Whisper Small Tr - CV 43h - LLR
This model is a fine-tuned version of openai/whisper-small on the Common Voice 16.1 dataset. It achieves the following results on the evaluation set:
- Loss: 0.2477
- Wer: 21.3892
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-06
- train_batch_size: 16
- eval_batch_size: 8
- seed: 42
- gradient_accumulation_steps: 2
- total_train_batch_size: 32
- 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 |
|---|---|---|---|---|
| 0.2468 | 0.37 | 500 | 0.2886 | 24.3238 |
| 0.2099 | 0.73 | 1000 | 0.2673 | 22.8161 |
| 0.1841 | 1.1 | 1500 | 0.2577 | 22.0433 |
| 0.1767 | 1.46 | 2000 | 0.2540 | 21.8600 |
| 0.1718 | 1.83 | 2500 | 0.2504 | 21.6444 |
| 0.1629 | 2.19 | 3000 | 0.2492 | 21.6120 |
| 0.1693 | 2.56 | 3500 | 0.2486 | 21.4161 |
| 0.1594 | 2.92 | 4000 | 0.2477 | 21.3892 |
Framework versions
- Transformers 4.39.3
- Pytorch 2.2.1+cu121
- Datasets 2.18.0
- Tokenizers 0.15.2