Benjamin Kramer [Wed, 11 Apr 2018 09:16:25 +0000 (02:16 -0700)]
[TF:XLA] Mark oom_test as optonly, it's really slow when compiled without optimization.
PiperOrigin-RevId:
192420481
Nick Felt [Wed, 11 Apr 2018 06:44:12 +0000 (23:44 -0700)]
Create FileWriter <-> tf.contrib.summary compatibility layer
This provides an implementation of FileWriter, activated by passing in a `session` parameter to the constructor, that is backed by session.run'ing graph ops that manipulate a tf.contrib.summary.create_file_writer() instance. Because tf.contrib.summary.SummaryWriters are backed by shared resources in the graph, this makes it possible to have a FileWriter and a tf.contrib.summary.SummaryWriter that both write to the same events file.
This change includes some related smaller changes:
- Factors out training_utils.py into a separate target to avoid a cyclic dep
- Moves contrib/summary/summary_ops.py to python/ops/summary_ops_v2.py
- Adds SummaryWriter.init(), .flush(), and .close() op-returning methods
- Changes create_file_writer() `name` arg to default to logdir prefixed by `logdir:` so shared resources are scoped by logdir by default
- Fixes a bug with tf.contrib.summary.flush() `writer` arg
- Makes create_file_writer()'s max_queue arg behave as documented
- Adds more testing for existing tf.contrib.summary API
PiperOrigin-RevId:
192408079
A. Unique TensorFlower [Wed, 11 Apr 2018 05:44:36 +0000 (22:44 -0700)]
[XLA] Fix the size of data buffer for sparse literals.
PiperOrigin-RevId:
192404543
A. Unique TensorFlower [Wed, 11 Apr 2018 05:29:13 +0000 (22:29 -0700)]
[XLA] Redesign: implement ComputeHost.
Also support convert from/to proto for ComputeHost.
PiperOrigin-RevId:
192403660
A. Unique TensorFlower [Wed, 11 Apr 2018 04:42:14 +0000 (21:42 -0700)]
[XLA] Redesgin: implement and test Gather, Conditional.
Also support convert from/to proto for Gather.
PiperOrigin-RevId:
192400659
A. Unique TensorFlower [Wed, 11 Apr 2018 03:48:57 +0000 (20:48 -0700)]
[XLA] Redesign: implment and test CrossReplicaSum.
PiperOrigin-RevId:
192397189
A. Unique TensorFlower [Wed, 11 Apr 2018 02:44:00 +0000 (19:44 -0700)]
Update ops-related pbtxt files.
PiperOrigin-RevId:
192392702
Justin Lebar [Wed, 11 Apr 2018 02:34:54 +0000 (19:34 -0700)]
[StreamExecutor] Merge StreamExecutor's and XLA's StatusOr classes.
StatusOr is a...complicated class to write. It's really not good to
have two copies of it. They've diverged (the XLA one is more
sophisticated), and this may be causing upstream build problems with
gcc6.
PiperOrigin-RevId:
192392111
A. Unique TensorFlower [Wed, 11 Apr 2018 02:31:05 +0000 (19:31 -0700)]
[XLA] Redesign: implement and test BatchNormXXX.
PiperOrigin-RevId:
192391748
Asim Shankar [Wed, 11 Apr 2018 02:20:58 +0000 (19:20 -0700)]
experimental C API: Fix compilation failure in Windows.
The functions added in
https://github.com/tensorflow/tensorflow/commit/
be917027e37c5e8f21f6ba07f24bdbf072cf6dfd
are temporary, and their existence breaks compilation in MSVC because of
https://docs.microsoft.com/en-us/cpp/c-language/maximum-string-length
and
https://docs.microsoft.com/en-us/cpp/error-messages/compiler-errors-1/compiler-error-c2026
So just disabling it in Windows for now.
PiperOrigin-RevId:
192391164
Allen Lavoie [Wed, 11 Apr 2018 02:01:33 +0000 (19:01 -0700)]
Checkpointable: remove colocation constraints from restore ops
Mystery solved thanks to log_device_placement.
PiperOrigin-RevId:
192389574
Francois Chollet [Wed, 11 Apr 2018 01:50:58 +0000 (18:50 -0700)]
Remove BN workaround for resource variable gradients bug that was recently fixed.
PiperOrigin-RevId:
192388867
Michael Case [Wed, 11 Apr 2018 01:44:13 +0000 (18:44 -0700)]
Merge changes from github.
PiperOrigin-RevId:
192388250
Francois Chollet [Wed, 11 Apr 2018 01:41:37 +0000 (18:41 -0700)]
Fix for users who were passing `Dimension` type as `units` arg in `Dense`.
PiperOrigin-RevId:
192387984
Allen Lavoie [Wed, 11 Apr 2018 01:04:20 +0000 (18:04 -0700)]
TFTS: De-flake the LSTM test
Disabling the value-based check for now. Hopefully the shapes are deterministic.
PiperOrigin-RevId:
192383553
Allen Lavoie [Wed, 11 Apr 2018 00:57:19 +0000 (17:57 -0700)]
Forward the status from LookupResource to GetInputTensorFromVariable rather than returning a generic error status
PiperOrigin-RevId:
192382499
A. Unique TensorFlower [Wed, 11 Apr 2018 00:49:51 +0000 (17:49 -0700)]
[XLA] Redesign: implement Conj.
PiperOrigin-RevId:
192381481
A. Unique TensorFlower [Wed, 11 Apr 2018 00:47:13 +0000 (17:47 -0700)]
[XLA] Redesign: implement SliceInDim.
PiperOrigin-RevId:
192381080
A. Unique TensorFlower [Wed, 11 Apr 2018 00:44:48 +0000 (17:44 -0700)]
[XLA] Redesign: implement ReduceAll.
PiperOrigin-RevId:
192380688
A. Unique TensorFlower [Wed, 11 Apr 2018 00:41:56 +0000 (17:41 -0700)]
[XLA] Redesign: implement and test SelectAndScatter.
PiperOrigin-RevId:
192380121
Austin Anderson [Wed, 11 Apr 2018 00:37:53 +0000 (17:37 -0700)]
Update LogToSTDErr for TF Lite usage
PiperOrigin-RevId:
192379483
Jacques Pienaar [Tue, 10 Apr 2018 23:47:08 +0000 (16:47 -0700)]
Add types to error message in case of mismatch. NFC.
PiperOrigin-RevId:
192370979
Michael Kuperstein [Tue, 10 Apr 2018 23:45:19 +0000 (16:45 -0700)]
Add a thread-safe producer-consumer queue.
PiperOrigin-RevId:
192370670
Alexandre Passos [Tue, 10 Apr 2018 23:35:58 +0000 (16:35 -0700)]
Fix issue with gradients of resource variables in cond.
PiperOrigin-RevId:
192369091
A. Unique TensorFlower [Tue, 10 Apr 2018 23:32:05 +0000 (16:32 -0700)]
[XLA] Redesign: implement and test ReduceWindow.
PiperOrigin-RevId:
192368401
A. Unique TensorFlower [Tue, 10 Apr 2018 23:31:46 +0000 (16:31 -0700)]
Update programmers guide
PiperOrigin-RevId:
192368335
A. Unique TensorFlower [Tue, 10 Apr 2018 23:26:05 +0000 (16:26 -0700)]
Supporting FakeQuant num_bits and getting the fake quant op matching tensorflow.
PiperOrigin-RevId:
192367307
Chris Leary [Tue, 10 Apr 2018 23:12:19 +0000 (16:12 -0700)]
[XLA] GRPC service definition.
PiperOrigin-RevId:
192364932
Francois Chollet [Tue, 10 Apr 2018 23:11:38 +0000 (16:11 -0700)]
Fix `nn` module RNN namespace issues.
PiperOrigin-RevId:
192364808
Mingxing Tan [Tue, 10 Apr 2018 22:59:39 +0000 (15:59 -0700)]
Allow passing allow_custom_ops for toco_convert.
PiperOrigin-RevId:
192362688
Alexandre Passos [Tue, 10 Apr 2018 22:49:03 +0000 (15:49 -0700)]
Cleaning up _distributed_apply now the device policy is unnecessary
PiperOrigin-RevId:
192360913
Akshay Agrawal [Tue, 10 Apr 2018 22:48:15 +0000 (15:48 -0700)]
In `get_variable`, nest the choice to use `ResourceVariable` under an
`init_scope`.
This makes sure that, when executing eagerly, calls to `get_variable` in a
`defun`-compiled function retrieve `ResourceVariable`s instead of `Variables`.
PiperOrigin-RevId:
192360775
A. Unique TensorFlower [Tue, 10 Apr 2018 22:45:37 +0000 (15:45 -0700)]
Small changes to testing code, plus a new binary to check diff from command line.
PiperOrigin-RevId:
192360373
A. Unique TensorFlower [Tue, 10 Apr 2018 22:39:37 +0000 (15:39 -0700)]
[XLA] Redesign: implement and test Conv.
PiperOrigin-RevId:
192359226
A. Unique TensorFlower [Tue, 10 Apr 2018 22:23:05 +0000 (15:23 -0700)]
Allow negative feature values in computation for `sum` combiner.
PiperOrigin-RevId:
192355950
A. Unique TensorFlower [Tue, 10 Apr 2018 22:01:49 +0000 (15:01 -0700)]
Add quantized LogSoftmax.
PiperOrigin-RevId:
192352432
A. Unique TensorFlower [Tue, 10 Apr 2018 21:51:54 +0000 (14:51 -0700)]
Run EvaluateNodes for ModelPrunerTest_StopGradientPruning. Also updated the test fixture to inherit from GrapplerTest.
PiperOrigin-RevId:
192350828
Yu-Cheng Ling [Tue, 10 Apr 2018 21:24:51 +0000 (14:24 -0700)]
ParseOpData returns kTfLiteError when error happens.
PiperOrigin-RevId:
192346224
A. Unique TensorFlower [Tue, 10 Apr 2018 21:22:33 +0000 (14:22 -0700)]
Bug fix for statistical_testing:
- Max/Min computations should be done over the sample dimension.
- Change dominate check to be greater_equal instead of greater (for matching dimensions).
PiperOrigin-RevId:
192345809
A. Unique TensorFlower [Tue, 10 Apr 2018 21:16:36 +0000 (14:16 -0700)]
Add missing import for RNNClassifier
PiperOrigin-RevId:
192344760
A. Unique TensorFlower [Tue, 10 Apr 2018 21:04:29 +0000 (14:04 -0700)]
[XLA] Redesign: implement and test Rev, BitcastConvertType, Map, and ReducePrecision.
PiperOrigin-RevId:
192342686
Suharsh Sivakumar [Tue, 10 Apr 2018 21:02:02 +0000 (14:02 -0700)]
Pad support for quantized zero.
PiperOrigin-RevId:
192342172
A. Unique TensorFlower [Tue, 10 Apr 2018 20:59:49 +0000 (13:59 -0700)]
test previously untested eval codepaths.
PiperOrigin-RevId:
192341561
A. Unique TensorFlower [Tue, 10 Apr 2018 20:57:00 +0000 (13:57 -0700)]
Handle duplicate features by coalescing them together into a single feature.
PiperOrigin-RevId:
192341065
Francois Chollet [Tue, 10 Apr 2018 20:49:37 +0000 (13:49 -0700)]
Refactor layers:
- tf.layers layers now subclasses tf.keras.layers layers.
- tf.keras.layers is now agnostic to variable scopes and global collections (future-proof). It also uses ResourceVariable everywhere by default.
- As a result tf.keras.layers is in general lower-complexity, with fewer hacks and workarounds. However some of current code is temporary (variable creation should be moved to Checkpointable, arguably, and there are some dependency issues that will require later refactors).
- The legacy tf.layers layers behavior is kept, with references to variable scopes and global collections injected in the subclassed tf.layers.base.Layer class (the content of tf.layers.base.Layer is the complexity differential between the old implementation and the new one).
Note: this refactor does slightly change the behavior of tf.layers.base.Layer, by disabling extreme edge-case behavior that either has long been invalid, or is dangerous and should most definitely be disabled. This will not affect any users since such behaviors only existed in the base Layer unit tests. The behaviors disabled are:
- Option to create reusable variables in `call` (already invalid for some time).
- Option to use a variable scope to create layer variables outside of the layer while not having the layer track such variables locally.
PiperOrigin-RevId:
192339798
Smit Hinsu [Tue, 10 Apr 2018 20:32:38 +0000 (13:32 -0700)]
Update declaration order in staging ops Buffer class according to C++ style guide
PiperOrigin-RevId:
192336966
Allen Lavoie [Tue, 10 Apr 2018 19:54:03 +0000 (12:54 -0700)]
Checkpointable: wrap restore ops in init_scope
This should make restore() work with defun-wrapped code, when variables are
created inside the function. Just lifts the restore code into the outer
context. Adds a test for it.
PiperOrigin-RevId:
192331065
A. Unique TensorFlower [Tue, 10 Apr 2018 19:28:56 +0000 (12:28 -0700)]
Employ array flat sizes more directly in reference_ops.
PiperOrigin-RevId:
192327464
Noah Eisen [Tue, 10 Apr 2018 19:28:04 +0000 (12:28 -0700)]
Upgrade gRPC version and fix file duplication
This bumps the gRPC version used in OSS Tensorflow to pick up grpc/grpc#14541, which exposes gRPC serialization classes which were previously hidden in an internal namespace. Using these files eliminates files duplicated from gRPC repo
PiperOrigin-RevId:
192327358
A. Unique TensorFlower [Tue, 10 Apr 2018 19:05:06 +0000 (12:05 -0700)]
Expand list of value-preserving ops. This will increase the number of graphs where we can apply the involution and mul->conv fusion optimizations.
PiperOrigin-RevId:
192323712
A. Unique TensorFlower [Tue, 10 Apr 2018 18:57:40 +0000 (11:57 -0700)]
Enable loop-invariant node motion in the Grappler loop optimizer.
Thanks to the team at Alibaba, who contributed the original version of this code.
PiperOrigin-RevId:
192322484
A. Unique TensorFlower [Tue, 10 Apr 2018 18:33:45 +0000 (11:33 -0700)]
Fix bug in TFLite Interpreter python interface
PiperOrigin-RevId:
192318426
A. Unique TensorFlower [Tue, 10 Apr 2018 18:20:38 +0000 (11:20 -0700)]
[XLA] Fix the size of the data returned from Literal for sparse literals.
PiperOrigin-RevId:
192315888
A. Unique TensorFlower [Tue, 10 Apr 2018 18:09:37 +0000 (11:09 -0700)]
Adds support for hoisting out common denominator in arithmetic_optimizer
PiperOrigin-RevId:
192314177
Nick Desaulniers [Tue, 10 Apr 2018 17:52:15 +0000 (10:52 -0700)]
[TF:XLA] fix a segfault in MakeFakeArguments, and add a test case.
PiperOrigin-RevId:
192310749
Asim Shankar [Tue, 10 Apr 2018 17:43:14 +0000 (10:43 -0700)]
Simplify test_util.run_in_graph_and_eager_modes
- Get rid of unnecessary options
- Update various resource variable tests so that they correctly exercise the cases where the variables are placed on GPU (these "with tf.device('/cpu:0')" blocks that were added for eager execution are no longer necessary)
PiperOrigin-RevId:
192309109
Yao Zhang [Tue, 10 Apr 2018 17:07:31 +0000 (10:07 -0700)]
Place data format op on CPU:0.
PiperOrigin-RevId:
192302833
A. Unique TensorFlower [Tue, 10 Apr 2018 16:50:02 +0000 (09:50 -0700)]
Fix markdown in a couple of tf.estimator docstrings.
PiperOrigin-RevId:
192299871
A. Unique TensorFlower [Tue, 10 Apr 2018 16:41:45 +0000 (09:41 -0700)]
Remove manifest_merger from tensorflow_demo. This is an internal-only attribute
that is being removed from bazel.
PiperOrigin-RevId:
192298746
Mark Daoust [Tue, 10 Apr 2018 16:40:25 +0000 (09:40 -0700)]
Fix capitalization (on master)
PiperOrigin-RevId:
192298563
Benjamin Kramer [Tue, 10 Apr 2018 16:26:59 +0000 (09:26 -0700)]
[XLA:GPU] Add infrastructure for unrolling kernels to improve bandwidth utilization.
We often have simple kernels that do very little actual work, duplicating that
can increase the used bandwidth.
This change introduces flags and infrastructure for unrolling kernels, it
doesn't include any cost heuristics and is disabled by default. Based on code
written by Bixia Zheng.
PiperOrigin-RevId:
192296781
A. Unique TensorFlower [Tue, 10 Apr 2018 15:55:24 +0000 (08:55 -0700)]
Added minimum op, better type support in maximum.
PiperOrigin-RevId:
192292693
A. Unique TensorFlower [Tue, 10 Apr 2018 15:51:18 +0000 (08:51 -0700)]
Update document
PiperOrigin-RevId:
192292160
A. Unique TensorFlower [Tue, 10 Apr 2018 14:47:14 +0000 (07:47 -0700)]
Suppress -Wself-assign in self-assignment tests, which triggers in newer clang
revisions.
PiperOrigin-RevId:
192284946
A. Unique TensorFlower [Tue, 10 Apr 2018 06:13:11 +0000 (23:13 -0700)]
[XLA] Redesign: implement and test custom call.
PiperOrigin-RevId:
192241311
A. Unique TensorFlower [Tue, 10 Apr 2018 05:04:04 +0000 (22:04 -0700)]
[XLA] Redesign: implement XlaBuilder::IsConstant, XlaBuidler::BuildConstantSubGraph, and Client::ComputeConstant(XlaComputation...).
- Since the builder no longer holds a client, we moved the ComputeConstant to the client side so that it can communicate with the service side. Now we add XlaBuilder::BuildConstantSubGraph, which is only responsible for building a subgraph that is compile-time constant.
- Before this change, every XlaBuilder has a unique id. Now since it also builds constant subgraph, we give every XlaComputation being built a global unique id, and uniquify instruction names when actually building the XlaComputation.
PiperOrigin-RevId:
192236997
A. Unique TensorFlower [Tue, 10 Apr 2018 05:01:59 +0000 (22:01 -0700)]
Implementation of ArgMax
PiperOrigin-RevId:
192236845
A. Unique TensorFlower [Tue, 10 Apr 2018 04:36:36 +0000 (21:36 -0700)]
Updating three tests in constant_folding_test.cc with PlaceHolders Nodes to check EvaluateNodes returns the same output for the original and optimized graph
PiperOrigin-RevId:
192235310
Michael Case [Tue, 10 Apr 2018 03:26:16 +0000 (20:26 -0700)]
Fix the GCS Kokoro build badge links.
The original badges I added were redirected/authenticated links (I think). This
resulted in broken images on the README. These new links should not be
redirected and should just link to the badge images.
PiperOrigin-RevId:
192230689
A. Unique TensorFlower [Tue, 10 Apr 2018 02:46:01 +0000 (19:46 -0700)]
Go: Update generated wrapper functions for TensorFlow ops.
PiperOrigin-RevId:
192227995
A. Unique TensorFlower [Tue, 10 Apr 2018 02:19:38 +0000 (19:19 -0700)]
Update ops-related pbtxt files.
PiperOrigin-RevId:
192226063
A. Unique TensorFlower [Tue, 10 Apr 2018 02:06:05 +0000 (19:06 -0700)]
[XLA] Attach a reference client to the client_library_test_base, and implement ComputeAndCompare for XlaBuilder.
- In ComputeAndCompare(XlaBuilder..), compute the reference result by executing on the interpreter backend.
- Also migrate some tests to use the new ComputeAndCompare(XlaBuilder..).
PiperOrigin-RevId:
192225152
A. Unique TensorFlower [Tue, 10 Apr 2018 01:53:22 +0000 (18:53 -0700)]
Include standard headers when libc++ requires them.
PiperOrigin-RevId:
192224104
A. Unique TensorFlower [Tue, 10 Apr 2018 01:46:09 +0000 (18:46 -0700)]
Add the total device time and total host time to tf_op_stats.
PiperOrigin-RevId:
192223560
A. Unique TensorFlower [Tue, 10 Apr 2018 01:08:13 +0000 (18:08 -0700)]
[XLA] Redesign: implement fft.
PiperOrigin-RevId:
192220070
Rachel Lim [Tue, 10 Apr 2018 01:03:57 +0000 (18:03 -0700)]
[tf.data] Add parameter for select_cols in tf.decode_csv for option to parse only a subset of CSV columns; added same parameter to make_csv_dataset function in tf.contrib.data.
PiperOrigin-RevId:
192219509
Younghee Kwon [Tue, 10 Apr 2018 00:45:13 +0000 (17:45 -0700)]
boosted_trees: early stop hooks are fixed to stop at the right moment
by reading tensor values in a separate session after train_op run.
PiperOrigin-RevId:
192217338
Skye Wanderman-Milne [Tue, 10 Apr 2018 00:40:46 +0000 (17:40 -0700)]
Remove extra #define EIGEN_USE_THREADS.
It breaks some builds with
error: "EIGEN_USE_THREADS" redefined [-Werror]
PiperOrigin-RevId:
192216796
A. Unique TensorFlower [Tue, 10 Apr 2018 00:36:41 +0000 (17:36 -0700)]
[slim] Allow passing timeout_fn to 'evaluation_loop' so that caller can have more control over what to do if checkpoints_iterator times out.
PiperOrigin-RevId:
192216302
Benoit Steiner [Tue, 10 Apr 2018 00:30:27 +0000 (17:30 -0700)]
Don't fail when running shape inference on a graph that contains functions.
PiperOrigin-RevId:
192215493
Patrick Nguyen [Mon, 9 Apr 2018 23:55:06 +0000 (16:55 -0700)]
Export recurrent and its RNN implementation in tf.contrib.
PiperOrigin-RevId:
192210794
A. Unique TensorFlower [Mon, 9 Apr 2018 23:43:02 +0000 (16:43 -0700)]
Internal Change
PiperOrigin-RevId:
192209093
A. Unique TensorFlower [Mon, 9 Apr 2018 23:24:56 +0000 (16:24 -0700)]
[XLA] Redesign: implement infeed and outfeed.
- XlaBuilder::Infeed is basically ComputationBuilder::Infeed + UserComputation::AddInfeedInstruction + ComputationLowerer::Visit + HloInstruction::CreateInfeed.
- Similar for Outfeed.
PiperOrigin-RevId:
192206502
Anna R [Mon, 9 Apr 2018 23:09:09 +0000 (16:09 -0700)]
Hide slide_dataset from the new API.
PiperOrigin-RevId:
192204209
Mark Heffernan [Mon, 9 Apr 2018 22:22:18 +0000 (15:22 -0700)]
Add bitcast-convert support to the evaluator and as a method on Literal.
PiperOrigin-RevId:
192197163
Michael Kuperstein [Mon, 9 Apr 2018 22:02:21 +0000 (15:02 -0700)]
[XLA] Use ThreadPool in a safer way.
ThreadPool joins its threads when it is destroyed, and there's no way to explicitly join. This means passing a ThreadPool and then scheduling in the callee is risky.
PiperOrigin-RevId:
192193752
A. Unique TensorFlower [Mon, 9 Apr 2018 22:00:17 +0000 (15:00 -0700)]
Don't run test flaky under TSAN.
PiperOrigin-RevId:
192193350
Alexandre Passos [Mon, 9 Apr 2018 21:56:18 +0000 (14:56 -0700)]
Acknowledges tape in graph mode doesn't like unknown shapes
PiperOrigin-RevId:
192192757
Allen Lavoie [Mon, 9 Apr 2018 21:52:53 +0000 (14:52 -0700)]
Tweak the context stack so init_scope works with eager Graphs
Previously breaking out into Graphs created with eager execution enabled would
enter the graph but not re-enable eager execution.
PiperOrigin-RevId:
192192109
Skye Wanderman-Milne [Mon, 9 Apr 2018 21:26:55 +0000 (14:26 -0700)]
Don't initialize global threadpool in GraphRunner.
TF_Graph creates a ShapeRefiner, which in
turn creates a GraphRunner, which prior to this change would eventually create a
LocalDevice that initialized the global eigen threadpool. This prevents
users from specifying a custom number of threads for the pool via a
ConfigProto.
This change introduces a new device class, SingleThreadedCpuDevice, that can
be used for light-weight computations without initializing the threadpool.
Addresses #18300.
PiperOrigin-RevId:
192188031
Guangda Lai [Mon, 9 Apr 2018 21:15:41 +0000 (14:15 -0700)]
Fix GPUDebugAllocator and GPUNanResetAllocator:
1. Eliminate the checks when the pointer is nullptr.
2. Fix nan array size. Some application can produce a zero size which would
cause out-of-range access error.
PiperOrigin-RevId:
192186224
A. Unique TensorFlower [Mon, 9 Apr 2018 21:12:05 +0000 (14:12 -0700)]
Doccumentation fix for LossSpec.
PiperOrigin-RevId:
192185646
A. Unique TensorFlower [Mon, 9 Apr 2018 21:08:50 +0000 (14:08 -0700)]
Fix bug in loop optimizer. Reuse existing constant nodes.
PiperOrigin-RevId:
192185091
Akshay Modi [Mon, 9 Apr 2018 21:07:12 +0000 (14:07 -0700)]
Automated g4 rollback of changelist
192160407
PiperOrigin-RevId:
192184809
A. Unique TensorFlower [Mon, 9 Apr 2018 21:02:30 +0000 (14:02 -0700)]
Adds __repr__ method to HParams.
PiperOrigin-RevId:
192184000
Shashi Shekhar [Mon, 9 Apr 2018 20:57:12 +0000 (13:57 -0700)]
Make node and registration getter const.
PiperOrigin-RevId:
192183067
Dimitris Vardoulakis [Mon, 9 Apr 2018 20:42:35 +0000 (13:42 -0700)]
Add opcode for new instruction that broadcasts degenerate dimensions.
Implicit broadcasts can be translated to the new instruction instead of a reshape-and-broadcast.
Follow-up CLs will add support in UserComputation and the various backends.
PiperOrigin-RevId:
192180356
A. Unique TensorFlower [Mon, 9 Apr 2018 20:24:03 +0000 (13:24 -0700)]
Add a utility that can detect the class that defined a method. This is useful when converting a e.g. a custom Keras model, to avoid recompiling the base model methods.
PiperOrigin-RevId:
192177117
A. Unique TensorFlower [Mon, 9 Apr 2018 20:21:34 +0000 (13:21 -0700)]
Adds a within_ops_fn parameter to get_forward_walk_ops and get_backward_walk_ops
that allows setting a condition on ops that are within or not within.
Also adds tests for these methods that were missing.
PiperOrigin-RevId:
192176693
A. Unique TensorFlower [Mon, 9 Apr 2018 19:41:03 +0000 (12:41 -0700)]
Implement DFS using a loop instead of recursion to avoid stack overflow.
PiperOrigin-RevId:
192169242