Whisper Tiny chinese - VingeNie
This model is a fine-tuned version of openai/whisper-tiny on the Common Voice 16.1 dataset. It achieves the following results on the evaluation set:
- Loss: 0.7681
- Cer Ortho: 38.5858
- Cer: 29.8372
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: 0.0001
- train_batch_size: 64
- eval_batch_size: 32
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: constant_with_warmup
- lr_scheduler_warmup_steps: 25
- training_steps: 1000
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Cer Ortho | Cer |
|---|---|---|---|---|---|
| 0.4971 | 0.5435 | 250 | 0.7639 | 39.3347 | 30.8173 |
| 0.3054 | 1.0870 | 500 | 0.7634 | 35.6305 | 29.4969 |
| 0.4034 | 1.6304 | 750 | 0.7576 | 38.6415 | 30.4015 |
| 0.2556 | 2.1739 | 1000 | 0.7681 | 38.5858 | 29.8372 |
Framework versions
- Transformers 4.40.0
- Pytorch 2.2.1+cu121
- Datasets 2.19.0
- Tokenizers 0.19.1
- Downloads last month
- 8