mms-1b-all-bigc_mini-drodat-gdro-eta-3.15e-4-dat-0.1454-no-sd-0.00-62
This model is a fine-tuned version of facebook/mms-1b-all on the BIGC_MINI - BEM dataset. It achieves the following results on the evaluation set:
- Loss: 35.5196
- Cer: 0.1249
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: 0.0003
- train_batch_size: 4
- eval_batch_size: 4
- seed: 42
- gradient_accumulation_steps: 2
- total_train_batch_size: 8
- optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 500
- num_epochs: 30.0
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Cer |
|---|---|---|---|---|
| 227.3277 | 0.9579 | 500 | 209.9947 | 0.9999 |
| 42.6672 | 1.9157 | 1000 | 42.5506 | 0.1416 |
| 40.3368 | 2.8736 | 1500 | 41.1025 | 0.1381 |
| 39.9799 | 3.8314 | 2000 | 40.1885 | 0.1356 |
| 38.024 | 4.7893 | 2500 | 39.3866 | 0.1383 |
| 36.4264 | 5.7471 | 3000 | 40.3026 | 0.1323 |
| 33.9292 | 6.7050 | 3500 | 38.1178 | 0.1404 |
| 33.2028 | 7.6628 | 4000 | 37.5152 | 0.1343 |
| 32.2362 | 8.6207 | 4500 | 37.4773 | 0.1479 |
| 35.1013 | 9.5785 | 5000 | 36.5141 | 0.1320 |
| 34.3974 | 10.5364 | 5500 | 37.0681 | 0.1379 |
| 31.6446 | 11.4943 | 6000 | 35.5192 | 0.1249 |
| 34.9507 | 12.4521 | 6500 | 36.2533 | 0.1333 |
| 31.2218 | 13.4100 | 7000 | 36.0195 | 0.1332 |
| 30.1093 | 14.3678 | 7500 | 35.9190 | 0.1362 |
Framework versions
- Transformers 4.53.0
- Pytorch 2.10.0+cu128
- Datasets 4.5.0
- Tokenizers 0.21.4