Whisper Small GA-EN Speech Translation
This model is a fine-tuned version of openai/whisper-small on the IWSLT-2023, FLEURS, BiteSize, SpokenWords, Tatoeba, and Wikimedia dataset. It achieves the following results on the evaluation set:
- Loss: 1.4107
- Bleu: 27.85
- Chrf: 46.91
- Wer: 73.4354
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 |
|---|---|---|---|---|---|---|
| 2.3549 | 0.1312 | 100 | 1.8335 | 7.17 | 24.71 | 135.7497 |
| 1.8906 | 0.2625 | 200 | 1.5173 | 15.56 | 34.19 | 91.5353 |
| 1.653 | 0.3937 | 300 | 1.3530 | 17.17 | 36.35 | 103.7371 |
| 1.4901 | 0.5249 | 400 | 1.3334 | 24.65 | 43.44 | 78.2530 |
| 1.3551 | 0.6562 | 500 | 1.2763 | 27.04 | 43.88 | 67.4471 |
| 1.2187 | 0.7874 | 600 | 1.2618 | 27.08 | 43.98 | 69.2031 |
| 1.0359 | 0.9186 | 700 | 1.2644 | 20.82 | 40.76 | 96.8483 |
| 0.5364 | 1.0499 | 800 | 1.3258 | 24.9 | 42.9 | 65.8262 |
| 0.4892 | 1.1811 | 900 | 1.3296 | 23.82 | 42.86 | 72.3098 |
| 0.4504 | 1.3123 | 1000 | 1.3001 | 25.78 | 43.72 | 75.5065 |
| 0.4161 | 1.4436 | 1100 | 1.2948 | 27.16 | 44.31 | 67.3120 |
| 0.3953 | 1.5748 | 1200 | 1.3261 | 29.14 | 44.65 | 65.5110 |
| 0.3509 | 1.7060 | 1300 | 1.3398 | 22.75 | 44.32 | 80.1441 |
| 0.2955 | 1.8373 | 1400 | 1.3077 | 26.29 | 42.89 | 74.8762 |
| 0.2801 | 1.9685 | 1500 | 1.3206 | 25.51 | 43.39 | 76.5871 |
| 0.1084 | 2.0997 | 1600 | 1.3609 | 28.01 | 45.59 | 68.1225 |
| 0.1003 | 2.2310 | 1700 | 1.3722 | 26.4 | 42.69 | 72.8501 |
| 0.1083 | 2.3622 | 1800 | 1.3776 | 3.81 | 19.2 | 396.1279 |
| 0.0939 | 2.4934 | 1900 | 1.3729 | 28.43 | 45.61 | 69.2031 |
| 0.0909 | 2.6247 | 2000 | 1.3834 | 27.12 | 43.39 | 67.4921 |
| 0.0772 | 2.7559 | 2100 | 1.4094 | 28.44 | 44.15 | 65.5110 |
| 0.0753 | 2.8871 | 2200 | 1.3825 | 30.5 | 46.21 | 64.9257 |
| 0.0438 | 3.0184 | 2300 | 1.4198 | 30.44 | 46.18 | 62.5844 |
| 0.0257 | 3.1496 | 2400 | 1.4033 | 31.03 | 46.67 | 63.6650 |
| 0.0252 | 3.2808 | 2500 | 1.4045 | 31.2 | 46.44 | 62.4043 |
| 0.0241 | 3.4121 | 2600 | 1.3971 | 32.42 | 48.21 | 61.1436 |
| 0.0208 | 3.5433 | 2700 | 1.4129 | 30.36 | 46.28 | 65.7362 |
| 0.0186 | 3.6745 | 2800 | 1.4076 | 31.14 | 47.73 | 64.4304 |
| 0.018 | 3.8058 | 2900 | 1.4151 | 27.67 | 45.87 | 73.5254 |
| 0.0193 | 3.9370 | 3000 | 1.4107 | 27.85 | 46.91 | 73.4354 |
Framework versions
- Transformers 4.41.2
- Pytorch 2.2.0+cu121
- Datasets 2.19.2
- Tokenizers 0.19.1