The capabilities of Multi-modal Large Language Models (MLLMs) in visual math problem-solvingremain insufficiently evaluated and understood. We investigate current benchmarks to incorporate excessive visual content within textual questions, which potentially assist MLLMs in deducing answers without truly interpreting the input diagrams.

To… See the full description on the dataset page: https://huggingface.co/datasets/AI4Math/MathVerse.