Microsoft has unveiled Mu, a compact language model optimized to run on Neural Processing Units (NPUs) like those found in Windows Settings for Copilot+ PCs. This development enhances users' ability to control system settings through natural language while minimizing cloud reliance. Boasting 330 million parameters, Mu achieves faster inference and lower memory usage compared to traditional models. Key performance metrics indicate a significant reduction in latency on Qualcomm's Hexagon NPU, and the model is trained on extensive examples for real-world applicability. Currently, it's accessible to Windows Insiders, showcasing its potential for interactive user experiences with immediate responses.
Microsoft has aimed to enhance user experience with Mu, a local language model, facilitating faster, real-time system setting adjustments without leaning heavily on cloud computing.
The integration of rotary positional embeddings and grouped-query attention within Mu contributes significantly to its reduction in latency, enabling users to interact flexibly with device settings.
Collection
[
|
...
]