vit-tiny_rvl_cdip
This model is a fine-tuned version of WinKawaks/vit-tiny-patch16-224 on the None dataset. It achieves the following results on the evaluation set:
- Loss: 0.1016
- Accuracy: 0.9025
- Brier Loss: 0.1427
- Nll: 1.7378
- F1 Micro: 0.9025
- F1 Macro: 0.9029
- Ece: 0.0142
- Aurc: 0.0141
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: 0.0001
- train_batch_size: 128
- eval_batch_size: 128
- 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 |
|---|---|---|---|---|---|---|---|---|---|---|
| 0.3377 | 1.0 | 2500 | 0.2693 | 0.8397 | 0.2295 | 2.0316 | 0.8397 | 0.8399 | 0.0140 | 0.0337 |
| 0.1962 | 2.0 | 5000 | 0.1745 | 0.8717 | 0.1835 | 1.9452 | 0.8717 | 0.8739 | 0.0122 | 0.0222 |
| 0.1359 | 3.0 | 7500 | 0.1380 | 0.8869 | 0.1643 | 1.8585 | 0.8869 | 0.8871 | 0.0089 | 0.0181 |
| 0.099 | 4.0 | 10000 | 0.1297 | 0.8920 | 0.1567 | 1.8113 | 0.8920 | 0.8921 | 0.0128 | 0.0168 |
| 0.068 | 5.0 | 12500 | 0.1253 | 0.8966 | 0.1520 | 1.7963 | 0.8966 | 0.8969 | 0.0120 | 0.0160 |
| 0.0475 | 6.0 | 15000 | 0.1153 | 0.8972 | 0.1487 | 1.7849 | 0.8972 | 0.8979 | 0.0136 | 0.0151 |
| 0.0341 | 7.0 | 17500 | 0.1110 | 0.8995 | 0.1460 | 1.7557 | 0.8995 | 0.8997 | 0.0151 | 0.0146 |
| 0.0238 | 8.0 | 20000 | 0.1059 | 0.9013 | 0.1438 | 1.7503 | 0.9013 | 0.9015 | 0.0120 | 0.0143 |
| 0.017 | 9.0 | 22500 | 0.1034 | 0.9022 | 0.1440 | 1.7344 | 0.9022 | 0.9026 | 0.0142 | 0.0143 |
| 0.0128 | 10.0 | 25000 | 0.1016 | 0.9025 | 0.1427 | 1.7378 | 0.9025 | 0.9029 | 0.0142 | 0.0141 |
Framework versions
- Transformers 4.26.1
- Pytorch 1.13.1.post200
- Datasets 2.9.0
- Tokenizers 0.13.2