Whisper Tiny GA-EN Speech Translation v.1.4
This model is a fine-tuned version of openai/whisper-tiny on an unknown dataset. The best model (this version) is at checkpoint 500, epoch 7.35, and it achieves the following results on the evaluation set:
- Loss: 2.5784
- Bleu: 17.0
- Chrf: 30.42
- Wer: 85.8622
Model description
More information needed
Intended uses & limitations
More information needed
Training and evaluation data
More information needed
Training procedure
Experiment
As this is a translation task into English, use language="English"
tokenizer = WhisperTokenizer.from_pretrained("openai/whisper-tiny",
cache_dir=cache_dir,
language="english",
task="translate")
Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 0.0001
- train_batch_size: 128
- eval_batch_size: 128
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 0.03
- training_steps: 1000
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Bleu | Chrf | Wer |
|---|---|---|---|---|---|---|
| 0.8117 | 1.47 | 100 | 2.1773 | 8.81 | 24.15 | 119.9910 |
| 0.3555 | 2.94 | 200 | 2.2596 | 11.81 | 27.7 | 110.8510 |
| 0.0872 | 4.41 | 300 | 2.4108 | 12.89 | 28.57 | 110.5808 |
| 0.063 | 5.88 | 400 | 2.5306 | 12.67 | 27.78 | 107.1139 |
| 0.034 | 7.35 | 500 | 2.5784 | 17.0 | 30.42 | 85.8622 |
| 0.0349 | 8.82 | 600 | 2.6201 | 16.64 | 29.98 | 86.8978 |
| 0.0198 | 10.29 | 700 | 2.7151 | 16.0 | 30.04 | 88.2936 |
| 0.0134 | 11.76 | 800 | 2.7159 | 14.16 | 30.03 | 105.0878 |
| 0.0088 | 13.24 | 900 | 2.7369 | 14.61 | 29.32 | 94.3269 |
| 0.0056 | 14.71 | 1000 | 2.7420 | 14.71 | 30.02 | 95.7226 |
Framework versions
- Transformers 4.39.2
- Pytorch 2.2.1+cu121
- Datasets 2.18.0
- Tokenizers 0.15.2