公开(公告)号：GB2604060A

公开(公告)日：2022-08-24

申请号：GB202206096

申请日：2020-09-29

Applicant: IBM

Inventor： SWAGATH VENKATARAMANI ,  VIJAYALAKSHMI SRINIVASAN ,  PHILIP HEIDELBERGER

IPC: G06N3/063

Abstract: Hybrid parallelism techniques where a mix of data and model parallelism techniques are used to split the workload of a layer across an array of processors are disclosed. When configuring the array, the bandwidth of the processors in one direction may be greater than the bandwidth in the other direction. Each layer is characterized according to whether they are more feature heavy or weight heavy. Depending on this characterization, the workload of an NN layer can be assigned to the array using a hybrid parallelism technique rather than using solely the data parallelism technique or solely the model parallelism technique. For example, if an NN layer is more weight heavy than feature heavy, data parallelism is used in the direction with the greater bandwidth (to minimize the negative impact of weight reduction) while model parallelism is used in the direction with the smaller bandwidth.