KronQ Optimizes LLM Quantization via Kronecker-Factored Hessian
July 7, 2026
KronQ is a post-training quantization framework that incorporates gradient covariance into the objective function. By using a Kronecker-factored Hessian approximation, it improves reconstruction accuracy over methods that only rely on input activation statistics.
HOW THIS AFFECTS YOU
●
builderYou can achieve higher quantization accuracy for LLM compression using second-order information.
●
researcherThis method improves PTQ by addressing the unequal contribution of output channels through gradient covariance.