Read original ↗
paperarXivTrust 82 · PrimaryPublished 5d agoLive · 4d ago

DSpark: Confidence-Scheduled Speculative Decoding with Semi-Autoregressive Generation

Speculative decoding accelerates Large Language Model (LLM) inference by decoupling draft generation from target verification. While recent parallel drafters efficiently propose long token sequences in a single forward pass, they suffer from rapid acceptance decay due to a lack of inter-token dependencies. Furthermore, indiscriminately verifying these extended blocks wastes critical batch capacity on tokens with high rejection risks, severely degrading throughput in high-concurrency serving systems. We introduce DSpark, a speculative decoding framework that unifies high-throughput parallel gen

Lineage graph

Paper → model → repo connections mined from source citations (Tier-1 exact match).

Why these links exist

  • Linked via arxiv authorXin Cheng

    DSpark: Confidence-Scheduled Speculative Decoding with Semi-Autoregressive Generation

  • Linked via arxiv authorXingkai Yu

    DSpark: Confidence-Scheduled Speculative Decoding with Semi-Autoregressive Generation

  • Linked via arxiv authorChenze Shao

    DSpark: Confidence-Scheduled Speculative Decoding with Semi-Autoregressive Generation

  • Linked via arxiv authorJiashi Li

    DSpark: Confidence-Scheduled Speculative Decoding with Semi-Autoregressive Generation

  • Linked via arxiv authorYunfan Xiong

    DSpark: Confidence-Scheduled Speculative Decoding with Semi-Autoregressive Generation

  • Linked via arxiv authorYi Qian

    DSpark: Confidence-Scheduled Speculative Decoding with Semi-Autoregressive Generation

  • Linked via arxiv authorJiaqi Zhu

    DSpark: Confidence-Scheduled Speculative Decoding with Semi-Autoregressive Generation

  • Linked via arxiv authorShirong Ma

    DSpark: Confidence-Scheduled Speculative Decoding with Semi-Autoregressive Generation

  • Linked via arxiv authorXiaokang Zhang

    DSpark: Confidence-Scheduled Speculative Decoding with Semi-Autoregressive Generation

  • Linked via arxiv authorJiasheng Ye

    DSpark: Confidence-Scheduled Speculative Decoding with Semi-Autoregressive Generation

  • Linked via arxiv authorQinyu Chen

    DSpark: Confidence-Scheduled Speculative Decoding with Semi-Autoregressive Generation

  • Linked via arxiv authorChengqi Deng

    DSpark: Confidence-Scheduled Speculative Decoding with Semi-Autoregressive Generation

  • Linked via arxiv authorJiping Yu

    DSpark: Confidence-Scheduled Speculative Decoding with Semi-Autoregressive Generation

  • Linked via arxiv authorDamai Dai

    DSpark: Confidence-Scheduled Speculative Decoding with Semi-Autoregressive Generation

  • Linked via arxiv authorZhengyan Zhang

    DSpark: Confidence-Scheduled Speculative Decoding with Semi-Autoregressive Generation

  • Linked via arxiv authorYixuan Wei

    DSpark: Confidence-Scheduled Speculative Decoding with Semi-Autoregressive Generation

  • Linked via arxiv authorYixuan Tan

    DSpark: Confidence-Scheduled Speculative Decoding with Semi-Autoregressive Generation

  • Linked via arxiv authorWenkai Yang

    DSpark: Confidence-Scheduled Speculative Decoding with Semi-Autoregressive Generation

  • Linked via arxiv authorRunxin Xu

    DSpark: Confidence-Scheduled Speculative Decoding with Semi-Autoregressive Generation

  • Linked via arxiv authorShanyu Wu

    DSpark: Confidence-Scheduled Speculative Decoding with Semi-Autoregressive Generation

  • Linked via arxiv authorZhean Xu

    DSpark: Confidence-Scheduled Speculative Decoding with Semi-Autoregressive Generation

  • Linked via arxiv authorXuanyu Wang

    DSpark: Confidence-Scheduled Speculative Decoding with Semi-Autoregressive Generation

  • Linked via arxiv authorMuyang Chen

    DSpark: Confidence-Scheduled Speculative Decoding with Semi-Autoregressive Generation

  • Linked via arxiv authorRui Tian

    DSpark: Confidence-Scheduled Speculative Decoding with Semi-Autoregressive Generation

  • Linked via arxiv authorXiao Bi

    DSpark: Confidence-Scheduled Speculative Decoding with Semi-Autoregressive Generation

  • Linked via arxiv authorZhewen Hao

    DSpark: Confidence-Scheduled Speculative Decoding with Semi-Autoregressive Generation

  • Linked via arxiv authorShaoyuan Chen

    DSpark: Confidence-Scheduled Speculative Decoding with Semi-Autoregressive Generation

  • Linked via arxiv authorHuanqi Cao

    DSpark: Confidence-Scheduled Speculative Decoding with Semi-Autoregressive Generation

  • Linked via arxiv authorWentao Zhang

    DSpark: Confidence-Scheduled Speculative Decoding with Semi-Autoregressive Generation

  • Linked via arxiv authorAnyi Xu

    DSpark: Confidence-Scheduled Speculative Decoding with Semi-Autoregressive Generation

  • Linked via arxiv authorHuishuai Zhang

    DSpark: Confidence-Scheduled Speculative Decoding with Semi-Autoregressive Generation

  • Linked via arxiv authorDongyan Zhao

    DSpark: Confidence-Scheduled Speculative Decoding with Semi-Autoregressive Generation

  • Linked via arxiv authorWenfeng Liang

    DSpark: Confidence-Scheduled Speculative Decoding with Semi-Autoregressive Generation

Implements

Covers

authored (incoming)

Related across the graph

Topics