OmniSQL: Synthesizing High-quality Text-to-SQL Data at Scale

TL;DR

This work tackles the scalability and generalization gaps in text-to-SQL by proposing a fully automatic, multi-stage data-synthesis framework that generates SynSQL-2.5M, a large, diverse, CoT-enabled dataset. Built upon this data, the authors train OmniSQL, an open-source text-to-SQL model in 7B, 14B, and 32B sizes, achieving state-of-the-art results across nine benchmarks and rivaling larger closed-source models while using fewer parameters. The framework decomposes data synthesis into web-table-driven database creation, complexity-aware SQL generation, stylized NL question synthesis, and CoT synthesis, with robust quality validation from GPT-4o and human experts. The work demonstrates strong cross-domain generalization, meaningful domain coverage, and clear practical impact for cost-effective, privacy-preserving, local deployment of open-source text-to-SQL systems.

Abstract

Text-to-SQL, the task of translating natural language questions into SQL queries, plays a crucial role in enabling non-experts to interact with databases. While recent advancements in large language models (LLMs) have significantly enhanced text-to-SQL performance, existing approaches face notable limitations in real-world text-to-SQL applications. Prompting-based methods often depend on closed-source LLMs, which are expensive, raise privacy concerns, and lack customization. Fine-tuning-based methods, on the other hand, suffer from poor generalizability due to the limited coverage of publicly available training data. To overcome these challenges, we propose a novel and scalable text-to-SQL data synthesis framework for automatically synthesizing large-scale, high-quality, and diverse datasets without extensive human intervention. Using this framework, we introduce SynSQL-2.5M, the first million-scale text-to-SQL dataset, containing 2.5 million samples spanning over 16,000 synthetic databases. Each sample includes a database, SQL query, natural language question, and chain-of-thought (CoT) solution. Leveraging SynSQL-2.5M, we develop OmniSQL, a powerful open-source text-to-SQL model available in three sizes: 7B, 14B, and 32B. Extensive evaluations across nine datasets demonstrate that OmniSQL achieves state-of-the-art performance, matching or surpassing leading closed-source and open-source LLMs, including GPT-4o and DeepSeek-V3, despite its smaller size. We release all code, datasets, and models to support further research.

Figures (4)

• Figure 1: Illustration of the proposed text-to-SQL data synthesis framework.
• Figure 2: Database domain and question visualizations.
• Figure 3: Quality evaluation of SynSQL-2.5M and BIRD judged by GPT-4o. Scores are computed using Equation \ref{['eq:score']}.
• Figure 4: Performance changes (measured by accuracy; higher is better) with increasing training data.