whisper-med_15k
This model was trained from scratch on five datasets. It achieves the following results on the evaluation set:
- Cer: 6.2657
- Cer Mecab: 6.5093
- Cer Ortho: 6.2657
- Loss: 0.1532
- Wer: 10.1273
- Wer Ortho: 10.1273
Model description
ADALORA test run
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: 0.000125
- train_batch_size: 4
- eval_batch_size: 2
- seed: 42
- gradient_accumulation_steps: 2
- total_train_batch_size: 8
- optimizer: Adafactor
- lr_scheduler_type: linear
- lr_scheduler_warmup_ratio: 0.01
- training_steps: 10000
Training results
Training Loss | Epoch | Step | Cer | Cer Mecab | Cer Ortho | Validation Loss | Wer | Wer Ortho |
---|---|---|---|---|---|---|---|---|
4.4492 | 0.03 | 300 | 306.2865 | 306.2865 | 306.2865 | 4.4784 | 442.4364 | 442.4364 |
1.0895 | 0.06 | 600 | 39.8357 | 41.7989 | 39.8427 | 0.9371 | 51.6909 | 51.5818 |
0.8748 | 0.09 | 900 | 33.7580 | 34.7327 | 33.7719 | 0.7186 | 47.4 | 47.4182 |
0.73 | 0.12 | 1200 | 27.1651 | 28.9265 | 27.1651 | 0.6159 | 37.2364 | 37.2364 |
0.6601 | 0.15 | 1500 | 20.8995 | 21.8950 | 21.0039 | 0.5812 | 31.2364 | 31.2364 |
0.606 | 0.18 | 1800 | 26.0164 | 27.2626 | 26.0164 | 0.5279 | 35.7273 | 35.7273 |
0.5825 | 0.21 | 2100 | 19.9109 | 20.6419 | 19.9039 | 0.5185 | 29.3455 | 29.3455 |
0.5231 | 0.24 | 2400 | 18.9710 | 19.9248 | 19.0128 | 0.4767 | 28.2364 | 28.2727 |
0.5058 | 0.27 | 2700 | 23.7121 | 25.1880 | 23.7190 | 0.4539 | 32.8727 | 32.8909 |
0.4752 | 0.3 | 3000 | 17.0217 | 18.2818 | 17.0217 | 0.4025 | 23.5091 | 23.5091 |
0.4351 | 0.33 | 3300 | 29.5879 | 30.1657 | 29.5879 | 0.4177 | 42.2364 | 42.2364 |
0.4392 | 0.36 | 3600 | 16.1933 | 16.7502 | 16.2002 | 0.3614 | 24.3636 | 24.3455 |
0.4123 | 0.39 | 3900 | 14.2648 | 15.0585 | 14.2648 | 0.3699 | 22.1273 | 22.1091 |
0.3981 | 0.42 | 4200 | 13.4851 | 14.0769 | 13.5060 | 0.3443 | 20.6727 | 20.7091 |
0.3985 | 0.45 | 4500 | 12.8168 | 13.2414 | 12.8168 | 0.3330 | 19.4000 | 19.4000 |
0.3521 | 0.48 | 4800 | 12.6636 | 13.2832 | 12.6636 | 0.3233 | 19.0545 | 19.0545 |
0.3453 | 0.51 | 5100 | 10.7212 | 11.3200 | 10.7212 | 0.2926 | 17.0909 | 17.0909 |
0.3026 | 0.54 | 5400 | 16.7850 | 18.4280 | 16.7920 | 0.2860 | 17.1818 | 17.1818 |
0.3408 | 0.57 | 5700 | 11.2434 | 11.7516 | 11.2434 | 0.2526 | 17.5636 | 17.5636 |
0.3101 | 0.6 | 6000 | 10.8605 | 11.4105 | 10.8674 | 0.2464 | 17.0 | 17.0182 |
0.2953 | 0.63 | 6300 | 10.5333 | 10.9997 | 10.5333 | 0.2389 | 16.1091 | 16.1091 |
0.2804 | 0.66 | 6600 | 10.9649 | 11.3478 | 10.9719 | 0.2305 | 16.6909 | 16.6909 |
0.2611 | 0.69 | 6900 | 9.9206 | 10.3523 | 9.9206 | 0.2216 | 15.5091 | 15.5091 |
0.2429 | 0.72 | 7200 | 8.7928 | 9.3498 | 8.7928 | 0.2070 | 13.5091 | 13.5091 |
0.2467 | 0.75 | 7500 | 8.1036 | 8.5352 | 8.1036 | 0.2019 | 12.8182 | 12.8182 |
0.253 | 0.78 | 7800 | 8.4099 | 8.8067 | 8.4099 | 0.1979 | 13.1455 | 13.1455 |
0.2407 | 0.81 | 8100 | 7.4283 | 7.6859 | 7.4283 | 0.1825 | 11.6000 | 11.6000 |
0.2206 | 0.84 | 8400 | 8.9042 | 9.1618 | 8.9042 | 0.1779 | 13.4727 | 13.4727 |
0.2123 | 0.87 | 8700 | 7.4909 | 7.7694 | 7.4909 | 0.1769 | 11.7273 | 11.7273 |
0.1976 | 0.9 | 9000 | 9.1131 | 9.4055 | 9.1131 | 0.1665 | 13.9273 | 13.9273 |
0.1757 | 1.0259 | 9300 | 6.6903 | 6.9618 | 6.6903 | 0.1590 | 10.5818 | 10.5818 |
0.1406 | 1.0559 | 9600 | 7.4561 | 7.7068 | 7.4561 | 0.1544 | 11.6545 | 11.6545 |
0.1422 | 1.0859 | 9900 | 6.2657 | 6.5093 | 6.2657 | 0.1532 | 10.1273 | 10.1273 |
Framework versions
- Transformers 4.42.0.dev0
- Pytorch 2.3.1+cu121
- Datasets 2.19.2
- Tokenizers 0.19.1
- Downloads last month
- 335
Inference API (serverless) does not yet support adapter-transformers models for this pipeline type.