Microsoft and Meta, two leading technology companies, have announced an expanded partnership to support the Llama 2 family of Large Language Models (LLMs) on Azure and Windows platforms. Llama 2 empowers developers and organizations to create innovative AI-powered tools and experiences. With a shared commitment to democratizing AI, Meta has chosen Microsoft as its preferred partner for the commercial release of Llama 2. This collaboration enables Azure customers to quickly deploy Llama 2 models on Azure, while Windows developers can optimize Llama for local execution. Let’s explore the details of this exciting partnership.
Democratizing AI with Llama 2
The partnership between Meta and Microsoft marks a significant step in democratizing AI and its benefits. Llama 2, a family of large language models, is now available on Azure, offering developers the flexibility to fine-tune and deploy models with parameters ranging from 7 billion to 70 billion. Azure, known for supporting frontier and open models, becomes an ideal platform for deploying these advanced models. Additionally, Llama 2 is optimized to run locally on Windows, providing Windows developers with seamless integration and the ability to leverage generative AI experiences within their applications.
A Longstanding Partnership
Meta and Microsoft have a history of collaboration in the field of AI. They previously collaborated to integrate ONNX Runtime with PyTorch on Azure, enabling a superior developer experience. Meta also selected Azure as its strategic cloud provider. This latest announcement further strengthens their partnership and accelerates innovation in the era of AI. With Azure’s purpose-built AI supercomputing platform, organizations can efficiently build, train, and deploy demanding AI workloads. By expanding their collaboration to include Llama 2 models, Microsoft solidifies its position as the world’s supercomputing platform for AI.
Expanding Azure AI Model Catalog and Windows Availability
Llama 2 joins Microsoft’s growing Azure AI model catalog, a hub for pre-built large AI models. This catalog simplifies the deployment and management of Llama 2 models, providing developers and ML professionals with a scalable solution for discovering, evaluating, customizing, and deploying these models. It offers support for model fine-tuning and evaluation, including optimization techniques such as DeepSpeed and ONNX Runtime, which significantly enhance the speed of model fine-tuning.
Windows developers also benefit from the partnership, as Llama 2 models are accessible via the GitHub repository. Leveraging the Windows Subsystem for Linux (WSL) and powerful GPUs, developers can fine-tune LLMs to meet their specific requirements directly on their Windows PCs. This availability positions Windows as the ideal platform for building tailored AI experiences.
Building Responsibly with Azure
Microsoft places a strong emphasis on responsible AI and ensures the safety and reliability of AI applications. Potential risks associated with large language models are mitigated through an iterative and layered approach. Azure AI customers can test Llama 2 models with their sample data to assess performance for their specific use cases. Additionally, prompt engineering and retrieval augmented generation (RAG) techniques can be employed to develop, evaluate, and optimize meta-prompts, resulting in safer and more reliable experiences for end-users.
Microsoft’s collaboration with Meta also integrates Meta’s safety techniques with Azure AI Content Safety. By default, deployments of Llama 2 models in Azure AI come with a layered safety approach, providing additional protection to ensure a safer online experience when using AI applications.