← Back to feed

Better and Faster Large Language Models via Multi-Token Prediction (2024)

AI · 2026-10-19 18:33

arXiv论文提出多词元预测训练方法，让大语言模型一次预测多个未来词元，在提升性能的同时加快训练与推理速度。

Source: arxiv.org

Auto-generated summary for reference only. RSS