setup.py should not contain import of torch about stable-fast HOT 4 CLOSED

If env e.g. docker has no installation of torch yet, this setup will fail.

Rather, torch should be imported lazily and the asserts performed only if torch is available

That cannot be achieved. Since stable-fast in fact is a CUDAExtension of torch, it must require torch to build. So torch must be installed at first, then stable-fast😊.

Its build system is on top of torch.

from stable-fast.

Yes. Indeed. Actually, the issue is that I can't seem to download the right whl, as I'm using torch nightly, whereas stable-fast wants torch.2.1, so it downloads the source code, thus running into this issue.

Is there any way I can force to use the whl?

from stable-fast.

Ok, I can just pip install https://github.com/chengzeyi/stable-fast/releases/download/v0.0.13.post3/stable_fast-0.0.13.post3+torch210cu121-cp310-cp310-manylinux2014_x86_64.whl

from stable-fast.

@jon-chuang You can always find a working demo on Colab now: https://colab.research.google.com/github/chengzeyi/stable-fast-colab/blob/main/stable_fast_colab.ipynb

from stable-fast.