Repeated generation

A model adds one token to the text. It then uses the new text to predict another token. It repeats these steps to build an answer. It stops when it reaches an end signal or a set limit.

One token becomes the next context

Simple generation loop · Stop at an end signal or a set limit.

Start with The sky is, predict and select a token, then append blue to the text. Repeat with the updated text until a stopping condition.

02 · Try

Put the steps for generating text in order.

Progress saves in this browser.