Microsoft and Hugging Face: A Strategic Collaboration for Open AI
At the Microsoft Build conference, a significant announcement was made regarding the strategic collaboration between Microsoft and Hugging Face. This partnership aims to streamline the use of open models and open-source AI across various platforms, particularly focusing on enhancing the experiences for AI builders. The collaboration encompasses a broad range of initiatives aimed at open science, open source, cloud applications, and developer tools, which are pivotal for the future of AI development.
A Collaboration for Cloud AI Builders
One of the most exciting aspects of this partnership is the introduction of enhanced experiences for building AI with open models on Microsoft Azure. The Hugging Face Collection in the Azure Model Catalog has seen immense popularity since its inception, being utilized by hundreds of Azure AI customers. With over a thousand open models already available, the collection is now expanding to include some of the most sought-after large language models (LLMs).
Expanded HF Collection in Azure Model Catalog
The Hugging Face Collection now features new models such as Llama 3 from Meta, Mistral 7B from Mistral AI, Command R Plus from Cohere, and Qwen 1.5 110B from Qwen. These additions are designed for direct, one-click deployment from Azure AI Studio, simplifying the process for users. To deploy these models within an Azure account, users can easily start from the model card on the Hugging Face Hub by selecting the “Deploy on Azure” option. Alternatively, models can be located directly in Azure AI Studio within the Hugging Face Collection, providing a user-friendly interface for deployment.
Build AI with the New AMD MI300X on Azure
The collaboration also brings forth new Azure ND MI300X virtual machines, powered by AMD Instinct MI300 GPUs. This innovation promises remarkable performance and cost efficiency for Hugging Face models. Through joint efforts with AMD and Microsoft, the integration of Hugging Face’s open-source library, Optimum-AMD, has been optimized for superior performance, ensuring that users can leverage the latest technology for their AI projects.
A Collaboration for Open Science
Microsoft has made significant strides in releasing popular open models on Hugging Face, with nearly 300 models now available under the Microsoft organization on the Hugging Face Hub. This includes the recently launched Phi-3 family of models, which are licensed under MIT and demonstrate exceptional performance, even with fewer parameters compared to larger models. For instance, the Phi-3 mini boasts only 3.8 billion parameters but outperforms many larger models, making it an excellent choice for on-device applications.
More Read
To showcase these capabilities, Hugging Face has integrated the Phi-3 mini model into Hugging Chat, a free consumer application that allows users to interact with cutting-edge open models and create personalized assistants.
A Collaboration for Open Source
The partnership between Hugging Face and Microsoft has spanned three years, focusing on making it easier to export and utilize Hugging Face models with ONNX Runtime through the open-source Optimum library. Recent developments include enhancements for local inference using WebGPU, integrating Transformers.js and ONNX Runtime Web. This collaboration has resulted in impressive performance metrics, as demonstrated by the Phi-3 model generating over 70 tokens per second in a browser environment.
A Collaboration for Developers
In addition to the advancements for cloud AI and open science, a new integration has been unveiled to enhance the developer experience. Hugging Face Spaces, which has seen over 500,000 AI demo applications created by the community, now features a new Spaces Dev Mode. This mode allows developers to effortlessly connect their Space to local VS Code or utilize a web-hosted VS Code environment, making it simpler than ever to build AI applications.
Currently, the Spaces Dev Mode is in beta and accessible to PRO subscribers, offering a seamless experience for developers looking to harness the power of AI.
What’s Next
The ongoing collaboration between Microsoft and Hugging Face is set to enhance the accessibility of open-source AI tools across various platforms. As more models are added to the Azure AI Studio model catalog and new features are introduced, the partnership promises to keep evolving, providing users with robust tools and resources to advance their AI projects. Stay tuned for further updates as this exciting collaboration unfolds.
Inspired by: Source
