Grounded_PRM is a grounded process supervision dataset designed for training and evaluating Process Reward Models (PRMs).The dataset focuses on step-level reasoning correctness, where each intermediate reasoning step is explicitly labeled to indicate whether it is logically valid and grounded toward solving the original problem.
The dataset is intended to support research on mathematical reasoning, chain-of-thought evaluation… See the full description on the dataset page: https://huggingface.co/datasets/Yuuuuuu98/Grounded_PRM.
