Artificial intelligence accelerator card heterogeneous cluster adaptation method, system and storage medium

Through the expansion of unified standardized adaptation modules and plug-ins, the independent adaptation and optimization of artificial intelligence accelerated card heterogeneous clusters is achieved, solving the problems of high development costs and inability to independently seek optimization, and improving the flexibility and efficiency of training.

CN119294454BActive Publication Date: 2025-03-25ZHEJIANG LAB
2 Cites 0 Cited by

Patent Information

Application Number
CN202411824198.5
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-12-12
Publication Date
2025-03-25
Estimated Expiration
2044-12-12

AI Technical Summary

Technical Problem

In the prior art, the cost of adaptation and development of heterogeneous clusters of artificial intelligence accelerator cards is high and cannot independently find the best, which makes it difficult to manage and collaborate.

Method used

By obtaining a unified standardized adaptation module registered according to the private keyword provided by the training framework, the hardware layer software stack of each heterogeneous acceleration card is connected to the factory mode, and the computing task distribution mechanism of the training framework is expanded through plug-ins, so that the training framework can independently find the target heterogeneous acceleration card.

Benefits of technology

It realizes unified adaptation of various heterogeneous accelerator cards without intrusion and modification of the training framework, reduces development costs, and improves the flexibility and efficiency of large-scale model training.

✦ Generated by Eureka AI based on patent content.
Patent Text Reader

Abstract

The present application relates to a method, system and storage medium for heterogeneous cluster adaptation of artificial intelligence accelerator cards, wherein the method for heterogeneous cluster adaptation of artificial intelligence accelerator cards includes: obtaining a unified standardized adaptation module registered according to a private keyword provided by a training framework; the training framework is a training framework for a heterogeneous cluster model; in the unified standardized adaptation module, accessing the hardware layer software stack of each heterogeneous accelerator card through a factory mode; extending the computing task distribution mechanism of the training framework in the form of a plug-in so that the training framework can autonomously optimize the target heterogeneous accelerator card; conditionally compiling according to the operating environment where the unified standardized adaptation module is located to access the target heterogeneous accelerator card and use it for autonomous optimization of the training framework; verifying the adaptation status of the unified standardized adaptation module and each target heterogeneous accelerator card according to a unified test module, thereby solving the problems of high development cost of heterogeneous cluster adaptation of artificial intelligence accelerator cards and inability to autonomously optimize.
Need to check novelty before this filing date? Find Prior Art