At the Microsoft Build Conference today, Satya Nadella expanded its collaboration with Haggingface to make it easier to deploy the wide variety of open models to Azure Secure Infrastructure.
Heading to Azure AI Foundry today, you’ll find a much expanded collection of over 10,000 hugged face models that can be deployed in powered AI applications with just a few clicks and using text, audio and images. And we’re just starting out!
It’s time to build – Extended collaboration
Two years ago, Microsoft and Hugging Face began a collaboration to make open models easier to access on Azure. At the time, there were 200,000 open models at the hub.
With nearly 2 million open models now covering a wide range of tasks, modalities, domains and languages, it’s time to collaborate to the next level. Announced today, the new collaboration creates a mutual success framework that will greatly expand how Azure customers can benefit from hugging their faces.
“Open source is a way for AI to move faster with transparency, choice and community. This collaboration represents our commitment to that momentum. Combining the vibrant model ecosystem of attacks with Azure’s secure enterprise infrastructure, we can choose the best model for developers.
-Asha Sharma, Corporate Vice President, Microsoft
To make Azure customers easier to access more open models, businesses can build AI applications and agents with complete control over technology and data for secure deployment alongside corporate private data.
Companies manage AI Destiny and enable them to safely deploy the best open models within their Azure accounts and build AI applications that can be trusted and verified.
– Clement Dellange, CEO and co-founder of Hug Face
How to use hug faces in Azure AI Foundry
Go to Azure AI Foundry and select your model catalog. Here you can find over 10,000 models under the collection of embracing faces.
The collection’s models include the most popular trend models for faces with a wide range of tasks for using text, audio, and images, including text extraction, fill masks, translation, sentence similarity, image classification, image generation, image generation, text to text to text, conversion to text, automatic speech recognition, speech classification, and more.
We only showcase models to create a face collection to hug in Azure AI Foundry Enterprise-Ready.
Pass face security tests and screen for vulnerabilities including Protectai Guardian and JFrog Security Scanner with model weights stored in the safetenser format, avoiding potential pickle vulnerabilities without remote code, and avoiding arbitrary code insertion at runtime.
Additionally, Microsoft and Hugging Face continually test inference containers for vulnerabilities that maintain and patch them as needed.
Now, let’s say you want to deploy the popular Microsoft PHI-4 Reasoning Plus Open Model.
First, select the model for the Azure AI Foundry Hugging Face Collection and click the (Expand) button. This form allows you to select VMs, instance counts, deployment parameters and start the deployment process with just another click.
Now, if you want a browsing model for your hub, you can also start from the model page. The Azure ML Deployment option moves to the same deployment option within Azure AI Machine Learning Studio.
More embracing faces to come to Azure AI Foundry
I’m really excited by all the new embracing face models and modalities available directly within Azure AI Foundry, but it won’t stop there!
In the coming weeks and months, you can expect the latest thunder lightning:
Day -0 Release – Embracing Face will work with Microsoft to continuously monitor trending models on Azure AI Foundry and hug ai badality with daily modalities on the same day as they land on Trending Models Updates, embracing new models from top model providers available on Azure AI Foundry. 3D, time series, protein, etc. Agents and Tools – Small, efficient and professional open models make them ideal for building powerful but secure AI agents and applications
If you’re using Azure, it’s time to build it with an open model!
