ResearchBenchmarksCommonsense Reasoning on HellaSwag (Accuracy, Drop)Commonsense Reasoning on HellaSwag (Accuracy, Drop)Follow41.1AccuracyBF1640.850440.915240.9841.0448May 25, 2026Evaluation resultsMethodMethodLinksAccuracyDropBF16shot-count=10-shot, Le...shot-count=10-shot, Learning Rate=10^-52026.0541.1—HiF8 QATshot-count=10-shot, Le...shot-count=10-shot, Learning Rate=10^-5, amax_algo=max, history=64, BF16 warmup=500 steps2026.0540.8658