Chi2 sampling on cuda

1

I want to generate a sample of torch.distributions.chi2.Chi2 in cuda.
Is there a way to create it directly from cuda instead of creating it on cpu and passing it to cuda?