Microsoft has unveiled Mu, its latest small language model (SML) designed specifically for Copilot+ PCs. This innovative model translates natural language queries into function calls within the Settings application of Windows 11, marking a significant step in the integration of AI into everyday computing tasks.
Windows Intelligence In Your Inbox
According to Vivek Pradeep, Microsoft’s vice president, Mu is engineered to handle complex input-output relationships efficiently. “Mu addresses scenarios that require inferring complex input-output relationships and has been designed to operate efficiently, delivering high performance while running locally,” he states. This model is fully integrated with the Neural Processing Unit (NPU), achieving an impressive response rate of over 100 tokens per second, which is essential for the user experience in the Settings context.
Pradeep elaborates that Mu’s development was informed by Phi Silica, Microsoft’s well-regarded SML. Mu is described as “micro-sized,” optimized for both performance and efficiency, and tailored for specific tasks. It employs a transformer encoder–decoder architecture that effectively separates input tokens from output tokens, thereby minimizing computational and memory demands.
While the technical intricacies may be complex, the essence of Mu lies in its evolution from Microsoft’s Phi models. The result is a base model that can be fine-tuned for specific tasks, offering performance comparable to Phi but at only one-tenth of its size.
The practical implications of Mu are what will resonate most with users. “Our goal was to create an AI-powered agent within Settings that understands natural language and changes relevant undoable settings seamlessly,” Pradeep explains. The integration of this agent into the existing search box aims to provide a smooth user experience, necessitating ultra-low latency for a multitude of possible settings. After evaluating various models, it was determined that Phi LoRA met precision goals but fell short in latency, leading to the decision to refine Mu for optimal performance within Windows Settings.
