RingSQL: Schema-Independent Synthetic Data Generation for Text-to-SQL Reinforcement Learning

Marko Sterbentz¹   Kevin Cushing²   Cameron Barrie¹   Kristian J. Hammond¹  

¹Northwestern University      ²Purdue University

📖 Overview

Recent advances in text-to-SQL have been driven by larger models, better datasets, and new training methods like RLVR. However, progress remains limited by scarce high-quality training data, a problem RLVR is especially sensitive to since noisy data can produce spurious rewards. Manual data creation is expensive, and existing synthetic methods trade off reliability for scalability: template-based approaches guarantee correct SQL but need schema-specific templates and lack diversity, while LLM-based generation scales easily but lacks quality guarantees. We introduce RingSQL, a hybrid framework for generating question-SQL pairs that combines schema-independent query templates with LLM-based paraphrasing of natural language questions. By grounding question generation in complete template questions, RingSQL preserves question-query correctness across all levels of query complexity, a property purely LLM-based methods fail to maintain. RingSQL also produces the only synthetic dataset that improves RLVR training performance across all tested model architectures and benchmarks, achieving 69.8% average accuracy and surpassing both the next-best synthetic dataset by 2.1% and human-annotated data from Spider and BIRD.

📊 Data

You can find the RingSQL data used in our experiments on HuggingFace:

Dataset	Size (train/dev)	HuggingFace Link
RingSQL-Gen	5000 (4000/1000)	🤗 Download

🤖 Model Weights

You can find the weights for models trained on data produced by RingSQL on HuggingFace:

Model	Size	HuggingFace Link
Qwen2.5-Coder-3B-Instruct-RingSQL-Gen	3B	🤗 Download
Qwen2.5-Coder-7B-Instruct-RingSQL-Gen	7B	🤗 Download

Installation

Satyrn Setup

Coming soon.

Data Generation Setup

Coming soon.

Training Setup

See the installation and usage instructions in the training directory.

📚 Citation

If you found RingSQL useful in your own work, please consider citing our paper:

@misc{sterbentz2026ringsql,
      title={RingSQL: Schema-Independent Synthetic Data Generation for Text-to-SQL Reinforcement Learning},
      author={Marko Sterbentz and Kevin Cushing and Cameron Barrie and Kristian J. Hammond},
      year={2026},
      eprint={2601.05451},
      archivePrefix={arXiv},
      url={https://arxiv.org/abs/2601.05451}
}