ResearchBenchmarksMassive Multitask Language Understanding on MMLU (Accuracy and Efficiency Profile)Massive Multitask Language Understanding on MMLU (Accuracy and Efficiency Profile)Follow67Normalized Log AccuracyLlama65.64865.99966.3566.701Mar 16, 2026Evaluation resultsMethodMethodLinksNormalized Log AccuracyMemory Usage (Bytes/Pos)Llamashots=5-shotshots=5-shot2026.03674.28T-Freeshots=5-shotshots=5-shot2026.0366.55.18HATifiedshots=5-shotshots=5-shot2026.0365.75.18