200-real-voices
This model is a fine-tuned version of openai/whisper-small on the 200-real-voices dataset. It achieves the following results on the evaluation set:
- Loss: 0.3355
- Wer: 16.4105
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: 8
- eval_batch_size: 4
- seed: 42
- optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
- lr_scheduler_type: linear
- num_epochs: 10
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|---|---|---|---|---|
| No log | 0 | 0 | 0.6890 | 41.0325 |
| 0.398 | 1.0 | 25 | 0.3067 | 19.6131 |
| 0.0985 | 2.0 | 50 | 0.2828 | 17.6403 |
| 0.0366 | 3.0 | 75 | 0.2957 | 17.1919 |
| 0.0209 | 4.0 | 100 | 0.3042 | 16.5129 |
| 0.0082 | 5.0 | 125 | 0.3147 | 16.4105 |
| 0.0048 | 6.0 | 150 | 0.3252 | 16.3976 |
| 0.0038 | 7.0 | 175 | 0.3309 | 16.4105 |
| 0.0028 | 8.0 | 200 | 0.3333 | 16.5257 |
| 0.0024 | 9.0 | 225 | 0.3351 | 16.3848 |
| 0.0022 | 10.0 | 250 | 0.3355 | 16.4105 |
Framework versions
- Transformers 4.49.0
- Pytorch 2.5.1+cu121
- Datasets 3.4.1
- Tokenizers 0.21.0