Fetching the paper…

DoLa: Decoding by Contrasting Layers Improves Factuality in Large Language Models · Around