COAL-SQL Framework Uses Coverage-Guided Augmentation for Text-to-SQL Training

The COAL-SQL framework improves open-source text-to-SQL performance through Coverage-Guided Augmentation (CGA) to identify structural data gaps and Failure-Driven Learning (FDL) to address training weaknesses. This method moves beyond standard supervised fine-tuning by dynamically targeting missing SQL structures.