Process Reward Model trained by OpenRLHF

- Dataset: Math-Shepherd (https://huggingface.co/datasets/peiyi9979/Math-Shepherd)
- Learning Rate: 1e-6
- Training Accuracy: 0.922