A LoRA-trained Whisper-Large-v3-turbo specifically in Yue-Chinese (lang_code: zh)
This model is a fine-tuned version of openai/whisper-large-v3-turbo on the Yue Dataset with Hong Kong style dataset. It achieves the following results on the evaluation set:
- Loss: 0.4710
- Cer: 28.9276
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: 0.0002
- train_batch_size: 16
- eval_batch_size: 16
- seed: 42
- gradient_accumulation_steps: 4
- total_train_batch_size: 64
- 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: 10
- training_steps: 500
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Cer |
|---|---|---|---|---|
| 1.1987 | 0.9324 | 100 | 0.6298 | 36.2358 |
| 0.5535 | 1.8578 | 200 | 0.5287 | 30.7884 |
| 0.4861 | 2.7832 | 300 | 0.4904 | 30.6960 |
| 0.4392 | 3.7086 | 400 | 0.4772 | 28.6151 |
| 0.4276 | 4.6340 | 500 | 0.4710 | 28.9276 |
Framework versions
- PEFT 0.18.0
- Transformers 4.57.1
- Pytorch 2.9.1+cu128
- Datasets 4.4.1
- Tokenizers 0.22.1