Figure 2.1: Autoregressive generation, the engine’s whole method
Figure 2.1 Autoregressive generation, the engine's whole method. The text so far
enters the model, which returns a ranked list of next-token guesses; one is
chosen, appended to the text, and the model runs again on the now-longer input.
The chosen token and its return path are in accent—there is no plan held in
reserve, only this local guess repeated.
