Generation

Producing a new sequence of tokens by sampling the model one step at a time. Each emitted token feeds back as input for the next step.