Updating the version from 1.2.0rc2 to 1.2.0.

Cherrypicks

Bugfix: Don't ignore logical names in single tensor predict signatures. User-supplied names in predict signatures were silently ignored and replaced with constants.

PiperOrigin-RevId: 158481505

Also redirect TensorBoard usage to use that endpoint.

This is part of my ongoing effort to have TensorBoard only
depend on TensorFlow via its public api, so that it can
be split into a project with a fast external build.

PiperOrigin-RevId: 157784552

Since TensorProtos are part of the TensorFlow API, it makes sense
to also include the methods that generate and parse them.

Similarly, we write out MetaGraphDef protos in the summary writer,
so we should provide the proto as well.

This is part of an ongoing effort to have TensorBoard only consume
TensorFlow methods through the public api.

PiperOrigin-RevId: 157657564

* Revert "Allow tensor as iou_threshold parameter to tf.image.non_max_suppression. (#9887)"

This reverts commit 513b1e4b.

It needs API review before going into an official release.

* Update goldens

Remove RewriterConfig from the Python API

Fixing the broken 1.2 branch tests.

* Fix numpy 1.13 incompatibilities

* Skip tests with numpy 1.13.0

Updating tag to 1.2.0-rc2.

Cherrypicks for 1.1.0-rc2

* Python 3.6 support on windows.

* Fix typo in README.md

* Make environment configurable for windows gpu build.

The issue is that python's GC has trouble collecting objects with __del__ methods.

The solution is two pronged:
* Keep track of usage state outside of the class, via a dict mapping
  id(object) => state
* Remove __del__ (this was the source: python's GC couldn't collect wrapped
  objects), and instead use weakref.finalize to emit warnings just as the object
  is being garbage collected.
* Added tests for garbage collection [they were failing before i fixed the issue]

PiperOrigin-RevId: 158042388

PiperOrigin-RevId: 158025196

This allows it to properly tile the initial wrapper state when using
BeamSearchDecoder with AttentionWrapper.  Unit tests updated to show this use.

PiperOrigin-RevId: 157903115

This prevents the op from consuming an inter-op thread pool thread
when blocked, and fixes a potential deadlock when many IteratorGetNext
ops are blocked. Fixes #10369.

PiperOrigin-RevId: 157878885

Re-enable tensor_array_ops_test in msan.

PiperOrigin-RevId: 157841785

PiperOrigin-RevId: 157781368

Add type error to start_queue_runners if given session is not a `tf.Session`. Due to semver, we suppress the error if a MonitoredSession is provided.

PiperOrigin-RevId: 157748375

PiperOrigin-RevId: 157722003

Snappy compression (and decompression) was enabled after the
1.1 release (in commit 63b2f999).
This means that checkpoints produced by the 1.2.0 release candidates
will cause TensorFlow 1.1 (and prior) binaries to crash as they
CHECK fail when trying to load snappy-compressed tables.

To ease transition, disable writing of compressed checkpoints in
1.2.0 for now.

Reconsider this in the next release.

PiperOrigin-RevId: 157675189

* colocate_with is now set properly when a TensorArray is passed through a
  while_loop
* added a new argument, "colocate_with_first_write" (default: True; this is
  the current behavior).  If False, the TensorArray is simply placed on the
  device from the context it's constructed in, and no colocation constraints
  are added.

PiperOrigin-RevId: 157643133

Motivation:
We'd like to evolve the checkpoint format over time (e.g., enable
different types of compression). Without this change, a TensorFlow
version that encounters a format that it doesn't understand would CHECK fail
with an unhelpful error message.

With this, it propagates a clearer error message up, giving the user some
hints about what could be wrong.

I don't have a unittest for this - I thought about writing a bundle and
then strategically corrupting the bytes on disk before reading it back,
but that seems a bit much. The intention of this change is to enable
graceful reporting of forward compatibility breakages. Ideas for an
appropriate unittest are appreciated.

PiperOrigin-RevId: 157620358

Updated unit tests to contain inputs of unknown shape (at graph build time).
Found an issue in the gather helper that stops it from properly propagating
the batch size of the output shape.  This caused problems with tf.while_loop.
Fixed.

PiperOrigin-RevId: 157533937

PiperOrigin-RevId: 157520123

PiperOrigin-RevId: 157487623

Implementation by Cinjon Resnick.  He can't push this since he's traveling.
I just copied the fix and added some small syntax tweaks to make the unit
tests pass.  More comprehensive unit tests will come in the near future.

Fixes at least part of #9904.

BeamSearchDecoder:
1. Fix the bug where we don't pass the next cell state through.
2. Gather the cell state (and attention if that's a part of the model
   as an AttentionWrapper on the cell) according to the next_beam_ids.
PiperOrigin-RevId: 157415564