Multi-Token Prediction: Why Must Language Models Generate One Token at a Time?
中文摘要
多词元预测探讨语言模型如何从逐一生成词元转变为预测未来多个步骤，以提升生成效率。
English Summary
Multi-Token Prediction explores shifting language models from generating one token at a time to predicting multiple future steps.
原文节选
What if an AI could look several steps into the future instead of asking “what comes next?” over and over again? Continue reading on Medium »