@ptrblck
I have one question and one request. Would you check it?
-
Question about Memory Usage for torch.utils.bottleneck with MNIST.
Training MNIST with --epochs 3 --no-cuda works fine, but failed with --epochs 4 --no-cuda on my server with 16GB-RAM machine. It seems like RAM memory (RES) consumption problem. I guess it needs 64GB-RAM for MNIST Training sample with torch.utils.bottleneck.
Is my guess correct? -
As for the document on torch.utils.bottleneck
It is helpful to write a document for trace customization (topk etc).
cp torch/utils/bottleneck/__main__.py bottleneck.py
pytorch bottleneck.py /script/path/to/source/script.py [args]