CBA

This repo will be updated and fixed within the next month. ———— 12/23

1. install envs

pip install -r requirements.txt

or

conda env create -f environment.yml

2. Run attack pipeline in each target model's directory.

Ethical Statement

Our research aims to raise community awareness of potential attacks on this open-source model and calls for further review mechanisms. Therefore, in this repository, we avoid providing the following:

• poisoned model weights version for specific target model
• directly available poison samples
• scripts that generate poison weights with a single click

This is to prevent potential malicious use.

Researchers in the field are encouraged to follow the tutorial step by step for data generation, causality analysis, and poison model training.