STACKQUADRANT

feifeibear/long-context-attention

Fine-tuning Tools

USP: Unified (a.k.a. Hybrid, 2D) Sequence Parallel Attention for Long Context Transformers Model Training and Inference

5.5
GitHub Metrics
Stars
687
Forks
84
Open Issues
14
Watchers
5
Contributors
24
Weekly Commits
0
Language
Python
License
Apache-2.0
Last Commit
May 21, 2026
Created
Mar 27, 2024
Latest Release
0.6.4
Release Date
Jan 13, 2026
Synced: Aug 28, 2026
Quality Scores
Documentation Qualityw: 20%
4.7

No dedicated docs site. Description: 119 chars. Stars signal: 687. Contributors: 24. Score: 4.7/10

Community Healthw: 20%
5.0

Stars: 687. Contributors: 24. Watchers: 5. Forks: 84. Issue ratio: 2.0%. Score: 5/10

Maintenance Velocityw: 15%
5.1

Last commit: 99d ago. Weekly commits: 0. Latest release: 0.6.4. Maturity bonus: 2.4y old. Score: 5.1/10

API Design & DXw: 20%
6.4

Stars/issues ratio: 49. Dynamic language: Python. No dedicated API docs. Permissive license: Apache-2.0. Popularity signal: 687 stars. Score: 6.4/10

Production Readinessw: 15%
5.6

Battle-tested: 687 stars. Peer review: 24 contributors. Versioned: 0.6.4. Licensed: Apache-2.0. Age: 2.4 years. Maintenance: last commit 99d ago. Score: 5.6/10

Ecosystem Integrationw: 10%
6.6

Fork interest: 84. Major ecosystem: Python. Integration-friendly: Apache-2.0. Adoption: 687 stars. Score: 6.6/10

Tags
attention-is-all-you-needdeepspeed-ulyssesllm-inferencellm-trainingpytorchring-attention
Radar
Documentation Quality
Community Health
Maintenance Velocity
API Design & DX
Production Readiness
Ecosystem Integration