Allen Lavoie [Fri, 20 Apr 2018 19:40:57 +0000 (12:40 -0700)]
Add a utility to visualize object-based checkpoints
Useful for generating a warm fuzzy feeling that everything you think should be saved was saved, and for explaining what object-based checkpointing is. (Also useful on the former front will be a planned "assert that all of this Graph's trainable variables are accessible from object X" function.)
Somewhat hacky since it generates strings rather than using the pydot bindings (and so works without a pydot dependency).
PiperOrigin-RevId:
193708003
A. Unique TensorFlower [Fri, 20 Apr 2018 19:37:39 +0000 (12:37 -0700)]
Changed tf_to_tflite build rule.
PiperOrigin-RevId:
193707628
Yangzihao Wang [Fri, 20 Apr 2018 19:05:11 +0000 (12:05 -0700)]
Add native dilated support for conv3d and its gradients in cudnn v>=6.
PiperOrigin-RevId:
193703316
Dimitris Vardoulakis [Fri, 20 Apr 2018 19:02:56 +0000 (12:02 -0700)]
[TF:XLA] Now that the compiler no longer introduces implicit broadcasts, forbid them in the HLO verifier.
PiperOrigin-RevId:
193702874
Anna R [Fri, 20 Apr 2018 19:01:21 +0000 (12:01 -0700)]
Prefix compat import with underscore in meta_graph_transform.py so that it
doesn't get exported as part of API: https://www.tensorflow.org/versions/r1.8/api_docs/python/tf/contrib/meta_graph_transform/meta_graph_transform
PiperOrigin-RevId:
193702570
A. Unique TensorFlower [Fri, 20 Apr 2018 18:43:48 +0000 (11:43 -0700)]
[XLA] Fix a bug in ToProto: don't add gather attributes twice.
PiperOrigin-RevId:
193699745
Akshay Modi [Fri, 20 Apr 2018 18:34:55 +0000 (11:34 -0700)]
Move the guts of TFE_Op into EagerOperation
PiperOrigin-RevId:
193698320
Akshay Modi [Fri, 20 Apr 2018 18:24:53 +0000 (11:24 -0700)]
Allow creating tensors from numpy arrays, and other various constants - try #2
Allow type-inference from a different input tensor, similar to args_to_matching_eager.
- Update TFE_Py_TensorShapeSlice to take tuples.
- Update int values to allow int/long in py2
END_PUBLIC
BEGIN_PUBLIC
Automated g4 rollback of changelist
192184809
PiperOrigin-RevId:
193696790
James Qin [Fri, 20 Apr 2018 18:23:29 +0000 (11:23 -0700)]
Temporarily set cudnn Rnn math precision to fp32.
Problem:
When calling cudnnGetRNNLinLayerMatrixParams(), return error CUDNN_STATUS_BAD_PARAM if:
* RNN descriptor set math precision = CUDNN_DATA_FLOAT
* input descriptor dataType = CUDNN_DATA_HALF
* weight descriptor dataType= CUDNN_DATA_HALF
If updating Rnn descriptor math precision to CUDNN_DATA_HALF, then no error.
cudnn 7.1.4 will fix the problem.
PiperOrigin-RevId:
193696566
akindyakov [Fri, 20 Apr 2018 18:23:15 +0000 (11:23 -0700)]
Speed up safe_strtod and safe_strtof functions by using double-conversion library
Closes #12102.
PiperOrigin-RevId:
193696537
Derek Murray [Fri, 20 Apr 2018 18:13:16 +0000 (11:13 -0700)]
Use CreateWorkerSession and DeleteWorkerSession for all distributed sessions.
This change adds a phase to the session creation protocol: the master now contacts all workers to register a session handle and create a "WorkerSession" on each worker before it first registers or runs a graph on any worker. Subsequent requests to a worker ensure that the worker has the session handle registered before performing the request, and an AbortedError is raised if the worker has not (e.g. because it restarted after a failure).
As a result, more failure cases are covered by the high-level APIs (tf.estimator, Slim, etc.) that recreate the session on receiving an AbortedError. Previously, there was a possible race condition in which a PS task could restart between variable initialization and the first step, leading to a FailedPreconditionError ("Attempting to use uninitialized value") that would not be handled by the high-level APIs.
PiperOrigin-RevId:
193694958
Sanjoy Das [Fri, 20 Apr 2018 18:10:56 +0000 (11:10 -0700)]
[XLA:CPU] Use Eigen for F64 dot operations
PiperOrigin-RevId:
193694613
A. Unique TensorFlower [Fri, 20 Apr 2018 17:34:49 +0000 (10:34 -0700)]
Disable constant folding and arithmetic optimizations for functions.
PiperOrigin-RevId:
193688466
Jacques Pienaar [Fri, 20 Apr 2018 17:32:24 +0000 (10:32 -0700)]
Adding Shape inference functions to outfeed enqueue ops.
PiperOrigin-RevId:
193688099
A. Unique TensorFlower [Fri, 20 Apr 2018 17:16:03 +0000 (10:16 -0700)]
Change the TF record reader to use 16MB buffering by default in order to improve performance.
PiperOrigin-RevId:
193685521
James Keeling [Fri, 20 Apr 2018 16:54:50 +0000 (09:54 -0700)]
Correct error in "Adding An Op" docs.
The macro `REGISTER_KERNEL_BUILDER` always declared a functor specialized on floats, instead of the type actually passed into the macro.
PiperOrigin-RevId:
193682519
Mark Daoust [Fri, 20 Apr 2018 16:31:08 +0000 (09:31 -0700)]
Don't delete inbound_nodes and outbound_nodes, these no longer exist.
PiperOrigin-RevId:
193679512
Saurabh Saxena [Fri, 20 Apr 2018 16:23:52 +0000 (09:23 -0700)]
Fix msan error in MapAndBatchDataset.
While checkpointing tensors in BatchResult.output save only the initialized slice.
If the final batch is short, the entire batch tensor may not be initialized.
PiperOrigin-RevId:
193678679
Sanjoy Das [Fri, 20 Apr 2018 16:20:36 +0000 (09:20 -0700)]
[TF:XLA] Bump open source llvm revision to r330313
PiperOrigin-RevId:
193678317
A. Unique TensorFlower [Fri, 20 Apr 2018 15:08:01 +0000 (08:08 -0700)]
Eliminate the guard around Winograd non-fused convolutions with cudnn7.
PiperOrigin-RevId:
193669636
A. Unique TensorFlower [Fri, 20 Apr 2018 13:36:56 +0000 (06:36 -0700)]
internal change
PiperOrigin-RevId:
193659701
Peter Hawkins [Fri, 20 Apr 2018 10:29:31 +0000 (03:29 -0700)]
[TF:XLA] Implement ClipByValue.
PiperOrigin-RevId:
193646890
Dimitris Vardoulakis [Fri, 20 Apr 2018 06:35:04 +0000 (23:35 -0700)]
[TF:XLA] Factor out the handling of while instructions to make HloVerifier::Run shorter.
PiperOrigin-RevId:
193626864
Asim Shankar [Fri, 20 Apr 2018 06:08:53 +0000 (23:08 -0700)]
Automated g4 rollback of changelist
193602050
PiperOrigin-RevId:
193625346
Roy Frostig [Fri, 20 Apr 2018 06:01:07 +0000 (23:01 -0700)]
[XLA] Rework the local XLA client's Shape class with separate array and tuple shape constructors.
PiperOrigin-RevId:
193624591
Yifei Feng [Fri, 20 Apr 2018 04:33:41 +0000 (21:33 -0700)]
Skip tests with no_oss tag in XLA builds.
PiperOrigin-RevId:
193619344
Derek Murray [Fri, 20 Apr 2018 03:00:21 +0000 (20:00 -0700)]
Always include the local worker in the list of filtered targets.
It is currently legal to specify a device filter that doesn't include the local worker.
In that case, the MasterSession includes all local devices regardless of the filter.
This change extends this behavior to the list of filtered workers, which will be crucial for backwards compatibility when we enable CreateWorkerSession for all MasterSessions, because we need to call CreateWorkerSession on all potential workers.
PiperOrigin-RevId:
193613313
Justin Lebar [Fri, 20 Apr 2018 02:27:34 +0000 (19:27 -0700)]
[XLA] Remove default argument on virtual function DeviceMemoryAllocator::Allocate().
Default args on virtual functions are disallowed by the Google style
guide, for good reason. They have the extremely surprising behavior
that the defaults you get when calling a function on a pointer depend
not on the underlying type of the object, but on whatever is the
semantic type of the pointer!
PiperOrigin-RevId:
193611213
Igor Saprykin [Fri, 20 Apr 2018 02:11:37 +0000 (19:11 -0700)]
Support variable parameter structure in TPU distribution strategy.
TPUStrategy is added to a few more tests.
There appears to be an issue with the batch norm test in minimize_loss_test where the moving averages stay at 0. I'm trying to resolve that separately as the next CL.
PiperOrigin-RevId:
193610264
Justin Lebar [Fri, 20 Apr 2018 02:03:03 +0000 (19:03 -0700)]
Automated g4 rollback of changelist
193593761
PiperOrigin-RevId:
193609407
Anna R [Fri, 20 Apr 2018 01:45:47 +0000 (18:45 -0700)]
Internal change.
PiperOrigin-RevId:
193608140
A. Unique TensorFlower [Fri, 20 Apr 2018 01:29:00 +0000 (18:29 -0700)]
[XLA] Fix a bug in the name_uniquer.
The problem happens because the name_uniquer stripped away the numeric suffix if it <=0. The solution is, if there was a numeric suffix, the result should also have a numeric suffix.
PiperOrigin-RevId:
193606838
A. Unique TensorFlower [Fri, 20 Apr 2018 01:17:02 +0000 (18:17 -0700)]
Optimize Graph function library.
PiperOrigin-RevId:
193605910
Derek Murray [Fri, 20 Apr 2018 01:12:57 +0000 (18:12 -0700)]
Respect any device filters in {Create,Delete}WorkerSessions().
This is another step towards enabling us to turn on explicit worker
sessions for all master sessions.
PiperOrigin-RevId:
193605565
Lukasz Kaiser [Fri, 20 Apr 2018 00:39:09 +0000 (17:39 -0700)]
internal
END_PUBLIC
BEGIN_PUBLIC
Automated g4 rollback of changelist
193571934
PiperOrigin-RevId:
193602050
Austin Anderson [Fri, 20 Apr 2018 00:30:49 +0000 (17:30 -0700)]
Internal testing changes
PiperOrigin-RevId:
193601134
A. Unique TensorFlower [Fri, 20 Apr 2018 00:27:04 +0000 (17:27 -0700)]
Update DecodeProtoOp so that it returns explicitly specified default values for missing fields.
PiperOrigin-RevId:
193600735
A. Unique TensorFlower [Fri, 20 Apr 2018 00:26:41 +0000 (17:26 -0700)]
Add support for non-Tensor args in recompute_grad
Previously, the function decorated by recompute_grad had to have a signature that contained only positional arguments, and all those arguments had to be Tensors. Most "layers" users define however have non-Tensor arguments (for example, various hyperparameters) and often have keyword arguments as well. This change allows a user to use whatever function signature they wish while being explicit about which arguments are Tensors.
PiperOrigin-RevId:
193600682
Jiri Simsa [Fri, 20 Apr 2018 00:21:50 +0000 (17:21 -0700)]
Minor code refactoring.
PiperOrigin-RevId:
193600173
Justin Lebar [Fri, 20 Apr 2018 00:18:10 +0000 (17:18 -0700)]
[XLA] De-unique_ptr-ify ShapedBuffer and ScopedShapedBuffer.
These are already notionally equivalent to T* and unique_ptr<T>, so
having a unique_ptr of a {Scoped,}ShapedBuffer is pretty redundant.
Also clean up the ScopedShapedBuffer API a bit.
PiperOrigin-RevId:
193599773
A. Unique TensorFlower [Fri, 20 Apr 2018 00:17:05 +0000 (17:17 -0700)]
Removed deprecated methods from tensorflow::StringPiece.
This will allow tensorflow::StringPiece to be more easily replaced with absl::string_view as absl::string_view does not contain those methods.
PiperOrigin-RevId:
193599651
A. Unique TensorFlower [Thu, 19 Apr 2018 23:53:14 +0000 (16:53 -0700)]
Run EvaluateNodes for ModelPruner test except for NoPruning.
PiperOrigin-RevId:
193596812
Mark Heffernan [Thu, 19 Apr 2018 23:35:40 +0000 (16:35 -0700)]
Improve error messages for LiteralTestUtil::Near.
Previously error messages for mismatches were difficult to read with much of the space taken by useless stack traces. This CL cleans up the message considerably and adds additional information including statistics about the values and mismatches.
PiperOrigin-RevId:
193594593
Benoit Steiner [Thu, 19 Apr 2018 23:30:26 +0000 (16:30 -0700)]
Cleaned up the handling of merge nodes
PiperOrigin-RevId:
193593810
Justin Lebar [Thu, 19 Apr 2018 23:30:02 +0000 (16:30 -0700)]
Make xla/service:cpu_plugin depend on the StreamExecutor host platform.
PiperOrigin-RevId:
193593761
A. Unique TensorFlower [Thu, 19 Apr 2018 22:56:17 +0000 (15:56 -0700)]
Automated g4 rollback of changelist
193564222
PiperOrigin-RevId:
193588935
Justin Lebar [Thu, 19 Apr 2018 22:55:53 +0000 (15:55 -0700)]
Internal-only change.
PiperOrigin-RevId:
193588868
Rohan Jain [Thu, 19 Apr 2018 22:32:37 +0000 (15:32 -0700)]
Simplifying cols_to_vars update
PiperOrigin-RevId:
193585237
Derek Murray [Thu, 19 Apr 2018 22:29:21 +0000 (15:29 -0700)]
Prevent a bool field from being accessed when uninitialized.
PiperOrigin-RevId:
193584746
A. Unique TensorFlower [Thu, 19 Apr 2018 22:27:19 +0000 (15:27 -0700)]
Convert a local variable and mutex to a struct so GUARDED_BY annotation works correctly.
PiperOrigin-RevId:
193584438
A. Unique TensorFlower [Thu, 19 Apr 2018 22:09:58 +0000 (15:09 -0700)]
Only generate floating points that are fractions like n / 256, since they are RGB pixels. This fixes RGBToHSVTest.testBatch on low-precision dtypes like bfloat16.
PiperOrigin-RevId:
193581652
Olivia Nordquist [Thu, 19 Apr 2018 21:59:25 +0000 (14:59 -0700)]
disabling opensource testing for failing xla test
PiperOrigin-RevId:
193579805
A. Unique TensorFlower [Thu, 19 Apr 2018 21:35:27 +0000 (14:35 -0700)]
Replaced calls to deprecated tensorflow::StringPiece methods with their
tensorflow::str_util equivalents.
This will allow the deprecated methods to be removed.
PiperOrigin-RevId:
193575992
Sanjoy Das [Thu, 19 Apr 2018 21:30:27 +0000 (14:30 -0700)]
The HLO element type converter must remove side effecting instructions like Rng
The CPU backend does not know how to lower bf16 typed RNG nodes so even unused
instances of these can't remain in the HLO IR.
HloComputation::ReplaceInstruction keeps these Rng nodes around since it doesn't
remove side effecting nodes.
PiperOrigin-RevId:
193575183
Asim Shankar [Thu, 19 Apr 2018 21:10:01 +0000 (14:10 -0700)]
RecordReader: Simplify interface contract and implementation.
Prior to this change, RecordReader had the following contract:
- Records can be read in any order, EXCEPT if compression or buffering was
enabled.
- If the underlying file is being concurrently written to
then calls to ReadRecord() may fail (because of an incomplete
record near the end of a file), but a retry may succeed (once the
record is written), EXCEPT if compression or buffering is enabled
(in which case the failure will be terminal).
This "retry-may-succeed" behavior is relied upon by tensorboard
(https://github.com/tensorflow/tensorboard/blob/1.7/tensorboard/backend/event_processing/event_file_loader.py#L55)
where one process (typically the model training process) is writing
tf.summary events to an event file and another process (tensorboard)
is concurrently reading it.
With this change, the intent is to remove the EXCEPTions and have the
same behavior irrespective of compression/buffering.
Additionally, fix a memory leak when ZlibInputStream::Reset() is invoked.
PiperOrigin-RevId:
193571934
A. Unique TensorFlower [Thu, 19 Apr 2018 21:05:06 +0000 (14:05 -0700)]
Make flat_transforms_to_matrices and matrices_to_flat_transforms public available.
PiperOrigin-RevId:
193571089
Asim Shankar [Thu, 19 Apr 2018 20:38:43 +0000 (13:38 -0700)]
tools/lib_package: Fix typo in README
PiperOrigin-RevId:
193566850
A. Unique TensorFlower [Thu, 19 Apr 2018 20:21:25 +0000 (13:21 -0700)]
Allow output has a different shape from input in the image.transform (#17011).
PiperOrigin-RevId:
193564222
Igor Saprykin [Thu, 19 Apr 2018 20:19:27 +0000 (13:19 -0700)]
Support various shapes in TPU DistributionStrategy.
PiperOrigin-RevId:
193563912
A. Unique TensorFlower [Thu, 19 Apr 2018 20:09:07 +0000 (13:09 -0700)]
Collective Ops Part 4
Add Broadcaster.
A few minor adjustments to CollectiveParams and RMA.
This change is part of a series of changes introducing infrastructure
for collective ops and initial implementations of reduction and broadcast.
PiperOrigin-RevId:
193562391
Sung Jin Hwang [Thu, 19 Apr 2018 19:44:21 +0000 (12:44 -0700)]
Make PmfToQuantizedCdf op to make adjustments if the sum of quantized pmf is
less than 2**precision.
Prior to the change, the op did nothing when the sum of quantized pmf was less
than 2**precision. While the produced CDF was valid for range coders,
adjustments to CDF could be made to achieve better compression rate.
PiperOrigin-RevId:
193558740
A. Unique TensorFlower [Thu, 19 Apr 2018 19:32:52 +0000 (12:32 -0700)]
Fix bug in ring_reducer.cc abort handling.
PiperOrigin-RevId:
193557334
Francois Chollet [Thu, 19 Apr 2018 18:58:04 +0000 (11:58 -0700)]
Add support for Dataset Iterators in Model training/eval methods in graph mode.
PiperOrigin-RevId:
193552275
A. Unique TensorFlower [Thu, 19 Apr 2018 18:57:48 +0000 (11:57 -0700)]
Internal cleanup.
PiperOrigin-RevId:
193552240
Benoit Steiner [Thu, 19 Apr 2018 18:57:36 +0000 (11:57 -0700)]
Avoid looking up the shape functions multiple times
Improved the handling of fed nodes
PiperOrigin-RevId:
193552210
A. Unique TensorFlower [Thu, 19 Apr 2018 18:55:46 +0000 (11:55 -0700)]
Change estimator to only log non-binary eval metrics, because logging binary metrics such as images will lead to crash.
PiperOrigin-RevId:
193551927
Rohan Jain [Thu, 19 Apr 2018 18:52:10 +0000 (11:52 -0700)]
Implements linear_model using _LinearModel. Added support for cols_to_vars in _LinearModel in order to make this possible. Also, made some fixes so that variable names come out the same as before.
PiperOrigin-RevId:
193551353
A. Unique TensorFlower [Thu, 19 Apr 2018 18:47:28 +0000 (11:47 -0700)]
Go: Update generated wrapper functions for TensorFlow ops.
PiperOrigin-RevId:
193550428
A. Unique TensorFlower [Thu, 19 Apr 2018 18:21:21 +0000 (11:21 -0700)]
Update ops-related pbtxt files.
PiperOrigin-RevId:
193546050
A. Unique TensorFlower [Thu, 19 Apr 2018 18:08:08 +0000 (11:08 -0700)]
Updating tests in constant_folding_test.cc so that they Evaluate the optimized and original graph and check if their outputs are the same.
PiperOrigin-RevId:
193543478
A. Unique TensorFlower [Thu, 19 Apr 2018 17:59:08 +0000 (10:59 -0700)]
Fix open source BUILD bugs for cloud profiler. Increment
version for releasing cloud_tpu_profiler 1.6 with pod profiling support.
PiperOrigin-RevId:
193541692
Derek Murray [Thu, 19 Apr 2018 17:57:55 +0000 (10:57 -0700)]
Allow non-isolated worker sessions to borrow `WorkerEnv::device_mgr`.
Without this change, a shared resource (e.g. an Iterator) could not be
created in one session `s1`, and used in a later session `s2` after
`s1` was closed, because the iterator might indirectly capture devices
from the previous session, and use them after they are freed when the
`WorkerSession` was deleted.
The current change only affects the singleton "legacy" WorkerSession,
which is never deleted, but this is necessary to switch all sessions
to use separate WorkerSession objects.
PiperOrigin-RevId:
193541426
A. Unique TensorFlower [Thu, 19 Apr 2018 17:33:42 +0000 (10:33 -0700)]
Fix undefined signed integer overflow by performing addition more carefully.
PiperOrigin-RevId:
193537461
Anjali Sridhar [Thu, 19 Apr 2018 17:26:54 +0000 (10:26 -0700)]
Add metric names to model.metrics_names in compile for keras models run in eager execution. This prevents us from dropping metrics when we run model.evaluate.
PiperOrigin-RevId:
193536341
Ian Langmore [Thu, 19 Apr 2018 17:20:43 +0000 (10:20 -0700)]
distribution_util moved into its own BUILD target, so linear_operator can depend on it.
PiperOrigin-RevId:
193535400
Jiri Simsa [Thu, 19 Apr 2018 17:14:09 +0000 (10:14 -0700)]
Prototype for tf.data writer API.
PiperOrigin-RevId:
193534333
Benoit Steiner [Thu, 19 Apr 2018 17:05:08 +0000 (10:05 -0700)]
Added support for concatenation and slicing of symbolic shapes
PiperOrigin-RevId:
193532769
Allen Lavoie [Thu, 19 Apr 2018 16:54:40 +0000 (09:54 -0700)]
Create a skeleton tf.contrib.checkpoint.
My plan for this is to incubate tools for working with object-based checkpoints:
- Tools for managing dependency graphs, e.g. checkpointable lists/dictionaries
- Inspecting/visualizing checkpoints
- Listing variables and gathering initializers from a Checkpointable object
and its dependencies
- Verifying all variables are accessible as dependencies, which should make
converting existing graph building Saver uses easier/safer.
This CL includes none of those things, it just moves the split_dependency tool
here instead of contrib/eager.
PiperOrigin-RevId:
193531292
Saurabh Saxena [Thu, 19 Apr 2018 16:37:20 +0000 (09:37 -0700)]
[tf.data] Add checkpointing support for MapAndBatchDataset.
PiperOrigin-RevId:
193528712
A. Unique TensorFlower [Thu, 19 Apr 2018 16:34:24 +0000 (09:34 -0700)]
Disable dlopen error of libneuralnetworks for non-Android platforms.
PiperOrigin-RevId:
193528346
A. Unique TensorFlower [Thu, 19 Apr 2018 16:23:35 +0000 (09:23 -0700)]
[TF:XLA] Change HloTestBase::ExecuteNoHloPasses to return a literal directly.
PiperOrigin-RevId:
193526900
A. Unique TensorFlower [Thu, 19 Apr 2018 16:13:47 +0000 (09:13 -0700)]
Pin pip to version 9.0.3
* This is because pip 10 is still unstable in some distros
* reference: https://github.com/pypa/pip/issues/5240
PiperOrigin-RevId:
193525542
A. Unique TensorFlower [Thu, 19 Apr 2018 14:13:03 +0000 (07:13 -0700)]
Teach the conditinal simplifier about sharding.
PiperOrigin-RevId:
193510638
A. Unique TensorFlower [Thu, 19 Apr 2018 13:58:34 +0000 (06:58 -0700)]
Update the Colorbot demo to use a Keras model in addition to the Estimator.
PiperOrigin-RevId:
193508874
A. Unique TensorFlower [Thu, 19 Apr 2018 12:11:30 +0000 (05:11 -0700)]
Support for converting entire class hierarchies:
* limit the methods being converted to those that have not been inherited from the superclass
* include the (possibly compiled) superclass in the definition of the compiled class
* either mark the superclass for conversion or generate an absolute aliased import line, depending on whether it's whitelisted
PiperOrigin-RevId:
193499204
A. Unique TensorFlower [Thu, 19 Apr 2018 11:27:38 +0000 (04:27 -0700)]
Handle corner case in Python 3: members annotated with @classmethod.
PiperOrigin-RevId:
193495506
A. Unique TensorFlower [Thu, 19 Apr 2018 09:58:31 +0000 (02:58 -0700)]
Set the random seed in on-demand mode.
PiperOrigin-RevId:
193488103
A. Unique TensorFlower [Thu, 19 Apr 2018 08:56:31 +0000 (01:56 -0700)]
Update feature_util's GetFeatures to show compile-time error for unsupported types instead of a link-time error.
PiperOrigin-RevId:
193480683
Yifei Feng [Thu, 19 Apr 2018 08:26:07 +0000 (01:26 -0700)]
Remove proto import in header files for core/kernels/boosted_trees.
Move implementations that requires declaration of TreeEnsemble to .cc files.
The goal is to make kernels mostly independent of proto headers, which will let us lock down our .so import
PiperOrigin-RevId:
193478404
Asim Shankar [Thu, 19 Apr 2018 07:28:33 +0000 (00:28 -0700)]
docs: Add a note on building the C and/or Java API binaries from source.
See #15290
PiperOrigin-RevId:
193473886
A. Unique TensorFlower [Thu, 19 Apr 2018 06:46:34 +0000 (23:46 -0700)]
Go: Update generated wrapper functions for TensorFlow ops.
PiperOrigin-RevId:
193471104
A. Unique TensorFlower [Thu, 19 Apr 2018 06:18:50 +0000 (23:18 -0700)]
Update ops-related pbtxt files.
PiperOrigin-RevId:
193469437
Derek Murray [Thu, 19 Apr 2018 05:59:01 +0000 (22:59 -0700)]
[tf.data] Add native implementation for `tf.contrib.data.unbatch()`.
The implementation has two main improvements:
1. Avoid relatively expensive (~15us) function invocation for each incoming batch.
2. Use std::move() where possible to avoid copying strings/variants into the unbatched
elements.
PiperOrigin-RevId:
193467856
Yuanzhong Xu [Thu, 19 Apr 2018 05:07:12 +0000 (22:07 -0700)]
[XLA] Fix BF16 propagation bug for while condition.
PiperOrigin-RevId:
193465140
Jacques Pienaar [Thu, 19 Apr 2018 04:15:41 +0000 (21:15 -0700)]
Initial addition of CLZ HLO
* Adds the HLO op and lowering on CPU/GPU/evaluator;
* This does not update the operation semantics;
PiperOrigin-RevId:
193461989
Eugene Brevdo [Thu, 19 Apr 2018 03:15:27 +0000 (20:15 -0700)]
Expose tf.contrib.training.{prepend_from_queue_and_padded_batch_dataset}
Also its helper method "enqueue_in_queue_dataset".
PiperOrigin-RevId:
193458095
Yao Zhang [Thu, 19 Apr 2018 03:04:56 +0000 (20:04 -0700)]
Use the host implementation of vec permute op if the input on the host. Note
that the op still needs to be placed on the GPU so that it stays within the
same partiion with the neighboring ops, and as a result, no unnecessary send
and recv are created.
PiperOrigin-RevId:
193457328
A. Unique TensorFlower [Thu, 19 Apr 2018 03:02:52 +0000 (20:02 -0700)]
Teach the reshape mover pass about sharding.
PiperOrigin-RevId:
193457083
A. Unique TensorFlower [Thu, 19 Apr 2018 02:49:12 +0000 (19:49 -0700)]
Go: Update generated wrapper functions for TensorFlow ops.
PiperOrigin-RevId:
193456151
A. Unique TensorFlower [Thu, 19 Apr 2018 02:18:33 +0000 (19:18 -0700)]
Update ops-related pbtxt files.
PiperOrigin-RevId:
193454093