dan_the_3rd

> Is there a specific boundary for the batch size we should be looking for I believe it would be easier for you to just benchmark without 24 sparsity and...

Yes that would be a solution. That would require a bit of refactoring of our `setup.py` file tho...

Hi, We already have such a PR opened here: https://github.com/facebookresearch/xformers/pull/743 The TLDR is that we can't do that at the moment, as using `pyproject` causes the build to happen in...

The issue is that if the isolated environment will have a version of PyTorch X, xFormers will be built for PyTorch version `X` only. If the environment in which the...

Hi, How did you install xformers? It looks like you didn't build the library

``` name: xformers-test ``` This does not look like an official package. The package we provide is called `xformers` and not `xformers-test`, so I'm not sure where this is coming...

> That's the name of the conda environment that gets created Oh right - my bad. > The problem is that the code I pointed out doesn't work when the...

Hi, Thanks for reporting this issue! Indeed we forgot one file in the release. We will fix that asap - cc @lw

> Could you please briefly explain the differences between the three types of kernels: fused, fused_no_wait, and fused_no_wait_no_memcpy? I believe only the `fused` kernel produces correct result. The other ones...

Hi, xFormers is compatible with CUDA 12, but you need to build it from source