Hugging Face Model Catalog on Azure Machine Learning
Hugging Face and Microsoft have integrated the Hugging Face Hub Model Catalog directly into Azure Machine Learning Studio. This collaboration allows users to deploy thousands of popular Transformers models from the Hugging Face Hub onto secure, scalable Azure infrastructure using managed endpoints with just a few clicks.
Native Integration for Enterprise Deployment
The Hugging Face Model Catalog is now a native part of the Azure Machine Learning Studio experience. This integration replaces a previous Azure Marketplace managed app solution, which Hugging Face describes as a promising initial step that had limitations that could only be solved through native integration.
This shift to a native experience is designed to address the specific needs of enterprise customers who require administrative control over their infrastructure to meet strict security, compliance, and privacy requirements. By deploying models within their own secure Azure environment, organizations can leverage open-source models while maintaining the necessary governance over their data and data flow.
Solving the Production Inference Challenge
Deploying Transformers to production typically involves a trade-off between simplicity and control. The integration addresses three primary challenges:
- Limited Model Selection: Standard cloud-based AI services often provide a limited set of models that may not support specific task types or allow for deep customization.
- Operational Complexity: While in-house platforms or general cloud ML services offer full control, they often increase the time, complexity, and cost associated with deploying and scaling production-grade inference APIs.
- Security and Compliance: Many organizations cannot use third-party AI services due to mandates requiring models to be deployed on infrastructure under their direct administrative control.
Deployment Workflow in Azure Machine Learning
The process for deploying a model from the Hugging Face Hub to Azure Machine Learning is streamlined into a few steps:
- Access Registry: Open the Hugging Face registry within Azure Machine Learning Studio.
- Browse Catalog: Access the Hugging Face Model Catalog and filter models by task or license.
- Select Model: Choose a specific model tile to open its detailed page.
- Configure Deployment: Select the real-time deployment option.
- Provision Infrastructure: Select an Azure instance type and initiate the deployment.
Once deployed, the model is available as an inference API that can be integrated into applications within minutes.
Availability
The Hugging Face Model Catalog on Azure Machine Learning is available in public preview across all Azure Regions where Azure Machine Learning is currently available.