Skip to content

Dataset · meta-math

Math Problems with Augmented Solutions

meta-math/MetaMathQA

Math question-answer pairs augmented from GSM8K and MATH train sets for reasoning training.

What's inside

The dataset contains textual math question-answer pairs augmented from the training portions of GSM8K and MATH; test data was not used. The card indicates a volume of 100k–1M examples and an original_question field to verify the source. The format is not described in the card.

What you can do with it

  • Fine-tuning Mistral-7B or LoRA for mathematical reasoning
  • Verifying model transfer to GSM8K and MATH
  • Training SFT models to solve step-by-step problems
Size
100K–1M rows
Downloads this month
87.3K
Likes
492

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("meta-math/MetaMathQA")