Invention Grant
- Patent Title: Hyper-square implementation of tree AllReduce algorithm for distributed parallel deep learning
-
Application No.: US16777731Application Date: 2020-01-30
-
Publication No.: US11620502B2Publication Date: 2023-04-04
- Inventor: Liang Han , Yang Jiao
- Applicant: ALIBABA GROUP HOLDING LIMITED
- Applicant Address: KY Grand Cayman
- Assignee: ALIBABA GROUP HOLDING LIMITED
- Current Assignee: ALIBABA GROUP HOLDING LIMITED
- Current Assignee Address: KY Grand Cayman
- Agency: Finnegan, Henderson, Farabow, Garrett & Dunner, LLP
- Main IPC: H04L67/00
- IPC: H04L67/00 ; H04L69/166 ; G06N3/063 ; G06N3/08 ; G06N3/04

Abstract:
The present disclosure provides a method for syncing data of a computing task across a plurality of groups of computing nodes. Each group including a set of computing nodes A-D, a set of intra-group interconnects that communicatively couple computing node A with computing nodes B and C and computing node D with computing nodes B and C, and a set of inter-group interconnects that communicatively couple each of computing nodes A-D with corresponding computing nodes A-D in each of a plurality of neighboring groups. The method comprises syncing data at a computing node of the plurality of groups of computing nodes using inter-group interconnects and intra-group interconnects along four different directions relative to the node; and broadcasting synced data from the node to the plurality of groups of computing nodes using inter-group interconnects and intra-group interconnects along four different directions relative to the node.
Public/Granted literature
- US20210241078A1 HYPER-SQUARE IMPLEMENTATION OF TREE ALLREDUCE ALGORITHM FOR DISTRIBUTED PARALLEL DEEP LEARNING Public/Granted day:2021-08-05
Information query