Searching for Best Practices in Retrieval-Augmented Generation
Topics: AI (Deep Learning), LLMO, Retrieval Augmented Generation (RAG)
This paper investigates optimal practices for implementing Retrieval-Augmented Generation (RAG) systems. The researchers from Fudan University conducted extensive experiments to identify the most effective combinations of RAG components, including query classification, document chunking, vector databases, retrieval methods, reranking, document repacking, and summarization. They aimed to balance performance and efficiency while providing practical recommendations for RAG implementation.