Awesome-LLM-RAG: A Panoramic Guide to Retrieval-Augmented Generation (RAG) Technology

Awesome-LLM-RAG is an open-source resource collection maintained by researchers from Johns Hopkins University, designed to provide systematic and comprehensive reference materials for researchers and practitioners in the RAG field. Adopting the classic "Awesome List" format, this project covers papers, tools, tutorials, and application cases, helping users systematically grasp the cutting-edge advancements in Retrieval-Augmented Generation technology.

Retrieval-Augmented Generation (RAG) is a significant technological breakthrough in the field of large language models (LLMs). By combining external knowledge retrieval with text generation, it addresses the limitations of pure parametric models in terms of knowledge timeliness, accuracy, and traceability. Simply put, RAG allows AI to "look up information" when answering questions, rather than relying solely on knowledge memorized during training.

The core content of the project is divided into four categories:

1. Academic Papers and Research Findings: Covers subfields such as retrieval-augmented language models (e.g., REALM, RAG), adaptive retrieval strategies (e.g., Self-RAG), long-text and memory mechanisms, RAG evaluation and optimization (e.g., RGB benchmark), etc.;
2. Open-Source Tools and Frameworks: Includes DSPy (declarative language model programming framework), ChunkTuner (text chunking optimization tool), Bernstein (multi-agent orchestrator), Agent Shadow Brain (AI coding agent), etc.;
3. Tutorials and Learning Resources: Recommends books like Build a Large Language Model (From Scratch), Retrieval Augmented Generation, The Seminal Papers, Enterprise RAG, Essential GraphRAG, etc.;
4. Academic Conferences and Workshops: Tracks events such as CIKM 2023 Generative AI Workshop, SIGIR 2023 Generative Information Retrieval Workshop, ACL 2023 Retrieval-Based Language Model Workshop, etc.

The development of RAG technology is divided into three phases:

1. Infrastructure Phase (2020-2022): Focuses on the integration of retrievers and generators, comparison between dense/sparse retrieval, trade-offs between end-to-end training and modular design. Representative works include Facebook's RAG model and Google's REALM;
2. Capability Enhancement Phase (2022-2023): Emphasizes adaptive retrieval, multi-hop reasoning, and instruction fine-tuning. Representative works include Self-RAG and Chain-of-Note;
3. System Optimization Phase (2023-Present): Shifts towards speculative decoding (e.g., REST technology), long-context processing, multimodal expansion, enterprise-level applications, etc.

Different user groups can benefit from it:

• Researchers: Quickly understand cutting-edge progress, find relevant papers and benchmark datasets, and avoid duplicate work;
• Engineers: Discover open-source tools suitable for production environments, learn industry best practices, and save research time;
• Learners: Obtain a complete learning path from beginner to advanced, including books, tutorials, code examples, and community resources.

The project is open-sourced under the MIT License, encouraging the community to submit papers or tools via Pull Requests to maintain the timeliness and comprehensiveness of the resources. In the future, RAG technology will evolve towards integration with Agent systems, multimodal expansion, real-time knowledge updates, personalized retrieval strategies, etc. Awesome-LLM-RAG will continue to serve as a resource hub to support community development.