arxiv:2503.24364

Query and Conquer: Execution-Guided SQL Generation

Published on Mar 31

· Submitted by

Borchmann on Apr 1

Upvote

Authors:

Łukasz Borchmann ,

Marek Wydmuch

Abstract

We propose a novel approach for generating complex outputs that significantly improves accuracy in text-to-SQL tasks. Our method leverages execution results to select the most semantically consistent query from multiple candidates, enabling smaller, cost-effective models to surpass computationally intensive reasoning methods such as o1, o3-mini, and DeepSeek R1 while reducing inference cost by as much as 30 times. It integrates effortlessly with existing models, offering a practical and scalable pathway to state-of-the-art SQL generation.

View arXiv page View PDF Add to collection

Community

Borchmann

Paper author Paper submitter 2 days ago

While the presented experiments focused on SQL generation, the underlying principle of leveraging execution-based similarity naturally extends to other programming languages and code-generation tasks. Supplementary results on text-to-code benchmarks provided in suggest the broad applicability of execution-guided self-consistency and highlight its potential in program synthesis across diverse domains.