AlioLeuchtmann/qwen-3.5-sql-27B-Distill-2B
AlioLeuchtmann/qwen-3.5-sql-27B-Distill-2B is a 2.3 billion parameter text-to-SQL model fine-tuned from Qwen3.5-2B. It specializes in converting natural language questions into SQLite queries, achieving 68.47% execution accuracy on the official BIRD benchmark. This model utilizes execution-verified structured chain-of-thought distillation from a Qwen 27B teacher, enabling it to outperform significantly larger models in its domain while running efficiently on consumer hardware.
Loading preview...
Overview
AlioLeuchtmann/qwen-3.5-sql-27B-Distill-2B is a 2.3 billion parameter text-to-SQL model, distilled from a Qwen 27B teacher, specifically designed for SQLite databases. It achieves an impressive 68.47% execution accuracy on the official BIRD benchmark, making it one of the smallest models to rank highly on this leaderboard. This model demonstrates that knowledge distillation can create highly competent, specialized small models that rival or exceed the performance of much larger general-purpose LLMs for specific tasks.
Key Capabilities
- High Accuracy Text-to-SQL: Achieves 68.47% EX on the BIRD benchmark, outperforming many models over 500 times its size, including frontier general-purpose LLMs like Claude 4.5 Sonnet and DeepSeek-R1.
- Efficient Distillation: Utilizes execution-verified structured chain-of-thought distillation, meaning only demonstrably correct reasoning traces are used for training, without reinforcement learning or complex test-time thinking.
- Resource-Friendly: At 2.3 billion parameters, it can run on a single consumer GPU or even CPU/edge hardware, offering significant cost and latency advantages over larger models.
- Robust Prompting: Designed to work with OmniSQL-style prompts, including schema DDL, example rows, questions, and evidence hints, and can leverage column descriptions for improved accuracy.
Good For
- Specialized Text-to-SQL Applications: Ideal for scenarios requiring accurate conversion of natural language to SQLite queries, particularly for analytical questions.
- Edge/On-Premise Deployment: Its small size makes it suitable for deployment in environments with limited computational resources or strict data privacy requirements.
- Cost-Effective Solutions: Offers a highly efficient alternative to large, general-purpose LLMs for well-defined, verifiable tasks like SQL generation, reducing API call costs and inference latency.
- Research into Distillation: Serves as a strong example of how knowledge distillation can create expert small models for domains where task competence is verifiable.