README.md

AS-ViT

Implement of Adaptive Sparse ViT (IJCAI23)

requirement

torch==1.9.0
torchvision==0.10.1
timm==0.4.12

training

sh finetune.sh

inference

python3 infer.py --model-path path_to_model.pth --batch-size 256 --arch deit_small --data-path path_to_imagenet