Installation on Linux for GPU usage

[AsyaOrlova]

Hello! I have a problem with creating environment for GPU usage.

I run this command:
conda env create -f conda_gpu.yml

And I get the error, which is depicted on the screenshots.
I suppose it is connected with pytorch-fast-transformers package.

Screenshots

System:

• OS: Linux
• Python version: 3.10.13
• GT4SD version: 1.4.3

[jannisborn]

HI @AsyaOrlova Thanks for your interest. Is this WSL? Because GT4SD does not support Windows

[AsyaOrlova]

@jannisborn Thanks for the quick answer! This is WSL, however I use it to connect to the remote server with Linux OS. So, I guess, this is not the problem, because I try to run gt4sd on server.

[jannisborn]

And you have GPU access on that machine? Otherwise you need conda_cpu_linux.yml

[AsyaOrlova]

Yes, I do have it. I also tried to create CPU environment and it failed with error as well.

[jannisborn]

Hard to debug from far, but your error suggests to pre-install torch into the env before attempting to install other deps

[AsyaOrlova]

I foun out I had problems specifically with gt4sd-molformer (something about a conflict between cuda and pytorch). So I created the environment without this package as a temporary workaround, since I don't actually need molformer for my purposes. Thank you for your help!

[jannisborn]

Closing as completed but feel free to reopn