whisper-small-bengali-5DB-3e6
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: 0.3547
- Wer: 77.3935
- Cer: 31.3745
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: 3e-06
- train_batch_size: 32
- eval_batch_size: 16
- seed: 42
- gradient_accumulation_steps: 2
- total_train_batch_size: 64
- 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: 100
- num_epochs: 10
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer | Cer |
|---|---|---|---|---|---|
| 4.3494 | 1.2675 | 100 | 1.0599 | 100.3706 | 62.1956 |
| 2.6315 | 2.5350 | 200 | 0.6637 | 85.7937 | 38.9817 |
| 1.9725 | 3.8025 | 300 | 0.5386 | 81.7789 | 34.8863 |
| 1.4710 | 5.0637 | 400 | 0.4409 | 79.6788 | 33.5581 |
| 1.0820 | 6.3312 | 500 | 0.3666 | 78.0729 | 32.9946 |
| 0.9373 | 7.5987 | 600 | 0.3586 | 77.0229 | 31.8273 |
| 0.8190 | 8.8662 | 700 | 0.3547 | 77.3935 | 31.3745 |
Framework versions
- Transformers 5.0.0
- Pytorch 2.10.0+cu128
- Datasets 4.8.5
- Tokenizers 0.22.2