# Google 等提出 DualSQL：多智能体 RL 训练 Text-to-SQL

- 来源：elvis (@omarsar0)
- 发布时间：2026-09-22 22:19
- AIHOT 分数：45
- AIHOT 链接：https://aihot.news/items/cmucs9itx0m72roed73v8a11q
- 原文链接：https://x.com/omarsar0/status/2102402295682224325

## AI 摘要

Google 等提出 DualSQL，用多智能体 RL 训练 Text-to-SQL，将任务拆成选表列与写 SQL 两个共享同一模型权重的智能体，推理时可经三个工具查询数据库。仅用 3,755 条样本训练，DualSQL-4B 在 BIRD dev 达 68.0% 执行准确率，追平此前 7B 模型；DualSQL-8B 达 71.1%，超过此前 32B 参数的单模型系统。

## 正文

Great paper from Google and colleagues.

Trains Text-to-SQL agents using multi-agent RL.

(bookmark it)

This work proposes DualSQL, which splits Text-to-SQL into two agents, one that links the question to the right tables and columns and one that writes the SQL.

Both agents run on the same model weights, so a single multi-agent RL run trains both roles together. The agents can query the database through three tools while they reason.

Multi-agent RL tends to collapse during training, so the authors add guardrails on rollouts and a new reward, robust execution match, that judges SQL correctness more accurately.

Trained on only 3,755 examples, DualSQL-4B reaches 68.0% execution accuracy on BIRD dev, matching earlier 7B models. DualSQL-8B reaches 71.1%, ahead of previous single-model systems with 32B parameters.

Paper: https://arxiv.org/abs/2609.18135

Chat with Paper: https://academy.dair.ai/papers/dualsql-text-to-sql-with-multi-agent-reinforcement-learning-2609.18135