llama.cpp has updated its support for Android Gpus, can we follow up? 