Microsoft and Hugging Face Expand Collaboration for Azure AI Foundry

Microsoft and Hugging Face have expanded their collaboration to integrate over 10,000 open-source models directly into Azure AI Foundry. This partnership allows enterprises to deploy a wide variety of open models on secure, enterprise-grade infrastructure, ensuring that companies maintain control over their technology and private data.

Enterprise-Grade Deployment of Open Models

Azure AI Foundry now features a "Hugging Face Collection" within its Model Catalog, providing immediate access to more than 10,000 models. These models support a broad spectrum of modalities and tasks, including:

  • Text: Generation, feature extraction, fill-mask, and translation.
  • Audio: Automatic speech recognition and audio classification.
  • Images: Classification, segmentation, text-to-image generation, and image-to-text conversion.
  • General: Identifying sentence similarity.

To ensure these models are enterprise-ready, Hugging Face and Microsoft apply strict security criteria. Only models that meet the following requirements are included in the collection:

  • Security Screening: Models must pass security tests using ProtectAI Guardian and JFrog security scanner to identify vulnerabilities.
  • Safe Format: Model weights must be stored in the safetensors format to eliminate potential Pickle vulnerabilities.
  • No Remote Code: Models cannot contain remote code to prevent arbitrary code insertion during runtime.

Additionally, Microsoft and Hugging Face continuously test and patch inference containers for vulnerabilities.

Deployment Workflow in Azure AI Foundry

Users can deploy open models through two primary pathways:

  1. Via Azure AI Foundry: Users select a model from the Hugging Face Collection in the Model Catalog, click "Deploy," and configure the VM, instance count, and deployment parameters.

  2. Via Hugging Face Hub: Users can select the "Deploy on Azure ML" option directly from a model page on the Hugging Face Hub, which redirects them to the deployment options within Azure AI Machine Learning Studio.

Future Roadmap and Integration

Microsoft and Hugging Face have outlined a rolling series of updates to further deepen the integration:

  • Day-0 Releases: New models from top providers will be made available in Azure AI Foundry on the same day they are released on the Hugging Face Hub.
  • Daily Trending Updates: Hugging Face will monitor trending models to enable them on Azure AI Foundry on a daily basis.
  • Expanded Modalities: Future updates will include support for additional modalities and domain-specific tasks, such as video, 3D, time series, and protein folding.
  • AI Agents: The collaboration aims to leverage small, efficient, specialized open models to facilitate the creation of secure AI agents and applications.

Executive Perspectives

"Open source is how AI moves faster—with transparency, choice, and community at its core. This collaboration represents our commitment to that momentum. By combining Hugging Face’s vibrant model ecosystem with Azure’s secure, enterprise-grade infrastructure, we’re giving developers the freedom to pick the best model for the job—and helping organizations innovate safely and at scale."

— Asha Sharma, Corporate Vice President at Microsoft

"We’re enabling companies to take control of their AI destiny, deploying the best open models securely within their Azure account, to build AI applications they can trust and verify."

— Clement Delangue, CEO and cofounder at Hugging Face

Sources