EAGLE-4 投机解码实测：27B 模型单请求达 863.8 tokens/s
EAGLE-4 投机解码提速实测
EAGLE-4 在 27B 模型单请求下达到 863.8 tokens/s，为 DFlash 2 的 1.28 倍、DSpark 的 1.40 倍。它保留“草稿—验证”循环，改进集中在草稿侧：自适应草稿结构和更廉价的草稿 LM head，由小模型预猜多个 token、目标模型一次前向校验，输出结果不变，测试环境约快 3 倍。
Rohan Paul· @rohanpaul_ai · X·2026-10-21 08:47· 1 小时前AI 评分43
AI 导读
EAGLE-4 在 27B 模型单请求下达到 863.8 tokens/s，为 DFlash 2 的 1.28 倍、DSpark 的 1.40 倍。它保留“草稿—验证”循环，改进集中在草稿侧：自适应草稿结构和更廉价的草稿 LM head，由小模型预猜多个 token、目标模型一次前向校验，输出结果不变，测试环境约快 3 倍。
正文 · 原文
Some of the most useful inference work right now is happening in the speculative decoding component.
It has become one of the most dependable ways to get more tokens per second out of a model you’ve already chosen.
EAGLE-4 decodes 863.8 tokens/s on a 27B model on a single request, 1.28X DFlash 2 and 1.40X DSpark.
EAGLE-4 is a speed-up add-on for running AI chatbots: a tiny helper model guesses the next few words, and the big model checks all those guesses at once instead of writing one word at a time. The answers come out identical, just about 3× faster on the tested setup.
EAGLE-4 keeps the draft-and-verify loop and puts its work into the drafting side, with an adaptive draft structure and a cheaper draft LM head.
The loop:
- A small draft model proposes several tokens ahead
- The target model checks all of them in one forward pass
- Matching tokens are kept, and the first mismatch is replaced with the target’s own token
- Repeat from there
EAGLE-4 is here. From EAGLE-1 to EAGLE-3, we’ve kept making inference faster. EAGLE-4 is our fastest yet. 961.6 tokens/s on @Alibaba_Qwen’s Qwen3.8-27B with TokenSpeed (BF16, 4× @NVIDIA B300, TP4, one request, T=0). Github: https://github.com/eagle-hq/EAGLE-4在 X 查看被引用的帖子
来源：Rohan Paul · x.com
来源：RadarAI · aihot.news