Dataset Overview

Basic Information

• Name: gretelai/synthetic_text_to_sql
• License: Apache-2.0
• Language: English
• Tags: synthetic, SQL, text‑to‑SQL, code
• Task Types: question answering, table QA, text generation
• Size Category: 100K < size < 1M

Dataset Content

• Number of Records: 105,851 (100,000 training, 5,851 test)
• Total Tokens: ~23 M, including ~12 M SQL tokens
• Covered Domains: 100 different domains/verticals
• SQL Task Types: data definition, retrieval, manipulation, analysis, reporting
• SQL Complexity: includes sub‑queries, single joins, multiple joins, aggregation, window functions, set operations
• Database Context: includes table and view creation statements
• Natural‑Language Explanation: explanations of the SQL queries
• Contextual Tags: used to optimize model training

Dataset Characteristics

• Diversity: broad range of SQL complexities and task types
• Quality: evaluated with LLM‑as‑a‑judge; scores higher than the b‑mc2/sql‑create‑context dataset on SQL standard compliance, correctness, and instruction adherence
• Applications: suitable for developers, researchers, and data enthusiasts building or refining text‑to‑SQL models

Dataset Structure

• Number of Fields: 11
• Example Fields: id, domain, domain_description, sql_complexity, sql_complexity_description, sql_task_type, sql_task_type_description, sql_prompt, sql_context, sql, sql_explanation

Data Quality Evaluation

• Evaluation Method: GPT‑4 scored 1,000 random samples and compared results with the b‑mc2/sql‑create‑context dataset
• Evaluation Results: outperformed the comparison dataset on multiple metrics

Citation Information

@misc{gretel‑synthetic‑text‑to‑sql-2024,
  author = {Meyer, Yev and Emadi, Marjan and Nathawani, Dhruv and Ramaswamy, Lipika and Boyd, Kendrick and Van Segbroeck, Maarten and Grossman, Matthew and Mlocek, Piotr and Newberry, Drew},
  title = {{Synthetic‑Text‑To‑SQL}: A synthetic dataset for training language models to generate SQL queries from natural language prompts},
  month = {April},
  year = {2024},
  url = {https://huggingface.co/datasets/gretelai/synthetic-text-to-sql}
}