Microsoft Expands Azure AI with Two New Phi-3 Models

Microsoft has announced multiple updates to its Azure AI platform, notably expanding the Phi-3 family of small language models (SLMs). The company introduced two new additions to the Phi-3 lineup: Phi-3.5-MoE and Phi-3.5-mini, aimed at boosting efficiency and accuracy.

Notable features of these new models include their extensive multilingual support, now surpassing 20 languages.

New Models Breakdown:

- Phi-3.5-MoE:

This Mixture of Experts model comprises 42 billion parameters by fusing 16 smaller models. This integration marries the speed and efficiency of smaller models with the high-quality output of larger ones.

- Phi-3.5-mini:

Smaller, at 3.8 billion parameters, this model extends its multilingual competence to various languages including Arabic, Chinese, Czech, and more. It's a substantial update over the Phi-3-mini model, incorporating user feedback since its release two months ago.

Beyond these, Microsoft has launched additional tools and services in Azure AI, aimed at simplifying insights extraction from unstructured data. Additionally, the AI21 Jamba 1.5 Large and Jamba 1.5 models will be introduced on Azure AI as a service, providing enhanced long-context processing abilities.

Further announcements include the general availability of the VS Code extension for Azure Machine Learning and the broader release of the Conversational PII Detection Service in Azure AI Language.

Corporate VP of the Azure AI Platform, Eric Boyd, emphasized Microsoft's ongoing investment in advancing Azure AI innovations to enable customers to develop, deploy, and scale AI solutions effectively and securely.

Earlier, SSP wrote that Robotaxi company Waymo boasted 100,000 successful weekly rides.