A Paradigm Shift in Machine Translation: Boosting Translation Performance of Large Language Models