* supporting more diverse tokenizers * Update llama.cpp --------- Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>