whisper-lt-finetune
This model is a fine-tuned version of openai/whisper-small on the Common Voice 11.0 dataset. It achieves the following results on the evaluation set:
- Loss: 0.2588
- Wer: 14.2619
Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 1e-05
- train_batch_size: 16
- 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: 250
- training_steps: 4000
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|---|---|---|---|---|
| 0.0783 | 1.3 | 1000 | 0.2478 | 15.5647 |
| 0.0287 | 2.6 | 2000 | 0.2441 | 14.3765 |
| 0.0087 | 3.9 | 3000 | 0.2516 | 14.3349 |
| 0.0021 | 5.19 | 4000 | 0.2588 | 14.2619 |
Framework versions
- Transformers 4.26.0.dev0
- Pytorch 1.12.1+cu113
- Datasets 2.7.1
- Tokenizers 0.13.2
