Recently, Meta and Microsoft announced that the Llama 2 family of LLMs would be supported on Azure and Windows. Llama 2 aims to help businesses and individuals create applications and experiences powered by generative AI. 

Azure is the platform for the most commonly adopted frontier and open models. Users can safely tune and deploy Llama 2 models with 7B, 13B, and 70B parameters. Furthermore, Llama will be streamlined for native Windows operation. With Llama, developers on Windows can offer generative AI experiences to their apps in a streamlined manner by targeting the DirectML execution provider via the ONNX Runtime.

ONNX runtime

From working together to integrate ONNX Runtime with PyTorch to create a wonderful developer experience for PyTorch on Azure to Meta's selection of Azure as a strategic cloud provider, Meta and Microsoft have long-time AI collaborators. This news strengthens Microsoft's position as the world's preeminent AI supercomputing platform and expands on the relationship to speed up innovation in the AI era.

In recent months, AI innovation has unleashed a surge of transformational potential, capturing the collective imagination with the promise of changing industries and employment. The facilities, hardware, and software of Azure's AI supercomputing platform were all developed to assist the world's top AI companies in developing, training, and deploying the most complex AI applications. Developers may take advantage of Azure AI's robust infrastructure for model training, fine-tuning, and inference, especially the capabilities that support AI safety, now that the Llama 2 models are available with Azure AI.

World-class tools

Windows Subsystem for Linux (WSL), Windows terminal, Microsoft Visual Studio, and VS Code are all world-class tools that are made even better by adding the Llama 2 models. It makes Windows the best place for developers to create AI experiences uniquely suited to their customers' needs.

Tech companies have been racing to build models to topple OpenAI's popular AI chatbot ChatGPT since its release in November last year. Meta has been moving at a snail's pace. Meta released the initial, limited version of LLaMA in February, simultaneously as Microsoft and Google revealed their artificial intelligence chatbots. The company is behind the curve but thinks publishing LLaMA 2 and making it open-source will help it catch up. The Azure AI model catalogue has been updated with the addition of Llama 2. Pre-built, large-scale AI models may now be readily discovered, evaluated, customised, and deployed thanks to the model catalogue, which is presently in public preview.

Using the catalogue, users may put Llama 2 into production without worrying about the many infrastructural dependencies it requires. Turnkey support for model fine-tuning and evaluation is provided, and it uses potent optimisation techniques like DeepSpeed and ONNX Runtime to speed up the process.

Llama 2

Llama 2, available on GitHub Repo, will make it simple for Windows developers to create innovative user experiences. Using Windows Subsystem for Linux and powerful graphics processing units (GPUs), programmers working on Windows computers can customise LLMs to their specifications.

Microsoft's AI strategy and partnership model is founded on the principle of responsible AI. Whether clients are constructing their models from scratch or utilising pre-built and configurable models from Microsoft, Meta, OpenAI, and the open-source ecosystem, the researchers have spent years investing extensively to make Azure the location for ethical, cutting-edge AI innovation.

Conclusion

Researchers at Microsoft adopt a layered, iterative strategy incorporating testing and measurement to reduce the hazards associated with using large language models. Customers of Microsoft's Azure AI service can put Llama 2 to the test with their data to evaluate how well it works in their specific scenario. Then, customers can employ prompt engineering and retrieval augmented generation (RAG) methods to create, test, and improve their app's meta-prompts, thereby providing users with a more secure and stable experience.

Azure's AI Content Safety service is just one example of how extra safeguards can make using AI-powered apps online less risky. Due to the cooperation with Meta, Azure AI now includes a layered safety mechanism for all Llama 2 model deployments by default. Furthermore, a significant step towards a responsible, open AI strategy was taken today with the addition of Llama 2 to the model catalogue and the partnership with Meta.

Image source: Unsplash