aboutsummaryrefslogtreecommitdiffhomepage
Commit message (Collapse)AuthorAge
* [XLA] Make --xla_dump_executions_to actually dump the HloSnapshot.Gravatar A. Unique TensorFlower2018-06-13
| | | | PiperOrigin-RevId: 200443383
* support int16-quantized data in TFLite interpreter.Gravatar A. Unique TensorFlower2018-06-13
| | | | PiperOrigin-RevId: 200442886
* [tf.data] Factor out a helper for creating flat args to `function.Defun`.Gravatar Derek Murray2018-06-13
| | | | | | The `defun_args()` helper flattens a nested structure down into the flat tuple of tensor types expected by `Defun`. PiperOrigin-RevId: 200441074
* Splits testLargeCase in metric_ops_test into a dedicated file for ↵Gravatar A. Unique TensorFlower2018-06-13
| | | | | | slow-running tests and re-enables it as a 'large' test. PiperOrigin-RevId: 200440883
* [TF:XLA] Replace bespoke NodeSlot class in subgraph encapsulation code with ↵Gravatar Peter Hawkins2018-06-13
| | | | | | | | | | InputTensor and OutputTensor classes from TF core. Add equality and hash methods to InputTensor and OutputTensor. No functional changes intended. PiperOrigin-RevId: 200440015
* Provide default name_scope in cond_v2.Gravatar Skye Wanderman-Milne2018-06-13
| | | | | | Otherwise passing in name="" results in trying to name the If op "". PiperOrigin-RevId: 200439070
* Initial application of runtime shapes to runtime kernels.Gravatar A. Unique TensorFlower2018-06-13
| | | | PiperOrigin-RevId: 200435608
* Switch Estimator from using DistributionStrategy.fetch() to .read_var().Gravatar A. Unique TensorFlower2018-06-13
| | | | PiperOrigin-RevId: 200434656
* Fix a build failure when cuda version is less than 9000.Gravatar A. Unique TensorFlower2018-06-13
| | | | PiperOrigin-RevId: 200432478
* fix md link formatGravatar Mark Daoust2018-06-13
| | | | PiperOrigin-RevId: 200431906
* Automated g4 rollback of changelist 199870879Gravatar A. Unique TensorFlower2018-06-13
| | | | PiperOrigin-RevId: 200431713
* Add a test that checks memory usage by running a model 100k times.Gravatar Akshay Modi2018-06-13
| | | | PiperOrigin-RevId: 200430314
* [TF:XLA] Bump open source llvm revision to r334593Gravatar Sanjoy Das2018-06-13
| | | | PiperOrigin-RevId: 200427133
* [tf.data] Factor out function argument restructuring into a helper.Gravatar Derek Murray2018-06-13
| | | | | | This cuts down on the amount of repeated (or near-repeated) code in Dataset wrappers. PiperOrigin-RevId: 200424152
* Reversible residual network example with manually built gradient computation inGravatar A. Unique TensorFlower2018-06-13
| | | | | | TensorFlow eager mode execution. PiperOrigin-RevId: 200422481
* Documenting capabilities and limitations of AutoGraphGravatar A. Unique TensorFlower2018-06-13
| | | | PiperOrigin-RevId: 200421914
* - Teaches zip tests about unzip location on AndroidGravatar A. Unique TensorFlower2018-06-13
| | | | | | | | - Passes use_nnapi from TestDriver to interpeter - Adds command line flag to generated tests for NNAPI - Fixes logic for allocating im2col tensor so that tests pass without NNAPI PiperOrigin-RevId: 200416472
* Disable failing zip_test_lstm targetGravatar A. Unique TensorFlower2018-06-13
| | | | PiperOrigin-RevId: 200414970
* [TF:XLA] Remove parallel checking support from encapsulate_subcomputations_pass.Gravatar Peter Hawkins2018-06-13
| | | | | | | | This support is unused and adds complexity to an already very complicated piece of code. No (observable) functional changes intended. PiperOrigin-RevId: 200411522
* Detect configurations that would be hitting a bug in cuBLAS and report an error.Gravatar A. Unique TensorFlower2018-06-13
| | | | PiperOrigin-RevId: 200411493
* Add AotCompilationMetadata field to variant of CompileAheadOfTime.Gravatar Jacques Pienaar2018-06-13
| | | | | | | | | | Add CompileAheadOfTime parameter that can optionally be populated during compilation process. This change is to allow populating metadata even if the CompileAheadOfTime fails. PiperOrigin-RevId: 200407917
* Fix for DumpGraphviz() failing on zero-sized arrays, which it should handle ↵Gravatar A. Unique TensorFlower2018-06-13
| | | | | | for debugging purposes. PiperOrigin-RevId: 200397151
* Correct name for _UnreadVariableGravatar Alexandre Passos2018-06-13
| | | | PiperOrigin-RevId: 200395171
* Clarify that SparseMatMul does not accept SparseTensor inputsGravatar A. Unique TensorFlower2018-06-13
| | | | PiperOrigin-RevId: 200392587
* [tf.data] Factor out `output_shapes` and `output_types` attr-setting code ↵Gravatar Derek Murray2018-06-13
| | | | | | into a helper. PiperOrigin-RevId: 200386950
* Remove uses of dynamic_cast, which is generally discouraged by the Google C++Gravatar A. Unique TensorFlower2018-06-13
| | | | | | | style guide, https://google.github.io/styleguide/cppguide.html#Run-Time_Type_Information__RTTI_ PiperOrigin-RevId: 200380532
* Add missing include of reshape_util.h to reshape_util.cc.Gravatar Derek Murray2018-06-13
| | | | PiperOrigin-RevId: 200378252
* Automated g4 rollback of changelist 200362771Gravatar A. Unique TensorFlower2018-06-13
| | | | PiperOrigin-RevId: 200370679
* Remove GANHead from GANEstimator.Gravatar A. Unique TensorFlower2018-06-13
| | | | PiperOrigin-RevId: 200362771
* Improve shape compatibility checking for MultiOutputFusionGravatar Adrian Kuegel2018-06-13
| | | | | | | | | We need to be careful how we compare reduce shapes. - If comparing against non-reduce shapes, we should compare the operand shape of a reduce with the other shape. - If comparing two reduce shapes, we need to compare both the operand shapes and the reduce shapes. Also, if we already have a multi-output fusion node, we should pick one of its reduce instructions for comparison, because it has more constraints than the other instructions. PiperOrigin-RevId: 200353595
* Documentation for Raspberry Pi installationGravatar Pete Warden2018-06-13
| | | | PiperOrigin-RevId: 200352941
* Split out HloGetTupleIndexInstruction and HloReducePrecisionInstruction as ↵Gravatar A. Unique TensorFlower2018-06-12
| | | | | | subclasses from HloInstruction. PiperOrigin-RevId: 200337508
* Update documentation for export_savedmodel().Gravatar A. Unique TensorFlower2018-06-12
| | | | PiperOrigin-RevId: 200334496
* Split out HloRngInstruction and HloParameterInstruction as subclasses from ↵Gravatar A. Unique TensorFlower2018-06-12
| | | | | | HloInstruction. PiperOrigin-RevId: 200333922
* [XLA] "Global value" tuple destructuring available via SWIG APIs.Gravatar Chris Leary2018-06-12
| | | | PiperOrigin-RevId: 200333639
* Move non-core Distributions and Bijectors to Tensorflow Probability (TFP), ↵Gravatar A. Unique TensorFlower2018-06-12
| | | | | | | | | | | | | and deprecate tf.contrib.distributions. This CL: - copies Distributions code into TFP - Updates it to use TFP idioms, e.g., 'tf.matmul' instead of 'math_ops.matmul'. - Creates a new `internal` subdirectory for modules that must be importable but not visible externally by way of `__init__.py`. - Adds a deprecation notice to tf.contrib.distributions BUILD files and to the class constructors of all tf.contrib Distributions and Bijectors. (and removes tests from the deprecated copies of `distribution_test` and `normal_conjugate_posteriors_test` whose `instance()` calls would be broken by the deprecation wrapper). - Updates miscellaneous references to distributions in TFP. It does not attempt to migrate *all* TFP code to use TFP Distributions; this will be done in a separate CL. PiperOrigin-RevId: 200333629
* Add a `run_metadata` keyword arg for `Session._make_callable_from_options()`.Gravatar Derek Murray2018-06-12
| | | | | | | | All callables returned from this private API now accept a "run_metadata" keyword argument whose behavior matches the `run_metadata` argument accepted by `Session.run()`. PiperOrigin-RevId: 200331667
* Do not count empty tuples as having one leaf node.Gravatar A. Unique TensorFlower2018-06-12
| | | | PiperOrigin-RevId: 200327849
* Leverage the standard error space by using tensorflow::StatusGravatar A. Unique TensorFlower2018-06-12
| | | | PiperOrigin-RevId: 200322035
* Add a test for using sparse variables when running mostly on TPUGravatar Igor Ganichev2018-06-12
| | | | | | | The test is primarily an example of what approaches currently work for sparse operations when we mostly want to run on TPU. PiperOrigin-RevId: 200320045
* Add checkpointing support for FileDataset in CacheDataset.Gravatar Saurabh Saxena2018-06-12
| | | | | | Checkpointing datasets with in-memory caching is not supported yet. PiperOrigin-RevId: 200316958
* Internal change.Gravatar Shashi Shekhar2018-06-12
| | | | PiperOrigin-RevId: 200314093
* [XLA] Delete StripDegenerateDimensions()Gravatar Michael Kuperstein2018-06-12
| | | | | | This is unused, and, as it turns out, is broken for sparse shapes. PiperOrigin-RevId: 200313641
* Automated g4 rollback of changelist 200292049Gravatar A. Unique TensorFlower2018-06-12
| | | | PiperOrigin-RevId: 200309129
* Checking that TPUEstimator model function features have static shapes.Gravatar A. Unique TensorFlower2018-06-12
| | | | PiperOrigin-RevId: 200306833
* Autotune bias_add_grad GPU kernel using two candidates: customized ↵Gravatar Mingxing Tan2018-06-12
| | | | | | bias_add_grad and reduce_sum. PiperOrigin-RevId: 200306546
* Temporarily disable Grappler memory optimization for fused_conv tests.Gravatar Jingyue Wu2018-06-12
| | | | PiperOrigin-RevId: 200294932
* [XLA] Handle kDynamicSlice and kDynamicUpdateSlice correctly in HloCostAnalysisGravatar Michael Kuperstein2018-06-12
| | | | | | | kDynamicSlice only needs to reads the part of the operand that is sliced out. kDynamicUpdateSlice needs to read the update, and write it out to the relevant part of the updated operand, but does not need to read the updated operand at all. PiperOrigin-RevId: 200293681
* PUBLIC: [TF:XLA] Eliminate more copies after HLO scheduling.Gravatar A. Unique TensorFlower2018-06-12
| | | | | | | | | | | | After scheduling HLOs it is very beneficial to try more copy elision: The sequential ordering from the schedule is stricter than the data-dependency ordering used during copy insertion. Also, allow more operands to share a buffer with their user. In particular, the user has to be element-wise only wrt to the specified operand, and not wrt to all operands. These two changes allow more copies to be eliminated. PiperOrigin-RevId: 200292049
* Fix a linkopt.Gravatar Shashi Shekhar2018-06-12
| | | | PiperOrigin-RevId: 200289114
generated by cgit v1.2.3 (git 2.39.1) at 2024-07-07 01:03:51 +0000