Model-J: DINO Model (model_idx_0105)
This model is part of the Model-J dataset, introduced in:
Learning on Model Weights using Tree Experts (CVPR 2025) by Eliahu Horwitz*, Bar Cavia*, Jonathan Kahana*, Yedid Hoshen
ProbeXModel Details
| Attribute | Value |
|---|
| Subset | DINO |
| Split | train |
| Base Model | facebook/dino-vitb16 |
| Dataset | CIFAR100 (50 classes) |
Training Hyperparameters
| Parameter | Value |
|---|
| Learning Rate | 7e-05 |
| LR Scheduler | constant_with_warmup |
| Epochs | 4 |
| Max Train Steps | 1332 |
| Batch Size | 64 |
| Weight Decay | 0.009 |
| Seed | 105 |
| Random Crop | True |
| Random Flip | True |
Performance
| Metric | Value |
|---|
| Train Accuracy | 0.9681 |
| Val Accuracy | 0.8592 |
| Test Accuracy | 0.8626 |
Training Categories
The model was fine-tuned on the following 50 CIFAR100 classes:
porcupine, caterpillar, castle, television, tank, baby, flatfish, poppy, hamster, snail, streetcar, maple_tree, crocodile, cup, orange, dolphin, table, man, snake, shark, can, woman, whale, bridge, leopard, sweet_pepper, dinosaur, pear, ray, beaver, bear, sea, bus, plate, rose, sunflower, boy, wolf, tiger, keyboard, cockroach, elephant, lawn_mower, crab, bowl, cattle, worm, bicycle, couch, motorcycle
