ResearchBenchmarksQuestion Answering on ARC Challenge DE (with Efficiency Metrics)Follow57.2Normalized Log AccuracyHATified32.65639.02845.451.772Mar 16, 2026Evaluation ResultsMethodMethodLinksNormalized Log AccuracyBytes per Sequence PositionHATifiedtraining=SFT, shots=25...training=SFT, shots=25-shot2026.0357.26.57Tülutraining=SFT, shots=25...training=SFT, shots=25-shot2026.0333.63.68