MA-RAG: Multi-Agent Retrieval-Augmented Generation via Collaborative Chain-of-Thought Reasoning

This repository contains the source code for the paper:
MA-RAG: Multi-Agent Retrieval-Augmented Generation via Collaborative Chain-of-Thought Reasoning.

---

🔧 Installation

To set up the environment, install the required dependencies using the requirements.txt file:

pip install -r requirements.txt

---

⚙️ Environment Configuration

Before running any experiments, set up your OpenAI API key and model name in a .env file:

OPENAI_API_KEY=your_openai_api_key_here
MODEL_NAME=gpt-4o

You can either export them manually or use a .env loader.

---

🧪 Steps to Run the Experiment

Step 1: Embed the Corpus

Run the following command to embed the corpus:

python corpus/embed_corpus.py

Step 2: Run the Main Experiment

Execute the main experiment with the following command:

python main.py --model gpt4omini --dataset hotpotqa --exp plan_rag_extract --gpus 0 1

---

📚 Citation

If you use this code, please cite the paper:

@article{marag2025,
      title={MA-RAG: Multi-Agent Retrieval-Augmented Generation via Collaborative Chain-of-Thought Reasoning}, 
      author={Thang Nguyen, Peter Chin, Yu-Wing Tai},
      year={2025},
      journal={arXiv preprint arXiv:2505.20096},
}