A small dataset that mirrors the MMMU-Pro benchmark distribution (30 academic subjects, college-level MCQs over images), generated synthetically. This testv1 set contains 78 questions × 3 variants = 234 rows. The three variants share the same underlying question but differ in presentation:

Variant Format File pattern

4opt Standard 4-option MCQ; image + text question + 4 options testv1/4opt/images/o4_XXXX.png

10opt Augmented… See the full description on the dataset page: https://huggingface.co/datasets/YiYang109/MMMU_distribution_simulation.