Skip to content
OpenTrain AIFor AI Companies

DuoAttention: Efficient Long-Context LLM Inference with Retrieval and Streaming Heads

Published Oct 1, 2024
arXiv PDF
Researcher verdict
Starting point
Use as implementation starting point
Benchmark evidence
Missing
Not verified yet
Time to first repro
A few hours
Fast first run
Risk flags
1
Review before use

Results and benchmarks

Freshness tier: cold
DuoAttention: Efficient Long-Context LLM Inference with Retrieval and Streaming Heads presents a transformer approach for retrieval / indexing.

Implementation

Best maintained implementation now

Recommended
Confidence: High
Reproducibility: Moderate
mit-han-lab/duo-attention

[ICLR 2025] DuoAttention: Efficient Long-Context LLM Inference with Retrieval and Streaming Heads

544 stars · 43 forks · Last push Feb 10, 2025 · MIT license

  • License
  • CI
  • Dependencies
  • Docker

Official implementation from Papers with Code · Repository link is mentioned in the paper metadata · Strong overlap with paper title keywords

Why this implementation
Confidence: high

mit-han-lab/duo-attention is the strongest maintained implementation based on ranking signals. License is declared (MIT). Dependency/environment manifests are present.

Open mit-han-lab/duo-attention
Reproduction risks
  • No CI workflows detected
  • Selected mit-han-lab/duo-attention as the strongest maintained implementation for new work.
  • Includes dependency/environment manifest signals.
  • Repository activity is within the last 24 months.

Compare implementation paths

Compare maintenance quality, reproducibility coverage, and evidence confidence before choosing a reproduction baseline.

mit-han-lab/duo-attention
best maintained
Maintenance
Stale
Confidence
High
Reproducibility
Moderate
Stars
544
Last push
Feb 10, 2025 (626d)

Official implementation from Papers with Code · Repository link is mentioned in the paper metadata

  • No push in 12+ months
  • No CI pipeline detected
  • No tagged releases
princeton-pli/prulong
alternative
Maintenance
Stale
Confidence
Low
Reproducibility
Limited
Stars
48
Last push
Jul 29, 2025 (457d)

Community adoption signal (48 stars)

  • No push in 12+ months
  • No CI pipeline detected
  • No tagged releases
devillove084/LLMZeroToHero
alternative
Maintenance
Recently updated
Confidence
Low
Reproducibility
Limited
Stars
12
Last push
Aug 3, 2026 (88d)

Matched via arXiv identifier search

  • No CI pipeline detected
  • No tagged releases
  • No Docker setup

Reproduction readiness

Time to first repro: hours
Last checked: Oct 28, 2026

Setup required

Dependencies pinned, manual setup needed

  • mit-han-lab/duo-attention has pyproject.toml but requires manual environment setup.
  • Last push was 626 days ago, so expect possible dependency version conflicts.
  • No Dockerfile, so you will set up the environment manually.
  • No CI pipeline, so test coverage is unknown.
Open mit-han-lab/duo-attention

Quick start

git clone https://github.com/mit-han-lab/duo-attention.git
pip install -e .

Repositories and ecosystem

No additional verified repositories beyond the primary recommendation.

Hugging Face artifacts

No direct paper-linked artifacts were found. Showing strongest curated related artifacts for faster exploration.

Models

Curated Related

Broaden model search

Transformer Retrieval / indexing Transformer Large Language Models Retrieval / indexing

Datasets

No trustworthy datasets matches right now.

Search datasets on Hugging Face

Spaces

No trustworthy spaces matches right now.

Search spaces on Hugging Face

Research context

Tasks

Retrieval / indexing

Methods

Transformer

Domains

Large Language Models, Information Retrieval

Evaluation and human feedback data

Open this paper in HFEPX to review benchmark signals, evaluation modes, and human-feedback protocol context.

Open in HFEPX
Explore similar papers

Jump to Paper2Code search queries derived from this paper's research context.

Retrieval / indexing Transformer Large Language Models Information Retrieval

Data includes links from Papers with Code ( CC-BY-SA-4.0 ).