Wang Merlyn

Duuuuuuude

为什么样本方差的分母是n-1

Finexus

MagBasics

ReSkin

MagX Review and Memo

Consistent Hashing

A Unified Architecture for Accelerating Distributed DNN Training in Heteogeneous GPU/CPU Clusters

Unified All Reduce and PS

What Is Bisection Bandwidth

Nccl Allreduce