Whisper smail en - Song train
This model is a fine-tuned version of Zzzkay1/whisper-small-en on the English songs * 25 dataset. It achieves the following results on the evaluation set:
- Loss: 2.0349
- Wer: 19.7354
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: 4
- eval_batch_size: 8
- seed: 42
- gradient_accumulation_steps: 4
- total_train_batch_size: 16
- 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: 200
- training_steps: 10000
- mixed_precision_training: Native AMP
- label_smoothing_factor: 0.1
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|---|---|---|---|---|
| 2.2808 | 0.6780 | 100 | 2.1672 | 29.8942 |
| 1.9497 | 1.3525 | 200 | 2.0260 | 25.5556 |
| 1.8937 | 2.0271 | 300 | 1.9647 | 22.4339 |
| 1.6921 | 2.7051 | 400 | 1.9518 | 22.0106 |
| 1.5627 | 3.3797 | 500 | 1.9594 | 20.3704 |
| 1.5424 | 4.0542 | 600 | 1.9702 | 20.0529 |
| 1.5121 | 4.7322 | 700 | 1.9520 | 19.5238 |
| 1.4968 | 5.4068 | 800 | 1.9692 | 30.8466 |
| 1.4735 | 6.0814 | 900 | 1.9945 | 20.2116 |
| 1.4767 | 6.7593 | 1000 | 1.9860 | 20.0 |
| 1.4672 | 7.4339 | 1100 | 1.9913 | 18.5714 |
| 1.4524 | 8.1085 | 1200 | 2.0059 | 19.9471 |
| 1.4523 | 8.7864 | 1300 | 2.0060 | 19.2593 |
| 1.4535 | 9.4610 | 1400 | 2.0232 | 19.9471 |
| 1.4464 | 10.1356 | 1500 | 2.0300 | 19.1534 |
| 1.4442 | 10.8136 | 1600 | 2.0349 | 19.7354 |
Framework versions
- Transformers 4.56.2
- Pytorch 2.7.1+cu118
- Datasets 4.1.1
- Tokenizers 0.22.1