Pytorch multiprocessing RuntimeError: Attempted to send CUDA tensor received from another process; this is not currently supported. Consider cloning before sending

Hi, I tried to copy many models and put them into a multiprocessing.manager.queue
If I pass a model as an arg to a subprocess and use queue.put(deepcopy(model))
It raise
RuntimeError: Attempted to send CUDA tensor received from another process; this is not currently supported. Consider cloning before sending.
Seems like this was becase copied model still have old model 's data
reproduce script and more description is under replenishing

I tried state_dict also, still don’t work