Stuck on an issue?

Lightrun Answers was designed to reduce the constant googling that comes with debugging 3rd party libraries. It collects links to all the places you might be looking at while hunting down a tough bug.

And, if you’re still stuck at the end, we’re happy to hop on a call to see how we can help out.

Confusion regarding DiffSharp-cuda-windows bundle

See original GitHub issue

I can’t get the DiffSharp-cuda-windows bundle to work. Three things that confuse me:

It depends on libtorch-cuda-10.2-linux-x64 (rather than libtorch-cuda-10.2-win-x64), which naturally doesn’t contain the necessary DLLs to run under windows
Any attempt to restore it seems to fail, and I don’t understand why; the DiffSharp-cpu bundle restores just fine, and looks mostly the same. The error is “Package ‘DiffSharp-cuda-windows’ is incompatible with ‘all’ frameworks in project”. This doesn’t seem to be transient or machine specific problem, but I don’t understand it, so it might just be me.
It targets netcoreapp3.0; I’d have expected it to be netstandard2.1 (per the nuspec); however, I see this was explicitly changed recently, so I assume there is a reason for this.

On a positive note, I have successfully got CUDA/GPU working under windows by manually adding the following dependencies (the newly added CI builds on nuget):

DiffSharp.Backends.Torch Version=“1.0.0-preview-243323135”
DiffSharp.Core Version=“1.0.0-preview-243323135”
libtorch-cuda-10.2-win-x64 Version=“1.5.6”

Incidentally, I’m probably going to end up using DiffSharp for significantly more work in the near future, and I recall something about testing under Windows GPU being a problem: I’d be happy to put some effort in if it’s something I can help with (my main machine is Win 10 with an RTX 2060 Super).