AGRO-SQL: Agentic Group-Relative Optimization with High-Fidelity Data Synthesis

TL;DR

This paper introduces AGRO-SQL, a comprehensive framework combining high-fidelity data synthesis and an agentic reinforcement learning approach to significantly improve Text-to-SQL systems in complex reasoning tasks.

Contribution

It presents a novel dual-centric framework that integrates high-quality data generation with an agentic RL method, advancing reasoning capabilities in Text-to-SQL models.

Findings

Achieves state-of-the-art results on BIRD and Spider benchmarks.

Demonstrates the effectiveness of high-fidelity data synthesis.

Shows improved reasoning in complex SQL queries.

Abstract

The advancement of Text-to-SQL systems is currently hindered by the scarcity of high-quality training data and the limited reasoning capabilities of models in complex scenarios. In this paper, we propose a holistic framework that addresses these issues through a dual-centric approach. From a Data-Centric perspective, we construct an iterative data factory that synthesizes RL-ready data characterized by high correctness and precise semantic-logic alignment, ensured by strict verification. From a Model-Centric perspective, we introduce a novel Agentic Reinforcement Learning framework. This framework employs a Diversity-Aware Cold Start stage to initialize a robust policy, followed by Group Relative Policy Optimization (GRPO) to refine the agent's reasoning via environmental feedback. Extensive experiments on BIRD and Spider benchmarks demonstrate that our synergistic approach achieves state-of-the-art performance among single-model methods.