Texar-PyTorch v0.1.2

New features

1. Integrated texar-pytorch with NNI and AdaptDL for distributed adaptive API (#331)
2. Integrated with NNI for hyperparameter tuning (#324)
3. Add Information Loss based on KL divergence (#328)
4. Allow WordpieceTokenizer to return the original character spans (#332)
5. Add a few modules: RNN Classifier (#303), SpanBERT (#300)

Feature improvements:

1. Fix a few documentation issues.

Fixes

1. Fix a type error in Beam Search top-k index (#330)
2. Fix a file operation error in Executor (#323)
3. Fix some evaluator bugs and related file handling (#320)
4. Fix a problem where metrics cannot be pickled (#319)