platform/upstream/tensorflow.git
6 years agoFix inconsistency in run_cond.
A. Unique TensorFlower [Tue, 27 Mar 2018 02:30:26 +0000 (19:30 -0700)]
Fix inconsistency in run_cond.

PiperOrigin-RevId: 190563114

6 years agoAdd "distribute" argument to tf.estimator.RunConfig() in anticipation
A. Unique TensorFlower [Tue, 27 Mar 2018 02:29:48 +0000 (19:29 -0700)]
Add "distribute" argument to tf.estimator.RunConfig() in anticipation
of upcoming DistributionStrategy support in Estimator.

PiperOrigin-RevId: 190563074

6 years ago[XLA] Redesign: implement and test ternary ops.
A. Unique TensorFlower [Tue, 27 Mar 2018 02:08:15 +0000 (19:08 -0700)]
[XLA] Redesign: implement and test ternary ops.

PiperOrigin-RevId: 190561679

6 years agoFinish deprecation of tf.contrib.bayesflow.{HMC,MetropolisHastings}.
Joshua V. Dillon [Tue, 27 Mar 2018 01:50:27 +0000 (18:50 -0700)]
Finish deprecation of tf.contrib.bayesflow.{HMC,MetropolisHastings}.
New home: https://github.com/tensorflow/probability/tree/master/tensorflow_probability/python/mcmc

PiperOrigin-RevId: 190560180

6 years ago[XLA] Redesign: handle metadata and sharding.
A. Unique TensorFlower [Tue, 27 Mar 2018 00:39:51 +0000 (17:39 -0700)]
[XLA] Redesign: handle metadata and sharding.

- Add a xla.OpSharding field to the HloInstructionProto.
- Metatdata handling is tested.

PiperOrigin-RevId: 190553731

6 years agoAdd missing parameter to OP_REQUIRES call.
A. Unique TensorFlower [Tue, 27 Mar 2018 00:02:55 +0000 (17:02 -0700)]
Add missing parameter to OP_REQUIRES call.

PiperOrigin-RevId: 190548854

6 years agoSave the last loss reduction method (for future use).
A. Unique TensorFlower [Mon, 26 Mar 2018 23:19:50 +0000 (16:19 -0700)]
Save the last loss reduction method (for future use).

PiperOrigin-RevId: 190543066

6 years agoRename convert_savedmodel to convert_saved_model
Suharsh Sivakumar [Mon, 26 Mar 2018 22:52:12 +0000 (15:52 -0700)]
Rename convert_savedmodel to convert_saved_model

to be consistent with export_saved_model

PiperOrigin-RevId: 190539064

6 years agoMakes tf.gather not silently snapshot resource variables.
Alexandre Passos [Mon, 26 Mar 2018 22:39:54 +0000 (15:39 -0700)]
Makes tf.gather not silently snapshot resource variables.

PiperOrigin-RevId: 190537320

6 years agoIn the experimental C API, parametrized batch_size for the generate dataset / iterato...
Mingsheng Hong [Mon, 26 Mar 2018 22:37:40 +0000 (15:37 -0700)]
In the experimental C API, parametrized batch_size for the generate dataset / iterator stack.

PiperOrigin-RevId: 190536945

6 years agoUpdate recompute_grad for TPU
A. Unique TensorFlower [Mon, 26 Mar 2018 22:34:21 +0000 (15:34 -0700)]
Update recompute_grad for TPU

PiperOrigin-RevId: 190536468

6 years ago[XLA] Redesign: implement and test unary and binary ops.
A. Unique TensorFlower [Mon, 26 Mar 2018 22:19:29 +0000 (15:19 -0700)]
[XLA] Redesign: implement and test unary and binary ops.

Also,
- Templatized ComputeAndCompareRX and CreateRXParameter so that they accept XlaBuilder and XlaOp.
- Clear data held by an XlaBuilder when Build() is called, otherwise errors will occur when the builder is reused.

PiperOrigin-RevId: 190534245

6 years agoFix some compiler warnings in MKL-DNN build.
A. Unique TensorFlower [Mon, 26 Mar 2018 22:07:10 +0000 (15:07 -0700)]
Fix some compiler warnings in MKL-DNN build.

PiperOrigin-RevId: 190532168

6 years agoClarify eager gradient doc strings
Igor Ganichev [Mon, 26 Mar 2018 21:33:10 +0000 (14:33 -0700)]
Clarify eager gradient doc strings

PiperOrigin-RevId: 190526387

6 years agoAdd description of shapes and a pointer to external tutorial notebook in `tf.distribu...
A. Unique TensorFlower [Mon, 26 Mar 2018 21:04:35 +0000 (14:04 -0700)]
Add description of shapes and a pointer to external tutorial notebook in `tf.distributions.Distribution`.

PiperOrigin-RevId: 190521666

6 years agoQuick fix to assign_moving_average documentation formatting.
A. Unique TensorFlower [Mon, 26 Mar 2018 20:41:35 +0000 (13:41 -0700)]
Quick fix to assign_moving_average documentation formatting.

PiperOrigin-RevId: 190517622

6 years agoGlobal rename of py2tf to autograph
A. Unique TensorFlower [Mon, 26 Mar 2018 20:30:17 +0000 (13:30 -0700)]
Global rename of py2tf to autograph

PiperOrigin-RevId: 190515509

6 years agoAdd additional protobuf imports.
Jiri Simsa [Mon, 26 Mar 2018 20:25:58 +0000 (13:25 -0700)]
Add additional protobuf imports.

PiperOrigin-RevId: 190514839

6 years agoInternal change.
Anna R [Mon, 26 Mar 2018 20:15:53 +0000 (13:15 -0700)]
Internal change.

PiperOrigin-RevId: 190512928

6 years agoAdd header guard to lstm_utils.
A. Unique TensorFlower [Mon, 26 Mar 2018 20:11:18 +0000 (13:11 -0700)]
Add header guard to lstm_utils.

PiperOrigin-RevId: 190512302

6 years agoOptimized ops, move code to early, common, section so that it can be shared.
A. Unique TensorFlower [Mon, 26 Mar 2018 20:08:54 +0000 (13:08 -0700)]
Optimized ops, move code to early, common, section so that it can be shared.

PiperOrigin-RevId: 190511964

6 years agoClarify doc strings on gradient methods
Igor Ganichev [Mon, 26 Mar 2018 19:47:47 +0000 (12:47 -0700)]
Clarify doc strings on gradient methods

PiperOrigin-RevId: 190508614

6 years agoInternal change.
Anna R [Mon, 26 Mar 2018 19:40:56 +0000 (12:40 -0700)]
Internal change.

PiperOrigin-RevId: 190507631

6 years agoInternal change.
Anna R [Mon, 26 Mar 2018 19:20:32 +0000 (12:20 -0700)]
Internal change.

PiperOrigin-RevId: 190504933

6 years agotfdbg CLI: Allow node exclusion with tensor filters
Shanqing Cai [Mon, 26 Mar 2018 19:16:34 +0000 (12:16 -0700)]
tfdbg CLI: Allow node exclusion with tensor filters

Fixes: #16619

See the referred GitHub issue for details, but users want to be
able to skip certain nodes when searching for inf/nans, because
some nodes generate inf/nans even in nominal conditions.

This CL adds a new optional flag `--filter_exclude_node_names`
(or `-fenn` for short), which allows users to do exactly that,
by using a regex for node names.

RELNOTES: tfdbg CLI: Allow exclusion of nodes by regular expressions
during tensor filter-enabled Session runs: see the new flags
`--filter_exclude_node_names` (or `-fenn` for short).
PiperOrigin-RevId: 190504225

6 years agoLSTM support: Support fused activation functions in int16 Add ops.
A. Unique TensorFlower [Mon, 26 Mar 2018 19:14:05 +0000 (12:14 -0700)]
LSTM support: Support fused activation functions in int16 Add ops.

PiperOrigin-RevId: 190503823

6 years agoExtended experimental C API with MNIST dataset/iterators support.
Mingsheng Hong [Mon, 26 Mar 2018 18:50:19 +0000 (11:50 -0700)]
Extended experimental C API with MNIST dataset/iterators support.

PiperOrigin-RevId: 190500020

6 years agoRevert to initializing number of threads when SetNumThreads is called. Requiring it
A. Unique TensorFlower [Mon, 26 Mar 2018 18:47:50 +0000 (11:47 -0700)]
Revert to initializing number of threads when SetNumThreads is called. Requiring it
to happen before OpInit() is way too confusing for users.

PiperOrigin-RevId: 190499644

6 years agoMake the CSE ("node deduping") pass in ArithmeticOptimizer more robust in the presenc...
A. Unique TensorFlower [Mon, 26 Mar 2018 18:44:19 +0000 (11:44 -0700)]
Make the CSE ("node deduping") pass in ArithmeticOptimizer more robust in the presence of ops that modify their inputs in-place: Do not dedup nodes if the underlying buffers for their outputs may be passed to an in-place op.

PiperOrigin-RevId: 190499037

6 years agoMade the NumElements function more accurate
Benoit Steiner [Mon, 26 Mar 2018 18:38:15 +0000 (11:38 -0700)]
Made the NumElements function more accurate

PiperOrigin-RevId: 190497916

6 years agotf.GradientTape: Clearly say that tf.while_loop and tf.cond are not supported
Asim Shankar [Mon, 26 Mar 2018 18:18:26 +0000 (11:18 -0700)]
tf.GradientTape: Clearly say that tf.while_loop and tf.cond are not supported
by tf.GradientTape.gradient() at this time.

PiperOrigin-RevId: 190494436

6 years agoBUGFIX: Fix failure-to-broadcast in Wishart.sample.
Joshua V. Dillon [Mon, 26 Mar 2018 18:15:37 +0000 (11:15 -0700)]
BUGFIX: Fix failure-to-broadcast in Wishart.sample.

PiperOrigin-RevId: 190493969

6 years agoFixes an issue with calling tf.contrib.seq2seq.dynamic_decode with an extended BasicD...
A. Unique TensorFlower [Mon, 26 Mar 2018 18:05:37 +0000 (11:05 -0700)]
Fixes an issue with calling tf.contrib.seq2seq.dynamic_decode with an extended BasicDecoder which for example returns a tf.contrib.seq2seq.AttentionWrapperState.
In this case the internal while-loop fails when trying to store an instance tf.contrib.seq2seq.AttentionWrapperState in the internal TensorArray.

PiperOrigin-RevId: 190491787

6 years ago[tf.data] Usability improvements to `tf.contrib.data.make_csv_dataset`.
Rachel Lim [Mon, 26 Mar 2018 17:52:46 +0000 (10:52 -0700)]
[tf.data] Usability improvements to `tf.contrib.data.make_csv_dataset`.

PiperOrigin-RevId: 190489086

6 years agoAdded experimental C APIs to build a stack of dataset + iterator nodes that
Mingsheng Hong [Mon, 26 Mar 2018 17:51:21 +0000 (10:51 -0700)]
Added experimental C APIs to build a stack of dataset + iterator nodes that
reads imagenet TFRecord files.

PiperOrigin-RevId: 190488817

6 years agoAdding a FunctionBufferingResourceReset Op that resets the state of the function...
Rohan Jain [Mon, 26 Mar 2018 17:25:46 +0000 (10:25 -0700)]
Adding a FunctionBufferingResourceReset Op that resets the state of the function buffering resource so that we can start using it with re-initializable iterators.

PiperOrigin-RevId: 190484110

6 years agoAutomated g4 rollback of changelist 190293303
A. Unique TensorFlower [Mon, 26 Mar 2018 17:00:20 +0000 (10:00 -0700)]
Automated g4 rollback of changelist 190293303

PiperOrigin-RevId: 190479555

6 years agoUpdated test (but not source) of https://www.tensorflow.org/api_docs/python/tf/contri...
A. Unique TensorFlower [Mon, 26 Mar 2018 15:52:53 +0000 (08:52 -0700)]
Updated test (but not source) of https://tensorflow.org/api_docs/python/tf/contrib/training/HParams to show that it allows '=' in the values.

PiperOrigin-RevId: 190470578

6 years agoDisable flaky prefetching_ops_test.
Gunhan Gulsoy [Mon, 26 Mar 2018 05:02:09 +0000 (22:02 -0700)]
Disable flaky prefetching_ops_test.

PiperOrigin-RevId: 190416108

6 years agoAlways cast `tf.distributions.Distribution` `_event_shape`, `_batch_shape`.
Joshua V. Dillon [Mon, 26 Mar 2018 04:57:09 +0000 (21:57 -0700)]
Always cast `tf.distributions.Distribution` `_event_shape`, `_batch_shape`.

PiperOrigin-RevId: 190415923

6 years agoMinor Error type and documentation fix.
Petros Mol [Mon, 26 Mar 2018 03:13:13 +0000 (20:13 -0700)]
Minor Error type and documentation fix.

PiperOrigin-RevId: 190411045

6 years agoUse compat.as_bytes() instead of str.encode().
Skye Wanderman-Milne [Mon, 26 Mar 2018 02:35:54 +0000 (19:35 -0700)]
Use compat.as_bytes() instead of str.encode().

PiperOrigin-RevId: 190409217

6 years agoAdd skeleton code for DebugStripper.
A. Unique TensorFlower [Sun, 25 Mar 2018 18:51:33 +0000 (11:51 -0700)]
Add skeleton code for DebugStripper.

PiperOrigin-RevId: 190391193

6 years agoRestore dependencies that are needed by the PIP package builder
A. Unique TensorFlower [Sun, 25 Mar 2018 16:38:54 +0000 (09:38 -0700)]
Restore dependencies that are needed by the PIP package builder

PiperOrigin-RevId: 190387090

6 years agoBuild and import rules for distributed strategy
Priya Gupta [Sun, 25 Mar 2018 06:42:08 +0000 (23:42 -0700)]
Build and import rules for distributed strategy

PiperOrigin-RevId: 190367484

6 years agoMisc typo fixes in the XLA sources and docs.
Dimitris Vardoulakis [Sat, 24 Mar 2018 07:24:50 +0000 (00:24 -0700)]
Misc typo fixes in the XLA sources and docs.

PiperOrigin-RevId: 190322644

6 years agoTrivial update of layer imports in eager execution examples, to reflect recommended...
Francois Chollet [Sat, 24 Mar 2018 06:04:03 +0000 (23:04 -0700)]
Trivial update of layer imports in eager execution examples, to reflect recommended practices.

PiperOrigin-RevId: 190319480

6 years ago[XLA] Redesign: implement Reshape and Transpose.
A. Unique TensorFlower [Sat, 24 Mar 2018 05:27:31 +0000 (22:27 -0700)]
[XLA] Redesign: implement Reshape and Transpose.

Also,
- Templatize ClientLibraryTestBase::CreateParameterAndTransferLiteral. The implementation is moved from .cc to .h because otherewise the linker complains.
- Migrate some reshape tests to use the XlaBuilder.

PiperOrigin-RevId: 190317960

6 years ago[XLA] Prevent using XlaOp from the wrong XlaBuilder.
A. Unique TensorFlower [Sat, 24 Mar 2018 03:20:49 +0000 (20:20 -0700)]
[XLA] Prevent using XlaOp from the wrong XlaBuilder.

PiperOrigin-RevId: 190312254

6 years agoA couple of small device-related utilities.
A. Unique TensorFlower [Sat, 24 Mar 2018 03:18:46 +0000 (20:18 -0700)]
A couple of small device-related utilities.

PiperOrigin-RevId: 190312148

6 years agoFix loop variable type and status propagation
A. Unique TensorFlower [Sat, 24 Mar 2018 02:24:04 +0000 (19:24 -0700)]
Fix loop variable type and status propagation

PiperOrigin-RevId: 190308776

6 years agoAdding support for iterating through a tf.data.Dataset for a single epoch.
A. Unique TensorFlower [Sat, 24 Mar 2018 02:02:37 +0000 (19:02 -0700)]
Adding support for iterating through a tf.data.Dataset for a single epoch.

PiperOrigin-RevId: 190307545

6 years agoWhen import_scoped_meta_graph is called within a name scope,
A. Unique TensorFlower [Sat, 24 Mar 2018 01:44:35 +0000 (18:44 -0700)]
When import_scoped_meta_graph is called within a name scope,
but called without an import_scope, the names of the created
variables are wrong, resulting in key not found errors when
adding these variables to their corresponding collections.

PiperOrigin-RevId: 190306555

6 years agoSwitch Android C++ compilation mode to "-std=c++11".
A. Unique TensorFlower [Sat, 24 Mar 2018 01:39:29 +0000 (18:39 -0700)]
Switch Android C++ compilation mode to "-std=c++11".

PiperOrigin-RevId: 190306256

6 years agoGo: Update generated wrapper functions for TensorFlow ops.
A. Unique TensorFlower [Sat, 24 Mar 2018 00:49:47 +0000 (17:49 -0700)]
Go: Update generated wrapper functions for TensorFlow ops.
PiperOrigin-RevId: 190302194

6 years agoSet the stream in TransformTensor.
Guangda Lai [Sat, 24 Mar 2018 00:28:10 +0000 (17:28 -0700)]
Set the stream in TransformTensor.

PiperOrigin-RevId: 190300166

6 years agoUpdate ops-related pbtxt files.
A. Unique TensorFlower [Sat, 24 Mar 2018 00:19:10 +0000 (17:19 -0700)]
Update ops-related pbtxt files.

PiperOrigin-RevId: 190299240

6 years agoMake _USE_C_API = True and _USE_C_SHAPES = False work with handle data.
Skye Wanderman-Milne [Fri, 23 Mar 2018 23:28:16 +0000 (16:28 -0700)]
Make _USE_C_API = True and _USE_C_SHAPES = False work with handle data.

This change makes _set_shapes_for_outputs_c_api fetch and set
Tensor._handle_data. This is necessary for running the
Python shape inference code on resource tensors.

PiperOrigin-RevId: 190293303

6 years agobug fix: evaluate nodes before swap the original graph
A. Unique TensorFlower [Fri, 23 Mar 2018 23:16:22 +0000 (16:16 -0700)]
bug fix: evaluate nodes before swap the original graph

PiperOrigin-RevId: 190291844

6 years agoMoves TensorHandleCopyToDevice to TensorHandle::CopyToDevice.
Alexandre Passos [Fri, 23 Mar 2018 23:15:55 +0000 (16:15 -0700)]
Moves TensorHandleCopyToDevice to TensorHandle::CopyToDevice.

PiperOrigin-RevId: 190291768

6 years ago[XLA] Redesign: implement local client and local service interface.
A. Unique TensorFlower [Fri, 23 Mar 2018 23:13:13 +0000 (16:13 -0700)]
[XLA] Redesign: implement local client and local service interface.

PiperOrigin-RevId: 190291400

6 years agoExtended scatter operations to work with a scalar update parameter and added scatter...
A. Unique TensorFlower [Fri, 23 Mar 2018 23:00:14 +0000 (16:00 -0700)]
Extended scatter operations to work with a scalar update parameter and added scatter-min and scatter-max operations.

PiperOrigin-RevId: 190289664

6 years ago[XLA:CPU] Update calls to IRBuilder::CreateMemCpy to the 2-alignment form.
Benjamin Kramer [Fri, 23 Mar 2018 22:52:35 +0000 (15:52 -0700)]
[XLA:CPU] Update calls to IRBuilder::CreateMemCpy to the 2-alignment form.

The single alignment version is going away.

PiperOrigin-RevId: 190288581

6 years ago[XLA] Don't CSE instructions which have side-effects
David Majnemer [Fri, 23 Mar 2018 22:34:47 +0000 (15:34 -0700)]
[XLA] Don't CSE instructions which have side-effects

PiperOrigin-RevId: 190285774

6 years agoAdding support for analyzing assignment info for nested tuples.
A. Unique TensorFlower [Fri, 23 Mar 2018 22:33:33 +0000 (15:33 -0700)]
Adding support for analyzing assignment info for nested tuples.

PiperOrigin-RevId: 190285584

6 years ago[TF:XLA] Fix PotentiallyImplementedAsEigenConvolution to use the correct shape as...
A. Unique TensorFlower [Fri, 23 Mar 2018 22:12:35 +0000 (15:12 -0700)]
[TF:XLA] Fix PotentiallyImplementedAsEigenConvolution to use the correct shape as the kernel shape

A small bug is found in accessing the kernel's shape of the convolution instruction in PotentiallyImplementedAsEigenConvolution. The bug was fixed and a new testcase is created to reveal the bug.

PiperOrigin-RevId: 190282385

6 years agoConvert the eager SPINN example to use tf.keras.Model and object-based checkpointing.
Allen Lavoie [Fri, 23 Mar 2018 22:12:21 +0000 (15:12 -0700)]
Convert the eager SPINN example to use tf.keras.Model and object-based checkpointing.

Uses a more recursive/functional tracking style which avoids numbering layers. Maybe this is too magical and we should adapt tf.keras.Sequential first? Let me know what you think.

PiperOrigin-RevId: 190282346

6 years agoadd EvaluateNodes to OpDedupping test.
A. Unique TensorFlower [Fri, 23 Mar 2018 22:10:56 +0000 (15:10 -0700)]
add EvaluateNodes to OpDedupping test.

PiperOrigin-RevId: 190282163

6 years agoPrepare the XLA and TF code to correctly behave once automatic device placement will...
A. Unique TensorFlower [Fri, 23 Mar 2018 21:52:44 +0000 (14:52 -0700)]
Prepare the XLA and TF code to correctly behave once automatic device placement will be enabled by default on computation shapes > 1.

PiperOrigin-RevId: 190278826

6 years agoBenchmarker for LPIRC CVPR2018.
A. Unique TensorFlower [Fri, 23 Mar 2018 21:41:02 +0000 (14:41 -0700)]
Benchmarker for LPIRC CVPR2018.

PiperOrigin-RevId: 190276899

6 years agoFix buffer overflow when fetching resources.
Alexandre Passos [Fri, 23 Mar 2018 21:19:37 +0000 (14:19 -0700)]
Fix buffer overflow when fetching resources.

PiperOrigin-RevId: 190273682

6 years agoFix behavior of bucket_by_sequence_length with tuple Dataset elements
A. Unique TensorFlower [Fri, 23 Mar 2018 21:00:39 +0000 (14:00 -0700)]
Fix behavior of bucket_by_sequence_length with tuple Dataset elements
Fixes #17932

PiperOrigin-RevId: 190270732

6 years ago[XLA] Implement the whole graph execution interface and make a test use XlaBuilder.
A. Unique TensorFlower [Fri, 23 Mar 2018 20:47:03 +0000 (13:47 -0700)]
[XLA] Implement the whole graph execution interface and make a test use XlaBuilder.
- Add Client::ExecuteGraph.
- Make client_library_test_base also (partially) support XlaBuilder by using template.
- Make one testcase in the axpy_simple_test use XlaBuilder. The test was slightly changed because currently the builder does not expend implicit broadcast automatically.

PiperOrigin-RevId: 190268658

6 years agoMoves TensorHandle to common_runtime
Alexandre Passos [Fri, 23 Mar 2018 20:23:35 +0000 (13:23 -0700)]
Moves TensorHandle to common_runtime

PiperOrigin-RevId: 190265301

6 years agoUpdates reduce_mean op.
Nupur Garg [Fri, 23 Mar 2018 20:20:13 +0000 (13:20 -0700)]
Updates reduce_mean op.

PiperOrigin-RevId: 190264873

6 years ago* Added links to Machine Learning Crash Course.
A. Unique TensorFlower [Fri, 23 Mar 2018 20:15:26 +0000 (13:15 -0700)]
* Added links to Machine Learning Crash Course.
* Removed a link to the script that diagnosed Windows installation errors
  because that script is now deprecated.

PiperOrigin-RevId: 190264199

6 years ago[XLA] Redesign: add public methods that create XlaOp.
A. Unique TensorFlower [Fri, 23 Mar 2018 19:32:42 +0000 (12:32 -0700)]
[XLA] Redesign: add public methods that create XlaOp.

This is basically copying and pasting the public builder methods from ComputationBuilder and leave the implemention blank.

PiperOrigin-RevId: 190258687

6 years ago[XLA:CPU] Task parallel assignment should not parallelize kRng, kInfeed, and kOutfeed.
Bixia Zheng [Fri, 23 Mar 2018 19:20:12 +0000 (12:20 -0700)]
[XLA:CPU] Task parallel assignment should not parallelize kRng, kInfeed, and kOutfeed.

Add two test cases.

PiperOrigin-RevId: 190257131

6 years ago[XLA] Allow reshape mover to move transpose across broadcast of a scalar value.
Bixia Zheng [Fri, 23 Mar 2018 19:02:57 +0000 (12:02 -0700)]
[XLA] Allow reshape mover to move transpose across broadcast of a scalar value.

This allows the simplification of pattern "transpose elementwise-ops inversed
transpose" to "elementwise-ops".

Add a test case.

PiperOrigin-RevId: 190254501

6 years agoUsing rounding when converting constants in BF16 propagation.
Yuanzhong Xu [Fri, 23 Mar 2018 18:41:05 +0000 (11:41 -0700)]
Using rounding when converting constants in BF16 propagation.

PiperOrigin-RevId: 190251081

6 years agoMove NullFileSystem to its own file.
A. Unique TensorFlower [Fri, 23 Mar 2018 18:27:56 +0000 (11:27 -0700)]
Move NullFileSystem to its own file.

PiperOrigin-RevId: 190249020

6 years agoMove security documentation to the main TensorFlow site for better visibility, and...
Frank Chen [Fri, 23 Mar 2018 18:05:14 +0000 (11:05 -0700)]
Move security documentation to the main TensorFlow site for better visibility, and leave a stub SECURITY.md pointing users there.

PiperOrigin-RevId: 190244853

6 years agoChanges loss_reduction default to SUM_OVER_BATCH_SIZE for multi_label_head.
A. Unique TensorFlower [Fri, 23 Mar 2018 18:02:14 +0000 (11:02 -0700)]
Changes loss_reduction default to SUM_OVER_BATCH_SIZE for multi_label_head.

PiperOrigin-RevId: 190244159

6 years agoOverride `__str__` and `__repr__` in `tf.distributions.Distribution`.
Joshua V. Dillon [Fri, 23 Mar 2018 18:01:25 +0000 (11:01 -0700)]
Override `__str__` and `__repr__` in `tf.distributions.Distribution`.

PiperOrigin-RevId: 190243989

6 years agoAdds support for SavedModel to TOCO.
Nupur Garg [Fri, 23 Mar 2018 17:53:49 +0000 (10:53 -0700)]
Adds support for SavedModel to TOCO.

PiperOrigin-RevId: 190242571

6 years agoSlightly tweak the whitelist checking mechanism.
A. Unique TensorFlower [Fri, 23 Mar 2018 17:52:57 +0000 (10:52 -0700)]
Slightly tweak the whitelist checking mechanism.

PiperOrigin-RevId: 190242423

6 years agoAutomated g4 rollback of changelist 190139303
Dimitris Vardoulakis [Fri, 23 Mar 2018 17:52:16 +0000 (10:52 -0700)]
Automated g4 rollback of changelist 190139303

PiperOrigin-RevId: 190242291

6 years agoRaise the timeout of tests that are somewhat close to the limit, to avoid them timing...
A. Unique TensorFlower [Fri, 23 Mar 2018 17:34:44 +0000 (10:34 -0700)]
Raise the timeout of tests that are somewhat close to the limit, to avoid them timing out.

PiperOrigin-RevId: 190239334

6 years agoOnly run GPU version of memory_optimizer_test.
A. Unique TensorFlower [Fri, 23 Mar 2018 17:28:17 +0000 (10:28 -0700)]
Only run GPU version of memory_optimizer_test.

PiperOrigin-RevId: 190238306

6 years agoCheck for the existence of an annotation before attempting to fetch it.
A. Unique TensorFlower [Fri, 23 Mar 2018 17:17:52 +0000 (10:17 -0700)]
Check for the existence of an annotation before attempting to fetch it.

PiperOrigin-RevId: 190236717

6 years agoSilence compilation warning in allocator.h
Derek Murray [Fri, 23 Mar 2018 17:12:05 +0000 (10:12 -0700)]
Silence compilation warning in allocator.h

Some builds report the following warning (even though LOG(FATAL) is terminal):

```
./tensorflow/core/framework/device_base.h(188): warning: missing return statement at end of non-void function "tensorflow::DeviceBase::GetScopedAllocator"
```

PiperOrigin-RevId: 190235863

6 years agoMerged commit includes the following changes:
A. Unique TensorFlower [Fri, 23 Mar 2018 16:15:58 +0000 (09:15 -0700)]
Merged commit includes the following changes:
190228434  by A. Unique TensorFlower:

    internal change only

--
190215102  by A. Unique TensorFlower:

    Remove placement strategy from Fisher estimator and add it in as Mixin.
    A new Kfac Optimizer sub class which runs covariance and inverse ops periodically. This should deprecate lazykfac.

--
190211603  by A. Unique TensorFlower:

    Internal only

--

PiperOrigin-RevId: 190228434

6 years agoFix HloSharding::ToString to be compatible with the HLO text parser
A. Unique TensorFlower [Fri, 23 Mar 2018 12:43:13 +0000 (05:43 -0700)]
Fix HloSharding::ToString to be compatible with the HLO text parser

The incompatibility made it impossible to parse HLO modules with
sharding attributes printed from a previous run of XLA. The previous
format generated by this function was ambigious so changing the HLO
parser only wouldn't be feasible.

PiperOrigin-RevId: 190208069

6 years agoUse int32 GPU kernel for StopGradient
A. Unique TensorFlower [Fri, 23 Mar 2018 11:41:47 +0000 (04:41 -0700)]
Use int32 GPU kernel for StopGradient

PiperOrigin-RevId: 190204238

6 years agoImplement operator<< for HloSharding
A. Unique TensorFlower [Fri, 23 Mar 2018 10:55:03 +0000 (03:55 -0700)]
Implement operator<< for HloSharding

The new operator makes the error messages coming from gtest more
readable as well as making logging easier.

PiperOrigin-RevId: 190200926

6 years agoInternal change
A. Unique TensorFlower [Fri, 23 Mar 2018 10:37:05 +0000 (03:37 -0700)]
Internal change

PiperOrigin-RevId: 190199534

6 years agoEnable fusion of element-wise ops using XLA (Off by default, can be enabled by settin...
A. Unique TensorFlower [Fri, 23 Mar 2018 06:25:37 +0000 (23:25 -0700)]
Enable fusion of element-wise ops using XLA (Off by default, can be enabled by setting the flag tf_xla_fusion_only=true).

PiperOrigin-RevId: 190179459

6 years agoAdd cast op support for bfloat16 in CPU.
A. Unique TensorFlower [Fri, 23 Mar 2018 04:46:31 +0000 (21:46 -0700)]
Add cast op support for bfloat16 in CPU.

PiperOrigin-RevId: 190173501

6 years agoAdded experimental C APIs to build a hard-coded stack of dataset + iterator.
Mingsheng Hong [Fri, 23 Mar 2018 03:51:51 +0000 (20:51 -0700)]
Added experimental C APIs to build a hard-coded stack of dataset + iterator.

PiperOrigin-RevId: 190170332

6 years agoImprovements to quantization matching code:
Suharsh Sivakumar [Fri, 23 Mar 2018 03:40:55 +0000 (20:40 -0700)]
Improvements to quantization matching code:

- Guarantee that we do the largest matches first.
- Don't allow matching layers multiple times.
- Don't allow adding quantization ops to the same node multiple times.
- Return a list of match results rather than yielding. This is much easier to reason about.
- Only require ReadVariableOp when matching resource variables, since the input to ReadVariableOps don't necessarily have to be a VarHandleOp.
- Place post activation bypass ops quantization nodes in the same post activation bypass op's name scope for better viewing.

PiperOrigin-RevId: 190169622