HAPMoE: Heterogeneity-Aware Automatic Parallelism Planning for Mixture-of-Experts Models Training
As model sizes continue to scale, distributed training has become inevitable. Automatic parallelization techniques can derive efficient training parallelism strategies at low cost while achieving superior performance. The difficulty of this problem is jointly determined by the complexity of the model and the underlying...
Meng-Yuan Fan, Pei-Zhuang Cong, Zi-Xiao Huang et al.
· 0 citations