Multi-agent system improves text to SQL query accuracy
DualSQL: Text-to-SQL with Multi-Agent Reinforcement Learning
Computation and LanguageArtificial IntelligenceDatabases
Summary
Turning English questions into database queries is hard because parts of the task depend on each other. The authors created DualSQL, which uses one model shared by two agents that work together and learn through reinforcement. They designed new tools to help the agents access databases better and introduced a new way to check if the generated queries are correct. Their approach achieves strong results with fewer training examples and smaller models than before.
What this means in practice
- •For software engineers: Integrate improved text-to-SQL components that jointly optimize schema linking and SQL generation for better query accuracy.
- •For business intelligence teams: Use enhanced natural language interfaces to generate more accurate SQL queries from user questions, speeding up data analysis.
Authors
Shijie Chen, Yu Gan, Yeounoh Chung, Jiani Zhang, Quannan Li, Sravan Babu Bodapati, Cody J. Greer, Yu Su, Fatma Ozcan
Abstract
State-of-the-art Text-to-SQL systems are typically multi-agent pipelines centered around two fundamental tasks: schema linking and SQL generation. However, existing work trains separate models for each task, failing to leverage the synergy between these interrelated tasks. In this work, we propose DualSQL, a new Text-to-SQL system consisting of two agents powered by a single model backbone. The agents share the same model weights and agentic scaffold, enabling joint optimization through a robust multi-agent reinforcement learning (RL) framework. We design three database access tools to facilitate effective multi-step reasoning grounded to interactions with the databases. To improve training and avoid model collapse, we introduce a set of rollout guardrail mechanisms that stabilizes multi-agent RL training, supporting DualSQL to keep improving during training. We also introduce a new SQL correctness metric, robust execution match (REX), to more accurately judge SQL correctness and assign reward signals. Being trained on only 3755 examples, DualSQL-4B achieves an impressive 68.0% execution accuracy on the BIRD development set, matching previous 7B models. DualSQL-8B further improves to 71.1%, outperforming previous state-of-the-art single-model solutions with 32B parameters. These results demonstrate the strength of joint multi-agent reinforcement learning for building high performance Text-to-SQL pipelines.