whisper-waxal-checkpoints
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: 0.5937
- Wer: 26.8326
- Cer: 9.9005
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: 1
- eval_batch_size: 1
- seed: 42
- gradient_accumulation_steps: 8
- 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
- training_steps: 4000
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Cer | Validation Loss | Wer |
|---|---|---|---|---|---|
| 3.2537 | 0.0625 | 250 | 30.9624 | 0.9343 | 63.2244 |
| 2.4887 | 0.125 | 500 | 26.6279 | 0.7615 | 53.3804 |
| 3.9830 | 0.1875 | 750 | 12.0267 | 0.7395 | 32.5904 |
| 3.9534 | 0.25 | 1000 | 11.6150 | 0.7234 | 31.8722 |
| 3.2836 | 0.3125 | 1250 | 11.4364 | 0.7015 | 30.9807 |
| 3.9759 | 0.375 | 1500 | 11.1741 | 0.6767 | 29.9282 |
| 3.7271 | 0.4375 | 1750 | 11.0702 | 0.6653 | 29.7424 |
| 4.4242 | 0.5 | 2000 | 10.7004 | 0.6491 | 29.0738 |
| 3.6541 | 0.5625 | 2250 | 10.7915 | 0.6335 | 28.6033 |
| 3.2283 | 0.625 | 2500 | 10.3815 | 0.6301 | 28.2442 |
| 2.9001 | 0.6875 | 2750 | 10.3961 | 0.6259 | 28.0461 |
| 3.0438 | 0.75 | 3000 | 10.3761 | 0.6171 | 27.7737 |
| 3.5500 | 0.8125 | 3250 | 10.0791 | 0.6093 | 27.1421 |
| 3.5308 | 0.875 | 3500 | 10.0080 | 0.5994 | 26.8945 |
| 3.6616 | 0.9375 | 3750 | 0.5933 | 26.7088 | 9.9297 |
| 3.7414 | 1.0 | 4000 | 0.5937 | 26.8326 | 9.9005 |
Framework versions
- Transformers 5.13.1
- Pytorch 2.11.0+cu128
- Datasets 5.0.0
- Tokenizers 0.22.2