vit-base_rvl_cdip-N1K_aAURC_8
This model is a fine-tuned version of jordyvl/vit-base_rvl-cdip on the None dataset. It achieves the following results on the evaluation set:
- Loss: 0.5639
- Accuracy: 0.8882
- Brier Loss: 0.2015
- Nll: 0.8676
- F1 Micro: 0.8882
- F1 Macro: 0.8882
- Ece: 0.0953
- Aurc: 0.0259
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
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_ratio: 0.1
- num_epochs: 10
Training results
| Training Loss | Epoch | Step | Validation Loss | Accuracy | Brier Loss | Nll | F1 Micro | F1 Macro | Ece | Aurc |
|---|---|---|---|---|---|---|---|---|---|---|
| 0.2121 | 1.0 | 2000 | 0.4148 | 0.862 | 0.2185 | 1.2055 | 0.8620 | 0.8637 | 0.0821 | 0.0286 |
| 0.1797 | 2.0 | 4000 | 0.4259 | 0.8738 | 0.2073 | 1.1106 | 0.8738 | 0.8729 | 0.0842 | 0.0269 |
| 0.0999 | 3.0 | 6000 | 0.4691 | 0.874 | 0.2161 | 1.0691 | 0.874 | 0.8739 | 0.0941 | 0.0287 |
| 0.0532 | 4.0 | 8000 | 0.5251 | 0.872 | 0.2218 | 1.1401 | 0.872 | 0.8726 | 0.0995 | 0.0287 |
| 0.0197 | 5.0 | 10000 | 0.5723 | 0.871 | 0.2303 | 1.0391 | 0.871 | 0.8710 | 0.1085 | 0.0297 |
| 0.0118 | 6.0 | 12000 | 0.5253 | 0.8845 | 0.2070 | 0.9140 | 0.8845 | 0.8847 | 0.0953 | 0.0246 |
| 0.0095 | 7.0 | 14000 | 0.5969 | 0.8718 | 0.2284 | 0.9640 | 0.8718 | 0.8717 | 0.1094 | 0.0249 |
| 0.0063 | 8.0 | 16000 | 0.5702 | 0.8848 | 0.2087 | 0.8868 | 0.8848 | 0.8846 | 0.1003 | 0.0258 |
| 0.0014 | 9.0 | 18000 | 0.5810 | 0.8825 | 0.2115 | 0.8615 | 0.8825 | 0.8828 | 0.0998 | 0.0271 |
| 0.0025 | 10.0 | 20000 | 0.5639 | 0.8882 | 0.2015 | 0.8676 | 0.8882 | 0.8882 | 0.0953 | 0.0259 |
Framework versions
- Transformers 4.33.3
- Pytorch 2.2.0.dev20231002
- Datasets 2.7.1
- Tokenizers 0.13.3