donut_f21_v4_ep3_edit_distance_loss_retrain
This model is a fine-tuned version of yuchiahung/donut_f21_v4_ep3_edit_distance_loss on the None dataset. It achieves the following results on the evaluation set:
- Loss: 19.6495
- Char Accuracy: 0.1456
- Exact Match Accuracy: 0.0
- Avg Pred Length: 1.456
- Avg Label Length: 10.0
- Length Ratio: 0.1456
- Avg Edit Distance: 8.544
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-06
- train_batch_size: 1
- eval_batch_size: 1
- seed: 132
- 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: cosine
- lr_scheduler_warmup_ratio: 0.1
- num_epochs: 3
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Char Accuracy | Exact Match Accuracy | Avg Pred Length | Avg Label Length | Length Ratio | Avg Edit Distance |
|---|---|---|---|---|---|---|---|---|---|
| 0.7473 | 1.0 | 5000 | 19.5055 | 0.1456 | 0.0 | 1.456 | 10.0 | 0.1456 | 8.544 |
| 1.6766 | 2.0 | 10000 | 19.5939 | 0.1458 | 0.0 | 1.458 | 10.0 | 0.1458 | 8.542 |
| 1.5029 | 3.0 | 15000 | 19.6495 | 0.1456 | 0.0 | 1.456 | 10.0 | 0.1456 | 8.544 |
Framework versions
- Transformers 4.57.6
- Pytorch 2.10.0+cu128
- Datasets 4.5.0
- Tokenizers 0.22.2