shreyasdesaisuperU/whisper-large-attempt1-with-silence-text

Whisper Large with Silence text SSD superU

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

• Loss: 3.8229
• Wer: 121.3467

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: 16
• eval_batch_size: 8
• seed: 42
• optimizer: Use OptimizerNames.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: 500
• training_steps: 2000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
2.548	3.125	100	2.5446	141.6905
1.4783	6.25	200	2.2643	150.7736
0.5697	9.375	300	2.6203	129.0831
0.1527	12.5	400	3.0222	129.9140
0.0793	15.625	500	3.0603	122.6361
0.0493	18.75	600	3.1006	139.5989
0.0373	21.875	700	3.2537	109.8281
0.0285	25.0	800	3.3265	130.6877
0.0132	28.125	900	3.3798	143.0659
0.012	31.25	1000	3.3797	147.6218
0.0069	34.375	1100	3.4862	105.7880
0.0028	37.5	1200	3.5677	133.3524
0.0009	40.625	1300	3.6265	133.0086
0.0004	43.75	1400	3.7364	128.3381
0.0003	46.875	1500	3.7760	126.2178
0.0004	50.0	1600	3.7947	122.4355
0.0003	53.125	1700	3.8080	122.8367
0.0002	56.25	1800	3.8165	124.1261
0.0002	59.375	1900	3.8211	120.9169
0.0002	62.5	2000	3.8229	121.3467

Framework versions

• Transformers 4.46.2
• Pytorch 2.2.2+cu121
• Datasets 3.1.0
• Tokenizers 0.20.3