DEEP LEARNING TASK SCHEDULING METHOD AND SYSTEM AND RELATED APPARATUS

摘要

A deep learning task scheduling method and system and a related apparatus. The method comprises: acquiring a task request for a deep learning task, the task request carrying a deep learning library type and a task type; determining, according to the deep learning library type and the task type, a target task description file template from a plurality of prestored task description file templates; determining, according to the deep learning library type and the task type, an identifier of a target task base mirror image from among identifiers of a plurality of prestored task base mirror images; generating a target task description file according to the target task description file template and the identifier of the target task base mirror image; sending the target task description file to a container scheduler; and the container scheduler selecting, according to the target task description file, the target task base mirror image from among the prestored task base mirror images, and creating at least one container for executing the task request. The method enables an improvement in the compatibility rate of deep learning task scheduling.