ProJudge is a comprehensive, multi-modal, multi-discipline, and multi-difficulty benchmark specifically designed for evaluating abilities of MLLM-based process judges.
It comprises 2,400 test cases and 50,118 step-level labels, spanning four scientific disciplines with diverse difficulty levels and multi-modal content.
In ProJudgeBench, each step is meticulously annotated… See the full description on the dataset page:
https://huggingface.co/datasets/julyai/ProJudgeBench.
