PyPI page
Home page
Author:
NVIDIA Corporation
Summary:
A unified PyTorch-native library for subquadratic alternatives to quadratic attention methods with optimized CUDA kernels.
Latest version:
0.2.0
Required dependencies:
datasets
|
einops
|
numpy
|
omegaconf
|
pillow
|
pyarrow
|
pytorch-lightning
|
rich
|
torch
|
torchvision
|
wandb
Optional dependencies:
megatron-core
|
nvidia-dali-cuda130
|
nvsubquadratic
|
quack-kernels
|
subquadratic-ops-torch-cu13
|
timm
Downloads last day:
15
Downloads last week:
117
Downloads last month:
500
7-day smoothing