PyPI page
Author:
Sriman Achanta
Summary:
FoldAttention declares softmax’s reference before execution, making attention additive and enabling up to 3.09× faster H100 decode and deterministic backward faster than nondeterministic kernels.
Latest version:
0.2.0
Required dependencies:
flash-attn-4
|
numpy
|
nvidia-cutlass-dsl
|
quack-kernels
|
torch
Optional dependencies:
flash-attn-4
|
nvidia-cutlass-dsl
|
quack-kernels
Downloads last day:
2
Downloads last week:
13
Downloads last month:
384
7-day smoothing