Whisper Base Vietnamese
This model is a fine-tuned version of openai/whisper-base on the mozilla-foundation/common_voice_16_0 vi dataset. It achieves the following results on the evaluation set:
- Loss: 0.7770
- Wer: 37.8024
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: 5e-07
- train_batch_size: 32
- eval_batch_size: 32
- seed: 42
- gradient_accumulation_steps: 2
- 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: 10000
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|---|---|---|---|---|
| 0.6043 | 33.0 | 500 | 0.9039 | 42.6408 |
| 0.2836 | 66.0 | 1000 | 0.7761 | 38.3106 |
| 0.1593 | 99.0 | 1500 | 0.7770 | 37.8024 |
| 0.0835 | 133.0 | 2000 | 0.8019 | 37.8634 |
| 0.0395 | 166.0 | 2500 | 0.8317 | 38.1582 |
| 0.0217 | 199.0 | 3000 | 0.8563 | 38.2395 |
| 0.0146 | 233.0 | 3500 | 0.8744 | 38.2801 |
| 0.0107 | 266.0 | 4000 | 0.8893 | 38.4733 |
| 0.0082 | 299.0 | 4500 | 0.9031 | 38.3310 |
| 0.0065 | 333.0 | 5000 | 0.9155 | 38.4326 |
| 0.0053 | 366.0 | 5500 | 0.9267 | 38.6156 |
| 0.0044 | 399.0 | 6000 | 0.9381 | 38.7579 |
| 0.0037 | 433.0 | 6500 | 0.9486 | 38.7782 |
| 0.0032 | 466.0 | 7000 | 0.9580 | 39.0120 |
| 0.0028 | 499.0 | 7500 | 0.9669 | 39.1441 |
| 0.0025 | 533.0 | 8000 | 0.9747 | 39.1746 |
| 0.0022 | 566.0 | 8500 | 0.9810 | 39.2864 |
| 0.0021 | 599.0 | 9000 | 0.9866 | 39.2763 |
| 0.002 | 633.0 | 9500 | 0.9899 | 39.3271 |
| 0.0019 | 666.0 | 10000 | 0.9911 | 39.3271 |
Framework versions
- Transformers 4.37.0.dev0
- Pytorch 2.1.2+cu121
- Datasets 2.16.2.dev0
- Tokenizers 0.15.0