GitLab

* Refactor to use distributed TF API; add driver script.

* Add explicit GPU flag so multiple cards can be used.

* Update docs to mention distributed.sh.

* Don't make use of HParams yet, since they're not available in TF1.0

Updates to adversarial_text model

Remove --no-preserve=all build_pip_package.py

This commit removes --no-preserve=all from
syntaxnet/dragnn/tools/build_pip_package.py. It was causing issues for
OSX developers, as that option is not available on OSX.

Open source release of Attention OCR

stack outputs instead of concat outputs

Open source release of adversarial crypto code.
(self-merging based upon email approval by martinwicke)

to Abadi & Andersen paper.

Fix num_residual_units for w28-10

Fix num_residual_units for w28-10

A script to generate a text file with FSNS URLs.

Fix a typo.

Improvements to several READMEs

* Updated slim README.md to include links to ResNetv2 models.

* Change v1 to V1, v2 to V2, and other minor comments.

Default is TensorFlow default of 'grpc' communication protocol.
If TensorFlow was complied with Verbs support 'grpc+verbs' can be
used to accelerate the tensor passing communication.

Variable summaries and the learning rate are added elsewhere in the
code.  A quick search also shows that this function is never called.

Add adversarial text model

Implementation for Cognitive Mapping and Planning paper.

'xrange' was used in multiple places. In this setting it provided no measurable gain in terms of performance or memory, but broke python3 compatibility. It was replaced with 'range'

The flag description for the momentum flag states that it is `The
momentum for the MomentumOptimizer and RMSPropOptimizer`, however its
not actually used in the RMSPropOptimizer.  Instead, a separate
`rmsprop_momentum` flag was used.  This deletes that flag for
simplicity.  It was not referenced anywhere else in the repo.