

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