1 paper
Shuang Ma, Yuyi Li, Yihan Zhang +12
Training and serving large language models (LLMs) rely heavily on high-performance GPU communication, yet implementing efficient GPU communication primitives requires deep expertis…