[Refactor]Add a switch for attention to return an unnormalized weight matrix. Move _get_attention_cell function position

[fierceX]

Description

Added the option for attention to return an unnormalized weight matrix.
This will be used for tinybert training.
Moved the position of the _get_attention_cell function from seq2seq_encoder_decoder to attention_cell,which I think makes more sense.

Checklist

Essentials

• PR's title starts with a category (e.g. [BUGFIX], [MODEL], [TUTORIAL], [FEATURE], [DOC], etc)
• Changes are complete (i.e. I finished coding on this PR)
• All changes have test coverage
• Code is well-documented

Changes

• Feature1, tests, (and when applicable, API doc)
• Feature2, tests, (and when applicable, API doc)

Comments

• If this change is a backward incompatible change, why must this change be made.
• Interesting edge cases to note here

cc @dmlc/gluon-nlp-team

[codecov]

Codecov Report

Merging #1007 into master will decrease coverage by 14.54%.
The diff coverage is 65%.

@@             Coverage Diff             @@
##           master    #1007       +/-   ##
===========================================
- Coverage   89.93%   75.39%   -14.55%     
===========================================
  Files          67       67               
  Lines        6340     6344        +4     
===========================================
- Hits         5702     4783      -919     
- Misses        638     1561      +923

Impacted Files	Coverage Δ
src/gluonnlp/model/seq2seq_encoder_decoder.py	81.81% <ø> (+6.43%)	⬆️
src/gluonnlp/model/transformer.py	91.34% <100%> (+0.02%)	⬆️
src/gluonnlp/model/attention_cell.py	68.85% <64.1%> (-27.38%)	⬇️
src/gluonnlp/data/batchify/embedding.py	17.69% <0%> (-80%)	⬇️
src/gluonnlp/data/candidate_sampler.py	16.32% <0%> (-75.52%)	⬇️
src/gluonnlp/optimizer/lamb.py	23.8% <0%> (-66.67%)	⬇️
src/gluonnlp/utils/parallel.py	34.04% <0%> (-63.83%)	⬇️
src/gluonnlp/model/sequence_sampler.py	11.49% <0%> (-63.07%)	⬇️
src/gluonnlp/optimizer/bert_adam.py	34.09% <0%> (-59.1%)	⬇️
src/gluonnlp/utils/parameter.py	23.72% <0%> (-57.63%)	⬇️
... and 28 more

[mli]

Job PR-1007/1 is complete.
Docs are uploaded to http://gluon-nlp-staging.s3-accelerate.dualstack.amazonaws.com/PR-1007/1/index.html

[leezu]

Thanks @fierceX. Are you aware of #910 ? Do you have any thoughts on integrating it?

[mli]

Job PR-1007/2 is complete.
Docs are uploaded to http://gluon-nlp-staging.s3-accelerate.dualstack.amazonaws.com/PR-1007/2/index.html

[mli]

Job PR-1007/3 is complete.
Docs are uploaded to http://gluon-nlp-staging.s3-accelerate.dualstack.amazonaws.com/PR-1007/3/index.html

[leezu]

@fierceX ping regarding integration with #910

[fierceX]

@leezu Okay, I'll take a look, but the follow-up to tinybert may not be too fast, because I'm a little busy right now.