Skip to content

Latest commit

 

History

History
35 lines (25 loc) · 773 Bytes

README.md

File metadata and controls

35 lines (25 loc) · 773 Bytes

ROW

This repository is the official implementation of

Learnability and Algorithm for Continual Learning. ICML, 2023 (paper)

****** IMPORTANT ****** Please download the pre-trained transformer network from

https://drive.google.com/file/d/1uEpqe6xo--8jdpOgR_YX3JqTHqj34oX6/view?usp=sharing

and save the file as ./deit_pretrained/best_checkpoint.pth

Requirements

Please install the following packages

pytorch==1.7.1
torchvision==0.8.2
cudatoolkit=11.0
ftfy
timm==0.4.12

Training

bash configs/<experiment.sh>

Substitute <experiment.sh> with the name of the experiment you want to run

Acknowledgement

The code format follows DER++, HAT

https://github.com/aimagelab/mammoth

https://github.com/joansj/hat