Whisper Small GA-EN Speech Translation Raw + warmup_ratio=0.01
This model is a fine-tuned version of openai/whisper-small on the IWSLT-2023, FLEURS, BiteSize, and SpokenWords dataset. It achieves the following results on the evaluation set:
- Loss: 1.6820
- Bleu: 30.14
- Chrf: 44.97
- Wer: 68.7528
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: 64
- eval_batch_size: 64
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_ratio: 0.01
- training_steps: 3000
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Bleu | Chrf | Wer |
|---|---|---|---|---|---|---|
| 2.0448 | 0.2155 | 100 | 1.7532 | 10.55 | 28.24 | 117.7848 |
| 1.5028 | 0.4310 | 200 | 1.5587 | 17.12 | 35.11 | 93.2913 |
| 1.3161 | 0.6466 | 300 | 1.5073 | 17.41 | 37.84 | 104.0973 |
| 1.0537 | 0.8621 | 400 | 1.4560 | 17.49 | 38.19 | 103.8721 |
| 0.4704 | 1.0776 | 500 | 1.4946 | 16.44 | 36.79 | 102.9266 |
| 0.4589 | 1.2931 | 600 | 1.5117 | 21.55 | 38.66 | 82.6204 |
| 0.4343 | 1.5086 | 700 | 1.5146 | 24.5 | 41.43 | 75.5966 |
| 0.4068 | 1.7241 | 800 | 1.4547 | 28.37 | 45.27 | 65.6011 |
| 0.3757 | 1.9397 | 900 | 1.4957 | 27.0 | 43.66 | 67.4471 |
| 0.1388 | 2.1552 | 1000 | 1.5642 | 26.72 | 41.74 | 66.5016 |
| 0.1472 | 2.3707 | 1100 | 1.5845 | 27.74 | 42.67 | 68.1675 |
| 0.1408 | 2.5862 | 1200 | 1.5932 | 28.95 | 44.07 | 65.4660 |
| 0.1436 | 2.8017 | 1300 | 1.5808 | 27.66 | 42.71 | 68.3476 |
| 0.1094 | 3.0172 | 1400 | 1.5684 | 27.1 | 43.29 | 71.3643 |
| 0.0677 | 3.2328 | 1500 | 1.6287 | 26.92 | 42.33 | 68.4376 |
| 0.0567 | 3.4483 | 1600 | 1.6431 | 23.58 | 42.85 | 81.7199 |
| 0.0594 | 3.6638 | 1700 | 1.6084 | 26.54 | 42.45 | 77.4426 |
| 0.0623 | 3.8793 | 1800 | 1.5817 | 29.29 | 45.85 | 67.8523 |
| 0.0323 | 4.0948 | 1900 | 1.6630 | 29.24 | 43.47 | 67.5822 |
| 0.0317 | 4.3103 | 2000 | 1.6494 | 26.43 | 43.4 | 73.3904 |
| 0.0325 | 4.5259 | 2100 | 1.6968 | 27.06 | 42.4 | 68.5277 |
| 0.025 | 4.7414 | 2200 | 1.6436 | 29.16 | 44.24 | 67.5371 |
| 0.0316 | 4.9569 | 2300 | 1.6412 | 29.9 | 46.15 | 66.5016 |
| 0.0159 | 5.1724 | 2400 | 1.6714 | 29.56 | 44.68 | 66.7717 |
| 0.0158 | 5.3879 | 2500 | 1.6458 | 29.56 | 45.46 | 65.8262 |
| 0.015 | 5.6034 | 2600 | 1.6595 | 29.97 | 44.9 | 68.0774 |
| 0.0145 | 5.8190 | 2700 | 1.6545 | 31.15 | 46.35 | 65.6461 |
| 0.0106 | 6.0345 | 2800 | 1.6724 | 30.24 | 45.36 | 66.8618 |
| 0.0076 | 6.25 | 2900 | 1.6834 | 30.25 | 45.13 | 67.9424 |
| 0.0049 | 6.4655 | 3000 | 1.6820 | 30.14 | 44.97 | 68.7528 |
Framework versions
- Transformers 4.41.2
- Pytorch 2.2.0+cu121
- Datasets 2.19.2
- Tokenizers 0.19.1