培训即服务 [英] Training as a service
问题描述
大家好,
尝试将我的pytorch培训脚本迁移到Azure机器学习服务。
trying to migrate my pytorch training script to Azure machine learning service.
在我看到的所有样本中培训实验是从笔记本电脑上启动的。
In all the samples I saw the training\experiment is launched from the notebook.
什么是自动启动培训的最佳方式?我有一个数据集生成服务,我想在数据集准备就绪后启动培训。
What would be the best way to automate the kick off of the training? I have a dataset generation service that I want to kick off the training once the dataset is ready.
谢谢,
Emanuel
Emanuel Shalev
Emanuel Shalev
推荐答案
Hello Emanuel,
Hello Emanuel,
您可以尝试将CI / CD用于Azure管道的机器学习项目。这是一个
lab ,它解释了设置可用于训练,比较模型和发布工件。一旦完成,CD也可以将其部署为web服务。
You can try using CI/CD for machine learning project with Azure Pipelines. Here is a lab that explains the setup which can be used to train, compare models and publish the artifacts. Once, this is done the CD can deploy it as a webservice too.
CI:
- 准备python环境
- 获取或创建AML服务的工作区
- 在远程DSVM /本地Python环境中提交培训作业
- 比较不同型号的性能并选择最佳
- 将模型注册到工作区
- 为评分Web服务创建Docker镜像
- 复制并发布工件到发布管道
CD:
- 将图像部署到ACI或AKS
您可以使用预定的运行当您的新数据集发布或在配置中有任何更改时运行管道时。
You can use scheduled runs when your new datasets are published or run the pipeline when there are any changes in the config.
这篇关于培训即服务的文章就介绍到这了,希望我们推荐的答案对大家有所帮助,也希望大家多多支持IT屋!