A model cannot take raw text. A tokenizer first splits the text into tokens, pieces of words, and gives each piece a number from a fixed vocabulary of perhaps tens of thousands of entries.
"The cat sat on the mat." might become seven tokens: The, cat, sat, on, the, mat and the full stop. A rare word such as "unbelievable" may split into pieces like un, believ and able. The splits shown are illustrative; every model has its own tokenizer.
💡Good to know
Limits, speed and prices for language models are usually counted in tokens, not words.