Using LLM to select the right SQL Query from candidates

This academic article discusses the use of text-to-SQL models to generate candidate SQL queries. However, the best query often isn’t at the top of the list. The authors propose a re-rank method that uses LLMs to predict the ground truth and select the correct SQL query from the candidate list. This method generates test cases and re-ranks the list based on their pass numbers on these test cases and their generation probabilities. The method has shown to improve the performance of some models by 3.6%.

Publication date: 5 Jan 2024
Project Page: Not provided
Paper: https://arxiv.org/pdf/2401.02115