whisper-small-accented-zh-4000steps
This model is a fine-tuned version of openai/whisper-small on the SELL-Corpus dataset. It achieves the following results on the evaluation set:
- Loss: 0.1960
- Wer: 6.1749
- Cer: 2.9133
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: 16
- eval_batch_size: 16
- 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
- lr_scheduler_warmup_steps: 500
- training_steps: 4000
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer | Cer |
|---|---|---|---|---|---|
| 0.0862 | 1.6393 | 1000 | 0.1556 | 6.6008 | 3.0774 |
| 0.0072 | 3.2787 | 2000 | 0.1739 | 6.1835 | 2.8870 |
| 0.0029 | 4.9180 | 3000 | 0.1851 | 6.1473 | 2.8853 |
| 0.0009 | 6.5574 | 4000 | 0.1960 | 6.1749 | 2.9133 |
Framework versions
- Transformers 4.46.3
- Pytorch 2.4.0
- Datasets 3.1.0
- Tokenizers 0.20.0