Topic: build device description

The ggml/llama.cpp project updated its codebase to shorten virtual device naming in CUDA and Metal, improving initialization processes for these platforms. These changes are crucial for enhancing the usability and performance of the project on GPU-based systems.