Daniel Falbel

That's because cuda12.8 is not a supported version for torch v0.15.1. I'm preparing a submission for v0.16.0 that will support cuda 12.8. You can still install torch for older cuda...

It's probably caused by some incompatible cudnn version. Can you share more about the environment you are running? Also, would be great to get a stack trace, maybe running outside...

Thanks @bhvieira , that's very helpful! Would you be able to look inside the directory returned by `system.file(package="torch")` and look for `cuda_` related files? Perhaps something like: ``` fs::dir_ls(system.file(package="torch"), recurse=TRUE,...

Interestingly we see: `torch/lib/cudnn_graph64_9.dll`. Wondering why it's not found. I'll need to log into a windows machine to further debug with something like https://www.dependencywalker.com but I don't currently have a...

It looks like it's not implemented yet in torch yet, unfortunately: https://github.com/pytorch/pytorch/blob/980f5ac0499447cd6d39b6241ae30ae30937a3e5/tools/autograd/derivatives.yaml#L1926-L1927