Whisper Medium GA-EN Speech Translation
This model is a fine-tuned version of openai/whisper-medium on an unknown dataset. It achieves the following results on the evaluation set:
- Loss: 1.5839
- Bleu: 26.1
- Chrf: 41.83
- Wer: 74.6511
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: 8
- eval_batch_size: 8
- 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: 4000
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Bleu | Chrf | Validation Loss | Wer |
|---|---|---|---|---|---|---|
| 3.5292 | 0.03 | 100 | 0.07 | 10.93 | 3.0255 | 296.9383 |
| 3.2384 | 0.05 | 200 | 0.63 | 9.99 | 2.9567 | 99.5047 |
| 2.9326 | 0.08 | 300 | 1.29 | 14.61 | 2.7918 | 159.8829 |
| 2.8396 | 0.11 | 400 | 3.7 | 17.65 | 2.7288 | 120.2611 |
| 3.3706 | 0.13 | 500 | 6.06 | 18.15 | 2.5552 | 110.8510 |
| 2.2285 | 0.16 | 600 | 2.9 | 17.34 | 2.6213 | 180.4593 |
| 2.452 | 0.19 | 700 | 7.91 | 22.0 | 2.3512 | 101.3958 |
| 2.3773 | 0.22 | 800 | 9.88 | 24.57 | 2.2666 | 105.7632 |
| 2.4751 | 0.24 | 900 | 7.46 | 21.25 | 2.1982 | 138.2710 |
| 1.9366 | 0.27 | 1000 | 7.76 | 25.49 | 2.1982 | 133.8136 |
| 1.9036 | 0.3 | 1100 | 12.02 | 28.01 | 2.0519 | 95.8127 |
| 2.072 | 0.32 | 1200 | 12.5 | 29.48 | 2.0200 | 105.4480 |
| 1.5439 | 0.35 | 1300 | 15.36 | 31.45 | 1.9203 | 93.3363 |
| 1.8711 | 0.38 | 1400 | 9.7 | 30.29 | 1.8251 | 144.3044 |
| 1.6849 | 0.4 | 1500 | 16.5 | 33.04 | 1.7918 | 92.7510 |
| 1.3787 | 0.43 | 1600 | 17.32 | 34.29 | 1.7533 | 87.4381 |
| 1.4595 | 0.46 | 1700 | 19.31 | 36.88 | 1.6776 | 86.5376 |
| 1.3406 | 0.49 | 1800 | 19.96 | 36.71 | 1.6488 | 81.6299 |
| 1.5607 | 0.51 | 1900 | 21.46 | 38.7 | 1.6280 | 78.3881 |
| 1.3687 | 0.54 | 2000 | 19.94 | 37.49 | 1.6109 | 84.5565 |
| 1.3761 | 0.57 | 2100 | 19.55 | 35.47 | 1.7206 | 83.8361 |
| 1.4642 | 0.59 | 2200 | 20.57 | 37.87 | 1.7014 | 82.9356 |
| 1.2263 | 0.62 | 2300 | 20.35 | 36.62 | 1.6773 | 82.1252 |
| 1.2511 | 0.65 | 2400 | 15.65 | 36.0 | 1.7131 | 111.1661 |
| 1.0471 | 0.67 | 2500 | 20.56 | 39.75 | 1.6460 | 83.5209 |
| 1.2797 | 0.7 | 2600 | 19.69 | 39.6 | 1.6122 | 90.8600 |
| 1.1708 | 0.73 | 2700 | 22.3 | 41.05 | 1.5805 | 79.4237 |
| 1.0686 | 0.76 | 2800 | 22.72 | 40.5 | 1.5689 | 78.2981 |
| 0.933 | 0.78 | 2900 | 20.89 | 41.06 | 1.5499 | 90.4548 |
| 0.9654 | 0.81 | 3000 | 21.94 | 40.62 | 1.5438 | 81.4048 |
| 1.1812 | 0.84 | 3100 | 1.5872 | 20.99 | 39.88 | 88.0234 |
| 1.0603 | 0.86 | 3200 | 1.6039 | 18.9 | 38.69 | 96.3530 |
| 0.964 | 0.89 | 3300 | 1.5899 | 23.92 | 41.25 | 76.0018 |
| 0.8625 | 0.92 | 3400 | 1.6323 | 19.96 | 39.15 | 87.7082 |
| 0.9115 | 0.94 | 3500 | 1.5222 | 23.64 | 41.67 | 83.1157 |
| 0.9932 | 0.97 | 3600 | 1.5477 | 24.09 | 41.78 | 81.8550 |
| 1.1792 | 1.0 | 3700 | 1.5282 | 24.85 | 41.57 | 78.7033 |
| 0.4436 | 1.02 | 3800 | 1.5823 | 24.1 | 40.77 | 80.3692 |
| 0.3277 | 1.05 | 3900 | 1.5852 | 24.6 | 41.34 | 78.5682 |
| 0.372 | 1.08 | 4000 | 1.5839 | 26.1 | 41.83 | 74.6511 |
Framework versions
- Transformers 4.39.3
- Pytorch 2.2.1+cu121
- Datasets 2.18.0
- Tokenizers 0.15.2
- Downloads last month
- 0