Fast, differentiable sorting and ranking in PyTorch
Full Changelog: https://github.com/teddykoker/torchsort/compare/v0.1.8...v0.1.9
Full Changelog: https://github.com/teddykoker/torchsort/compare/v0.1.7...v0.1.8
Add half-precision (fp16) support.
Ensure cuda code is added to source release on PyPi.
Remove -arch="compute_50"
from extra args (affected building when TORCH_CUDA_ARCH_LIST is used).
Google colab installation fix.
Fixes CUDA memory leak with torchsort.soft_rank
with regularization="kl"
.
Adds validation for correct user input for torchsort.soft_rank
and torchsort.soft_sort
.