llama.cpp

server : use common_token_to_piece instead of common_detokenize

#11740

Merged