Efficient Guided Generation for Large Language Models

Brandon T. Willard, Rémi Louf

In this article we show how the problem of neural text generation can be constructively reformulated in terms of transitions between the states of a finite-state machine. This fram…