vit-base_rvl_cdip-N1K_ce_32
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: 0.5671
- Accuracy: 0.8915
- Brier Loss: 0.1895
- Nll: 0.9175
- F1 Micro: 0.8915
- F1 Macro: 0.8919
- Ece: 0.0850
- Aurc: 0.0200
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: 32
- eval_batch_size: 32
- 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 |
|---|---|---|---|---|---|---|---|---|---|---|
| 0.1771 | 1.0 | 500 | 0.4121 | 0.8885 | 0.1719 | 1.2085 | 0.8885 | 0.8888 | 0.0509 | 0.0203 |
| 0.134 | 2.0 | 1000 | 0.4415 | 0.8882 | 0.1782 | 1.1210 | 0.8882 | 0.8886 | 0.0626 | 0.0212 |
| 0.0682 | 3.0 | 1500 | 0.4722 | 0.8855 | 0.1847 | 1.0778 | 0.8855 | 0.8858 | 0.0740 | 0.0213 |
| 0.0325 | 4.0 | 2000 | 0.4851 | 0.8905 | 0.1796 | 1.0195 | 0.8905 | 0.8911 | 0.0712 | 0.0213 |
| 0.0145 | 5.0 | 2500 | 0.5409 | 0.8842 | 0.1946 | 1.0096 | 0.8842 | 0.8850 | 0.0860 | 0.0217 |
| 0.0082 | 6.0 | 3000 | 0.5378 | 0.8872 | 0.1886 | 0.9573 | 0.8872 | 0.8879 | 0.0858 | 0.0206 |
| 0.0059 | 7.0 | 3500 | 0.5446 | 0.8895 | 0.1870 | 0.9288 | 0.8895 | 0.8897 | 0.0844 | 0.0206 |
| 0.0046 | 8.0 | 4000 | 0.5580 | 0.8885 | 0.1874 | 0.9153 | 0.8885 | 0.8889 | 0.0859 | 0.0203 |
| 0.0043 | 9.0 | 4500 | 0.5675 | 0.8905 | 0.1903 | 0.9313 | 0.8905 | 0.8910 | 0.0864 | 0.0201 |
| 0.004 | 10.0 | 5000 | 0.5671 | 0.8915 | 0.1895 | 0.9175 | 0.8915 | 0.8919 | 0.0850 | 0.0200 |
Framework versions
- Transformers 4.33.3
- Pytorch 2.2.0.dev20231002
- Datasets 2.7.1
- Tokenizers 0.13.3