llama.cpp

convert : automatically fall back to HfVocab if tokenizer.model doesn't exist

#5821

Merged