Mediatek announced on Monday that it has now optimized several of its mobile platforms for Microsoft’s PHI-3.5 Artificial Intelligence (AI) model. The PHI-3.5 series of small language models (SLM), including a mixture of PHI-3.5 Expert (MOE), PHI-3.5 MINI and PHI-3.5 Vision, was released in August. Open source AI models are available on the hug face. These are not typical conversational models, but require the user to enter a specific description to obtain the indication model of the desired output.
MediAtek announced in a blog post that its Dimenisty 9400, Dimenty 9300 and Dimente 8300 chipsets are now optimized for the PHI-3.5 AI model. Therefore, these mobile platforms can efficiently process and infer AI tasks generated for internal devices using Mediatek’s neural processing units (NPUs).
Optimizing chipsets for specific AI models involves tailoring the hardware design, architecture, and operations of the chipset to effectively support the processing power, memory access patterns, and data flow of that particular model. After optimization, the AI model will show reduced latency and power consumption and increase throughput.
Mediatek emphasizes that its processor is optimized not only for Microsoft’s PHI-3.5 MOE, but also for the Phi-3.5 Mini, which provides multilingual support and PHI-3.5 vision with multi-frame image understanding and reasoning.
It is worth noting that PHI-3.5 MOE has 16×3.8 billion parameters. However, when using two experts (typical use cases), only 6.6 billion are active parameters. On the other hand, PHI-35 has 4.2 billion parameters and image encoder, and Phi-3.5 Mini has 3.8 billion parameters.
In terms of performance, Microsoft claims that the PHI-3.5 MOE is superior to both the Gemini 1.5 Flash and GPT-4O Mini AI models on the Squality benchmark, which tests readability and accuracy when summarizing text blocks.
While developers can directly leverage Microsoft Phi-3.5 through the Embrace or Azure AI Model Directory, Mediatek’s Neuropilot SDK Toolkit also provides access to these SLMs. The latter will enable developers to build optimized device applications that can generate AI inference using AI models on the above mobile platforms.
