Model parameter fusion method and apparatus

A model parameter fusion apparatus, applied to a machine learning system, wherein the machine learning system comprises M model parameter fusion apparatuses and the apparatus comprises:

• at least one processor;
• and
• a non-transitory computer-readable storage medium coupled to the at least one processor and storing programming instructions for execution by the at least one processor, wherein the programming instructions instruct the at least one processor to:
• receiving fusion indication information from a fusion controller;
• divide a model parameter of the model parameter fusion apparatus into N blocks according to the fusion indication information,
• wherein N is a quantity of model parameter fusion apparatuses that participate in a fusion and that are of the M model parameter fusion apparatuses, and
• wherein an ith block of the N blocks resulting from the division of the model parameter is an ith model parameter block, and 1≤i≤N≤M;
• receive ith model parameter blocks respectively sent by model parameter fusion apparatuses of the N model parameter fusion apparatuses other than the model parameter fusion apparatus;
• fuse the ith model parameter block of the model parameter fusion apparatus and the ith model parameter blocks respectively sent by the other model parameter fusion apparatuses, so as to obtain an ith general model parameter block;
• and
• distribute the ith general model parameter block to the other model parameter fusion apparatuses of the N model parameter fusion apparatuses than the model parameter fusion apparatus.