ResearchBenchmarksQuestion Answering on Science Q&A Materials (test)Follow79.1Mean@16SC-SDPO34.946.37557.8569.325May 26, 2026Evaluation ResultsMethodMethodLinksMean@16Majority@16SC-SDPOBackbone=OLMo-3-7B, Tr...Backbone=OLMo-3-7B, Training Method=SC-SDPO, Alpha (α)=0.5, Sampling Count (k)=162026.0579.180.7SDPOBackbone=OLMo-3-7B, Tr...Backbone=OLMo-3-7B, Training Method=SDPO, Sampling Count (k)=162026.057880OLMo-3-7BBackbone=OLMo-3-7B, Tr...Backbone=OLMo-3-7B, Training Method=Pre-trained, Sampling Count (k)=162026.0536.630.9