TeCoD: Template Constrained Decoding Boosts Text-to-SQL Accuracy by 36%

Researchers have introduced Template Constrained Decoding (TeCoD), a system that improves the accuracy and reliability of Text-to-SQL generation using large language models. TeCoD converts historical natural language-SQL pairs into reusable templates and employs a fine-tuned natural language inference model to select the appropriate template for a given query. During SQL generation, TeCoD enforces the template through grammar-constrained decoding with a novel partitioned strategy, ensuring syntactic validity and efficiency. The system achieves up to 36% higher execution accuracy compared to existing methods, particularly in complex or unseen schemas. This addresses key challenges in real-world deployment of LLM-based Text-to-SQL systems, such as inconsistent accuracy and invalid SQL generation.