Initialises the llama backend and returns a guard that frees it on scope exit.
Loads all available ggml backends (CPU, CUDA, Metal, etc.).
Calls llama_backend_free when it goes out of scope.
See Source File