combined_whisper_sanlish_lr2e5_newsplit
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: 0.6368
- Wer: 34.7012
- Cer: 8.4391
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: 2e-05
- train_batch_size: 8
- eval_batch_size: 4
- seed: 42
- gradient_accumulation_steps: 2
- total_train_batch_size: 16
- 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: 25
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer | Cer |
|---|---|---|---|---|---|
| 10.2229 | 1.0 | 117 | 1.9545 | 77.7877 | 21.9067 |
| 3.6557 | 2.0 | 234 | 0.8444 | 51.7395 | 13.2614 |
| 2.5242 | 3.0 | 351 | 0.7170 | 43.2649 | 11.1516 |
| 1.9480 | 4.0 | 468 | 0.6318 | 39.8751 | 9.8985 |
| 1.4886 | 5.0 | 585 | 0.6287 | 39.2507 | 9.6764 |
| 1.2852 | 6.0 | 702 | 0.6565 | 36.8421 | 9.4702 |
| 0.9506 | 7.0 | 819 | 0.6290 | 35.8608 | 9.1529 |
| 0.8017 | 8.0 | 936 | 0.6124 | 35.0580 | 8.7405 |
| 0.7469 | 9.0 | 1053 | 0.6339 | 34.4335 | 8.6612 |
| 0.6137 | 10.0 | 1170 | 0.6147 | 34.1659 | 8.4867 |
| 0.4990 | 11.0 | 1287 | 0.6087 | 35.2364 | 8.5977 |
| 0.4281 | 12.0 | 1404 | 0.6068 | 34.7012 | 8.5343 |
| 0.3966 | 13.0 | 1521 | 0.6225 | 34.3443 | 8.5343 |
| 0.3693 | 14.0 | 1638 | 0.6183 | 34.3443 | 8.5819 |
| 0.3288 | 15.0 | 1755 | 0.6368 | 34.7012 | 8.4391 |
Framework versions
- Transformers 5.16.1
- Pytorch 2.11.0+cu128
- Datasets 4.8.5
- Tokenizers 0.23.1