Deep learning training execution method and related device

Abstract

The invention discloses a deep learning training execution method. The method comprises the steps of obtaining a deep learning training task by a computing node; importing a data set corresponding to the deep learning training task into a data cache resource from a data storage node; and executing the deep learning training task based on the data set in the data cache resource. A data set corresponding to an obtained deep learning task is imported into a data cache resource from a data storage node, and then a deep learning training task is executed based on the data set in the data cache resource instead of executing the task based on the data set in the data storage node. The data storage node is prevented from slowing down the execution efficiency of the training task, meanwhile, the data reading and writing efficiency is improved through the data cache resource, and the time delay is reduced. The invention also discloses a deep learning training execution device, a server and a computer readable storage medium, which have the above beneficial effects.

Description

technical field [0001] The present application relates to the field of computer technology, and in particular to a deep learning training execution method, execution device, server, and computer-readable storage medium. Background technique [0002] With the continuous development of artificial intelligence technology, the scale and number of model training tasks related to it are also increasing. In order to adapt to the current training tasks, multiple storage systems are used to form a storage system to store various data of training tasks in order to cope with massive training data. [0003] In related technologies, generally multiple storage servers are used to build an independent storage system to solve the problem of data set storage. However, independent storage servers generally use relatively cheap hardware, and the efficiency of data reading and writing is not high. The storage system is generally independent of the computing nodes of the deep learning training...

AI Technical Summary

Problems solved by technology

However, independent storage servers generally use relatively cheap hardware, and the efficiency of data reading and writing is not high

The storage system is generally independent of the computing nodes of the deep learning training platform. The computing program needs to go through a relatively long network path to read and write data, which also limits the efficiency of data usage.

Moreover, under large-scale training, there will be instantaneous read and write pressure on the storage system that far exceeds the carrying capacity of the storage system, which will affect the training effect of deep learning tasks.

It can be seen that the data efficiency of the data cache acceleration process is low, resulting in a large delay

Images