GluonNLP 0.4 release tasks

[szha]

We're targeting GluonNLP 0.4 release on 8/17 in preparation for the upcoming KDD tutorials.
@sxjscience @szhengac @leezu
@cgraywang @astonzhang @eric-haibin-lin
@piiswrong @mli

Here's the proposed scope. Please comment if you'd like to add or drop any item:

• Add transformer notebook #279 (@cgraywang)
• reproducing AWD model #199 Add nt-asgd for language model #170 (@cgraywang)
• new language model parameters on wikitext2 (@cgraywang)
• word_language_model.py is broken on StandardRNN #298 fix docs #299 (@cgraywang @szha)
• Verify machine translation scripts and examples (@sxjscience @szhengac)
• Simplify shared dataloader based on recent update on mxnet dataloader #285 (@szhengac)
• update gbw model after transform API refactoring #294 (@eric-haibin-lin)
• Sharded dataloader causing CI hangs #274 Update sharded dataloader based on the change of dataloader #280 (@szhengac)
• Add symbolic beam search #233 (@junrushao1994)
• Unify bptt_batchify interface between LanguageModelStream and LanguageModelDataset #245 Unifybpttbatchifydatasetstream #246 (@leezu)
• Verify language model scripts and examples (@cgraywang, @eric-haibin-lin, @leezu)
• Verify embedding scripts and examples (@leezu)
• Unify bptt_batchify interface between LanguageModelStream and LanguageModelDataset #245 Don't treat language model datasets special #128 Transform refactor #282 (@leezu)
  • GBW vocab needs update #291 (@leezu)
• Move custom cython operator to MXNet #262 Replace cython sampler with mxnet sampler to improve user experience #270 (@eric-haibin-lin)
• fix big rnn eval hybridization and sparse_grad=False #277 Unify BigRNN and SampledLogits interface #278 (@eric-haibin-lin)
• Update contributors list (@szha)
• apply_weight_drop required changes to integrate mxnet 1.3 #264 fix weight drop and add tests #268 (@szha)
• Pypi Readme doesn't render well #205 fix relative links that pypi doesn't handle #293 (@szha)
• Sampler in stochastic beam search / sampling #172 (@hhexiy @szha)

To our contributors: if you have work you'd like to get into GluonNLP 0.4 and you can commit to finish by 8/16, you can comment here and let us know too. @sravanbabuiitm @xiaotinghe @dofuer99 @junrushao1994 @congchan

[junrushao]

@szha Symbolic beam search #233 is done and have passed unittests. It will be ready to be merged once #264 is done.

[szha]

We got text classification based on fasttext thanks to @sravanbabuiitm

[szha]

Pinging everyone again. The cutoff is today.

[szha]

0.4 has been released. We will track the unfinished items separately.