← Discover
Discover · GitHub · FlashKDA↑ +91 today

High performance cores for some Delta Attention

Developed by Moonshot AI, FlashKDA offers high-performance kernels for the Some Delta Attention mechanism. This CUDA-based technology aims to accelerate attention calculations in large language models.

What you get

  • CUDA-based accelerated attention calculations
  • Working efficiently on large language models
  • Kernel structure optimized with CUTLASS

Installation

Basic setup
git clone https://github.com/MoonshotAI/FlashKDA.git flash-kda
cd flash-kda
git submodule update --init --recursive
pip install -v --no-build-isolation .
Build for all architectures
FLASH_KDA_CUDA_ARCHS=all pip install -v --no-build-isolation .

Running it

Using FLA as a backend
pip install -U flash-linear-attention

If you don't write code

🤖 Paste this into your AI agent (Claude Code · Codex · Antigravity)

I want to speed up some Delta Attention calculations using the FlashKDA tool. How can I optimize my model's attention mechanism by using the chunk_kda function under torch.inference_mode(), integrated with the flash-linear-attention library? Please create an application example, taking into account the necessary parameters and hardware requirements I need to pay attention to.

Related dictionary terms

KernelsAttentionArtificial Intelligence
Who it is forIt is suitable for developers who want to accelerate attention calculations on large language models on CUDA.
LicenseMIT

Links

TreScout did not build this tool · we found it in GitHub trends and wrote it up. This page describes the repository as of 2026-07-30: The star count and our text belong to that day, the repository may have changed since. Check the repository link for the current state. This page was machine-translated from the Turkish original · the Turkish version prevails.