tracked author

Zhijian Liu

DFlash corresponding author. Homepage identifies Zhijian Liu as UC San Diego Assistant Professor, Z Lab director, and NVIDIA Research Scientist, with MIT PhD advised by Song Han; it links Google Scholar, GitHub zhijian-liu, and X zhijianliu_.

1 archived notes X: high HomepageGitHubX

Related Notes

按论文归档时间排序,展示该作者在本站已经出现的材料。

Jun 28, 2026

DFlash: Block Diffusion for Flash Speculative Decoding

DFlash 的核心贡献是把 diffusion LLM 放到 speculative decoding 的 draft stage:用目标自回归模型的多层 hidden features 做条件,把融合特征注入 draft model 每层 KV cache,再用 block diffusion 一次并行预测整块候选 token。它在 Qwen3 4B/8B 非 thinking 模式下报告约 4.0x 4.9x 平均加速,在 re...

2602.06036-dflash-block-diffusion-speculative-decoding Speculative DecodingMulti-Token Prediction