vit-base_rvl_cdip-N1K_ce_16
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: 0.6681
- Accuracy: 0.89
- Brier Loss: 0.2001
- Nll: 0.9073
- F1 Micro: 0.89
- F1 Macro: 0.8905
- Ece: 0.0923
- Aurc: 0.0219
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: 16
- eval_batch_size: 16
- 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 |
|---|---|---|---|---|---|---|---|---|---|---|
| 0.209 | 1.0 | 1000 | 0.4595 | 0.8775 | 0.1885 | 1.1949 | 0.8775 | 0.8784 | 0.0616 | 0.0237 |
| 0.1707 | 2.0 | 2000 | 0.4835 | 0.881 | 0.1887 | 1.1366 | 0.881 | 0.8803 | 0.0720 | 0.0237 |
| 0.0893 | 3.0 | 3000 | 0.5434 | 0.8808 | 0.1991 | 1.0313 | 0.8808 | 0.8805 | 0.0830 | 0.0237 |
| 0.0442 | 4.0 | 4000 | 0.5746 | 0.8845 | 0.1964 | 0.9971 | 0.8845 | 0.8850 | 0.0858 | 0.0234 |
| 0.0176 | 5.0 | 5000 | 0.6168 | 0.8802 | 0.2062 | 1.0035 | 0.8802 | 0.8799 | 0.0935 | 0.0241 |
| 0.0098 | 6.0 | 6000 | 0.6533 | 0.882 | 0.2074 | 0.9667 | 0.882 | 0.8829 | 0.0953 | 0.0237 |
| 0.0066 | 7.0 | 7000 | 0.6557 | 0.8838 | 0.2041 | 0.9568 | 0.8838 | 0.8833 | 0.0942 | 0.0235 |
| 0.0049 | 8.0 | 8000 | 0.6557 | 0.8878 | 0.1995 | 0.9076 | 0.8878 | 0.8883 | 0.0934 | 0.0220 |
| 0.0027 | 9.0 | 9000 | 0.6693 | 0.8882 | 0.2024 | 0.9127 | 0.8882 | 0.8888 | 0.0939 | 0.0222 |
| 0.0031 | 10.0 | 10000 | 0.6681 | 0.89 | 0.2001 | 0.9073 | 0.89 | 0.8905 | 0.0923 | 0.0219 |
Framework versions
- Transformers 4.33.3
- Pytorch 2.2.0.dev20231002
- Datasets 2.7.1
- Tokenizers 0.13.3