WhatThe model never sees letters. Your text is cut into tokens: whole words, pieces of words, spaces with a word, punctuation.
HowEach token is an entry in a fixed list of about 200,000 and has an ID number. Common words are one token; long or rare words fall apart into several.
Why it mattersTokens are the only thing the model reads and predicts. It sees pieces, not letters, which is why counting letters is hard for it.