ResearchBenchmarksKnowledge Evaluation on Winogrande (Evaluation)Follow58AccuracyDisagreement56.4456.84557.2557.655Jan 31, 2026Evaluation ResultsMethodMethodLinksAccuracyDisagreementProtocol=Training-timeProtocol=Training-time2026.0158ActDecProtocol=Inference-timeProtocol=Inference-time2026.0157.5BaselineArchitecture=LoRA, Los...Architecture=LoRA, Loss=Vanilla CE, Protocol=Training-time2026.0157.3MEProtocol=Training-timeProtocol=Training-time2026.0157.3CADProtocol=Inference-timeProtocol=Inference-time2026.0156.9GAME-LoRAProtocol=Training-timeProtocol=Training-time2026.0156.5