Huawei details how Kirin 9050 runs a 30B-parameter model on device (technode.com)

Huawei has announced that its Kirin 9050 series can efficiently run a complex 30-billion-parameter AI model directly on a device, without necessitating an internet connection. This advancement is made possible by the chip's new Da Vinci neural processing unit, which employs low-bit quantization techniques and optimizes the coordination of memory and computing resources. By activating approximately 2 billion parameters at any given time through a mixture-of-experts approach, the Kirin 9050 is capable of performing sophisticated tasks directly on smartphones. This development is significant for the AI/ML community as it indicates a move towards more powerful, on-device AI processing, reducing dependence on cloud computing and enhancing privacy and response times for users. The Kirin 9050 not only supports impressive machine learning capabilities but also enables innovative features like AI screen reading and eye-tracked page turning, showcasing its potential to enhance user experience. This technology could pave the way for smarter, more interactive mobile applications and is likely to influence future advancements in mobile AI.