ResearchBenchmarksLarge Language Model Evaluation on LLaMA-3 8BFollow6.13PPLbaseline-18,833.6248108,334.7201235,503.065362,671.4099Mar 9, 2026Evaluation ResultsMethodMethodLinksPPLZero-Shot ScoreMMLU Scorebaseline#Bits=FP16#Bits=FP162026.036.1367.1662.13SERQ#Bits=W4A4#Bits=W4A42026.037.7562.4153.8SpinQuant#Bits=W4A4#Bits=W4A42026.038.2661.7549.93QuaRot#Bits=W4A4#Bits=W4A42026.038.4159.1247.29SmoothQ(g128)#Bits=W4A4#Bits=W4A42026.0317.2648.9729.3SmoothQ#Bits=W4A4#Bits=W4A42026.03471,00036.3425.17