智能算法模型部署平台设计

Design of Intelligent Algorithm Model Deployment Platform

针对智能算法模型差异大导致的部署效率低的问题,基于Docker设计了一种轻量级的智能算法模型部署平台。通过将模型打包为Docker镜像,利用镜像仓库托管镜像,利用数据库管理模型部署信息,基于RemoteAPI远程控制Docker节点部署容器,实现系统中所有智能算法模型的整体部署和统一管理。以BERT中文预训练模型部署为例,对平台的智能算法模型部署能力进行了实验验证。结果表明,平台满足智能服务的部署要求,且极大地提高了部署效率。

Aiming at the problem of low deployment efficiency caused by large differences in intelligent algorithm models,a lightweight intelligent algorithm model deployment platform is designed based on Docker.By packaging the model as a docker image,hosting the image using the image warehouse,managing the model deployment information using the database,and remotely controlling the Docker node to run container through its Remote API,the overall deployment and unified management of all the intelligent algorithm models in the system are realized.Taking the deployment of Bert Chinese pre-trained model as an example,the deployment ability of the platform is experimentally verified.The results show that the platform meets the deployment requirements of intelligent services and greatly improves the deployment efficiency.