resnet50_rvl-cdip
This model is a fine-tuned version of microsoft/resnet-50 on the None dataset. It achieves the following results on the evaluation set:
- Loss: 0.8368
- Accuracy: 0.7503
- Brier Loss: 0.3458
- Nll: 3.2289
- F1 Micro: 0.7503
- F1 Macro: 0.5224
- Ece: 0.0166
- Aurc: 0.0739
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: 64
- eval_batch_size: 64
- seed: 42
- gradient_accumulation_steps: 16
- total_train_batch_size: 1024
- 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 |
|---|---|---|---|---|---|---|---|---|---|---|
| No log | 1.0 | 312 | 3.9686 | 0.1994 | 0.9069 | 6.9377 | 0.1994 | 0.0036 | 0.0434 | 0.6332 |
| 5.3078 | 2.0 | 625 | 1.5040 | 0.5644 | 0.5696 | 4.9767 | 0.5644 | 0.0480 | 0.0330 | 0.2052 |
| 5.3078 | 3.0 | 937 | 1.1500 | 0.6602 | 0.4588 | 4.0574 | 0.6602 | 0.1527 | 0.0193 | 0.1309 |
| 1.3983 | 4.0 | 1250 | 1.0174 | 0.6961 | 0.4132 | 3.6856 | 0.6961 | 0.2658 | 0.0184 | 0.1053 |
| 1.0466 | 5.0 | 1562 | 0.9439 | 0.7167 | 0.3862 | 3.5182 | 0.7167 | 0.3477 | 0.0150 | 0.0921 |
| 1.0466 | 6.0 | 1875 | 0.9042 | 0.7302 | 0.3717 | 3.3972 | 0.7302 | 0.3345 | 0.0160 | 0.0854 |
| 0.9333 | 7.0 | 2187 | 0.8713 | 0.7395 | 0.3593 | 3.3567 | 0.7395 | 0.4236 | 0.0162 | 0.0801 |
| 0.8757 | 8.0 | 2500 | 0.8550 | 0.7444 | 0.3531 | 3.2398 | 0.7444 | 0.4113 | 0.0150 | 0.0772 |
| 0.8757 | 9.0 | 2812 | 0.8389 | 0.7487 | 0.3468 | 3.1800 | 0.7487 | 0.4613 | 0.0149 | 0.0745 |
| 0.8509 | 9.98 | 3120 | 0.8368 | 0.7503 | 0.3458 | 3.2289 | 0.7503 | 0.5224 | 0.0166 | 0.0739 |
Framework versions
- Transformers 4.36.0.dev0
- Pytorch 2.2.0.dev20231112+cu118
- Datasets 2.14.5
- Tokenizers 0.14.1