Whisper Small JA - Lorenzo Concina
This model is a fine-tuned version of [SVJ Japanese dataset](https://huggingface.co/SVJ Japanese dataset) on the Common Voice 11.0 dataset. It achieves the following results on the evaluation set:
- Loss: 0.5596
- Cer: 17.7261
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: 4
- eval_batch_size: 8
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 500
- training_steps: 5000
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Cer |
|---|---|---|---|---|
| 0.0682 | 0.33 | 1000 | 0.6098 | 19.9133 |
| 0.0229 | 0.67 | 2000 | 0.5501 | 18.1911 |
| 0.0457 | 1.0 | 3000 | 0.5745 | 19.3174 |
| 0.0123 | 1.34 | 4000 | 0.5315 | 18.2546 |
| 0.0238 | 1.67 | 5000 | 0.5596 | 17.7261 |
Framework versions
- Transformers 4.26.0.dev0
- Pytorch 1.13.0
- Datasets 2.7.1
- Tokenizers 0.11.0
- Downloads last month
- 0