Explaining and Improving Contrastive Decoding by Extrapolating the Probabilities of a Huge and Hypothetical LM

Open in new window