AI RESEARCH
Pretraining large language models with MXFP4
arXiv CS.AI
•
ArXi:2605.09825v1 Announce Type: cross Why does full-pipeline FP4