Skip to content

Dataset · approximatelabs

627 million tables from GitHub and Common Crawl

approximatelabs/tablib-v1-full

Table corpus with provenance metadata and context for training tabular models.

What's inside

The dataset contains 627 million tables (69 TiB) extracted from HTML, PDF, CSV, TSV, Excel, and SQLite in GitHub and Common Crawl. These are lightly preprocessed tables plus 867 billion tokens of contextual metadata: provenance, filename, preceding/following text, and HTML metadata. A 0.1% subsample is available; the storage format is not specified in the card.

What you can do with it

  • Pre-train a table understanding model
  • Train table and context extraction
  • Use metadata for table retrieval
Size
10B–100B rows
Downloads this month
19.5K
Likes
70
Access
on request on Hugging Face

How to load it

from datasets import load_dataset

ds = load_dataset("approximatelabs/tablib-v1-full")