Although line 381 had this: device = torch.device('cuda:0' if torch.cuda.is_available() else 'cpu'), devices without a connected CUDA graphics card would fail in line 49, because of get_device_properties. This PR fixes that and adds a warning, if the CPU fallback occurs. It also adds a cuda_device argument, so the use of the CPU or other graphics card can be forced.
I also slightly improved the set up instructions in the readme and added a section for using an AMD graphics card. However, I actually don't have a supported AMD graphics card and haven't tested this yet.
Although line 381 had this:
device = torch.device('cuda:0' if torch.cuda.is_available() else 'cpu'), devices without a connected CUDA graphics card would fail in line 49, because ofget_device_properties. This PR fixes that and adds a warning, if the CPU fallback occurs. It also adds a cuda_device argument, so the use of the CPU or other graphics card can be forced.I also slightly improved the set up instructions in the readme and added a section for using an AMD graphics card. However, I actually don't have a supported AMD graphics card and haven't tested this yet.