PET-SQL: A Prompt-Enhanced Two-Round Refinement of Text-to-SQL with Cross-consistency

TL;DR

This paper introduces PET-SQL, a two-round refinement framework for Text-to-SQL tasks that leverages cross-consistency among LLMs and enhanced prompts to improve accuracy on complex database queries.

Contribution

It proposes a novel prompt representation and a two-stage refinement process with cross-consistency, significantly advancing Text-to-SQL performance.

Findings

Achieves state-of-the-art 87.6% execution accuracy on Spider benchmark.

Introduces reference-enhanced prompts with schema and sampled cell values.

Utilizes cross-consistency across LLMs for post-refinement.

Abstract

Recent advancements in Text-to-SQL (Text2SQL) emphasize stimulating the large language models (LLM) on in-context learning, achieving significant results. Nevertheless, they face challenges when dealing with verbose database information and complex user intentions. This paper presents a two-stage framework to enhance the performance of current LLM-based natural language to SQL systems. We first introduce a novel prompt representation, called reference-enhanced representation, which includes schema information and randomly sampled cell values from tables to instruct LLMs in generating SQL queries. Then, in the first stage, question-SQL pairs are retrieved as few-shot demonstrations, prompting the LLM to generate a preliminary SQL (PreSQL). After that, the mentioned entities in PreSQL are parsed to conduct schema linking, which can significantly compact the useful information. In the second stage, with the linked schema, we simplify the prompt's schema information and instruct the LLM to produce the final SQL. Finally, as the post-refinement module, we propose using cross-consistency across different LLMs rather than self-consistency within a particular LLM. Our methods achieve new SOTA results on the Spider benchmark, with an execution accuracy of 87.6%.