Constant GPU_LAYERS 

const GPU_LAYERS: u32 = 999;

Offload everything; llama.cpp clamps this to the layers that exist and silently does nothing without an accelerated backend compiled in.