DistributedSampler — Expected a ‘cuda’ device type for generator when generating indices

Question

Performing distributed training, I have the following code like this:

training_sampler = DistributedSampler(training_set, num_replicas=2, rank=0)
training_gen