Skip to content

Dataset · b-mc2

Text-to-SQL with CREATE TABLE context

b-mc2/sql-create-context

78,577 examples of natural language queries, SQL CREATE TABLE, and SQL queries for text-to-SQL.

What's inside

The dataset is compiled from WikiSQL and Spider and contains 78,577 examples: a natural language query, an SQL CREATE TABLE statement, and an answering SQL query. Data is in English, in text format. CREATE TABLE provides table and column names as context to reduce hallucinations in text-to-SQL.

What you can do with it

  • Train text-to-SQL LLMs
  • Perform LoRA fine-tuning for schema-based SQL generation
  • Evaluate table and column name hallucinations
Tasks
Text generation, Question answering, Table question answering
Languages
English
Size
10K–100K rows
Downloads this month
9.3K
Likes
518

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("b-mc2/sql-create-context")