This repository contains the code for our paper "Towards Reliable, Generalizable, and Specific In-Context Knowledge Editing via Multi-Objective Reinforcement Learning". The code is organized into several directories, each containing specific components of our framework.

• utils/ - Utility functions for data processing, in-context learning construction, and frozen LLM API interactions
• eval.py - Evaluation code for assessing the performance of the knowledge editing
• train.py - Training code for the MO-IKE framework
• Datasets/ - Contains the counterfact.json file used for training and evaluation. You may evaluate the performance of MO-IKE on other datasets by replacing the counterfact.json file with your desired dataset
• requirements.txt - Required Python packages for running the code

To install the required packages, run the following command:

git clone https://github.com/xuzhongwm/MO-IKE.git
cd MO-IKE
pip install -r requirements.txt

python train.py

python eval.py