ResearchBenchmarksMathematical Reasoning on GSM8K (Accuracy, Drop)Mathematical Reasoning on GSM8K (Accuracy, Drop)Follow1.59AccuracyBF161.2781.3591.441.521May 25, 2026Evaluation resultsMethodMethodLinksAccuracyDropBF16shot-count=5-shot, Lea...shot-count=5-shot, Learning Rate=10^-52026.051.59—HiF8 QATshot-count=5-shot, Lea...shot-count=5-shot, Learning Rate=10^-5, amax_algo=max, history=64, BF16 warmup=500 steps2026.051.29—