Search papers, labs, and topics across Lattice.
The paper introduces Intergroup Relative Preference Modeling (IRPM), a novel method for training pointwise Generative Reward Models (GRMs) from pairwise preference data. IRPM addresses the computational bottleneck of pairwise GRMs in RLHF by using intergroup comparisons of chosen vs. rejected samples to derive pointwise rewards, enabling O(n) reward evaluation. Experiments demonstrate that IRPM achieves state-of-the-art performance among pointwise GRMs on RM-Bench, JudgeBench, and RewardBench, and shows significant improvements in post-training evaluations.
Pointwise reward models can finally compete with pairwise models in RLHF, thanks to a new intergroup comparison method that scales linearly with the number of candidates.
Generative Reward Models (GRMs) have demonstrated strong performance in reward modeling, due to their interpretability and potential for refinement through reinforcement learning (RL). However, widely used pairwise GRMs create a computational bottleneck in reinforcement learning from human feedback (RLHF), when calibrating or aggregating preference signals over n candidates, often incurring O(n^2) pairwise judgments. To address this issue, we propose Intergroup Relative Preference Modeling (IRPM), an RL-based method that extends the Bradley--Terry preference-learning paradigm via intergroup comparisons to train pointwise GRMs from pairwise preference data. IRPM derives pointwise reward for each response by contrasting groups of chosen vs. rejected samples, enabling pointwise scores comparable across candidate sets and O(n) reward evaluation for a variable number of candidates during RL training, while preserving interpretability and scalability. Experiments show that IRPM achieves state-of-the-art performance among pointwise GRMs on RM-Bench, JudgeBench and RewardBench, and approaches the performance of leading pairwise GRMs. In addition, IRPM achieves substantial gains in post-training evaluations, demonstrating its effectiveness.
