Every Token Becomes a Number

Tokens are not stored as text inside the model. Each token is looked up in a fixed vocabulary and replaced by a numeric token ID, so the input sequence becomes a sequence of integers. The vocabulary is finite, which means every possible token already has an assigned number.