llama.cpp

support loading vocab from fast tokenizer config in convert.py

#3633

Merged