ResearchBenchmarksHumor on HH-RLHF (test)Follow2.481RewardPPO2.357242.389372.42152.45363Aug 19, 2025Evaluation ResultsMethodMethodLinksRewardKL DivergencePPOModel Size=13BModel Size=13B2025.082.48112.14MAVISModel Size=13BModel Size=13B2025.082.4659.2MAVISModel Size=7BModel Size=7B2025.082.3763.78PPOModel Size=7BModel Size=7B2025.082.36210.43