Can llama.cpp support run lora on cpu and base model on gpu?

[LCJD99]

I want to achieve heterogeneous inference on Snapdragon 8 Gen 3 using llama.cpp, that is, to make the LoRA run on the CPU and the base model run on the GPU. However, I haven't found the feasibility so far. I'd like to ask where I should start to implement parallel computing on different devices.