中文说明 | 日本語

🎉 News

• [2025.02.27]🎯📢Now you can use pip install minirag-hku to run our code!
• [2025.02.14]🎯📢Now MiniRAG supports 10+ heterogeneous graph databases, including Neo4j, PostgreSQL, TiDB, etc. Happy valentine's day!🌹🌹🌹
• [2025.02.05]🎯📢Our team has released VideoRAG understanding extremely long-context videos.
• [2025.02.01]🎯📢Now MiniRAG supports API&Docker deployment. see This for more details.

TLDR

MiniRAG is an extremely simple retrieval-augmented generation framework that enables small models to achieve good RAG performance through heterogeneous graph indexing and lightweight topology-enhanced retrieval.

Abstract

The growing demand for efficient and lightweight Retrieval-Augmented Generation (RAG) systems has highlighted significant challenges when deploying Small Language Models (SLMs) in existing RAG frameworks. Current approaches face severe performance degradation due to SLMs' limited semantic understanding and text processing capabilities, creating barriers for widespread adoption in resource-constrained scenarios. To address these fundamental limitations, we present MiniRAG, a novel RAG system designed for extreme simplicity and efficiency. MiniRAG introduces two key technical innovations: (1) a semantic-aware heterogeneous graph indexing mechanism that combines text chunks and named entities in a unified structure, reducing reliance on complex semantic understanding, and (2) a lightweight topology-enhanced retrieval approach that leverages graph structures for efficient knowledge discovery without requiring advanced language capabilities. Our extensive experiments demonstrate that MiniRAG achieves comparable performance to LLM-based methods even when using SLMs while requiring only 25% of the storage space. Additionally, we contribute a comprehensive benchmark dataset LiHua-World for evaluating lightweight RAG systems under realistic on-device scenarios with complex queries.

MiniRAG Framework

MiniRAG employs a streamlined workflow built on the key components: heterogeneous graph indexing and lightweight graph-based knowledge retrieval. This architecture addresses the unique challenges faced by on-device RAG systems, optimizing for both efficiency and effectiveness.

Install

• Install from source (Recommend)

cd MiniRAG
pip install -e .

• Install from PyPI (Our code is based on LightRAG, so you can install it directly)

pip install lightrag-hku

Quick Start

• All the code can be found in the ./reproduce.
• Download the dataset you need.
• Put the dataset in the ./dataset directory.
• Note: We have already put the LiHua-World dataset in ./dataset/LiHua-World/data/ as LiHuaWorld.zip. If you want to use other dataset, you can put it in the ./dataset/xxx.

Then use the following bash command to index the dataset:

python ./reproduce/Step_0_index.py
python ./reproduce/Step_1_QA.py

Or, use the code in ./main.py to initialize MiniRAG.

Overall Performance Table

In the table, / means the method struggles to generate effective responses.

Reproduce

All the code can be found in the ./reproduce directory.

Code Structure

├── dataset
│   └── LiHua-World
│       ├── README.md
│       ├── README_CN.md
│       ├── data
│       │   ├── LiHuaWorld.zip
│       └── qa
│           ├── query_set.csv
│           └── query_set.json
├── minirag
│   ├── kg
│   │   ├── __init__.py
│   │   ├── neo4j_impl.py
│   │   └── oracle_impl.py
│   ├── __init__.py
│   ├── base.py
│   ├── exceptions.py
│   ├── llm.py
│   ├── minirag.py
│   ├── operate.py
│   ├── prompt.py
│   ├── storage.py
│   └── utils.py
├── reproduce
│   ├── Step_0_index.py
│   └── Step_1_QA.py
├── LICENSE
├── main.py
├── README.md
├── README_CN.md
├── requirements.txt
├── setup.py

Dataset: LiHua-World

LiHua-World is a dataset specifically designed for on-device RAG scenarios, containing one year of chat records from a virtual user named LiHua. The dataset includes three types of questions: single-hop, multi-hop, and summary, with each question paired with manually annotated answers and supporting documents. For more details, please refer to the README of LiHua-World dataset.

Star History

Contribution

Thank you to all our contributors!

Acknowledgements

You may refer to related work that serves as foundations for our framework and code repository, nano-graphrag and LightRAG. Thanks for their wonderful works.

🌟Citation

@article{fan2025minirag,
  title={MiniRAG: Towards Extremely Simple Retrieval-Augmented Generation},
  author={Fan, Tianyu and Wang, Jingyuan and Ren, Xubin and Huang, Chao},
  journal={arXiv preprint arXiv:2501.06713},
  year={2025}
}

Thank you for your interest in our work!

🚀 Quick Start

cd MiniRAG
pip install -e .