Published at 10/21/2026, 12:08:48 AM · 20 views
📰 Eagle releases EAGLE-4, a speculative-decoding drafter that adapts to batch size and hardware; it reports 961.6 tok/s on Qwen3.8-27B for a single request
On Oct 20, inference company Eagle released EAGLE-4, the fourth generation of the EAGLE speculative-decoding family. One drafter resizes its speculative tree for the batch size, hardware and workload while keeping the target model's output distribution.
💬 What's your take? Join the discussion in the comments.
Read the original Chinese version
📰 Eagle 发布 EAGLE-4 推测解码草稿模型：一个 drafter 按批量与硬件自适应，官方称 Qwen3.8-27B 单请求 961.6 tok/s
10 月 20 日，推理加速公司 Eagle 发布 EAGLE-4，这是 EAGLE 推测解码系列的第四代。它用一个草稿模型按批量、硬件和负载动态调整推测树，同时保持目标模型输出分布不变。官方口径下，Qwen3.8-27B 在 4 张 B300 上单请求达 961.6 tokens/s。
💬 你怎么看？欢迎在评论区讨论。
#speculative-decoding #inference #vLLM
0 Comments