Overview

The XiYanSQL-QwenCoder series, developed by XGenerationLab, focuses on advancing large language models for text-to-SQL applications. This 32 billion parameter model is part of a family that includes 3B, 7B, and 14B versions, catering to diverse developer needs. It is engineered to directly perform text-to-SQL tasks or serve as a robust base for further fine-tuning of SQL models.

Key Capabilities

• High-Performance SQL Generation: The XiYanSQL-QwenCoder-32B achieves a 69.03% EX score on the BIRD TEST set, establishing a new state-of-the-art for single fine-tuned models in this domain.
• Multi-Dialect Support: The model supports major SQL dialects such as SQLite, PostgreSQL, and MySQL, enhancing its applicability across various database environments.
• Benchmark Excellence: Across the series, models maintain leading positions in their respective size categories, demonstrating strong capabilities on both BIRD and Spider benchmarks, outperforming models like Codellama-34b, Deepseek-coder-33b, and even some larger commercial models in specific text-to-SQL metrics.

Good For

• Direct Text-to-SQL Conversion: Ideal for applications requiring accurate SQL query generation from natural language inputs.
• SQL Model Fine-tuning: Provides an excellent starting point for developers looking to fine-tune SQL models for specific use cases or proprietary databases.
• Multi-Database Environments: Suitable for projects that interact with different SQL database types due to its dialect support.