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: 0.7947
- Wer: 0.3971
- Cer: 0.1473
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 |
|---|---|---|---|---|---|
| 0.8839 | 1.0 | 266 | 0.7407 | 0.4685 | 0.1726 |
| 0.5422 | 2.0 | 532 | 0.6039 | 0.4253 | 0.1538 |
| 0.3711 | 3.0 | 798 | 0.5960 | 0.4095 | 0.1490 |
| 0.2373 | 4.0 | 1064 | 0.6073 | 0.4032 | 0.1464 |
| 0.1341 | 5.0 | 1330 | 0.6548 | 0.4016 | 0.1470 |
| 0.0791 | 6.0 | 1596 | 0.6893 | 0.4035 | 0.1457 |
| 0.0429 | 7.0 | 1862 | 0.7218 | 0.4055 | 0.1495 |
| 0.0199 | 8.0 | 2128 | 0.7657 | 0.3949 | 0.1421 |
| 0.0132 | 9.0 | 2394 | 0.7834 | 0.3979 | 0.1450 |
| 0.0066 | 10.0 | 2660 | 0.7947 | 0.3971 | 0.1473 |
Framework versions
- Transformers 4.57.3
- Pytorch 2.8.0+cu128
- Datasets 3.6.0
- Tokenizers 0.22.2