Azure Machine Learning - Public Preview for Build
Published date: May 23, 2023
Expand AzureML’s Responsible AI dashboard to support text/image classification scenarios: You can now create and generate Responsible AI dashboards for text and image models from CLI and SDK.
Link Azure Machine Learning workspace to Purview catalog: You can now automatically push relevant metadata of assets, such as models, jobs, and datasets to the Purview catalog.
Hugging Face foundation models in AzureML: You can now build and operationalize open source SOTA models at scale.
AzureML prompt flow: You can now create AI workflows that connect to various language models and data sources.
Managed Feature Store: You can now experiment and ship models faster, increase reliability of your models and reduce your operational costs.
Perform continuous model monitoring: You can now proactively find and resolve issues faster, and continuously improve models for enhanced quality and compliance.
Manage Network Isolation: You can now streamline your network isolation experience, speed up your workspace setup, and free yourself from the hassles of virtual network management.
Track, compare, and visualize your training jobs with our improved experiment tracking tools: You can now quickly investigate, compare, and summarize your experimentation results with various chart types and markdown functionality that you can customize to your desired preference in a new dashboard view.
Model Training with Serverless Compute: You can now focus on your job spec without having to learn about compute and how to set it up.
Import data from external sources for training in AzureML platform: You can now import data from various external sources right from AzureML without dependency on other tools or teams.
Connect compute instances to Visual Studio Code for the Web: You can now continue your work directly in the browser connected to an AzureML compute instance without needing to download or install an application.
Deploy pipelines and components under Batch Endpoints: You can now deploy complex compute graphs under batch endpoints to perform batch inference.