Deep learning training platform

Abstract

The invention discloses a deep learning training platform. The platform comprises: a plurality of software instances; deep learning training clusters; the service management end that is configured to receive a user request and distribute the user request to at least one of the plurality of software instances for processing; the cluster adapter that is configured to receive the processing results of the plurality of software instances and submit the processing results to the deep learning training cluster for training, wherein the service management end is configured to allocate a user request to other software instances during upgrading in response to confirmation of upgrading of a plurality of software instances in the plurality of software instances. The problem that when a user uses a deep learning training platform, the deep learning platform needs to interrupt service under the condition of software upgrading is solved. And software upgrading is realized under the conditions of not influencing the use of the user and not interrupting the service. In addition, the upgrading effectiveness can be verified in the upgrading process, and rollback of software upgrading operation can be carried out when problems are found.

Description

technical field [0001] The invention relates to the technical field of deep learning, in particular to a deep learning training platform. Background technique [0002] At present, artificial intelligence technology has developed rapidly, and various industries are rapidly undergoing intelligent transformation. As a representative of artificial intelligence, deep learning technology has produced a large number of constantly changing and rapidly developing needs for deep learning training in various fields. In the face of constantly innovative technologies and changing user needs, how to upgrade the deep learning platform has become a problem to be solved in the industry. Deep learning is a computing-intensive training task. If the service is interrupted during the software upgrade process, the training effect and user experience will be greatly affected, which in turn will affect the user's income from using the product. In the deep learning training platform, in order to e...

AI Technical Summary

Problems solved by technology

[0004] In view of this, the purpose of the embodiment of the present invention is to propose a deep learning training platform, which solves the problem that the deep learning platform needs to perform software upgrades and interrupt services when users are using the deep learning training platform, so as to achieve , Without interrupting the service, the software upgrade can be realized, and the effectiveness of the upgrade can be verified during the upgrade process, and the software upgrade operation can be rolled back if any problem is found

Images