OPPO Implements First On-Device MoE Architecture

OPPO, a leading global smart device brand, has become the first company to implement the Mixture of Experts (MoE) architecture on-device. This breakthrough enhances AI processing efficiency and expands the potential for advanced on-device AI, setting the stage for future innovations integrating AI with mobile technology.

As AI evolves, performing tasks on-device demands substantial computational power. This can affect performance in devices with limited hardware resources. To address this, OPPO collaborated with chipset providers to implement the MoE architecture. This method dynamically activates specialized sub-models to optimize task performance, improving processing efficiency by 40% in lab tests, and reducing resource needs and energy consumption.

The MoE architecture allows a wider range of devices to perform complex AI tasks, advancing AI adoption across the industry. OPPO is committed to expanding access to AI technology, with over 5,860 patent applications in AI and the 2024 establishment of OPPO's AI Center. These efforts aim to provide high-quality AI experiences globally, ensuring broader accessibility to advanced AI capabilities.