Instructions to use 1999xia/ViT_Fast with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- timm
How to use 1999xia/ViT_Fast with timm:
import timm model = timm.create_model("hf_hub:1999xia/ViT_Fast", pretrained=True) - Notebooks
- Google Colab
- Kaggle
Model Weights
Checkpoints are available on Hugging Face: https://huggingface.co/YOUR_USERNAME/ViT_Fast
Checkpoint Structure
- cifar100_vit_b16_ft/best_model.pth — CIFAR-100 Baseline (91.69%)
- oxford_pets_vit_b16_ft/best_model.pth — Oxford Pets Baseline (93.32%)
- food101_vit_b16_ft/best_model.pth — Food-101 Baseline (91.37%)
- cifar100_mae_patchsel_b16_keep75_distill/best_model.pth — MAE+Router 75% CIFAR (91.10%)
- cifar100_vit_b16_img168/best_model.pth — Downsample 168x168 CIFAR (91.56%)
- oxford_pets_vit_b16_img168/best_model.pth — Downsample 168x168 Oxford (90.65%)
- router_distill_cifar100/router.pth — Distilled Router CIFAR
- router_distill_food101/router.pth — Distilled Router Food-101
Download
See Hugging Face repo for download links.