基于容器云的高校人工智能实验平台构建

Construction of Experimental Platform for University Artificial Intelligence Based on Container Cloud

针对人工智能实验教学存在的算力资源不足、实验环境搭建复杂等问题,本文基于CPU+GPU协同计算架构,利用Containerd和Kubernetes构建底层容器云基础设施,并在其上部署可定制化的实验环境镜像,设计和实现功能模块,从而形成人工智能实验平台。本文重点阐述了Kubernetes集群搭建、Jupyter Hub部署以及实验镜像制作方法等,可为多层次实验案例灵活提供Pytorch、TensorFlow等不同的实验运行环境。通过两个深度学习实验,证明了本文构建的平台具备良好的实用性和可行性,是一种有效的人工智能实验教学解决方案。

To address the issues of insufficient computing resources and complex experimental environment setup in artificial intelligence(AI)experimental teaching,a CPU+GPU collaborative computing architecture was employed to construct a container-based cloud infrastructure using Containerd and Kubernetes.Customizable experimental environment mirrors are deployed on top of the infrastructure,and functional modules are designed and implemented to form an AI experimental platform.The emphasis was placed on explaining the construction of Kubernetes clusters,the deployment of Jupyter hub,and the method for creating experimental environment mirrors,which can flexibly provide different experimental running environments such as Python and TensorFlow for multilevel experimental cases.The platform's practicality and feasibility are demonstrated through two deep learning experiments,making it an effective solution for AI experimental teaching.