DeepSeek announces the open-source optimization of parallel strategies
ChainCatcher news, according to Jinshi reports, Deepseek announced the open-source optimization of parallel strategies on the fourth day of Open Source Week, including DualPipe------a bidirectional pipeline parallel algorithm for computing-communication overlap in V3/R1 training; EPLB------an expert parallel load balancer for V3/R1; and analysis of computing-communication overlap in V3/R1.
Предупреждение о рисках Предупреждение о рисках
Связанные теги
