cff-version: 1.2.0
message: "If you use eagle-harness in your work, cite the EAGLE-4 paper below."
type: software
title: "eagle-harness"
abstract: >-
  EAGLE-4 for SGLang, vLLM and experimental TokenSpeed: cost-model-driven dynamic-tree speculative decoding with EAGLE-4 drafts, as an
  out-of-tree engine plugin.
authors:
  - name: "Eagle AI Team"
repository-code: "https://github.com/eagle-hq/EAGLE-4"
url: "https://eagle.ai"
keywords:
  - speculative decoding
  - EAGLE
  - LLM inference
  - sglang
  - vLLM
preferred-citation:
  type: generic
  title: "EAGLE-4: One Flexible Drafter for All Batch Sizes and Hardware"
  authors:
    - name: "EAGLE-4 Team"
  year: 2026
  notes: "Preprint"
  url: "https://github.com/eagle-hq/EAGLE-4/blob/main/EAGLE4_Paper.pdf"
references:
  - type: conference-paper
    title: "EAGLE: Speculative Sampling Requires Rethinking Feature Uncertainty"
    authors:
      - given-names: Yuhui
        family-names: Li
      - given-names: Fangyun
        family-names: Wei
      - given-names: Chao
        family-names: Zhang
      - given-names: Hongyang
        family-names: Zhang
    collection-title: "International Conference on Machine Learning"
    year: 2024
  - type: conference-paper
    title: "EAGLE-2: Faster Inference of Language Models with Dynamic Draft Trees"
    authors:
      - given-names: Yuhui
        family-names: Li
      - given-names: Fangyun
        family-names: Wei
      - given-names: Chao
        family-names: Zhang
      - given-names: Hongyang
        family-names: Zhang
    collection-title: "Empirical Methods in Natural Language Processing"
    year: 2024
  - type: conference-paper
    title: "EAGLE-3: Scaling up Inference Acceleration of Large Language Models via Training-Time Test"
    authors:
      - given-names: Yuhui
        family-names: Li
      - given-names: Fangyun
        family-names: Wei
      - given-names: Chao
        family-names: Zhang
      - given-names: Hongyang
        family-names: Zhang
    collection-title: "Annual Conference on Neural Information Processing Systems"
    year: 2025