s3: You Don't Need That Much Data to Train a Search Agent via RL

TL;DR

s3 is a lightweight, model-agnostic RL framework that trains a searcher to improve generation accuracy with minimal data, outperforming larger-data baselines across diverse QA tasks.

Contribution

It introduces s3, a novel decoupled searcher training method using a Gain Beyond RAG reward, requiring significantly less data than existing approaches.

Findings

s3 outperforms baselines trained on 70x more data

Achieves better downstream QA performance across multiple benchmarks

Requires only 2.4k training samples

Abstract

Retrieval-augmented generation (RAG) systems empower large language models (LLMs) to access external knowledge during inference. Recent advances have enabled LLMs to act as search agents via reinforcement learning (RL), improving information acquisition through multi-turn interactions with retrieval engines. However, existing approaches either optimize retrieval using search-only metrics (e.g., NDCG) that ignore downstream utility or fine-tune the entire LLM to jointly reason and retrieve-entangling retrieval with generation and limiting the real search utility and compatibility with frozen or proprietary models. In this work, we propose s3, a lightweight, model-agnostic framework that decouples the searcher from the generator and trains the searcher using a Gain Beyond RAG reward: the improvement in generation accuracy over naive RAG. s3 requires only 2.4k training samples to outperform baselines trained on over 70x more data, consistently delivering stronger downstream performance across six general QA and five medical QA benchmarks.