MoonshotAI

    MoonshotAI/FlashKDA

    FlashKDA: high-performance Kimi Delta Attention kernels

    Cuda
    MIT
    1.3K stars
    131 forks
    1.3K GitHub watchers
    Updated Oct 26, 2026
    View on GitHub

    Build with Backblaze B2

    SDKs, agent skills, IDE extensions, and reference pipelines from Backblaze Labs. All open source.

    Explore Backblaze Labs

    Loading star history...

    Where it stands

    Rank, growth pace and release cadence relative to everything GStars tracks.

    The project lists 5 contributors on GitHub.

    Use Cases & Benefits

    • Implements high-performance Kimi Delta Attention kernels optimized for CUDA architectures SM90 and above.
    • Delivers accelerated attention computations by leveraging CUTLASS and CUDA 12.9+, outperforming standard PyTorch implementations.
    • Use for integrating efficient KDA kernels in PyTorch models requiring fast inference on NVIDIA GPUs.
    • Use for research and development of novel attention mechanisms benefiting from custom CUDA kernel optimizations.
    • Use for building scalable sequence models that demand low-latency and high-throughput attention operations on modern GPU hardware.

    Should you depend on it?

    License terms, project hygiene and risks, from GitHub metadata and GStars' snapshots.

    License

    MIT

    Permissive: commercial use, modification and redistribution are allowed with attribution.

    Community files

    • README: present

    Watch-outs

    • No GitHub releases; pin a commit or tag instead of a release.

    General guidance from repository metadata, not legal advice. How we score

    From the README

    FlashKDA: Flash Kimi Delta Attention — high-performance KDA kernels built on CUTLASS

    Excerpt from the MoonshotAI/FlashKDA README on GitHub (MIT license). © its authors.Read the full README on GitHub

    Daily snapshots

    Stars gained per day, and how forks and open issues moved, from GStars' daily snapshots

    Jul 30: +100

    +337 stars over 90 days across 8 snapshot days; best day +100 on Jul 30.

    Forks

    131+38

    Open issues

    27+9

    Compared with peers in Cuda

    Fastest-growing repositories GStars tracks in this cohort for the week of Nov 1, by 7-day star growth

    MoonshotAI/FlashKDA compared with Cuda repositories by stars, 7-day star growth, health score and last release
    RepositoryStars+7dHealthLast release
    deepseek-ai/DeepEP10.4K—35416 days ago
    deepseek-ai/DeepGEMM9.2K—46106 days ago
    flashinfer-ai/flashinfer6.8K—601 day ago
    MoonshotAI/FlashKDA1.3K—8—
    Peer median9.2K—46—

    Also by MoonshotAI

    Other repositories from this maintainer that GStars tracks

    Discover Repositories

    Search across tracked repositories by name or description