vit-base_rvl_cdip-N1K_ce_4
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: 0.9480
- Accuracy: 0.8792
- Brier Loss: 0.2240
- Nll: 1.0075
- F1 Micro: 0.8793
- F1 Macro: 0.8794
- Ece: 0.1101
- Aurc: 0.0274
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: 4
- eval_batch_size: 4
- 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 |
|---|---|---|---|---|---|---|---|---|---|---|
| 0.4172 | 1.0 | 4000 | 0.6321 | 0.8475 | 0.2427 | 1.1862 | 0.8475 | 0.8484 | 0.0957 | 0.0352 |
| 0.3421 | 2.0 | 8000 | 0.6729 | 0.8645 | 0.2301 | 1.1766 | 0.8645 | 0.8642 | 0.1020 | 0.0295 |
| 0.2091 | 3.0 | 12000 | 0.7988 | 0.854 | 0.2563 | 1.1608 | 0.854 | 0.8555 | 0.1183 | 0.0352 |
| 0.1319 | 4.0 | 16000 | 0.8683 | 0.861 | 0.2503 | 1.1575 | 0.861 | 0.8617 | 0.1188 | 0.0354 |
| 0.0673 | 5.0 | 20000 | 0.9057 | 0.8642 | 0.2479 | 1.1524 | 0.8643 | 0.8635 | 0.1195 | 0.0314 |
| 0.0333 | 6.0 | 24000 | 0.9553 | 0.8605 | 0.2524 | 1.1006 | 0.8605 | 0.8600 | 0.1226 | 0.0366 |
| 0.0223 | 7.0 | 28000 | 0.9393 | 0.8708 | 0.2350 | 1.1027 | 0.8708 | 0.8713 | 0.1159 | 0.0274 |
| 0.0194 | 8.0 | 32000 | 1.0108 | 0.8705 | 0.2407 | 1.0850 | 0.8705 | 0.8704 | 0.1169 | 0.0309 |
| 0.0015 | 9.0 | 36000 | 0.9412 | 0.876 | 0.2291 | 1.0136 | 0.8760 | 0.8763 | 0.1123 | 0.0270 |
| 0.004 | 10.0 | 40000 | 0.9480 | 0.8792 | 0.2240 | 1.0075 | 0.8793 | 0.8794 | 0.1101 | 0.0274 |
Framework versions
- Transformers 4.33.3
- Pytorch 2.2.0.dev20231002
- Datasets 2.7.1
- Tokenizers 0.13.3