TY  - RPRT
TI  - Full-Stack FP4: Stable LLM Pretraining with Quantized Projections, Optimizers, and Attention
AU  - Siyu Ding
AU  - Mingchuan Ma
AU  - Jiabo Tong
AU  - Xingrun Xing
AU  - Ziming Wang
AU  - Guoqi Li
PY  - 2026
UR  - https://arxiv.org/abs/2607.04422
ID  - 2607.04422
ER  -