linkedin / linkedin/Liger-Kernel
DeepSeek Native Sparse Attention (NSA) Kernel
Open
Nobody has claimed this yet.
feature
fun
help wanted
- Dominant language
- Python
- Stars
- 6.6k
- Forks
- 603
- Avg merge
- 1d 20h
- Merged PRs (30d)
- 47
Description
🚀 The feature, motivation and pitch
Native Sparse Attention: Hardware-Aligned and Natively Trainable Sparse Attention https://arxiv.org/abs/2502.11089
Potentially useful python reference https://github.com/dhcode-cpp/NSA-pytorch
Alternatives
No response
Additional context
No response
Contributor guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Research direction
Start by reading the Native Sparse Attention paper and the linked NSA-pytorch reference. Then inspect Liger-Kernel's existing Triton kernel entry points to determine the integration scope; done means a defined and working DeepSeek NSA kernel, with validation criteria agreed for the implementation.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- python
- Domain
- machine-learning
- Issue type
- Feature
- Difficulty
- 5/5
- Estimated time
- Over a week
- Activity status
- Quiet
- Clarity
- Needs clarification
- Newbie friendliness
- 35/100