Model-J: MAE Model (model_idx_0322)

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

🌐 Project | 📃 Paper | 💻 GitHub | 🤗 Dataset

Model Details

Attribute	Value
Subset	MAE
Split	test
Base Model	`facebook/vit-mae-base`
Dataset	CIFAR100 (50 classes)

Training Hyperparameters

Parameter	Value
Learning Rate	7e-05
LR Scheduler	constant_with_warmup
Epochs	3
Max Train Steps	999
Batch Size	64
Weight Decay	0.007
Seed	322
Random Crop	False
Random Flip	True

Performance

Metric	Value
Train Accuracy	0.9444
Val Accuracy	0.8579
Test Accuracy	0.8576

Training Categories

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

bus, hamster, rocket, beaver, palm_tree, snake, skyscraper, sweet_pepper, girl, cloud, dolphin, clock, mushroom, keyboard, kangaroo, snail, flatfish, crocodile, lizard, butterfly, tiger, pear, bear, sunflower, porcupine, mountain, rabbit, plate, fox, tractor, woman, cockroach, streetcar, otter, orchid, leopard, pickup_truck, bridge, camel, couch, crab, mouse, dinosaur, bee, shark, cattle, forest, raccoon, elephant, shrew