build At Microsoft's Build developer conference on Tuesday, CEO Satya Nadella announced that Redmond's AI Studio is moving to general availability.
“Using Azure AI Studio, we will provide an end-to-end tooling solution for developing and securing the copilot apps you build,” he said in his opening keynote. “This is an end-to-end development environment for building, training, fine-tuning, and responsibly running AI models.”
Redmond is making OpenAI's latest multimodal model, GPT-4o, available in Azure Studio and as an API. The “o” in the model name stands for “omni” and refers to its ability to accept different types of input (audio, text, visual material) and respond in any of those modes.
Windows Business also announced Phi-3-vision, a family of multimodal models optimized for deployment to personal devices. Phi-3-vision is suitable for applications that can answer questions about input images or generate charts.
Nadella also said Microsoft is adding the ability to create custom models to Azure AI. “We recognize that there are very specific use cases and advanced applications that require further customization of these models,” he explained.
Screenshot of Satya Nadella at Microsoft Build 2024 – click to enlarge
The goal of all this is to enable developers to create what Microsoft calls “copilots,” or apps that incorporate generative AI.
Nabila Babar, principal program manager for Microsoft AI Platform, said in a blog post that AI Studio makes it easier to configure and manage AI services and oversee their safety and quality.
“The platform includes familiar tools from Azure Machine Learning, including Azure AI services such as Azure OpenAI Service and Azure AI Search, and prompt flows for guided experiences for rapid prototyping,” said Babar. says Mr. “It supports a code-first SDK and CLI and integrates with the Azure Developer (AZD) CLI and AI Toolkit for Visual Studio Code to provide the scalability you need as your demands grow.”
Azure AI Studio offers over 1,600 models in its model catalog from vendors such as Meta, Mistral, Microsoft, and OpenAI.
Additionally, with the newly announced partnership with HuggingFace, the Azure AI model catalog now also includes 20 models from AI startups, including David Kim's Rhea-72B-v0.5 and MTSAIR's Multiverse-70B. I did.
This agreement will enable Hugging Face Hub users to take advantage of Azure's AMD-based infrastructure, specifically ND MI300X v5 virtual machines with AMD GPUs.
Additionally, Microsoft has devised a “Development Mode” feature for Hugging Face Spaces to create and host models. This allows developers to connect Visual Studio Code to a space and edit code locally or directly in the browser. ®
