# Retrieval-Augmented Generation for Large Language Models: A Survey

**Authors:** Yunfan Gao, Yun Xiong, Xinyu Gao, Kangxiang Jia, Jinliu Pan, Yuxi Bi, Yi Dai, Jiawei Sun, Haofen Wang  
**arXiv:** [2312.10997](https://arxiv.org/abs/2312.10997)  
**Date:** 2023-12-18  
**Summary page:** https://www.summarizepaper.com/en/paper/2312.10997v1/

## Key points

- Large language models (LLMs) face challenges in practical applications such as hallucinations, slow knowledge updates, and lack of transparency in answers.
- Retrieval-Augmented Generation (RAG) has emerged as a promising approach to address these issues.
- RAG involves retrieving relevant information from external knowledge bases before generating answers with LLMs.
- RAG significantly improves answer accuracy and reduces model hallucination, especially for knowledge-intensive tasks.
- Citing sources allows users to verify the accuracy of answers and build trust in the model's outputs.
- RAG facilitates knowledge updates and enables the incorporation of domain-specific knowledge.
- It effectively combines the parameterized knowledge of LLMs with non-parameterized external knowledge bases.
- Three paradigms of RAG are discussed: Naive RAG, Advanced RAG, and Modular RAG.
- The three main components of RAG are retriever, generator, and augmentation methods.
- Key technologies within each component are explored.
- Two evaluation methods are emphasized for assessing effectiveness: key metrics and abilities for evaluation.
- An automatic evaluation framework is introduced to aid in accurately assessing RAG models.
- Potential future research directions include vertical optimization, horizontal scalability, and technical stack & ecosystem enhancement for RAG.

## Summary

Large language models (LLMs) have shown impressive capabilities but still face challenges in practical applications such as hallucinations, slow knowledge updates and lack of transparency in answers. To address these issues, Retrieval-Augmented Generation (RAG) has emerged as a promising approach. RAG involves retrieving relevant information from external knowledge bases before generating answers with LLMs. It has been proven to significantly improve answer accuracy and reduce model hallucination, particularly for knowledge-intensive tasks. By citing sources, users can verify the accuracy of answers and build trust in the outputs of the model. Additionally, RAG facilitates knowledge updates and enables the incorporation of domain-specific knowledge. It effectively combines the parameterized knowledge of LLMs with non-parameterized external knowledge bases making it one of the most important methods for implementing large language models. This paper provides an overview of the development paradigms of RAG in the era of LLMs. It summarizes three paradigms: Naive RAG, Advanced RAG and Modular RAG; organizing and summarizing their three main components: retriever, generator and augmentation methods; discussing key technologies within each component; exploring how to evaluate their effectiveness with two evaluation methods emphasizing key metrics and abilities for evaluation; introducing an automatic evaluation framework to aid in assessing RAG models accurately; and outlining potential future research directions from three perspectives: vertical optimization, horizontal scalability and technical stack & ecosystem enhancement for RAG. Overall this survey paper provides a comprehensive understanding of Retrieval-Augmented Generation (RAG) for Large Language Models (LLMs). It highlights its significance in addressing challenges faced by LLMs in practical applications while providing insights into its development paradigms, components organization, evaluation methods and future research directions.