SQL-PaLM: A Comprehensive Framework for Enhancing Text-to-SQL Performance with Large Language Models
SQL-PaLM is a comprehensive framework that adapts large language models, specifically PaLM-2, for enhancing Text-to-SQL performance through few-shot prompting and instruction fine-tuning. The framework explores key aspects such as diversifying training data coverage, incorporating synthetic data, integrating query-specific database content, and efficient column selection to enable scaling to real-world databases.