ResearchBenchmarksMathematical Reasoning on ORZ57K Suite (test)Follow37.1Final Test AccuracyGRPO35.43635.86836.336.732Feb 17, 2025Evaluation ResultsMethodMethodLinksFinal Test AccuracySpeed-up (Steps)GRPOLILO Integration=trueLILO Integration=true2025.0237.11.5GRPOLILO Integration=falseLILO Integration=false2025.0235.5—