Skip to content

Dataset · TIGER-Lab

Massive corpus of web instructions for LLM training

TIGER-Lab/WebInstructFull

Tens of millions of English question-answer pairs from the web collected for the MAmmoTH2 project.

What's inside

The full version of the dataset from TIGER-Lab used to train the MAmmoTH2 models. It contains 10M to 100M English text examples in an instruction-response format scraped from the web. Distributed under a CC BY-NC 4.0 license, it is restricted to non-commercial research.

What you can do with it

  • Large-scale SFT fine-tuning of language models to improve reasoning and question-answering skills
  • Training open research models using the MAmmoTH2 pipeline
Tasks
Question answering
Languages
English
Size
10M–100M rows
Downloads this month
1.3K
Likes
36

How to load it

from datasets import load_dataset

ds = load_dataset("TIGER-Lab/WebInstructFull")