Teddy Koker

Hi! See this passage from the [original paper](https://cdn.openai.com/papers/Generative_Pretraining_from_Pixels_V2.pdf): >An IR of 32^2 × 3 is still quite computationally intensive. While working at even lower resolutions is tempting, prior work has...

Hey @pilgrimygy. I added python 3.8, 3.9, and 3.10 to the CI tests; all tests are passing. Have you tried reinstalling with the new python version?

Hi @zimonitrome, this is not possible with this library

Nice work! What would you think about having the current `soft_sort` function return indices in the same way [torch.sort](https://pytorch.org/docs/stable/generated/torch.sort.html) does. This way you could have the differentiable sorted values of...

Thats a great point! Having a toy example would certainly be beneficial in determining the best way to proceed.

This is an interesting find! I have tried this now and am seeing a similar phenomenon:  Running `torchsort_2`, `_3`, `_4` (all just calling the exact same function repeatedly) seem...

In the paper they follow the method described in [Differentiable Ranking and Sorting using Optimal Transport ](https://papers.nips.cc/paper/2019/hash/d8c24ca8f23c562a5600876ca2a550ce-Abstract.html): I believe this is what you would need to do: ```python import torch...

Alright training on CIFAR10 was quite straightforward, I uploaded an example to [extra/cifar10.py](https://github.com/teddykoker/torchsort/blob/main/extra/cifar10.py). Neither this paper nor the previously mentioned one give exact details about the architecture they use (kernel...

Thanks for the comment @vltanh. Following you advice I have made a few changes to the cifar10 script (still a work in progress on the [fix_topk](https://github.com/teddykoker/torchsort/tree/fix_topk) branch). Adding a softmax...

Here is what I got with no augmentation; this looks similar to what you mention: