vit-base_rvl_cdip-N1K_AURC_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: 0.3096
- Accuracy: 0.883
- Brier Loss: 0.2014
- Nll: 0.9150
- F1 Micro: 0.883
- F1 Macro: 0.8832
- Ece: 0.0891
- Aurc: 0.0256
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 |
|---|---|---|---|---|---|---|---|---|---|---|
| 0.0736 | 1.0 | 2000 | 0.2385 | 0.8592 | 0.2172 | 1.2230 | 0.8592 | 0.8608 | 0.0734 | 0.0313 |
| 0.0594 | 2.0 | 4000 | 0.2561 | 0.8712 | 0.2047 | 1.2297 | 0.8713 | 0.8716 | 0.0678 | 0.0283 |
| 0.0421 | 3.0 | 6000 | 0.2432 | 0.867 | 0.2104 | 1.1813 | 0.867 | 0.8679 | 0.0749 | 0.0303 |
| 0.0256 | 4.0 | 8000 | 0.2882 | 0.8632 | 0.2199 | 1.1103 | 0.8632 | 0.8635 | 0.0847 | 0.0310 |
| 0.0147 | 5.0 | 10000 | 0.4246 | 0.8515 | 0.2466 | 1.1118 | 0.8515 | 0.8489 | 0.1059 | 0.0360 |
| 0.0105 | 6.0 | 12000 | 0.2747 | 0.8668 | 0.2220 | 1.0335 | 0.8668 | 0.8691 | 0.0986 | 0.0278 |
| 0.004 | 7.0 | 14000 | 0.2954 | 0.878 | 0.2034 | 0.9467 | 0.878 | 0.8783 | 0.0865 | 0.0264 |
| 0.0034 | 8.0 | 16000 | 0.3339 | 0.8708 | 0.2185 | 0.9551 | 0.8708 | 0.8713 | 0.0969 | 0.0286 |
| 0.0017 | 9.0 | 18000 | 0.3125 | 0.8748 | 0.2099 | 0.9454 | 0.8748 | 0.8761 | 0.0953 | 0.0265 |
| 0.0009 | 10.0 | 20000 | 0.3096 | 0.883 | 0.2014 | 0.9150 | 0.883 | 0.8832 | 0.0891 | 0.0256 |
Framework versions
- Transformers 4.33.3
- Pytorch 2.2.0.dev20231002
- Datasets 2.7.1
- Tokenizers 0.13.3