ResearchBenchmarksCommonsense Reasoning on Wino-X DE 5-shot (test)Commonsense Reasoning on Wino-X DE 5-shot (test)Follow80.3Normalized Log AccuracyT-Free76.34877.37478.479.426Mar 16, 2026Evaluation resultsMethodMethodLinksNormalized Log AccuracyBytes per Sequence PositionT-Freeshots=5-shot, evaluati...shots=5-shot, evaluation_mode=Pre-training2026.0380.35.63HATifiedshots=5-shot, evaluati...shots=5-shot, evaluation_mode=Pre-training2026.0378.95.63Llamashots=5-shot, evaluati...shots=5-shot, evaluation_mode=Pre-training2026.0376.53.67