Whisper Small GA-EN Speech Translation
This model is a fine-tuned version of openai/whisper-small on an unknown dataset. The best model (this version) is at checkpoint 1400, epoch 1.51, and it achieves the following results on the evaluation set:
- Loss: 1.3989
- Bleu: 28.53
- Chrf: 44.93
- Wer: 68.1675
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: 0.0001
- train_batch_size: 32
- eval_batch_size: 32
- 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: 1500
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Bleu | Chrf | Validation Loss | Wer |
|---|---|---|---|---|---|---|
| 2.2789 | 0.11 | 100 | 9.07 | 25.39 | 2.0838 | 102.2963 |
| 1.9858 | 0.22 | 200 | 12.68 | 29.42 | 1.7854 | 101.1706 |
| 1.6904 | 0.32 | 300 | 11.93 | 31.4 | 1.6522 | 148.2215 |
| 1.4934 | 0.43 | 400 | 16.44 | 35.2 | 1.5699 | 95.3174 |
| 1.371 | 0.54 | 500 | 15.89 | 34.46 | 1.5181 | 100.9455 |
| 1.1806 | 0.65 | 600 | 20.62 | 40.11 | 1.4475 | 91.8955 |
| 1.0781 | 0.76 | 700 | 18.55 | 40.22 | 1.4067 | 99.5948 |
| 0.9166 | 0.86 | 800 | 26.87 | 43.16 | 1.4104 | 71.3192 |
| 0.848 | 0.97 | 900 | 25.95 | 42.61 | 1.3556 | 75.6866 |
| 0.3712 | 1.08 | 1000 | 22.4 | 41.02 | 1.3936 | 87.2580 |
| 0.4415 | 1.19 | 1100 | 28.13 | 43.0 | 1.4157 | 68.0324 |
| 0.4166 | 1.29 | 1200 | 27.75 | 44.39 | 1.4206 | 71.1391 |
| 0.387 | 1.4 | 1300 | 28.48 | 44.44 | 1.4083 | 69.4282 |
| 0.3714 | 1.51 | 1400 | 28.53 | 44.93 | 1.3989 | 68.1675 |
| 0.3695 | 1.62 | 1500 | 26.13 | 43.65 | 1.4049 | 76.9923 |
Framework versions
- Transformers 4.39.3
- Pytorch 2.2.1+cu121
- Datasets 2.18.0
- Tokenizers 0.15.2