feifeibear / long-context-attention
USP: Unified (a.k.a. Hybrid, 2D) Sequence Parallel Attention for Long Context Transformers Model Training and Inference
Why this job: works in the areas we cover, but does not say which job it does
What it does
This repo provides a sequence parallel approach that synergizes the strengths of two popular distributed attentions, i.e. DeepSpeed-Ulysses-Attention and Ring-Attention, delivering a more general and stronger versatility and better performance. The project is built on zhuzilin/ring-flash-attention and refers to the DeepSpeed-Ulysses.
Get started
No single install command. Read the steps in the README (opens in new tab)
- Maintenance
- Active
- Last commit 55 days ago
- Stars
- 696
- On GitHub
- Licence
- Apache-2.0
- Allows business use
- Downloads for
- None listed
- No release files
- Latest version
- 0.6.4
- Released 10 months ago
- Built with
- Python
- 98% of the code
Is it looked after?
- Contributors
- 3
- Last 90 days
- Issues
- 72% closed
- 11 still open
- Pull requests
- 129 merged
- 4 waiting
- Releases
- 18
- 1 in the last year
- No contributing guide
- No security policy
- No code of conduct
More: languages, topics and project record
Languages
- Python 98%
- Shell 2%
Topics
- attention-is-all-you-need
- llm-inference
- llm-training
- pytorch
- ring-attention
- deepspeed-ulysses
Project record
- JobOther AI tools
- StatusListed
- Commits265
- Forks86
- Watching5
- Stars in 28 days0
- CreatedMar 27, 2024
- First seen by usNov 7, 2026
- CheckedNov 10, 2026