SQLPrompt: In-Context Text-to-SQL with Minimal Labeled Data

Anonymous

Abstract: Text-to-SQL aims to automate the process of generating SQL queries on a database from natural language text. In this paper, we propose SQLPrompt, a novel method to push the state-of-the-art of Text-to-SQL with in-context learning, leveraging LLM's zero-shot and few-shot adaptation capability. Our method comprises a novel prompt design approach to efficiently consider the database information; execution-based consistency decoding; and employing mixture of prompts and/or LLMs. We show that SQLPrompt outperforms previous state-of-the-art for in-context learning with zero labeled data by a large margin, closing the gap with finetuning state-of-the-art with thousands of labeled data.

Paper Type: short

Research Area: NLP Applications