Mitigating the Backdoor Effect for Multi-Task Model Merging via Safety-Aware Subspace （ICLR2025)

Introduction

We propose a novel defense-aware merging algorithm to tackle backdoor risks in model merging during CLIP Safety-Fine-tuning, based on orthogonal subspaces guided by safety, co-optimizing performance and safety across downstream tasks

The code base is orgainized based on the https://github.com/tanganke/fusion_bench.

Installation

Install from PyPI:

pip install fusion-bench

or install the latest version in development from github repository

git clone https://github.com/tanganke/fusion_bench.git
cd fusion_bench

pip install -e . # install the package in editable mode

Install with Language Model Evaluation Harness

pip install "fusion-bench[lm-eval-harness]"

or install from local directory

pip install -e ".[lm-eval-harness]"

This will install the latest version of fusion-bench and the dependencies required for LM-Eval Harness. Documentation for using LM-Eval Harness within FusionBench framework can be found at this online documentation or in the docs/taskpool/lm_eval_harness.md markdown file.

Project Structure

The project is structured as follows:

• fusion_bench/: the main package of the benchmark.
  • method: contains the implementation of the fusion methods.

    naming convention: fusion_bench/method/{method_name}/{variant}.py contains the implementation of the specific method or its variants. For example, fusion_bench/method/regmean/clip_regmean.py contains the implementation of the RegMean algorithm for CLIP vision models.

  • modelpool: contains the implementation of the model pool, responsible for managing the models and dataset to be loaded.
  • taskpool: contains the implementation of the task pool, responsible for evaluating the performance of models returned by the algorithm.
• config/: configuration files for the benchmark. We use Hydra to manage the configurations.
  • method: configuration files for the fusion methods.

    naming convention: config/method/{method_name}/{variant}.yaml contains the configuration for the specific method or its variants.

  • modelpool: configuration files for the model pool.
  • taskpool: configuration files for the task pool.
  • model: configuration files for the models.
  • dataset: configuration files for the datasets.
• docs/: documentation for the benchmark. We use mkdocs to generate the documentation. Start the documentation server locally with mkdocs serve. The required packages can be installed with pip install -r mkdocs-requirements.txt.
• examples/: example scripts for running some of the experiments.

  naming convention: examples/{method_name}/ contains the files such as bash scripts and jupyter notebooks for the specific method.

• tests/: unit tests for the benchmark.## Project Structure

Citation

@article{yang2024mitigating,
  title={Mitigating the Backdoor Effect for Multi-Task Model Merging via Safety-Aware Subspace},
  author={Yang, Jinluan and Tang, Anke and Zhu, Didi and Chen, Zhengyu and Shen, Li and Wu, Fei},
  journal={arXiv preprint arXiv:2410.13910},
  year={2024}
}

Acknowledgements

Some of the code in this repository is based on the following repositories:

• https://github.com/tanganke/fusion_bench
• https://github.com/tanganke/pytorch_classification
• https://github.com/mxzheng/TrojViT
• https://github.com/SCLBD/BackdoorBench