This article is compiled from the speech delivered by ByteDance infrastructure R&D engineer Shan Jixi at the ArchSummit Global Architect Summit. It mainly introduces the three stages and key nodes of ByteDance offline training development, as well as two very important ones in cloud-native offline training. Part – computing scheduling and data orchestration, and finally will combine the first two parts to share 4 cases of byte beating in practice.

#Daily #blog #100w #cores #450w #cores #ByteDances #ultralargescale #cloudnative #offline #training #practice