Skip to content

DeepSeek Native Sparse Attention (NSA) Kernel #656

@qingquansong

Description

@qingquansong

🚀 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

Metadata

Metadata

Assignees

No one assigned

    Labels

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions