Add support for llama.cpp

[yairallouche]

Add support for llama.cpp as an inference backend for Granite Switch models.