RMBench

RMBench is a benchmark dataset used to evaluate the performance of reward models in aligning Large Language Models with human preferences.

Papers using RMBench (8)

PaTaRM: Bridging Pairwise and Pointwise Signals via Preference-Aware Task-Adaptive Reward Modeling2025 · 4 citesCDRRM: Contrast-Driven Rubric Generation for Reliable and Interpretable Reward Modeling2026Outcome Accuracy is Not Enough: Aligning the Reasoning Process of Reward Models2026C2: Scalable Rubric-Augmented Reward Modeling from Binary Preferences2026HelpSteer3-Preference: Open Human-Annotated Preference Data across Diverse Tasks and Languages2025Error Typing for Smarter Rewards: Improving Process Reward Models with Error-Aware Hierarchical Supervision2025HelpSteer3-Preference: Open Human-Annotated Preference Data across Diverse Tasks and Languages2025Efficient Process Reward Model Training via Active Learning2025