ReSearch: Learning to Reason with Search for LLMs via Reinforcement Learning
This addresses a key bottleneck in enhancing LLM reasoning for tasks requiring multiple retrieval steps, offering a novel training approach with potential broad impact.
The paper tackles the challenge of integrating reasoning with external search processes in LLMs for complex multi-hop questions, proposing ReSearch, a reinforcement learning framework that trains models to reason with search without supervised data, achieving strong generalizability across benchmarks.
Large Language Models (LLMs) have shown remarkable capabilities in reasoning, exemplified by the success of OpenAI-o1 and DeepSeek-R1. However, integrating reasoning with external search processes remains challenging, especially for complex multi-hop questions requiring multiple retrieval steps. We propose ReSearch, a novel framework that trains LLMs to Reason with Search via reinforcement learning without using any supervised data on reasoning steps. Our approach treats search operations as integral components of the reasoning chain, where when and how to perform searches is guided by text-based thinking, and search results subsequently influence further reasoning. We train ReSearch on Qwen2.5-7B(-Instruct) and Qwen2.5-32B(-Instruct) models and conduct extensive experiments. Despite being trained on only one dataset, our models demonstrate strong generalizability across various benchmarks. Analysis reveals that ReSearch naturally elicits advanced reasoning capabilities such as reflection and self-correction during the reinforcement learning process.