PyPI page
Home page
Author:
Allen Institute for AI
Summary:
CuTe/Triton kernels for linear-attention ops (KDA chunk kernel, KDA short conv), as drop-in replacements for flash-linear-attention, on Blackwell
Latest version:
0.2.0
Required dependencies:
fla-core
|
torch
|
triton
Optional dependencies:
cuda-python
|
nvidia-cutlass-dsl
Downloads last day:
23
Downloads last week:
267
Downloads last month:
587