Tokenization is a vital method in machine learning, acting as the initial point for preparing text data. Essentially, it involves breaking down a larger block of information – like a sentence – into smaller units called “pieces". These tokens can be copyright, punctuation marks, or even sub-lexical