README.md

---
base_model: facebook/dino-vitb16
library_name: transformers
pipeline_tag: image-classification
tags:
  - probex
  - model-j
  - weight-space-learning
---

# Model-J: DINO Model (model_idx_0854)

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

<p align="center">
    🌐 <a href="https://horwitz.ai/probex" target="_blank">Project</a> | 📃 <a href="https://arxiv.org/abs/2410.13569" target="_blank">Paper</a> | 💻 <a href="https://github.com/eliahuhorwitz/ProbeX" target="_blank">GitHub</a> | 🤗 <a href="https://huggingface.co/ProbeX" target="_blank">Dataset</a>
</p>

![ProbeX](https://raw.githubusercontent.com/eliahuhorwitz/ProbeX/main/imgs/poster.png)

## Model Details

| Attribute | Value |
|---|---|
| **Subset** | DINO |
| **Split** | test |
| **Base Model** | `facebook/dino-vitb16` |
| **Dataset** | CIFAR100 (50 classes) |

## Training Hyperparameters

| Parameter | Value |
|---|---|
| Learning Rate | 5e-05 |
| LR Scheduler | constant |
| Epochs | 7 |
| Max Train Steps | 2331 |
| Batch Size | 64 |
| Weight Decay | 0.03 |
| Seed | 854 |
| Random Crop | False |
| Random Flip | False |

## Performance

| Metric | Value |
|---|---|
| Train Accuracy | 0.9896 |
| Val Accuracy | 0.8827 |
| Test Accuracy | 0.8754 |

## Training Categories

The model was fine-tuned on the following 50 CIFAR100 classes:

`squirrel`, `pear`, `television`, `tulip`, `couch`, `leopard`, `mountain`, `sea`, `plain`, `plate`, `possum`, `hamster`, `cockroach`, `bottle`, `seal`, `keyboard`, `wardrobe`, `dolphin`, `aquarium_fish`, `turtle`, `shrew`, `lamp`, `camel`, `bus`, `sunflower`, `snail`, `sweet_pepper`, `boy`, `crocodile`, `fox`, `castle`, `otter`, `rocket`, `skunk`, `elephant`, `streetcar`, `baby`, `cloud`, `table`, `man`, `mouse`, `worm`, `poppy`, `girl`, `pine_tree`, `kangaroo`, `caterpillar`, `telephone`, `orange`, `lion`