N-Grams Uncovered: A Key Component of Large Language Models