Microsoft Mu and why Windows will need super processors

Last year, Microsoft released Copilot+ computers that feature a dedicated Neural Processing Unit (NPU) capable of over 40 TOPS. These Copilot+ computers came with Phi-Silica, an SLM built into the device to provide language intelligence capabilities to Microsoft and third-party apps.

Today, Microsoft unveiled Mu, a new language model built into Windows 11. Microsoft's goal was to create an AI agent within the Settings app that can understand a user's natural language queries. It has integrated this AI agent into the existing search box for a seamless user experience.

See more articles from iGuRu.gr when you search for news on Google.

iGuRu.gr preferred source

So Microsoft will use the new Mu model to power the new AI agent feature in the Settings app, which was recently released to Windows Insiders on the Dev Channel with Copilot+ PCs.

Like Phi-Silica, Mu is designed to run on NPUs, delivering over 100 tokens per second while running locally. In the official Microsoft white paper, the company explains how it designed and trained the Mu language model.

Here is an overview of the Mu language model:

• Mu is a 330M encoder-decoder language model.
• The encoder-decoder approach achieved approximately 47% lower first token latency and 4,7 times higher decoding speed compared to a decoder-only model of similar size.
• Mu uses weight distribution on certain elements to reduce the total number of parameters.
• Microsoft trained Mu using NVIDIA A100 GPUs in Azure Machine Learning.
• The Mu is almost comparable in performance to a similarly improved Phi-3.5-mini, despite being one-tenth the size.
• The model is best suited for multi-word queries. So, for short or partial words, the Settings app will continue to display lexical and semantic search results in the search box.

Interested users and those with the hardware can download Windows 26120.3964 build 5058496 (KB11) or later to check out this new AI agent in the Windows Settings app.

newsbot

Although the press releases will range from very select to rare, I said I'd pass...because sometimes the editors hide.