List your company

Matrix Multiplication Latency on Llama-3 8B

152.69Kernel-level latency (µs)CodeGEMM
132.9772266.0386399.1532.1614Dec 19, 2025

Evaluation results

MethodLinks
CodeGEMMconfig=m1v4g1282025.12152.69
LUTGEMMquantization=q2-g1282025.12160.1
QuIP#config=e8p2025.12162.63
CodeGEMMconfig=m2v8g1282025.12172.18
QTIPconfig=r22025.12189.94
AQLMconfig=2x82025.12250.12
cuBLASprecision=fp162025.12332.45
AQLMconfig=1x162025.12645.51