cycloneboy/CscSQL-Grpo-XiYanSQL-QwenCoder-7B-2502

TEXT GENERATIONConcurrency Cost:1Model Size:7.6BQuant:FP8Ctx Length:32kLicense:cc-by-nc-4.0Architecture:Transformer0.0K Open Weights Cold

The cycloneboy/CscSQL-Grpo-XiYanSQL-QwenCoder-7B-2502 is a 7.6 billion parameter model developed by CycloneBoy, specifically fine-tuned for text-to-SQL generation. This model utilizes a novel Corrective Self-Consistency (CSC-SQL) method, integrating Self-Consistency and Self-Correction, and is further enhanced by Group Relative Policy Optimization (GRPO) via reinforcement learning. It excels at translating natural language questions into accurate SQL queries, achieving 71.72% execution accuracy on the BIRD private test set.

Loading preview...

Overview

This model, CscSQL-Grpo-XiYanSQL-QwenCoder-7B-2502, is a 7.6 billion parameter language model developed by CycloneBoy, specifically designed for advanced text-to-SQL generation. It is part of the CSC-SQL framework, which introduces a novel approach combining Self-Consistency and Self-Correction to improve SQL query accuracy. The model is fine-tuned using Group Relative Policy Optimization (GRPO) via reinforcement learning, significantly enhancing its ability to generate correct and robust SQL queries from natural language.

Key Capabilities

  • Corrective Self-Consistency (CSC-SQL): Integrates Self-Consistency and Self-Correction by selecting the two most frequent outputs from parallel sampling and feeding them into a merge revision model for correction.
  • Reinforcement Learning Fine-tuning: Employs the GRPO algorithm to fine-tune both the SQL generation and revision models, leading to enhanced output quality.
  • High Accuracy: Achieves 71.72% execution accuracy on the BIRD private test set with its 7B variant, demonstrating strong performance in complex text-to-SQL tasks.
  • Robust Error Handling: Addresses both syntactic errors (via Self-Correction) and suboptimal output selection (via Self-Consistency) more effectively than traditional methods.

Good For

  • Text-to-SQL Applications: Ideal for systems requiring accurate translation of natural language questions into SQL queries for relational databases.
  • Database Interaction: Useful for developers building interfaces that allow users to query databases using natural language.
  • Research in LLM-based SQL Generation: Provides a strong baseline and methodology for further research in improving the reliability and accuracy of LLMs for database interactions.

Finetunes
1 models