Plan*RAG: Efficient Test-Time Planning for Retrieval Augmented Generation
This addresses inefficiencies in RAG systems for multi-hop reasoning tasks, offering a practical, modular solution that can be integrated with existing methods.
The paper tackles the problem of plan fragmentation and execution failures in retrieval-augmented generation (RAG) by introducing Plan*RAG, a framework that isolates reasoning plans as directed acyclic graphs to enable structured multi-hop reasoning, achieving consistent improvements over methods like RQ-RAG and Self-RAG on standard benchmarks.
We introduce Plan*RAG, a novel framework that enables structured multi-hop reasoning in retrieval-augmented generation (RAG) through test-time reasoning plan generation. While existing approaches such as ReAct maintain reasoning chains within the language model's context window, we observe that this often leads to plan fragmentation and execution failures. Our key insight is that by isolating the reasoning plan as a directed acyclic graph (DAG) outside the LM's working memory, we can enable (1) systematic exploration of reasoning paths, (2) atomic subqueries enabling precise retrievals and grounding, and (3) efficiency through parallel execution and bounded context window utilization. Moreover, Plan*RAG's modular design allows it to be integrated with existing RAG methods, thus providing a practical solution to improve current RAG systems. On standard multi-hop reasoning benchmarks, Plan*RAG consistently achieves improvements over recently proposed methods such as RQ-RAG and Self-RAG, while maintaining comparable computational costs.