RAG (Retrieval-Augmented Generation)
Technique combining information retrieval from a knowledge base with LLM-powered response generation.
Technique combining information retrieval from a knowledge base with LLM-powered response generation.
Définition détaillée
RAG is an architecture that enhances LLM capabilities by providing contextual information from an external knowledge base. The process has two steps: (1) a query retrieves relevant documents from a vector database via embeddings, (2) these documents are injected into the LLM prompt for a contextualized response. RAG solves the problem of outdated or missing LLM knowledge and enables source citation.
Cas d'usage
HR document chatbot: employees query their internal policies in natural language. RAG retrieves relevant passages and the LLM generates a precise, sourced answer.
Termes associés
En savoir plus
Questions fréquentes
What's the difference between RAG and fine-tuning?
RAG adds dynamic context without modifying the model, ideal for evolving knowledge. Fine-tuning adapts the model deeply to a domain, better for specific style or tone.
Is RAG compatible with data privacy?
Yes, RAG can be deployed entirely on-premises with an open-source LLM and a private vector database, ensuring no data leaves your infrastructure.