ReZero: Enhancing LLM search ability by trying one-more-time
This addresses the issue of search persistence for LLMs in knowledge-intensive tasks, representing an incremental improvement over existing methods.
The paper tackles the problem of LLMs prematurely halting after failed searches in RAG by introducing ReZero, an RL framework that rewards retrying queries, achieving 46.88% accuracy compared to a 25% baseline.
Retrieval-Augmented Generation (RAG) improves Large Language Model (LLM) performance on knowledge-intensive tasks but depends heavily on initial search query quality. Current methods, often using Reinforcement Learning (RL), typically focus on query formulation or reasoning over results, without explicitly encouraging persistence after a failed search. We introduce ReZero (Retry-Zero), a novel RL framework that directly rewards the act of retrying a search query following an initial unsuccessful attempt. This incentivizes the LLM to explore alternative queries rather than prematurely halting. ReZero demonstrates significant improvement, achieving 46.88% accuracy compared to a 25% baseline. By rewarding persistence, ReZero enhances LLM robustness in complex information-seeking scenarios where initial queries may prove insufficient.