ndizi-whisper-small-optimized
This model is a fine-tuned version of openai/whisper-small on an unknown dataset. It achieves the following results on the evaluation set:
- Loss: 0.8405
- Wer: 0.4023
- Cer: 0.1522
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
- gradient_accumulation_steps: 2
- total_train_batch_size: 8
- 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: 200
- num_epochs: 10
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer | Cer |
|---|---|---|---|---|---|
| 1.5773 | 1.0 | 137 | 1.0763 | 0.5543 | 0.2140 |
| 0.8057 | 2.0 | 274 | 0.7256 | 0.4444 | 0.1627 |
| 0.4219 | 3.0 | 411 | 0.6550 | 0.4226 | 0.1669 |
| 0.27 | 4.0 | 548 | 0.6721 | 0.4086 | 0.1583 |
| 0.1647 | 5.0 | 685 | 0.6997 | 0.4299 | 0.1664 |
| 0.0989 | 6.0 | 822 | 0.7398 | 0.4098 | 0.1526 |
| 0.0636 | 7.0 | 959 | 0.7776 | 0.4101 | 0.1561 |
| 0.0376 | 8.0 | 1096 | 0.8047 | 0.3984 | 0.1502 |
| 0.0207 | 9.0 | 1233 | 0.8293 | 0.4025 | 0.1509 |
| 0.0114 | 10.0 | 1370 | 0.8405 | 0.4023 | 0.1522 |
Framework versions
- Transformers 4.57.3
- Pytorch 2.8.0+cu128
- Datasets 3.6.0
- Tokenizers 0.22.2