Meta Llama 3 support added to Dimensity platforms

In mid-April Meta introduced Llama 3, the next generation of its open-source Large Language Model (LLM) for AI applications. It achieves a major leap over the previous generation due to improvements in pre-training and post-training procedures, alignment, and boosted diversity in model responses, which elevates Generative AI capability to meet the demands of its increasing popularity.

Building upon the framework MediaTek established with Meta Llama 2 last year, we are again working to deliver a combination of compelling hardware and developer-friendly software tools with potent capability. MediaTek will be officially supporting Llama 3 8B on Dimensity platforms, including the Dimensity 9300, Dimensity 8300, and future Gen-AI-enabled platforms. The ability to run Llama 3 on-device provides several benefits, including more responsiveness, better privacy, security, and reliability, and the ability to work in areas with little to no connectivity, and with lower operation cost.

Models like Llama 3 8B are ideal for mobile platforms and edge-AI devices. Llama 3 provides state-of-the-art performance, and excels at language nuances, contextual understanding, and complex tasks like translation and dialogue generation. Llama 3 can handle multi-step tasks effortlessly, while its refined post-training processes significantly lower false refusal rates, improve response alignment, and boost diversity in model answers. Additionally, it drastically elevates capabilities like reasoning, code generation, text understanding, instruction following, and world knowledge.

Developers can embrace Llama 3 through MediaTek's NeuroPilot SDK. NeuroPilot is a toolkit that enables and optimises on-device GenAI inference capabilities across MediaTek's diverse product portfolio, including mobile platforms and Edge-AI capable devices.

The opportunity to provide developers with tools to utilise Llama 3 on the Dimensity platform is a testament to MediaTek's vision to push the limits of what is possible with AI.