Eagle Corp 发布开源灵活 drafter EAGLE-4，可根据 batch size、硬件和工作负载自适应调整投机解码策略，同时保持目标模型的输出分布。实测搭配 Qwen3.8-27B 在 4x NVIDIA B300 上达到 961.6 tokens/s。Eagle 是服务 NVIDIA、Meta、Intel、AMD、Perplexity 等公司的 AI 推理平台。
Eagle Corp released EAGLE-4, its latest open-source flexible drafter, for accelerating LLM inference.
> At its core, “One drafter adapts its speculative decoding strategy to the batch size, the hardware, and the workload, and keeps the target model's output distribution.”
> Qwen3.8-27B with its EAGLE-4 draft is the measured target pair, achieving 961.6 tokens/s while running on 4x NVIDIA B300.
Eagle is an AI inference platform that powers leading companies like NVIDIA, Meta, Intel, AMD, and Perplexity.
Inference testing time 👀
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：🚨 AI News | TestingCatalog · x.com