How do PyTorch consume memory when dynamically constructing the training graph and the testing graph? e.g. Does it save every middle level Variable ? How to estimate the usage of the network? Is there any function for me to view the memory usage?
How do PyTorch consume memory when dynamically constructing the training graph and the testing graph? e.g. Does it save every middle level Variable ? How to estimate the usage of the network? Is there any function for me to view the memory usage?