PRM-800K

The PRM-800K dataset is a benchmark containing process labels used to evaluate the performance of process reward models (PRMs) in generating verification chains-of-thought for step-level supervision in various challenging tasks.

Papers using PRM-800K (3)

Process Reward Models That Think2025 · 1 citesSPC: Evolving Self-Play Critic via Adversarial Games for LLM Reasoning2025Towards Hierarchical Multi-Step Reward Models for Enhanced Reasoning in Large Language Models2025