Skip to content

Dataset · trl-lab

Large question-SQL query pair corpus for model training

trl-lab/SQaLe-2-text-to-SQL-Queries

Over 1.4 million English questions and 176k SQL queries based on real database schemas.

What's inside

A semi-synthetic English text dataset created for training compact specialized text-to-SQL models. Contains 1,408,056 natural language questions and 176,761 unique SQL queries across 9,259 populated SQLite databases. Table schemas are taken from the real SchemaPile collection.

What you can do with it

  • Training and SFT of compact SQL generation models
  • Building table-based question-answering search systems
  • Benchmarking text-to-SQL conversion quality on complex schemas
Tasks
Text generation, Table question answering
Languages
English
Size
100K–1M rows
Downloads this month
530
Likes
6

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("trl-lab/SQaLe-2-text-to-SQL-Queries")