Bridging the Gap Between Promise and Performance for Microscaling FP4 Quantization

Vage Egiazarian, Roberto Castro, Denis Kuznedelev, Andrei Panferov, Eldar Kurtic, Shubhra Pandit, Alexandre Marques, Mark Kurtz, Saleh Ashkboos, Torsten Hoefler, Dan Alistarh
September, 2025
PDF Cite Code Paper arXiv

Abstract

A study of MXFP4 and NVFP4 inference, introducing Micro-Rotated-GPTQ and GPU kernels tailored to the constraints of microscaling formats.

Type
Conference paper
Publication
ICLR 2026
Andrei Panferov
Andrei Panferov
PhD Candidate in Computer Science

I work on efficient large language models, with a focus on low-precision training, quantization, and scaling laws.