whisper-small-santali-sanlish-revised2_211sen
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: 1.1669
- Wer: 62.5
- Cer: 16.2162
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: 4
- seed: 42
- optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED 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: 25
- num_epochs: 20
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer | Cer |
|---|---|---|---|---|---|
| 2.2761 | 1.0 | 42 | 1.1518 | 85.0 | 23.6036 |
| 0.6258 | 2.0 | 84 | 0.9264 | 58.75 | 18.1982 |
| 0.2154 | 3.0 | 126 | 0.9527 | 61.2500 | 18.1982 |
| 0.0842 | 4.0 | 168 | 0.9710 | 55.0000 | 16.5766 |
| 0.0357 | 5.0 | 210 | 1.0940 | 72.5 | 20.3604 |
| 0.0351 | 6.0 | 252 | 1.0147 | 70.0 | 18.0180 |
| 0.0120 | 7.0 | 294 | 1.0036 | 51.2500 | 16.0360 |
| 0.0047 | 8.0 | 336 | 1.1085 | 62.5 | 16.3964 |
| 0.0074 | 9.0 | 378 | 1.0849 | 55.0000 | 15.4955 |
| 0.0039 | 10.0 | 420 | 1.1564 | 65.0 | 17.4775 |
| 0.0009 | 11.0 | 462 | 1.1331 | 60.0 | 15.8559 |
| 0.0005 | 12.0 | 504 | 1.1476 | 60.0 | 15.8559 |
| 0.0004 | 13.0 | 546 | 1.1567 | 60.0 | 16.3964 |
| 0.0004 | 14.0 | 588 | 1.1608 | 60.0 | 16.3964 |
| 0.0003 | 15.0 | 630 | 1.1669 | 62.5 | 16.2162 |
Framework versions
- Transformers 5.13.1
- Pytorch 2.11.0+cu128
- Datasets 4.0.0
- Tokenizers 0.22.2