// ⚡ Inference Acceleration & Serving

EAGLE

Speculates at the feature level for higher acceptance rates.

2024 Advanced core

IN PLAIN WORDS

A light network predicts future second-to-top-layer features from current features and the next token, then decodes drafts — more accurate than Medusa.

Used in: EAGLE speculative decoding in vLLM and SGLang

TIMELINE · prerequisites → this → what builds on it

2022 Speculative D…2024 EAGLE

01Original paper

EAGLE: Speculative Sampling Requires Rethinking Feature Uncertainty

Yuhui Li, Fangyun Wei, Chao Zhang, et al. · 2024

arXiv ↗ ✓ Verified · arXiv API

02Where it sits

Learn the left column first; the right column builds on this one.

Learn first

Speculative Decoding

EAGLE

Then you can learn

Nothing in the atlas builds on it yet

Related:Medusa

Left: prerequisites · right: builds on this · dashed: related. Full map →

03Learning resources

Official docs, explainers, courses and reference code. All links checked.

Reference code

github.com ↗EN

04Where the course teaches it

U62 Serving Engines PagedAttention, continuous batching, prefix caching, speculative decoding, prefill/decode disaggregation and gateway routing. 8 h7 algorithms4 papers

For AI / data

Verification: verified, checked 2026-10-21.
JSON: /api/learn/algorithms.json (id: eagle)