llama.cpp

f7f468a9 - gguf-py : fail fast on nonsensical special token IDs (#4489)