Hi,
I was training on a Windows machine with num_workers > 0 and found in the task manager that only one worker has disk I/O.

So i inserted this block to collate_fn and saw multiple workers working.
pid = os.getpid()
wid = torch.utils.data.get_worker_info().id
print(f"[PID={pid}/WID={wid}]")
Is this a bug in task manager, or is there more to it?