Skip to content

Dataset · TIGER-Lab

Expanded math problem dataset for LLMs

TIGER-Lab/MATH-plus

Over 100k English math problems for training reasoning models.

What's inside

An English text dataset combining MetaMath, MATH-orca samples, and additional MATH-based problems generated using GPT-4. The total size ranges from 100k to 1 million records. The set was used in training the MAmmoTH2-plus mathematical model.

What you can do with it

  • SFT training and creating LoRA to improve LLM mathematical capabilities
  • Training models in step-by-step resolution of complex English problems
Tasks
Text generation
Languages
English
Size
100K–1M rows
Downloads this month
654
Likes
46

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("TIGER-Lab/MATH-plus")