fn backend() -> Result<&'static LlamaBackend, NormalizeError>Expand description
llama.cpp’s backend is global to the process and may be initialized exactly once.
Log routing must precede the backend starting, or ggml’s Metal device-init block has already reached stderr. Ordering here is documented rather than enforced: swapping the two lines below still compiles.
LlamaBackend::void_logs is not an alternative. It binds llama’s sink
only, leaving ggml’s untouched.