$R^3$: "This is My SQL, Are You With Me?" A Consensus-Based Multi-Agent System for Text-to-SQL Tasks
This addresses the problem of improving SQL query generation from natural language for database users, representing an incremental advance in multi-agent methods.
The paper tackles the Text-to-SQL task by proposing R³, a consensus-based multi-agent system that outperforms existing single LLM and multi-agent systems by 1.3% to 8.1% on Spider and Bird benchmarks, with Llama-3-8B showing over 20% improvement over chain-of-thought prompting.
Large Language Models (LLMs) have demonstrated strong performance on various tasks. To unleash their power on the Text-to-SQL task, we propose $R^3$ (Review-Rebuttal-Revision), a consensus-based multi-agent system for Text-to-SQL tasks. $R^3$ outperforms the existing single LLM Text-to-SQL systems as well as the multi-agent Text-to-SQL systems by $1.3\%$ to $8.1\%$ on Spider and Bird. Surprisingly, we find that for Llama-3-8B, $R^3$ outperforms chain-of-thought prompting by over 20\%, even outperforming GPT-3.5 on the development set of Spider.