DataGpt-SQL-7B: An Open-Source Language Model for Text-to-SQL

An effective text-to-SQL methodology significantly enhances the efficiency of data analysis and retrieval, empowering non-expert users to interact with databases through their native language, democratizing access to and analysis of data. However, the reliance on powerful, close-sourced Large Language Models (LLM) and sophisticated prompting techniques to address these tasks may introduce risks associated with openness, privacy, and substantial costs. In light of these considerations, we propose a series of compact, fine-tuned models to tackle text-to-SQL problems. Schema linking, particularly the alignment of table and column information, is a recurrent issue in text-to-SQL tasks. We proposed the cross-DataBase and Inner-DataBase methods to construct samples, improving the performance of model to identify the correct schema and columns from context.