MiniRAG: Towards Extremely Simple Retrieval-Augmented Generation

TL;DR

MiniRAG is a simple, efficient retrieval-augmented generation system that enables small language models to perform comparably to larger models with significantly reduced storage needs, suitable for resource-constrained environments.

Contribution

MiniRAG introduces a novel graph-based indexing and retrieval approach that enhances lightweight RAG systems' performance without complex semantic understanding.

Findings

MiniRAG achieves comparable performance to LLM-based methods with only 25% of storage.

It demonstrates effective knowledge retrieval using a simple, graph-structured approach.

The system is validated on a new benchmark dataset for lightweight RAG scenarios.

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 for evaluating lightweight RAG systems under realistic on-device scenarios with complex queries. We fully open-source our implementation and datasets at: https://github.com/HKUDS/MiniRAG.