Code For Our Papaer

🛠️1. Setup

🔍 1.1 LLM Environment

We use LLAMA-3-8B as the high-level semantic planner. Please install it via Ollama.

🐍 1.2 Python Environment

Create and activate a Python environment:

conda create -n LERL python=3.10.8
conda activate LERL
pip install torch==1.13
pip install numpy==2.3.2
pip install pandas==1.23.5

📁 2. Download the Data

You can manually download the compressed dataset from the following link: 🔗 Download from Google Drive

🚀 3. Running the Code

🧠 3.1 Launching LLM Instances

Start three concurrent LLAMA-3-8B servers (each in the background):

nohup python run_llama/run_llama1.py &
nohup python run_llama/run_llama2.py &
nohup python run_llama/run_llama3.py &

👤 3.2 Training the User Model

On KuaiRand:

cd code
bash scripts/run_multibehavior.sh

On KuaiRec:

bash scripts/run_multibehavior_rec.sh

🎯 3.3 Running the Policy

On KuaiRand：

bash scripts/train_lerlc.sh

On KuaiRec：

bash scripts/train_lerlc_rec.sh