mms-1b-all-bigc_mini-train-drodat-gdro-1.15e-5-dat-0.207-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: 34.8514
- Cer: 0.1197
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 |
|---|---|---|---|---|
| 182.7749 | 0.9579 | 500 | 159.7604 | 0.8344 |
| 41.9502 | 1.9157 | 1000 | 42.2198 | 0.1406 |
| 40.0927 | 2.8736 | 1500 | 40.8279 | 0.1369 |
| 39.5556 | 3.8314 | 2000 | 39.7877 | 0.1352 |
| 37.0724 | 4.7893 | 2500 | 38.3159 | 0.1349 |
| 35.907 | 5.7471 | 3000 | 38.5418 | 0.1305 |
| 33.388 | 6.7050 | 3500 | 42.1111 | 0.1759 |
| 32.4594 | 7.6628 | 4000 | 37.1892 | 0.1393 |
| 32.0805 | 8.6207 | 4500 | 37.1478 | 0.1394 |
| 34.924 | 9.5785 | 5000 | 36.1114 | 0.1268 |
| 33.7696 | 10.5364 | 5500 | 36.8895 | 0.1402 |
| 30.8225 | 11.4943 | 6000 | 35.8162 | 0.1226 |
| 33.9841 | 12.4521 | 6500 | 35.8530 | 0.1306 |
| 30.8205 | 13.4100 | 7000 | 35.4646 | 0.1244 |
| 29.6223 | 14.3678 | 7500 | 35.7283 | 0.1330 |
| 28.4497 | 15.3257 | 8000 | 35.2760 | 0.1275 |
| 27.3041 | 16.2835 | 8500 | 34.8516 | 0.1197 |
| 30.4223 | 17.2414 | 9000 | 34.9730 | 0.1242 |
| 30.2937 | 18.1992 | 9500 | 35.5532 | 0.1308 |
| 28.4706 | 19.1571 | 10000 | 34.9966 | 0.1208 |
Framework versions
- Transformers 4.53.0
- Pytorch 2.10.0+cu128
- Datasets 4.5.0
- Tokenizers 0.21.4