动察 Beating AI Flash News: Huawei has officially deployed a 30B total parameter, 2B activated parameter on-device MoE large model into its new-generation tri-fold smartphone, the Mate XT 2. Huawei claims this is the "industry's first on-device MoE full-modal large model." The model runs on the newly released Kirin 9050 Pro, with the Da Vinci architecture NPU providing on-device computing power.

This 30B/2B on-device model was actually previewed as early as June. At that time, Huawei stated that it was specifically optimized for Kirin chips, utilizing technologies such as expert prediction to reduce memory usage and improve inference throughput, with plans to bring it to Kirin chips and smartphones in the fall.