ResearchBenchmarksReward Modeling on D_dish real policy rollouts (test)Follow0.67Rollout ρSARM-0.37-0.10.170.44Sep 29, 2025Evaluation ResultsMethodMethodLinksRollout ρSE CountPSE CountFE CountSARM2025.090.6710911ReWiND2025.090.551297w/o Rdescription=SARM witho...description=SARM without Rewind augmentation2025.090.51278LIV2025.090.397126GVL2025.0900612VLC2025.09-0.331200