The paper "DTS-SQL: Decomposed Text-to-SQL with Small Large Language Models" by Mohammadreza Pourreza and Davood Rafiei from the University of Alberta introduces a novel two-stage fine-tuning approach to improve the performance of small open-source models in the text-to-SQL task. The approach decomposes the task into two simpler components: schema linking and SQL generation, aiming to enhance the alignment between the models and the task requirements. - **Problem**: Leading models for text-to-SQL heavily rely on proprietary Large Language Models (LLMs), which pose concerns over data privacy and cost. - **Solution**: The proposed two-stage fine-tuning method uses two smaller LLMs (7 billion parameters each) to improve execution accuracy by 3 to 7 percent compared to conventional single-step fine-tuning methods. - **Evaluation**: The method is evaluated on two large cross-domain datasets (Spider and Spider-SYN) and two small LLMs (DeepSeek 7B and Mistral 7B). The results show that the two-stage approach achieves comparable performance to state-of-the-art methods using GPT-4 with few-shot learning and well-designed prompts. - **Methodology**: The two-stage approach involves: - **Schema Linking Fine-Tuning**: Identifying relevant tables and columns from natural language queries. - **SQL Generation Fine-Tuning**: Constructing SQL queries based on the identified schema. - **Results**: The method achieves state-of-the-art performance on the Spider development set and comparable performance to larger models on the test set. However, there is still room for improvement, particularly in the schema-linking stage. - **Conclusion**: The two-stage fine-tuning approach enables small open-source models to rival larger models in the text-to-SQL task, addressing the reliance on proprietary models and improving performance. - The paper emphasizes the importance of ethical considerations in research, adhering to the ACL Ethics Policy and upholding ethical principles. - The authors acknowledge the limitations of the current approach, particularly in schema-linking, and suggest further research in this area. - The paper provides detailed experimental results and comparisons with various baseline approaches, demonstrating the effectiveness of the proposed method.The paper "DTS-SQL: Decomposed Text-to-SQL with Small Large Language Models" by Mohammadreza Pourreza and Davood Rafiei from the University of Alberta introduces a novel two-stage fine-tuning approach to improve the performance of small open-source models in the text-to-SQL task. The approach decomposes the task into two simpler components: schema linking and SQL generation, aiming to enhance the alignment between the models and the task requirements. - **Problem**: Leading models for text-to-SQL heavily rely on proprietary Large Language Models (LLMs), which pose concerns over data privacy and cost. - **Solution**: The proposed two-stage fine-tuning method uses two smaller LLMs (7 billion parameters each) to improve execution accuracy by 3 to 7 percent compared to conventional single-step fine-tuning methods. - **Evaluation**: The method is evaluated on two large cross-domain datasets (Spider and Spider-SYN) and two small LLMs (DeepSeek 7B and Mistral 7B). The results show that the two-stage approach achieves comparable performance to state-of-the-art methods using GPT-4 with few-shot learning and well-designed prompts. - **Methodology**: The two-stage approach involves: - **Schema Linking Fine-Tuning**: Identifying relevant tables and columns from natural language queries. - **SQL Generation Fine-Tuning**: Constructing SQL queries based on the identified schema. - **Results**: The method achieves state-of-the-art performance on the Spider development set and comparable performance to larger models on the test set. However, there is still room for improvement, particularly in the schema-linking stage. - **Conclusion**: The two-stage fine-tuning approach enables small open-source models to rival larger models in the text-to-SQL task, addressing the reliance on proprietary models and improving performance. - The paper emphasizes the importance of ethical considerations in research, adhering to the ACL Ethics Policy and upholding ethical principles. - The authors acknowledge the limitations of the current approach, particularly in schema-linking, and suggest further research in this area. - The paper provides detailed experimental results and comparisons with various baseline approaches, demonstrating the effectiveness of the proposed method.