Skip to content

Dataset · Zhiqiang007

Multimodal math dataset with 360k pairs

Zhiqiang007/MathV360K

40,000 images and 360,000 question-answer pairs for training vision-language models on problem-solving.

What's inside

Created by the authors of Math-LLaVA, this dataset combines 40,000 images from 24 sources and 360,000 English question-answer pairs. Total volume spans between 100,000 and 1,000,000 records. Using these data, Math-LLaVA achieved 46.6% accuracy on the MathVista benchmark and 15.69% on MathVision.

What you can do with it

  • Training multimodal models (MLLMs) in mathematical reasoning over images
  • Fine-tuning vision-language assistants to recognize and solve chart and diagram problems
Languages
English
Size
100K–1M rows
Downloads this month
1.8K
Likes
35

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("Zhiqiang007/MathV360K")