llama.cpp

e8b8d32e - server : fix incorrect num_tokens_predicted (#3480)