ResearchBenchmarksFaithfulness on Summarization (test)Follow-0.268RewardPPO-0.54672-0.47436-0.402-0.32964Aug 19, 2025Evaluation ResultsMethodMethodLinksRewardKL DivergencePPOModel Size=13BModel Size=13B2025.08-0.2685.71MAVISModel Size=13BModel Size=13B2025.08-0.3013.1MAVISModel Size=7BModel Size=7B2025.08-0.3462.25PPOModel Size=7BModel Size=7B2025.08-0.5363.93