Integrating Reasoning and Generalization in Text-to-SQL via Self-Enhanced Fine-Tuning

The article introduces CoTE-SQL, a novel approach to text-to-SQL generation that enhances reasoning and generalization in large language models. Key innovations include self-enhanced reasoning traces, structured chain-of-thought prompting, and error-aware revisions based on SQL execution feedback. CoTE-SQL achieves state-of-the-art results on the Bird benchmark (53.39% EX / 59.02 VES) and strong performance on the Spider benchmark (79.60% EX / 77.19 VES), particularly excelling in handling complex queries, which is crucial for practitioners aiming to improve the robustness of LLM applications in database querying.