llama.cpp

9a3b91c6 - server : fix usage.completion_tokens undercount with n > 1 (#29486)