llama.cpp

4d4d2366 - convert : automatically fall back to HfVocab if tokenizer.model doesn't exist (#5821)