Token
A small chunk of an input sequence, often a word or part of a word, that a transformer processes as one unit. The input is broken into tokens, each is turned into numbers, and attention then works out how the tokens relate.
A hospital budgets a hosted model that drafts discharge letters, priced per token, by multiplying letters per month by average words per letter. The bill comes in higher. The notes sent in are billed along with the letter that comes out, and tokens are often pieces of words, so long drug names, abbreviations and words in languages other than English tend to split into several tokens each. Before signing, ask the vendor to run a de-identified sample of your real notes, in every language you document in, through its tokenizer and budget from those counts.
Terms like this come up in real clinical scenarios across the HelloAI courses: bite-sized modules with verifiable certificates. An account takes one minute, no password needed.
Sign in →