DDP freez becasue of torchsummary.summary

[mehran66]

sharing some learning with Distributed Data Parallel (DDP). my training code was freezing at some point during the training loop and took me so much time to find the issue; at somepoint before the training loop, I was using torchsummary.summary(model, (3, input_size, input_size)) to print the model summary for my log. this function does not support ddp and creates a deadlock.