In the scripts exemplify the use of PyTorch in combination with POT, the torch.Generator() is always created on the CPU, which leads to an error if a GPU is available.

To reproduce the error run the example script 'plot_sliced_wass_grad_flow_pytorch.py' with an available GPU.

This is easily fixed by changing
gen = torch.Generator()
to
gen = torch.Generator(device=device)
in lines 77 and 139 of the script.
The same has to be done in the example jupyter notebook.