[論文解讀] Kangaroo: Lossless Self-Speculative Decoding via Double Early Exiting

Last Updated on 2024-07-25 by Clay 前言 這是華為諾亞方舟實驗室所提出加速框 … 閱讀全文 [論文解讀] Kangaroo: Lossless Self-Speculative Decoding via Double Early Exiting