Skip to content

Dataset · MathLLMs

MATH-Vision benchmark for multimodal mathematics

MathLLMs/MathVision

Visual math problems with images and text for evaluating VLMs.

What's inside

MATH-Vision is a benchmark for measuring multimodal mathematical reasoning. It includes question-answering, multiple-choice, visual-question-answering, and image-to-text tasks; modalities are image and text, language is English, size is 1K<n<10K. There is a MATH-Vision-Wild variant: the same testmini tasks photographed on paper, iPads, laptops, and projectors under varying lighting and angles.

What you can do with it

  • Evaluating VLMs on visual math problems
  • Comparing models on MATH-Vision leaderboards
  • Testing VLM robustness to real-world problem photos
Tasks
Question answering, Multiple choice, Visual question answering, Text generation, Image → text, Image text → text
Languages
English
Size
1K–10K rows
Downloads this month
15.7K
Likes
184

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("MathLLMs/MathVision")