SYNCHRONOUS TRAINING METHOD AND CLUSTER FOR CONVOLUTIONAL NEURAL NETWORK MODEL, AND READABLE STORAGE MEDIUM

摘要

A synchronous training method and cluster for a convolutional neural network model, and a readable storage medium. The method comprises: when training a convolutional neural network, allocating training samples to GPUs according to a training data volume. The difference between the times consumed by GPUs for training the same convolutional neural network according to the allocated training data volume is reduced, and the training waiting time between the GPUs can be shortened, so that the training efficiency is thus improved. In addition, in the method, the types of GPUs for training the convolutional neural network can be different, so that the number of idle GPUs can be reduced, and hardware costs can also be reduced.