Building LLM apps with more complex logical flows can be challenging with smaller, local LLMs. Graphs offer one way to tackle this, laying out the logic flow as a graph and using local LLMs for reasoning steps within specific nodes. Here, we show how to build complex reasoning flows with local LLMs using LangGraph. We walk through the process of building Corrective RAG from scratch, a recent paper that uses self-reflection to improve RAG performance.

Paper:
[ Ссылка ]

Code:
[ Ссылка ]

свернуть