Here is the English translation:

According to Wall Street CN, the parameter scale of on-device mobile models is rapidly expanding. On September 22, Qualcomm, together with StepFun and others, fully deployed a 30B MoE model on the sixth-generation Snapdragon 8 Elite Extreme Gen 6 mobile platform, achieving local on-device inference. Earlier, on September 16, vivo announced preliminary R&D of a BlueLM 30B MoE on-device model. On October 13, ModelBest released the open-source 35B-A3B model MiniCPM-V 4.7 and ran it on an actual iPhone 17 Pro Max. This marks the beginning of models with tens of billions of parameters being deployed on mobile devices, made possible largely by technologies such as MoE (Mixture of Experts), which allow the total parameter scale of a model to be partially decoupled from the computational load of a single inference run.