whisper-kor_noising_4
This model is a fine-tuned version of openai/whisper-small on the whisper-kor_noising_4 dataset. It achieves the following results on the evaluation set:
- Loss: 0.4072
- Wer: 23.5379
- Cer: 10.7578
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: 1e-05
- train_batch_size: 16
- eval_batch_size: 16
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 500
- training_steps: 4000
- mixed_precision_training: Native AMP
Training results
Training Loss | Epoch | Step | Validation Loss | Wer | Cer |
---|---|---|---|---|---|
0.2602 | 0.05 | 100 | 0.4040 | 24.4742 | 10.9617 |
0.2721 | 0.09 | 200 | 0.4070 | 24.3302 | 11.0676 |
0.2355 | 0.14 | 300 | 0.4146 | 24.9640 | 11.3655 |
0.2238 | 0.18 | 400 | 0.4275 | 25.5690 | 11.5615 |
0.2185 | 0.23 | 500 | 0.4297 | 26.2460 | 11.9614 |
0.209 | 0.28 | 600 | 0.4323 | 26.5918 | 12.6671 |
0.2203 | 0.32 | 700 | 0.4270 | 28.4788 | 13.3963 |
0.2003 | 0.37 | 800 | 0.4344 | 26.4333 | 12.0790 |
0.2297 | 0.42 | 900 | 0.4305 | 28.6949 | 15.6584 |
0.1892 | 0.46 | 1000 | 0.4314 | 25.3961 | 12.1731 |
0.2137 | 0.51 | 1100 | 0.4292 | 25.4394 | 11.3812 |
0.2058 | 0.55 | 1200 | 0.4291 | 24.9784 | 11.3459 |
0.2139 | 0.6 | 1300 | 0.4257 | 25.2377 | 12.0477 |
0.2002 | 0.65 | 1400 | 0.4252 | 25.0792 | 11.4910 |
0.2151 | 0.69 | 1500 | 0.4218 | 25.2233 | 11.4478 |
0.2298 | 0.74 | 1600 | 0.4168 | 28.5941 | 14.9763 |
0.2138 | 0.78 | 1700 | 0.4157 | 25.5258 | 11.8752 |
0.2282 | 0.83 | 1800 | 0.4140 | 24.8055 | 12.0124 |
0.2476 | 0.88 | 1900 | 0.4121 | 28.5508 | 14.0628 |
0.2122 | 0.92 | 2000 | 0.4090 | 25.5114 | 13.0200 |
0.2742 | 0.97 | 2100 | 0.4083 | 24.2869 | 11.7497 |
0.1406 | 1.02 | 2200 | 0.4137 | 24.5462 | 11.0401 |
0.0976 | 1.06 | 2300 | 0.4184 | 24.6183 | 11.2126 |
0.1235 | 1.11 | 2400 | 0.4186 | 24.5462 | 11.1852 |
0.1237 | 1.15 | 2500 | 0.4156 | 24.4598 | 11.1342 |
0.108 | 1.2 | 2600 | 0.4164 | 24.2581 | 11.0911 |
0.1228 | 1.25 | 2700 | 0.4140 | 24.5751 | 11.2126 |
0.1292 | 1.29 | 2800 | 0.4144 | 24.2293 | 11.0558 |
0.1306 | 1.34 | 2900 | 0.4161 | 24.2149 | 11.0166 |
0.14 | 1.39 | 3000 | 0.4137 | 24.0709 | 11.0479 |
0.1298 | 1.43 | 3100 | 0.4139 | 24.5174 | 11.0519 |
0.1326 | 1.48 | 3200 | 0.4139 | 24.2581 | 10.9852 |
0.1342 | 1.52 | 3300 | 0.4132 | 25.8859 | 12.6632 |
0.1362 | 1.57 | 3400 | 0.4116 | 23.9268 | 10.8480 |
0.1607 | 1.62 | 3500 | 0.4106 | 23.5955 | 11.3577 |
0.1256 | 1.66 | 3600 | 0.4099 | 23.9556 | 10.8558 |
0.1276 | 1.71 | 3700 | 0.4084 | 23.7396 | 10.7696 |
0.1453 | 1.75 | 3800 | 0.4087 | 23.5811 | 10.7343 |
0.1614 | 1.8 | 3900 | 0.4077 | 23.6243 | 10.7892 |
0.1435 | 1.85 | 4000 | 0.4072 | 23.5379 | 10.7578 |
Framework versions
- Transformers 4.35.2
- Pytorch 2.1.1+cu118
- Datasets 2.15.0
- Tokenizers 0.15.0
- Downloads last month
- 8