[FIX] Update BERTLayerNorm Implementation

[eric-haibin-lin]

Description

apache/mxnet#13732
There was a typo in the nn.LayerNorm documentation. It turns out that we can safely reuse nn.LayerNorm which is faster than the implementation in GluonNLP.
As a sanity check, I run compare_tf_gluon_model.py again with bert_24_1024_16/bert_12_768_12 on book_corpus_wiki_en_uncased/book_corpus_wiki_en_cased and the test passes.

@fierceX @haven-jeon

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

[mli]

Job PR-485/3 is complete.
Docs are uploaded to http://gluon-nlp-staging.s3-accelerate.dualstack.amazonaws.com/PR-485/3/index.html

[codecov]

Codecov Report

Merging #485 into master will increase coverage by 0.43%.
The diff coverage is 100%.

@@            Coverage Diff             @@
##           master     #485      +/-   ##
==========================================
+ Coverage   71.02%   71.46%   +0.43%     
==========================================
  Files         119      118       -1     
  Lines       10045     9937     -108     
==========================================
- Hits         7134     7101      -33     
+ Misses       2911     2836      -75

Flag	Coverage Δ
#PR470	?
#PR482	?
#PR485	71.46% <100%> (?)
#master	?
#notserial	46.91% <66.66%> (+0.2%)	⬆️
#py2	71.21% <100%> (+0.43%)	⬆️
#py3	71.25% <100%> (-0.07%)	⬇️
#serial	56.97% <100%> (+0.5%)	⬆️

[codecov]

Codecov Report

Merging #485 into master will increase coverage by 0.43%.
The diff coverage is 100%.

@@            Coverage Diff             @@
##           master     #485      +/-   ##
==========================================
+ Coverage   71.02%   71.46%   +0.43%     
==========================================
  Files         119      118       -1     
  Lines       10045     9937     -108     
==========================================
- Hits         7134     7101      -33     
+ Misses       2911     2836      -75

Flag	Coverage Δ
#PR470	?
#PR482	?
#PR485	71.46% <100%> (?)
#master	?
#notserial	46.91% <66.66%> (+0.2%)	⬆️
#py2	71.21% <100%> (+0.43%)	⬆️
#py3	71.25% <100%> (-0.07%)	⬇️
#serial	56.97% <100%> (+0.5%)	⬆️

[codecov]

Codecov Report

Merging #485 into master will increase coverage by 0.43%.
The diff coverage is 100%.

@@            Coverage Diff             @@
##           master     #485      +/-   ##
==========================================
+ Coverage   71.02%   71.46%   +0.43%     
==========================================
  Files         119      118       -1     
  Lines       10045     9937     -108     
==========================================
- Hits         7134     7101      -33     
+ Misses       2911     2836      -75

Flag	Coverage Δ
#PR470	?
#PR482	?
#PR485	71.46% <100%> (?)
#master	?
#notserial	46.91% <66.66%> (+0.2%)	⬆️
#py2	71.21% <100%> (+0.43%)	⬆️
#py3	71.25% <100%> (-0.07%)	⬇️
#serial	56.97% <100%> (+0.5%)	⬆️

[szha]

LGTM assuming that the specialized layer norm block is for extension purpose