Whisper-small-lg-finetuned
This model is a fine-tuned version of openai/whisper-small on the Grain dataset. It achieves the following results on the evaluation set:
- Loss: 0.0014
- Wer: 0.0040
- Cer: 0.0013
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: 8
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_ratio: 0.1
- num_epochs: 80
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer | Cer |
|---|---|---|---|---|---|
| 1.9276 | 1.0 | 1296 | 0.4841 | 1.1593 | 0.4773 |
| 0.2693 | 2.0 | 2592 | 0.0967 | 1.2498 | 0.5777 |
| 0.0668 | 3.0 | 3888 | 0.0300 | 1.1842 | 0.5634 |
| 0.0234 | 4.0 | 5184 | 0.0162 | 0.8390 | 0.3714 |
| 0.0117 | 5.0 | 6480 | 0.0108 | 0.8158 | 0.3744 |
| 0.0072 | 6.0 | 7776 | 0.0094 | 0.4896 | 0.2288 |
| 0.0051 | 7.0 | 9072 | 0.0086 | 0.2434 | 0.1106 |
| 0.005 | 8.0 | 10368 | 0.0090 | 0.2317 | 0.1230 |
| 0.0041 | 9.0 | 11664 | 0.0064 | 0.1364 | 0.0608 |
| 0.0029 | 10.0 | 12960 | 0.0064 | 0.0704 | 0.0215 |
| 0.0025 | 11.0 | 14256 | 0.0053 | 0.0756 | 0.0495 |
| 0.0018 | 12.0 | 15552 | 0.0059 | 0.0699 | 0.0313 |
| 0.0022 | 13.0 | 16848 | 0.0036 | 0.0238 | 0.0095 |
| 0.0018 | 14.0 | 18144 | 0.0053 | 0.0426 | 0.0195 |
| 0.0013 | 15.0 | 19440 | 0.0051 | 0.0203 | 0.0059 |
| 0.0017 | 16.0 | 20736 | 0.0028 | 0.0255 | 0.0124 |
| 0.0009 | 17.0 | 22032 | 0.0031 | 0.0254 | 0.0116 |
| 0.001 | 18.0 | 23328 | 0.0038 | 0.0105 | 0.0031 |
| 0.0014 | 19.0 | 24624 | 0.0022 | 0.0109 | 0.0034 |
| 0.001 | 20.0 | 25920 | 0.0015 | 0.0108 | 0.0037 |
| 0.0009 | 21.0 | 27216 | 0.0036 | 0.0170 | 0.0047 |
| 0.0005 | 22.0 | 28512 | 0.0014 | 0.0091 | 0.0032 |
| 0.0007 | 23.0 | 29808 | 0.0014 | 0.0101 | 0.0031 |
| 0.001 | 24.0 | 31104 | 0.0020 | 0.0108 | 0.0035 |
| 0.0004 | 25.0 | 32400 | 0.0015 | 0.0093 | 0.0030 |
| 0.0006 | 26.0 | 33696 | 0.0022 | 0.0174 | 0.0076 |
| 0.0007 | 27.0 | 34992 | 0.0020 | 0.0122 | 0.0079 |
| 0.0006 | 28.0 | 36288 | 0.0016 | 0.0081 | 0.0029 |
| 0.0004 | 29.0 | 37584 | 0.0020 | 0.0110 | 0.0031 |
| 0.0007 | 30.0 | 38880 | 0.0015 | 0.0106 | 0.0037 |
| 0.0005 | 31.0 | 40176 | 0.0025 | 0.0116 | 0.0032 |
| 0.0005 | 32.0 | 41472 | 0.0016 | 0.0097 | 0.0027 |
| 0.0003 | 33.0 | 42768 | 0.0010 | 0.0087 | 0.0034 |
| 0.0004 | 34.0 | 44064 | 0.0015 | 0.0116 | 0.0062 |
| 0.0002 | 35.0 | 45360 | 0.0010 | 0.0047 | 0.0020 |
| 0.0001 | 36.0 | 46656 | 0.0009 | 0.0052 | 0.0020 |
| 0.0006 | 37.0 | 47952 | 0.0027 | 0.0097 | 0.0031 |
| 0.0003 | 38.0 | 49248 | 0.0017 | 0.0054 | 0.0016 |
| 0.0002 | 39.0 | 50544 | 0.0013 | 0.0066 | 0.0023 |
| 0.0003 | 40.0 | 51840 | 0.0023 | 0.0072 | 0.0023 |
| 0.0002 | 41.0 | 53136 | 0.0012 | 0.0044 | 0.0018 |
| 0.0003 | 42.0 | 54432 | 0.0035 | 0.0075 | 0.0031 |
| 0.0003 | 43.0 | 55728 | 0.0035 | 0.0073 | 0.0024 |
| 0.0001 | 44.0 | 57024 | 0.0014 | 0.0047 | 0.0016 |
| 0.0 | 45.0 | 58320 | 0.0014 | 0.0040 | 0.0013 |
Framework versions
- Transformers 4.45.2
- Pytorch 2.1.0+cu118
- Datasets 3.0.1
- Tokenizers 0.20.1