w2v-bert-grain-lg_grn_only
This model is a fine-tuned version of facebook/w2v-bert-2.0 on the None dataset. It achieves the following results on the evaluation set:
- Loss: 0.1053
- Wer: 0.0336
- Cer: 0.0113
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: 5e-05
- train_batch_size: 16
- eval_batch_size: 8
- seed: 42
- gradient_accumulation_steps: 2
- total_train_batch_size: 32
- optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
- lr_scheduler_type: linear
- num_epochs: 100
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer | Cer |
|---|---|---|---|---|---|
| 0.992 | 1.0 | 482 | 0.1349 | 0.1317 | 0.0273 |
| 0.178 | 2.0 | 964 | 0.1184 | 0.0983 | 0.0213 |
| 0.12 | 3.0 | 1446 | 0.1047 | 0.0833 | 0.0185 |
| 0.091 | 4.0 | 1928 | 0.0996 | 0.0742 | 0.0175 |
| 0.0721 | 5.0 | 2410 | 0.0946 | 0.0712 | 0.0175 |
| 0.0593 | 6.0 | 2892 | 0.1013 | 0.0686 | 0.0168 |
| 0.0479 | 7.0 | 3374 | 0.0908 | 0.0614 | 0.0148 |
| 0.0421 | 8.0 | 3856 | 0.0956 | 0.0649 | 0.0159 |
| 0.0371 | 9.0 | 4338 | 0.1026 | 0.0694 | 0.0170 |
| 0.0328 | 10.0 | 4820 | 0.1046 | 0.0592 | 0.0145 |
| 0.031 | 11.0 | 5302 | 0.0912 | 0.0529 | 0.0134 |
| 0.0255 | 12.0 | 5784 | 0.0870 | 0.0547 | 0.0140 |
| 0.0224 | 13.0 | 6266 | 0.1073 | 0.0588 | 0.0146 |
| 0.0207 | 14.0 | 6748 | 0.0963 | 0.0493 | 0.0136 |
| 0.0212 | 15.0 | 7230 | 0.1016 | 0.0484 | 0.0149 |
| 0.0183 | 16.0 | 7712 | 0.0994 | 0.0456 | 0.0125 |
| 0.0185 | 17.0 | 8194 | 0.1107 | 0.0495 | 0.0134 |
| 0.0181 | 18.0 | 8676 | 0.1012 | 0.0482 | 0.0136 |
| 0.0153 | 19.0 | 9158 | 0.0947 | 0.0506 | 0.0140 |
| 0.0131 | 20.0 | 9640 | 0.0890 | 0.0475 | 0.0121 |
| 0.0113 | 21.0 | 10122 | 0.0884 | 0.0475 | 0.0126 |
| 0.0114 | 22.0 | 10604 | 0.1205 | 0.0597 | 0.0147 |
| 0.0117 | 23.0 | 11086 | 0.0864 | 0.0404 | 0.0111 |
| 0.0107 | 24.0 | 11568 | 0.0939 | 0.0401 | 0.0122 |
| 0.0094 | 25.0 | 12050 | 0.0997 | 0.0404 | 0.0119 |
| 0.0078 | 26.0 | 12532 | 0.0952 | 0.0399 | 0.0121 |
| 0.0088 | 27.0 | 13014 | 0.1014 | 0.0417 | 0.0116 |
| 0.0077 | 28.0 | 13496 | 0.0954 | 0.0380 | 0.0110 |
| 0.0072 | 29.0 | 13978 | 0.1035 | 0.0427 | 0.0124 |
| 0.0084 | 30.0 | 14460 | 0.0977 | 0.0401 | 0.0119 |
| 0.0082 | 31.0 | 14942 | 0.0929 | 0.0378 | 0.0117 |
| 0.0084 | 32.0 | 15424 | 0.0966 | 0.0397 | 0.0119 |
| 0.0055 | 33.0 | 15906 | 0.0967 | 0.0401 | 0.0115 |
| 0.006 | 34.0 | 16388 | 0.0899 | 0.0354 | 0.0107 |
| 0.006 | 35.0 | 16870 | 0.0954 | 0.0351 | 0.0107 |
| 0.0049 | 36.0 | 17352 | 0.0988 | 0.0484 | 0.0128 |
| 0.0073 | 37.0 | 17834 | 0.0947 | 0.0349 | 0.0107 |
| 0.0049 | 38.0 | 18316 | 0.0893 | 0.0343 | 0.0104 |
| 0.0036 | 39.0 | 18798 | 0.0909 | 0.0317 | 0.0097 |
| 0.0049 | 40.0 | 19280 | 0.0875 | 0.0328 | 0.0099 |
| 0.0061 | 41.0 | 19762 | 0.1071 | 0.0371 | 0.0114 |
| 0.0059 | 42.0 | 20244 | 0.0979 | 0.0380 | 0.0114 |
| 0.0043 | 43.0 | 20726 | 0.0914 | 0.0347 | 0.0102 |
| 0.0034 | 44.0 | 21208 | 0.0946 | 0.0321 | 0.0100 |
| 0.004 | 45.0 | 21690 | 0.0905 | 0.0338 | 0.0097 |
| 0.0038 | 46.0 | 22172 | 0.0967 | 0.0312 | 0.0104 |
| 0.0023 | 47.0 | 22654 | 0.0986 | 0.0336 | 0.0104 |
| 0.0025 | 48.0 | 23136 | 0.0873 | 0.0299 | 0.0095 |
| 0.0027 | 49.0 | 23618 | 0.1071 | 0.0349 | 0.0111 |
| 0.003 | 50.0 | 24100 | 0.0968 | 0.0293 | 0.0098 |
| 0.0033 | 51.0 | 24582 | 0.1058 | 0.0404 | 0.0120 |
| 0.0034 | 52.0 | 25064 | 0.1020 | 0.0367 | 0.0113 |
| 0.0031 | 53.0 | 25546 | 0.0950 | 0.0302 | 0.0093 |
| 0.0016 | 54.0 | 26028 | 0.0988 | 0.0315 | 0.0100 |
| 0.0027 | 55.0 | 26510 | 0.0868 | 0.0297 | 0.0096 |
| 0.003 | 56.0 | 26992 | 0.0955 | 0.0332 | 0.0103 |
| 0.002 | 57.0 | 27474 | 0.0930 | 0.0315 | 0.0102 |
| 0.0022 | 58.0 | 27956 | 0.1053 | 0.0336 | 0.0113 |
Framework versions
- Transformers 4.46.1
- Pytorch 2.1.0+cu118
- Datasets 3.1.0
- Tokenizers 0.20.1