With the recent launch of Microsoft Phi-3.5, MediaTek has announced that it has optimized the new Small Language Model (SLM) for its generative AI-capable NPUs, which includes the Dimensity 9400, 9300, 8300 and future generative AI-enabled chipsets.
As part of the launch, Microsoft also introduced Phi-3.5-mini (featuring enhanced multi-lingual support) and Phi-3.5-vision (improved multi-frame image understanding and reasoning). These latest additions are designed to be both highly capable and cost-effective, surpassing larger models in various benchmarks, including language, reasoning, coding, and math, while excelling in performance, power efficiency, latency, and safety.
This follows MediaTek’s continued support for the Microsoft SLM, optimized for on-device tasks, further demonstrating MediaTek’s dedication to advancing edge AI on Dimensity platforms. Recently, Sameer Sharma, AVP and GM of the AI IoT BU (Americas & Europe) at MediaTek, presented a session at Microsoft Ignite on how to optimize Phi and SLM performance. The presentation and slides are available to download.
Developers can leverage Microsoft Phi-3.5 through the Azure AI Model Catalog as well as MediaTek's NeuroPilot SDK, a toolkit that enables and optimizes on-device Gen-AI inference across MediaTek’s broad product range, particularly mobile platforms with built-in NPUs.
Read more about the differences and capabilities of Microsoft Phi-3.5.
