Add option for weight tying on TPU's

[lezwon]

What does this PR do?

Fixes #2705

Before submitting

• Was this discussed/approved via a GitHub issue? (not for typos and docs)
• Did you read the contributor guideline, Pull Request section?
• Did you make sure your PR does only one thing, instead of bundling different changes together?
• Did you make sure to update the documentation with your changes? (if necessary)
• Did you write any new necessary tests? (not for typos and docs)
• Did you verify new and existing tests pass locally with your changes?
• Did you update the CHANGELOG? (not for typos, docs, test updates, or internal minor changes/refactorings)

PR review

Anyone in the community is free to review the PR once the tests have passed.
Before you start reviewing make sure you have read Review guidelines. In short, see the following bullet-list:

• Is this pull request ready for review? (if not, please submit in draft mode)
• Check that all items from Before submitting are resolved
• Make sure the title is self-explanatory and the description concisely explains the PR
• Add labels and milestones (and optionally projects) to the PR so it can be classified
• Check that target branch and milestone match!

Did you have fun?

Make sure you had fun coding 🙃

[pep8speaks]

Hello @lezwon! Thanks for updating this PR.

There are currently no PEP 8 issues detected in this Pull Request. Cheers! 🍻

Comment last updated at 2021-02-17 23:42:15 UTC

[codecov]

Codecov Report

Merging #5441 (3605c0b) into master (6a409c7) will decrease coverage by 2%.
The diff coverage is 100%.

@@           Coverage Diff           @@
##           master   #5441    +/-   ##
=======================================
- Coverage      93%     91%    -2%     
=======================================
  Files         160     160            
  Lines       11358   11720   +362     
=======================================
+ Hits        10569   10648    +79     
- Misses        789    1072   +283     

[rohitgr7]

this should go to release/1.2-dev.

[carmocca]

Removed ready-to-go since this is now blocked by the accelerator refactor

[kaushikb11]

@lezwon Could you help me out on what this check means?

[lezwon]

@kaushikb11 if the layer weights are not tied while on tpu, then the layer count on the tpu will show as no_of_layers + 1, as the xla library will make a copy of the layer weights. This check makes sure the layer count matches after moving the model to the device. 😊👍

[kaushikb11]

@lezwon Yup, but I don't think it's necessary to do parameter_validation for every module to call? wdyt?

Just a fyi: Currently I am doing this https://github.com/PyTorchLightning/pytorch-lightning/blob/tpu_spawn_added/pytorch_lightning/plugins/training_type/tpu_spawn.py#L172, by using xla's MpModelWrapper. This will make the test fail. Maybe I could add this check specific to TPU acclerators.

[lezwon]

Yes, you could maybe add it specifically to TPU accelerators. Also do check it works well when using xla with GPU's.