whisper-small-santali-sanlish-revised2_211sen_withouthyphen
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: 1.1378
- Wer: 51.7241
- Cer: 15.4955
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 |
|---|---|---|---|---|---|
| 2.2762 | 1.0 | 42 | 1.1519 | 82.7586 | 22.5225 |
| 0.6259 | 2.0 | 84 | 0.9254 | 52.5862 | 17.2973 |
| 0.2153 | 3.0 | 126 | 0.9534 | 53.4483 | 17.2973 |
| 0.0843 | 4.0 | 168 | 0.9699 | 49.1379 | 15.4955 |
| 0.0386 | 5.0 | 210 | 1.0676 | 66.3793 | 18.0180 |
| 0.0310 | 6.0 | 252 | 1.0169 | 54.3103 | 15.6757 |
| 0.0122 | 7.0 | 294 | 0.9931 | 49.1379 | 15.8559 |
| 0.0041 | 8.0 | 336 | 1.0642 | 53.4483 | 16.2162 |
| 0.0088 | 9.0 | 378 | 1.0821 | 52.5862 | 15.8559 |
| 0.0010 | 10.0 | 420 | 1.1472 | 54.3103 | 16.3964 |
| 0.0008 | 11.0 | 462 | 1.1281 | 53.4483 | 16.0360 |
| 0.0005 | 12.0 | 504 | 1.1378 | 51.7241 | 15.4955 |
Framework versions
- Transformers 5.13.1
- Pytorch 2.11.0+cu128
- Datasets 4.0.0
- Tokenizers 0.22.2