adhere to lazy import rules (#806)

[fbgheith]

Summary:

Lazy import changes Python import semantics, specifically when it comes to initialization of packages/modules: https://www.internalfb.com/intern/wiki/Python/Cinder/Onboarding/Tutorial/Lazy_Imports/Troubleshooting/

For example, this pattern is not guaranteed to work:

import torch.optim
...
torch.optim._multi_tensor.Adam   # may fail to resolve _multi_tensor

And this is guaranteed to work:

import torch.optim._multi_tensor
...
torch.optim._multi_tensor.Adam   # will always work

A recent change to PyTorch changed module initialization logic in a way that exposed this issue.

But the code has been working for years? This is the nature of undefined behavior, any change in the environment (in this the PyTorch code base can make it fail.

Differential Revision: D58881291

[facebook-github-bot]

This pull request was exported from Phabricator. Differential Revision: D58881291

[facebook-github-bot]

This pull request was exported from Phabricator. Differential Revision: D58881291