Exciting New Developments in Generative AI
Today, we’re thrilled to announce that Microsoft and Hugging Face are taking our partnership to the next level! Building on our previous collaborations, we’ve deepened our alliance to make AI more accessible and powerful through new product updates and integrations.
Key Areas of Collaboration
At Microsoft Build 2024, we’re excited to share four major advancements:
1. New Hugging Face Models in Azure AI
- 20 New Models: We’re adding models like Rhea-72B-v0.5 and Multiverse-70B to the Azure AI Model Catalog. These models expand the variety of choices for our customers.
- Enhanced User Experience: These models offer advanced features like Text Generation Inference (TGI) and Text Embedding Inference (TEI) for efficient deployment and serving.
2. Upgraded Azure Infrastructure with AMD
- ND MI300X v5: In partnership with AMD, we’re enhancing Azure AI infrastructure with AMD Instinct™ MI300X GPUs.
- Seamless Integration: Hugging Face models can now leverage AMD’s ROCm™ ecosystem on Azure, allowing users to run over 10,000 pre-trained models without needing to rewrite their applications.
3. Phi-3-mini on HuggingChat
- Broadened Reach: The Phi-3-mini model is now available on the HuggingChat playground, offering a great starting point for developers to experiment with small models.
4. Hugging Face Spaces with VS Code
- New Dev Mode: This feature streamlines the development process, allowing developers to edit their code directly in Visual Studio Code (VS Code) and see changes in real-time without using git.
Azure AI Model Catalog: 20 New Hugging Face Models
The Azure AI Model Catalog is the go-to hub for discovering, deploying, and fine-tuning a wide selection of generative AI models. Alongside other providers like Cohere, Meta, and Mistral, the Hugging Face collection now includes:
- Popular Models: Smaug-72B-v0.1 from Abacus AI and Fugaku-LLM-13B from Fugaku-LLM.
- Enhanced Features: Advanced runtime optimizations like Flash Attention and Paged Attention for high performance and efficient deployment.
Partnering on Azure’s Latest AI Infrastructure with AMD
At Microsoft Build, we introduced the GA of Azure’s new AI infrastructure, the ND MI300X v5, powered by AMD Instinct™ MI300X GPUs. Hugging Face is one of the first to utilize this infrastructure, achieving impressive performance benchmarks. This collaboration allows Hugging Face users to fully leverage AMD’s ROCm™ open software ecosystem on Azure.
Phi-3-mini on HuggingChat
The Phi-3-mini model is now part of the HuggingChat playground, making it easier for developers to start experimenting with small models. This integration brings together the power of open platforms on Hugging Face with enterprise-grade offerings on Azure AI.
Visual Studio Code Integration with Hugging Face Spaces
Our new “Dev Mode” feature for Hugging Face Spaces simplifies the development process for AI developers. Here’s what you can do:
- Real-Time Editing: Edit your Space directly in VS Code and see changes immediately.
- User-Friendly: No need to push changes using git. Just refresh your Space to see updates.
- Efficient Development: Commit and merge your changes once you’re satisfied, making the development process more streamlined.
If you’re interested in joining the private preview program for these features, you can sign up here. To start using Hugging Face models on Azure AI, check out these Python samples.
By combining Microsoft’s powerful cloud infrastructure with Hugging Face’s leading AI models, we’re paving the way for innovative and scalable AI solutions. Stay tuned for more updates and happy coding!
