whisper_med_alex.en
This model is a fine-tuned version of crossdelenna/whisper_med_alex.en on an unknown dataset. It achieves the following results on the evaluation set:
- Loss: 0.0002
- Wer: 5.3233
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-06
- train_batch_size: 24
- eval_batch_size: 20
- seed: 42
- optimizer: Use OptimizerNames.ADAMW_TORCH 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: 10
- training_steps: 7501
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|---|---|---|---|---|
| 0.3202 | 2.6087 | 300 | 0.1248 | 10.6277 |
| 0.1006 | 5.2174 | 600 | 0.0287 | 6.0217 |
| 0.0297 | 7.8261 | 900 | 0.0127 | 5.0118 |
| 0.0135 | 10.4348 | 1200 | 0.0041 | 6.0500 |
| 0.0048 | 13.0435 | 1500 | 0.0025 | 5.5970 |
| 0.0034 | 15.6522 | 1800 | 0.0022 | 5.5687 |
| 0.0027 | 18.2609 | 2100 | 0.0015 | 5.3610 |
| 0.0018 | 20.8696 | 2400 | 0.0012 | 5.2950 |
| 0.0016 | 23.4783 | 2700 | 0.0011 | 5.3044 |
| 0.0013 | 26.0870 | 3000 | 0.0009 | 5.2761 |
| 0.0011 | 28.6957 | 3300 | 0.0008 | 5.2478 |
| 0.0009 | 31.3043 | 3600 | 0.0006 | 5.3044 |
| 0.0008 | 33.9130 | 3900 | 0.0005 | 5.3044 |
| 0.0007 | 36.5217 | 4200 | 0.0005 | 5.2855 |
| 0.0006 | 39.1304 | 4500 | 0.0004 | 5.3327 |
| 0.0005 | 41.7391 | 4800 | 0.0004 | 5.3705 |
| 0.0005 | 44.3478 | 5100 | 0.0004 | 5.3516 |
| 0.0005 | 46.9565 | 5400 | 0.0003 | 5.2855 |
| 0.0004 | 49.5652 | 5700 | 0.0003 | 5.3327 |
| 0.0004 | 52.1739 | 6000 | 0.0003 | 5.2950 |
| 0.0003 | 54.7826 | 6300 | 0.0002 | 5.2950 |
| 0.0003 | 57.3913 | 6600 | 0.0002 | 5.3327 |
| 0.0003 | 60.0 | 6900 | 0.0002 | 5.3138 |
| 0.0002 | 62.6087 | 7200 | 0.0002 | 5.3233 |
| 0.0002 | 65.2174 | 7500 | 0.0002 | 5.3233 |
Framework versions
- Transformers 4.48.3
- Pytorch 2.5.1+cu124
- Datasets 3.3.2
- Tokenizers 0.21.0