tracked author

Yossi Matias

Fast Inference from Transformers via Speculative Decoding coauthor. Google Research profile identifies him as Vice President, Google and Head of Google Research and lists speculative decoding among recent research breakthroughs. X profile snippets and external Google-related mentions align with the same role.

1 archived notes X: high HomepageX

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 24, 2026

Fast Inference from Transformers via Speculative Decoding

这篇论文把 speculative execution 推到随机采样场景:小 draft model 先自回归猜 $\gamma$ 个 token,大 target model 一次并行验证这些前缀,并用 $\min(1,p/q)$ 接受概率与 residual distribution 校正拒绝位置,从而在无需改模型、无需重训且保持 target 输出分布不变的前提下,把大模型串行 decode 的目标调用数降低到每步平均生成多个 ...

2211.17192-fast-inference-transformers-speculative-decoding Speculative Decoding