AI RESEARCH

Pretraining large language models with MXFP4

arXiv CS.AI

ArXi:2605.09825v1 Announce Type: cross Why does full-pipeline FP4