BudgetDraft: Acceptance-Aware Multi-View Training for Sparse-KV Speculative Decoding 文章

ArXiv CS.AI2026-06-02NEWSen作者: Liang He, Jingbo Wen, Qishi Zhan, Yixiong Chen, Kangning Cui, Qizhen Lan, Xilu Wang

BudgetDraft: Acceptance-Aware Multi-View Training for Sparse-KV Speculative Decoding · 相关技术