ResearchBenchmarksCommon Sense Reasoning on Wino-X DEFollow76.3Normalized Log AccuracyHATified65.27668.1387173.862Mar 16, 2026Evaluation ResultsMethodMethodLinksNormalized Log AccuracyBytes per Sequence PositionHATifiedtraining=SFT, shots=5-...training=SFT, shots=5-shot2026.0376.35.63Tülutraining=SFT, shots=5-...training=SFT, shots=5-shot2026.0365.73.67