I want to create a complete End to End ML system, where I train several models and deploy the best.

Currently, I have an End-to-End System where I used The Azure ML SDK with Azure DevOps.

Architecture currently in use:

Here all the Training jobs are run using the Azure ML SDK. They are launched within Azure DevOps.

Now, I want to use the Azure ML Pipeline for Model Training and Registering instead of using the Azure ML SDK. Azure ML Training Pipeline:

How can I use the models trained in the Azure ML Studio Pipelines and deploy it using Azure DevOps?

Currently, I pass the metadata generated from the Azure Training jobs to the CD pipeline as artifacts, and then from there, I can deploy the best model.

Now with Azure Pipeline, how can that be done?

CodePudding user response：

We can deploy using batch or real-time endpoint. Created a sample application with multiple models in the single pipeline. We need to get the individual model result into registration and then click on deploy in the model's section to get deployed from the pipelines perspective.

We can create any type of clusters or instances for deployment.

Using any kind of real-time endpoint and deploy the model.