Qwen-2.5-coder-Arctic-ExCoT-32B
| Property | Value |
|---|---|
| Base Model | Qwen-2.5-coder |
| Parameters | 32B |
| Developer | Snowflake |
| Model URL | Hugging Face |
What is Qwen-2.5-coder-Arctic-ExCoT-32B?
Qwen-2.5-coder-Arctic-ExCoT-32B is a state-of-the-art Text-to-SQL model developed by Snowflake's AI research team. It represents the first model in the Arctic Text2SQL family, implementing the novel ExCoT framework that combines Chain-of-Thought prompting with SQL execution-based DPO optimization.
Implementation Details
The model leverages execution results rather than human preferences as feedback signals, enabling scalable and high-quality optimization without requiring expensive human annotations. Built on the Qwen-2.5-coder architecture with 32 billion parameters, it achieves remarkable performance on the BIRD benchmark.
- Achieves 68.25% execution accuracy on BIRD-dev set
- Maintains 68.19% accuracy on BIRD-test set
- Outperforms major models like GPT-4 and Claude 3.5 by significant margins
- Trained on public datasets (BIRD and Spider)
Core Capabilities
- Advanced Text-to-SQL conversion with high accuracy
- Execution-guided reasoning through Chain-of-Thought
- Robust performance on complex database queries
- Zero requirement for human preference data in optimization
Frequently Asked Questions
Q: What makes this model unique?
The model's ExCoT framework sets it apart by using SQL execution results as feedback for optimization, eliminating the need for human annotations while achieving state-of-the-art performance on Text-to-SQL tasks.
Q: What are the recommended use cases?
The model is specifically designed for converting natural language queries into SQL, making it ideal for database query applications, data analysis tools, and business intelligence systems requiring natural language interfaces.
