Gradient Inversion Attacks on Parameter-Efficient Fine-Tuning
December 11, 2025 ยท View on GitHub
This repository contains the code associated with the attack propsed in the paper.
How to Run
- Run the file Adapter_attack.ipynb. That will call functions from the following files.
- Transformer_Model_neuron.py contains the architecture with inserted adapter module
- Design_Model_Adapter.py contains the code for malicious design of the parameters
- Processing_v2.py contains the code for creating patch statistics' intervals according to public dataset
- Recover_Adapter.py recovers the patches from adapter gradients
- Clustering.py groups the recovered patches to create the entire images (motivated by Fowl et. al., 2023)
- The folder "Optimization Attack" contains code for optimization-based attack (Geiping et. al., 2020)