llama.cpp

Minor improvements in GPT2 tokenizer

#3567

Merged