Is there support for newer GPUs (e.g., RTX 4090) with updated PyTorch version?

[cicichena]

Hi, thank you for your great work on this project!

I'm trying to run the code on a machine with an NVIDIA RTX 4090, but the current PyTorch version (1.8.1) specified in the repository doesn't support this GPU.

Do you have any plans to update the codebase to support newer GPUs (e.g., using PyTorch 1.13+ or a compatible version)?

Alternatively, could you suggest which versions of PyTorch, PyTorch Lightning, and other dependencies might work best with this project on CUDA 11.7 or newer?

Thanks a lot!

[iprapas]

Hi!

Currently, we don't have plans to support newer versions of the libraries.

I think the changes needed are not so major. For example, using the new pytorch lightning import import from lightning instead of import from pytorch_lightning, etc

If you end up doing it eventually, feel free to submit a pull request

[cicichena]

Thanks for reply, I updated the dependencies to the following version, and it works. "pytorch-lightning>=1.2.0", "hydra-core>=1.0.6", "torch==1.13.0+cu117", "torchvision==0.14.0+cu117", "fastai==2.7.10"