As AI models become increasingly vital for computing functionality and user experience, the challenge lies in effectively integrating them into smaller devices like personal computers without major resource utilization. Microsoft has developed a solution to this challenge with the introduction of Phi Silica, a small language model (SLM) designed to work with the Neural Processing Units (NPUs) in their new Copilot+ PCs.
Existing AI models consume sizeable computational power which can impact the overall performance of personal devices. Typically, these models we heavily reliant on central processing units (CPU) and graphics processing units (GPU), causing a slowdown in the operation of other tasks and subsequently reducing device efficiency.
Phi Silica stands apart with its 3.3 billion parameters, making it the smallest model in the Phi family. Despite its compactness, Phi Silica delivers impressive performance with a latency of 650 tokens per second and power consumption of only 1.5 Watts. It is optimized to operate efficiently, ensuring CPUs and GPUs aren’t overtaxed, thereby improving the overall device performance.
The Phi Silica API can be accessed via the Windows App SDK, enabling developers to leverage AI capabilities within the Windows ecosystem. This integration brings additional functionalities such as Optical Character Recognition (OCR), Studio Effects, Live Captions, and Recall User Activity APIs to the table, thus inspiring the creation of innovative user experiences. Microsoft also plans to launch Vector Embedding, RAG API, and Text Summarization, further expanding developer capabilities.
Phi Silica stands along with other models within the Phi-3 series, each boasting a varying range of parameters from 3.8 billion to 14 billion. Unique among them, Phi Silica is the first state-of-the-art SLM to ship with Windows, marking a significant stride towards the direct delivery of advanced AI capabilities to users.
Following Microsoft’s announcement of the Copilot+ PC equipped with dedicated AI processors, it is anticipated that the first of these PCs will launch mid-June. They will feature Qualcomm’s Snapdragon X Elite and Plus chips and will be made available in collaboration with major PC manufacturers throughout the summer. Intel is also developing a proprietary Copilot+ PC processor called Lunar Lake, expected to release in 2024’s third quarter.
In summary, Microsoft’s Phi Silica presents a solution to optimally utilize AI models within personal devices. It offers high performance at low resource usage, ensuring smoother operation for other applications. By providing developers with substantial tools without compromise on system performance, it paves the way for an enhanced computing experience and innovative applications. It’s revolutionary design and integration with Copilot+ PCs points towards a transformation in the efficiency and performance of personal computing.
