tweak running examples without cuda

[mattip]

Tweak the run_python_examples.sh script to complete in ~5 minutes (on my machine) without CUDA. Once this is merged, we can consider adding #704 to actually run the examples as part of CI.

TBD: how do we get a pre-release version of torch/torchvision in order to test the latest HEAD? Perhaps it should be a pytorch/pytorch cron job that runs once a day or once a week?

ping @jlin27

[brianjo]

You can install and run the nightly build using:

pip install --pre torch torchvision -f https://download.pytorch.org/whl/nightly/cu102/torch_nightly.html

Per pytorch.org install matrix. You might have to install nightlies for audio, vision and others to get this to work.

[brianjo]

You can install and run the nightly build using:

pip install --pre torch torchvision -f https://download.pytorch.org/whl/nightly/cu102/torch_nightly.html

Per pytorch.org install matrix. You might have to install nightlies for audio, vision and others to get this to work.

Sorry, for non cuda nightly builds on Linux, the pip command is:

pip install numpy
pip install --pre torch torchvision -f https://download.pytorch.org/whl/nightly/cpu/torch_nightly.html

Cheers

[soumith]

the changes to mnist and mnist_hogwild are too interfering to the spirit of the example. These are examples first, and CI tests second.

Can we find a way to make the changes to be 6 lines or less, even if it means they run for a minute each or something?

[mattip]

@soumith reworked dry_run handling in d0447a1

[soumith]

lgtm, did you locally check that no breakages got introduced?

[mattip]

I checked that this PR did not introduce breakage by running the tests locally. There is a problem with dcgan and PyTorch 1.7: it is using a deprecated construct. When I do a PR to add CI, that will need to be fixed to make the CI pass.

Edit: qualify blanket statement, maybe I made a mistake.