- Fixed using an average layer for only one attention head. - Fixed issue with many weights not being trained due to kwargs being passed to the parent object in the layer definitions.
1.0.2
===========
- Added RaggedTensorFeeder utensil that is able to feed a ragged tensor as data to an RNN model that can only train on non ragged batches.
1.0.1
===========
- fixing naming bug where we could not use more than one AttentionMechanism due to all such mechanisms having the same name but names need to be unique.
1.0.0
===========
- All good. Ready for release.
0.9.3
===========
- Final upload to real pypi and final test
0.9.2
===========
- fixing README.md - removed pyscaffold spam from setup.cfg - uploading to TestPyPi and testing if it installs correctly