Beyond Length Scaling: Synergizing Breadth and Depth for Generative Reward Models
Open paper
Benchmark
Model
Rank
Results
reward-model-evaluation-on-rewardbench
Mix-GRM (SFT+RLVR)
#2
Accuracy: 91.8
arithmetic-reasoning-on-gsm8k
Mix-GRM (used in DPO)
#177
Accuracy: 77.6
Rank counts only results with a code link.