We are thrilled to announce the General Availability of Azure Machine Learning registries today. We launched the public preview at Ignite in October 2022 and have added support for network isolation and data asset since then.
What are Azure Machine Learning registries?
Azure Machine Learning registries serve as organization-level repositories for machine learning assets including models, environments, components and data. They offer a centralized platform for cataloging and operationalizing machine learning models, accommodating the multiple teams, individuals, and environments involved in the machine learning process. By providing a centralized platform for sharing and discovering machine learning models and pipelines, registries encourage enhanced collaboration among data science teams.
They make multi-environment MLOps simple by eliminating the need of manually copying a model (or any asset) to another workspace.
Why should you care?
Azure Machine Learning users can create organization registries in their tenant. They have two main value propositions:
Improved collaboration within your organization
Creating the components, models, environments, and datasets in registries makes them accessible in any workspace within your Organization (Azure Active Directory tenant). Data scientists across diverse teams can utilize registries in different workspaces to efficiently share assets. Azure Machine Learning registries streamline the reuse of work within teams, eliminating the need for manual asset duplication and preserving the lineage of the data. Here are collaboration scenarios supported by registries:
- Share model solving a specific problem to potential reuse by other teams.
- Host a central catalog of models and associated training components that can be used to retrain for finetune the models with different datasets.
- Create training environments and components once and use them to run reproducible training jobs in any workspace.
Enabling multi-environment MLOps
Generally, data science teams use one environment (workspace) to develop and evaluate the model but another environment (workspace) to deploy the models for security and compliance reasons. This process required downloading the model and registering it to the production workspace, which often resulted in broken lineage as it was not easy to keep track the data and parameters used for training. Azure Machine Learning registries makes it easy to share a model from workspace to registry and deploy it from registry to another workspace along with maintaining the lineage with job and datasets used for training the model.
What is new for General availability?
- Support for network isolation
Azure Machine Learning registries can be configured with a private endpoint to meet the enterprise security requirements. Follow this article to learn more.
- Support for sharing data assets
Now you can share datasets as well using Azure Machine Learning registries. The capability to share data assets using registries is currently in public preview. Follow this article to learn more.
- Ability to view assets from the Azure Machine Learning registry in workspace using Azure Machine Learning Studio
If you are using Azure Machine Learning studio, you can view the assets from registry in relevant sections in studio making it easy to discover the assets from registries you have access to.
For example, in the model section of workspace you can view the models from the registries you have access to.
Get started today:
- Learn more about registries here: blog and Azure Machine Learning registries
- Follow these tutorials to create and share assets in registries: Share models, components, and environments across workspaces and Share data across workspaces with registries
- Build ML Pipelines: Create ML pipelines using Designer (UI)
- Watch the Azure Machine Learning Breakout sessions: