Apple Silicon and macOS VMs: Faster LLM Inference with llama.cpp

The percentage is based on available heat signal, not comment count or independent people.