Installing on macOS not possible

[sniklaus]

Installing on macOS not possible

[sniklaus]

That sounds more like an issue related to PyTorch, the official discussion forum might be a better place to seek help in this regard. It seems like people who experience this issue are encouraged to do the inference on the CPU without using CUDA at all. However, the separable convolution layer is unfortunately only implemented in CUDA so far.

[sniklaus]

You could try commenting out this line to disable cuDNN and see whether the errors still occur. I am afraid that I only work with Linux and can thus likely not provide further guidance.

[clausbertels]

The problem is that I had to install pytorch from source. Doing this, I disabled CUDA support because I kept getting build errors due to cuDNN not being available.

Now when I run run.py I get an AttributeError: 'module' object has no attribute '_cuda_getDevice'. So I guess I won't be able to run sepconv anytime soon on macOS?