tracked author

Fangyun Wei

EAGLE, EAGLE-2 and EAGLE-3 coauthor at Microsoft Research. His Microsoft Research profile records a Peking University master's degree and Microsoft Research Asia employment, and its publication list includes the EAGLE series.

1 archived notes X: not-found Homepage

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。

归档 Jul 24, 2026

EAGLE 3: Scaling up Inference Acceleration of Large Language Models via Training Time Test

EAGLE 3 去除特征回归约束，用多层目标特征融合和训练时多步自生成展开直接训练草稿 token 分布；LLaMA 3.1 8B 在 MT Bench 从 EAGLE 2 的 3.16 倍提高到 4.40 倍，6.47 倍是单批量研究原型峰值，SGLang 批量 64 为 1.38 倍。

待审阅 2503.01840-eagle-3-training-time-test Speculative Decoding Multi-Token Prediction Serving Runtime