Whisper largeV2 French MLS
This model is a fine-tuned version of openai/whisper-large-v2 on the facebook/multilingual_librispeech french dataset. It achieves the following results on the evaluation set:
- Loss: 0.0903
- Wer: 4.5616
Model description
The model is fine-tuned for 4000 updates/steps on multilingual librispeech French train data.
- Zero-shot - 7.3 (MLS French test)
- Fine-tune MLS French train - 4.56 (MLS French test) (-37.5%)
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: 8
- eval_batch_size: 8
- seed: 42
- gradient_accumulation_steps: 8
- total_train_batch_size: 64
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- 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 |
|---|---|---|---|---|
| 0.1303 | 0.25 | 1000 | 0.1219 | 6.3618 |
| 0.0751 | 0.5 | 2000 | 0.1033 | 5.3905 |
| 0.0613 | 0.75 | 3000 | 0.0970 | 4.9193 |
| 0.0796 | 1.0 | 4000 | 0.0903 | 4.5616 |
Framework versions
- Transformers 4.26.0.dev0
- Pytorch 1.13.0+cu117
- Datasets 2.7.1.dev0
- Tokenizers 0.13.2