Daily Arxiv

This is a page that curates AI-related papers published worldwide.
All content here is summarized using Google Gemini and operated on a non-profit basis.
Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.

Auto prompt sql: a resource-efficient architecture for text-to-sql translation in constrained environments

Created by
  • Haebom

Author

Zetong Tang, Qian Ma, Di Wu

Outline

This paper aims to present an efficient Text-to-SQL model in resource-constrained environments. To overcome the reliance on existing resource-intensive, large-scale open-source models, we propose a novel architecture called Auto Prompt SQL (AP-SQL). AP-SQL decomposes the Text-to-SQL conversion task into three steps: schema filtering, context-based, example-based search-enhanced Text-to-SQL generation, and prompt-based schema association and SQL generation. To improve schema selection accuracy, we fine-tune a large-scale language model, and through prompt engineering techniques utilizing Chain-of-Thought (CoT) and Graph-of-Thought (GoT) templates, we enhance the model's inference ability and achieve accurate SQL generation. We demonstrate the effectiveness of AP-SQL through comprehensive evaluation results using the Spider benchmark.

Takeaways, Limitations

Takeaways:
Presenting the possibility of implementing an efficient Text-to-SQL system in a resource-constrained environment.
A new architecture proposal that allows large-scale models to leverage their performance in small-scale models.
Demonstrating the effectiveness of model performance improvement through prompt engineering techniques (CoT, GoT).
AP-SQL's superior performance is verified on the Spider benchmark.
Limitations:
Lack of performance validation on datasets other than the Spider benchmark
Lack of detailed description of the types and characteristics of large-scale language models used.
Further research is needed on the scalability and generalization performance of AP-SQL.
Additional verification of performance and stability in real-world application environments is required.
👍