Measuring Multimodal Mathematical Reasoning
with the MATH-Vision Dataset