Seed token

The first token (or short prompt) you hand to a language model to start generation. Everything after is sampled from the model.