COAL-SQL: Coverage-Guided Augmentation and Failure-Driven Learning for Text-to-SQL Post-Training

COAL-SQL, a unified post-training framework combining Coverage-Guided Augmentation (CGA) and Failure-Driven Learning (FDL), achieves 64.9% execution accuracy on the BIRD development set using only 12,600 distinct post-training examples, according to the arXiv paper 2609.20842v1. CGA uses greedy selection to identify SQL structures missing from the original dataset and constructs complementary examples, while FDL retains GRPO as the main optimization objective and adds step-level SFT on verified reasoning traces from a strong LLM plus epoch-level retrieval of structurally related examples for accumulated failures. The authors report COAL-SQL outperforms baselines trained at comparable scale, with code available at https://github.com/TechNomad-ds/COAL-SQL.

arXiv:2609.20842v1 Announce Type: new Abstract: Text-to-SQL translates natural-language questions into executable SQL queries, but open-source large language models still require task-specific post-training for complex, real-world SQL generation. Effective post-training requires both training data that cover the capabilities demanded by the target task and a learning strategy that enables the model to acquire them. Existing datasets provide valuable supervision but incompletely cover SQL structures, while augmentation methods typically expand data without identifying structural gaps. Moreover, supervised fine-tuning SFT or reinforcement learning RL alone cannot dynamically address weaknesses exposed during training. We propose COAL-SQL, a unified framework combining Coverage-Guided Augmentation CGA and Failure-Driven Learning FDL . CGA uses greedy selection to identify SQL structures missing from the original dataset and constructs complementary examples, improving structural coverage. FDL retains GRPO as the main optimization objective while supplying targeted supervision for unsolved examples. At the step level, it applies SFT to verified reasoning traces generated by a strong LLM for accumulated failures. At the epoch level, it retrieves structurally related examples based on accumulated failures to create targeted practice, helping the model acquire the corresponding SQL capabilities. With only 12,600 distinct post-training examples, COAL-SQL achieves 64.9% execution accuracy on the BIRD development set and outperforms baselines trained at comparable scale. The code is available at https://github.com/TechNomad-ds/COAL-SQL.