Hi, you can use
-
torch.multiprocessing.SimpleQueue
to let the child processes to put the results in the queue. -
point-to-point communication functions to send tensors between different distributed processes.
You may want to refer to this thread for more explanation.