This repository provides a comprehensive survey of Retrieval-Augmented Generation (RAG) for Large Language Models, targeting researchers and practitioners. It aims to offer a more flexible, intuitive, and dynamically updated knowledge base than traditional static survey papers, enabling customized analysis and summarization of RAG-related resources.

How It Works

The project leverages Notion databases to organize RAG knowledge, including academic papers, cutting-edge readings, benchmarks, datasets, and scholars. Notion's "Relation" property allows for bidirectional linking between databases, facilitating cross-analysis (e.g., linking papers to datasets or institutions). Users can duplicate the Notion workspace to create a personalized, interactive RAG knowledge base.

Quick Start & Requirements

• Access: Visit the Notion homepage linked in the README.
• Usage: Duplicate the Notion workspace to your local Notion account to make modifications and create custom views.
• Requirements: A Notion account is required.

Highlighted Details

• Comprehensive coverage of RAG, including academic papers, tools, datasets, and key researchers.
• Detailed explanations of RAG paradigms (Naive, Advanced, Modular) and augmentation strategies.
• Comparison of RAG with fine-tuning and discussion of evaluation methods and benchmarks.
• Exploration of RAG's future prospects, including multimodal expansion and ecosystem growth.

Maintenance & Community

The project is initiated by researchers from Tongji University and Fudan University. Updates are ongoing, with recent additions including OpenRAG Base and modular RAG concepts. Contact information for contributors is provided for questions and collaboration.

Licensing & Compatibility

The project itself is not explicitly licensed in the README. The survey paper is available on arXiv (arXiv:2312.10997). Notion's terms of service apply to the workspace.

Limitations & Caveats

The Notion workspace is a static view unless duplicated locally. Some content, like "Quick reading" guides, may be in Chinese by default, though English options are mentioned. The project is still under active development with "more content coming soon."