Saved in:
Bibliographic Details
Main Authors: Dao, Alan, Le, Thinh
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2504.11001
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866908320437108736
author Dao, Alan
Le, Thinh
author_facet Dao, Alan
Le, Thinh
contents 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.
format Preprint
id arxiv_https___arxiv_org_abs_2504_11001
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle ReZero: Enhancing LLM search ability by trying one-more-time
Dao, Alan
Le, Thinh
Computation and Language
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.
title ReZero: Enhancing LLM search ability by trying one-more-time
topic Computation and Language
url https://arxiv.org/abs/2504.11001