trocr-warm-3style-hh104-hh1-hh2
This model is a fine-tuned version of cyttic/exp2-frozen-benyehuda-cont on an unknown dataset. It achieves the following results on the evaluation set:
- Loss: 0.5464
- Cer: 0.0324
- Wer: 0.0866
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: 2e-05
- train_batch_size: 8
- eval_batch_size: 8
- seed: 42
- gradient_accumulation_steps: 2
- total_train_batch_size: 16
- optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED 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: 0.1
- num_epochs: 2
Training results
| Training Loss | Epoch | Step | Validation Loss | Cer | Wer |
|---|---|---|---|---|---|
| 5.0234 | 0.1290 | 2000 | 2.2049 | 0.2156 | 0.4321 |
| 3.4243 | 0.2581 | 4000 | 1.5238 | 0.1321 | 0.2891 |
| 2.8799 | 0.3871 | 6000 | 1.2677 | 0.0986 | 0.2294 |
| 2.2127 | 0.5161 | 8000 | 1.0814 | 0.0793 | 0.1918 |
| 2.1143 | 0.6452 | 10000 | 0.9444 | 0.0644 | 0.1596 |
| 1.7773 | 0.7742 | 12000 | 0.8380 | 0.0568 | 0.1433 |
| 1.6481 | 0.9032 | 14000 | 0.7657 | 0.0503 | 0.1300 |
| 1.3575 | 1.0323 | 16000 | 0.7091 | 0.0468 | 0.1201 |
| 1.1907 | 1.1613 | 18000 | 0.6691 | 0.0418 | 0.1103 |
| 1.1625 | 1.2903 | 20000 | 0.6404 | 0.0393 | 0.1049 |
| 1.2172 | 1.4194 | 22000 | 0.6136 | 0.0373 | 0.0990 |
| 1.0258 | 1.5484 | 24000 | 0.5880 | 0.0357 | 0.0944 |
| 1.0359 | 1.6774 | 26000 | 0.5711 | 0.0342 | 0.0915 |
| 1.0539 | 1.8065 | 28000 | 0.5588 | 0.0337 | 0.0896 |
| 1.0278 | 1.9355 | 30000 | 0.5478 | 0.0323 | 0.0866 |
| 0.9686 | 2.0 | 31000 | 0.5464 | 0.0324 | 0.0866 |
Framework versions
- Transformers 5.14.1
- Pytorch 2.11.0+cu128
- Datasets 5.0.0
- Tokenizers 0.22.2