antgroup/Agentar-Scale-SQL-Generation-32B
1768
1---2license: apache-2.03base_model:4- seeklhy/OmniSQL-32B5---6 7 8## Important Links9 10[](https://arxiv.org/abs/2509.24403)11[](https://github.com/antgroup/Agentar-Scale-SQL)12[](https://bird-bench.github.io/)13[](https://huggingface.co/collections/antgroup/agentar-scale-sql)14[](https://modelscope.cn/collections/Agentar-Scale-SQL-0c368e98f73f41)15 16## Introduction17 18We are excited to release the **Agentar-Scale-SQL-Generation-32B**, the core **Reasoning SQL Generator** used in our SOTA framework, **Agentar-Scale-SQL**. Our framework achieved **81.67% execution accuracy** on the challenging BIRD benchmark, ranking first on the official leaderboard.19 20This model is a key component of our "Orchestrated Test-Time Scaling" strategy and has several key features:21 22 - **Base Model:** It is fine-tuned from `Omni-SQL-32B`.23 - **RL-Enhanced Reasoning:** The model was further trained using an execution-grounded **Reinforcement Learning** framework (GRPO) to enhance its intrinsic reasoning capabilities.24 - **Deep Reasoning:** It is engineered to conduct deep, step-by-step reasoning and construct complex, high-accuracy SQL queries.25 26This model is one of the two main generators in the `Agentar-Scale-SQL` framework's "Diverse Synthesis" step, working in parallel with an ICL generator to produce a robust pool of SQL candidates.27 28## Model Downloads29 30| **Model** | **Role** | 31|-----------------------------------|----------------|32| **Agentar-Scale-SQL-Generation-32B** | **SQL Generator** |33| Agentar-Scale-SQL-Selection-32B | SQL Selector |34 35## Performance36 37The performance metrics below reflect the **entire Agentar-Scale-SQL framework**, which uses this Generation model as a key component. The results demonstrate our SOTA performance on the BIRD benchmark.38 39| Methods | EX (Dev) | **EX (Test)** | R-VES (%) |40|:-----------------------------|:---:|:---:|:---------:|41| **Agentar-Scale-SQL (Ours)** | **74.90** | **81.67** | **77.00** |42| AskData + GPT-4o | 76.14 | 80.88 | 76.24 |43| LongData-SQL | 74.32 | 77.53 | 71.89 |44| CHASE-SQL + Gemini | 74.90 | 76.02 | 69.94 |45| JoyDataAgent-SQL | 74.25 | 75.74 | 70.16 |46| TCDataAgent-SQL | 74.12 | 75.74 | - |47| Contextual-SQL | 73.50 | 75.63 | 70.02 |48| XiYan-SQL | 73.34 | 75.63 | 71.41 |49 50 51## Prompt Template52 53````python54PROMPT_TEMPLATE = """Task Overview:55You are a data science expert. Below, you are provided with a database schema and a natural language question. Your task is to understand the schema and generate a valid SQL query to answer the question.56 57Database Engine:58{{ dialect }}59 60Database Schema:61{{ db_schemas }}62This schema describes the database's structure, including tables, columns, primary keys, foreign keys, and any relevant relationships or constraints.63{% if matched_contents %}64Matched contents:65{{ matched_contents }}66Matched contents presents values related to the question, together with their source table and column, for your reference in SQL generation.67{% endif %}68Question:69{%- if hint %}70{{ hint }}71{{ question }}72{%- else %}73{{ question }}74{%- endif %}75 76Instructions:77- If Matched contents is provided, you can use it as reference when generating the SQL query.78- Make sure you only output the information that is asked in the question. If the question asks for a specific column, make sure to only include that column in the SELECT clause, nothing more.79- The generated query should return all of the information asked in the question without any missing or extra information.80- Before generating the final SQL query, please think through the steps of how to write the query.81 82Output Format:83In your answer, please enclose the generated SQL query in a code block:84```sql85-- Your SQL query86```87 88Take a deep breath and think step by step to find the correct SQL query.89"""90````91 92## Acknowledgments93 94If you find our work useful, please cite the Agentar-Scale-SQL paper:95 96```bibtex97@misc{wang2025agentarscalesqladvancingtexttosqlorchestrated,98 title={Agentar-Scale-SQL: Advancing Text-to-SQL through Orchestrated Test-Time Scaling}, 99 author={Pengfei Wang and Baolin Sun and Xuemei Dong and Yaxun Dai and Hongwei Yuan and Mengdie Chu and Yingqi Gao and Xiang Qi and Peng Zhang and Ying Yan},100 year={2025},101 eprint={2509.24403},102 archivePrefix={arXiv},103 primaryClass={cs.CL},104 url={https://arxiv.org/abs/2509.24403}, 105}106```107 108