[doc] Fix closure in manual optimization

[akihironitta]

What does this PR do?

In most cases, zero_grad can be called outside the closure, but zero_grad needs to be called inside the closure when using an optimizer like LBFGS which requires reevaluation of the loss in optimizer.step(closure).

This PR fixes the closure in the docs so that it works in all cases above.

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

Did you have fun?

Make sure you had fun coding 🙃

[codecov]

Codecov Report

Merging #6374 (b1721dc) into master (4f391bc) will decrease coverage by 7%.
The diff coverage is n/a.

@@           Coverage Diff            @@
##           master   #6374     +/-   ##
========================================
- Coverage      93%     86%     -7%     
========================================
  Files         161     161             
  Lines       11464   12596   +1132     
========================================
+ Hits        10706   10840    +134     
- Misses        758    1756    +998     

[rohitgr7]

quick question:
is zero_grad and its corresponding hooks called before backward and optimizer.step in case of automatic_optimization??

[carmocca]

quick question:
is zero_grad and its corresponding hooks called before backward and optimizer.step in case of automatic_optimization??

Yes: #6147

[akihironitta]

@rohitgr7 Yes, but internally, we call those methods inside optimizer.step:

optimizer.step(closure=closure)
  closure()
    forward()
    zero_grad()
    backward()
  update_params()

but this is equivalent to:

forward()
zero_grad()
backward()
optimizer.step(closure=None)
  update_params()

All the difference is whether forward(), zero_grad(), and backward() are called inside or outside the closure. We do that because some optimizers, such as LBFGS, need to call them inside the closure. https://pytorch.org/docs/master/optim.html#torch.optim.LBFGS