ResearchBenchmarksMathematical Reasoning on GSM8k (Accuracy, Calibration Runtime, Delta)Follow78.09AccuracyDense68.074870.674973.27575.8751Mar 17, 2026Evaluation ResultsMethodMethodLinksAccuracyCalibration Runtime (s)DeltaDenseModel=Llama-3.1-8B-Ins...Model=Llama-3.1-8B-Instruct, Quantization=Full Precision2026.0378.09——ZipCalModel=Llama-3.1-8B-Ins...Model=Llama-3.1-8B-Instruct, Calibration Category=Mean, Quantization=GPTQ W4A162026.037010.72.39COLAModel=Llama-3.1-8B-Ins...Model=Llama-3.1-8B-Instruct, Calibration Category=Math, Quantization=GPTQ W4A162026.0368.4636—