Disable quantization aware training observers

[manipopopo]

What does this PR do?

Fixes #8507

With this PR, the observers belonging to fake-quantization modules (torch.quantization.FakeQuantizeBase) will be disabled by QuantizationAwareTraining during validating, testing and predicting stages by default. The fake-quantization modules can use observers to do calibration only in the stages listed in observer_enabled_stages, which is ("train",) by default.

QuantizationAwareTraining don't disable observers during the sanity check as the model hasn't been calibrated with training data yet. After the sanity check, the fake-quantization modules are restored to initial states.

Note that we only handle observers belonging to fake-quantization modules.

Does your PR introduce any breaking changes ? If yes, please list them.

The quantization aware training observers belonging to fake-quantization modules (torch.quantization.FakeQuantizeBase) will be disabled during validating, testing and predicting stages by default. The old behavior can be recovered by passing observer_enabled_stages=("train", "validate", "test", "predict") to pytorch_lightning.callbacks.QuantizationAwareTraining.

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)
• Did you list all the breaking changes introduced by this pull request?

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 #8540 (fdf161b) into master (f95ba20) will increase coverage by 0%.
The diff coverage is 100%.

@@          Coverage Diff           @@
##           master   #8540   +/-   ##
======================================
  Coverage      89%     89%           
======================================
  Files         182     182           
  Lines       16111   16165   +54     
======================================
+ Hits        14273   14327   +54     
  Misses       1838    1838           

[Borda]

pls add typing to all impemented hooks

[Borda]

last suggestion, then it looks good to me 🐰