Different memory of pytorch 0.4.1 and 0.1.12

I want to know whether the pytorch 0.4.1 have a better memory optimization for gpu. For the same model, the pytorch 0.4.1 use less memory than pytorch 0.1.12. So is it right?


Since PyTorch is being improved continuously you can assume that the execution speed and memory footprint both will be improved in newer versions.
Regarding the two versions you are mentioning, it seems this PR improved the memory usage for large models greatly.

However, I’m not sure if you are using the GPU or run your code on CPU, since newer versions of CUDA/cuDNN etc. also play an important role here.

Thank you very much. I got it