aboutsummaryrefslogtreecommitdiffhomepage
path: root/tensorflow/contrib/tpu
Commit message (Collapse)AuthorAge
* Changed Adam algorithm variant formula from sqrt(max(v, epsilon**2)) to ↵Gravatar A. Unique TensorFlower2018-10-08
| | | | | | sqrt(v + epsilon**2) and changed flag name accordingly. PiperOrigin-RevId: 216240045
* Mark tensorflow/contrib/tpu:datasets_test flakyGravatar Todd Wang2018-10-06
| | | | | | It fails 1/1000 runs in OSS builds. PiperOrigin-RevId: 216050192
* Profiler collects the number of replicas and num cores per replica used in ↵Gravatar A. Unique TensorFlower2018-10-05
| | | | | | the model. PiperOrigin-RevId: 215947354
* - Don't set tpu optimizer parameter variable during weight initialization if ↵Gravatar Jing Li2018-10-05
| | | | | | | | the optimizer isn't set, e.g. loading weights and then predict. - Add load_weights for `KerasTpuModel`. PiperOrigin-RevId: 215920993
* Skip numeric checking in BROADCAST mode.Gravatar Youlong Cheng2018-10-04
| | | | PiperOrigin-RevId: 215752559
* Re-add proto fields temporarily for internal compatibility.Gravatar A. Unique TensorFlower2018-10-03
| | | | PiperOrigin-RevId: 215585187
* Remove initial accumulator (and other auxiliary parameter) values fromGravatar A. Unique TensorFlower2018-10-02
| | | | | | optimization parameter protos and removed uses of that functionality in tests. PiperOrigin-RevId: 215494433
* Upgrade cloud tpu profiler to 1.12.0.Gravatar A. Unique TensorFlower2018-10-02
| | | | PiperOrigin-RevId: 215454323
* Add mode_override to the TPU embedding enqueue ops. This allows the mode to beGravatar A. Unique TensorFlower2018-10-01
| | | | | | | overridden at runtime allowing dynamic switching between inference and training modes. Not fully implemented yet. PiperOrigin-RevId: 215325071
* Make Keras/TPU more robust to closed TF sessions.Gravatar Russell Power2018-10-01
| | | | PiperOrigin-RevId: 215313156
* [tf.data] Deprecate `tf.contrib.data` and introduce `tf.data.experimental` ↵Gravatar Derek Murray2018-10-01
| | | | | | | | | | | | | | | | | | | | to replace it. This change prepares `tf.data` for TensorFlow 2.0, where `tf.contrib` will no longer exist. It retains the pre-existing endpoints in `tf.contrib.data` with deprecation warnings. Note there are some exceptions to the move: * Deprecated symbols in `tf.contrib.data` have not been moved to `tf.data.experimental`, because replacements already exist. * `tf.contrib.data.LMDBDataset` has not been moved, because we plan to move it to a SIG-maintained repository. * `tf.contrib.data.assert_element_shape()` has not yet been moved, because it depends on functionality in `tf.contrib`, and it will move in a later change. * `tf.contrib.data.AUTOTUNE` has not yet been moved, because we have not yet determined how to `tf_export()` a Python integer. * The stats-related API endpoints have not yet appeared in a released version of TensorFlow, so these are moved to `tf.data.experimental` without retaining an endpoint in `tf.contrib.data`. In addition, this change includes some build rule and ApiDef refactoring: * Some of the "//third_party/tensorflow/python:training" dependencies had to be split in order to avoid a circular dependency. * The `tf.contrib.stateless` ops now have a private core library for the generated wrappers (and accordingly are hidden in their ApiDef) so that `tf.data.experimental.sample_from_datasets()` can depend on them. PiperOrigin-RevId: 215304249
* Small tweaks to comments and documentation strings.Gravatar A. Unique TensorFlower2018-10-01
| | | | PiperOrigin-RevId: 215297961
* Added option (off by default) to enable a higher-performance variant of theGravatar A. Unique TensorFlower2018-10-01
| | | | | | Adam optimizer's variable update formula. PiperOrigin-RevId: 215290881
* Expose tpu_host_placement_function().Gravatar Youlong Cheng2018-10-01
| | | | PiperOrigin-RevId: 215259803
* internal change onlyGravatar A. Unique TensorFlower2018-10-01
| | | | PiperOrigin-RevId: 215248985
* Automated rollback of commit 3f4423fad57694bc8d7adc427d65e5a18c8592b2Gravatar A. Unique TensorFlower2018-10-01
| | | | PiperOrigin-RevId: 215200418
* Add learning_rates input to the SendTPUEmbeddingGradients op. This allows theGravatar A. Unique TensorFlower2018-09-28
| | | | | | learning rate to be modified at runtime. The implementation is not yet complete. PiperOrigin-RevId: 215030536
* Fix bad reference to self._name in TPUEstimator infeed loop.Gravatar Russell Power2018-09-28
| | | | PiperOrigin-RevId: 215029224
* Move TPU variables to the TPU device in TPUStrategy.Gravatar Jonathan Hseu2018-09-28
| | | | PiperOrigin-RevId: 215027511
* Added flag to enable non-lazy Adam optimizer implementation for TPU embeddingsGravatar A. Unique TensorFlower2018-09-28
| | | | | | | (actual implementation is pending). Added comments with pointers to C++ implementations of optimizers. PiperOrigin-RevId: 215026002
* Add option to disable initialization/shutdown of the TPU.Gravatar Russell Power2018-09-28
| | | | PiperOrigin-RevId: 215016286
* Internal changes only.Gravatar A. Unique TensorFlower2018-09-28
| | | | PiperOrigin-RevId: 215009955
* Puts the keras optimizer weights on device.Gravatar Jianwei Xie2018-09-28
| | | | PiperOrigin-RevId: 214974535
* internal change onlyGravatar A. Unique TensorFlower2018-09-28
| | | | PiperOrigin-RevId: 214967868
* Remove @{} api_links and ban "@{}" from python and md files.Gravatar Mark Daoust2018-09-28
| | | | PiperOrigin-RevId: 214964988
* internal change onlyGravatar A. Unique TensorFlower2018-09-28
| | | | PiperOrigin-RevId: 214941829
* Adding to tpu_lib depenencies to optimization_parameters_py, ↵Gravatar Daryl Ng2018-09-27
| | | | | | tpu_embedding_configuration_py, and tpu_embedding_output_layout_py. PiperOrigin-RevId: 214879168
* Fix heartbeat probing.Gravatar Russell Power2018-09-27
| | | | PiperOrigin-RevId: 214846488
* Enable worker heartbeat polling for all available workers.Gravatar Russell Power2018-09-27
| | | | PiperOrigin-RevId: 214831772
* Fix custom getter handling in tpu.rewrite() and friends.Gravatar A. Unique TensorFlower2018-09-26
| | | | | | It used to save the existing custom getter then overwrites the custom getter. That means the previous custom getter will never be called inside "computation". It now create a new custom getter that calls the previous custom getter. PiperOrigin-RevId: 214715720
* Adding per table load and retrieve ops and additional enqueue operations. ↵Gravatar Daryl Ng2018-09-26
| | | | | | Other additional refactoring. PiperOrigin-RevId: 214715083
* Add Mirrored distribution strategy support for new metrics with Keras and ↵Gravatar Pavithra Vijay2018-09-26
| | | | | | | | Estimator Add support for stateful metrics in model to estimator PiperOrigin-RevId: 214714322
* Fixed the bug which slows the TPU traning.Gravatar Jianwei Xie2018-09-26
| | | | PiperOrigin-RevId: 214702243
* internal change onlyGravatar A. Unique TensorFlower2018-09-26
| | | | PiperOrigin-RevId: 214698827
* Add experimental asynchronous checkpoint hook.Gravatar Russell Power2018-09-26
| | | | | | | | | This triggers checkpoints in a separate thread while allowing training to continue. This can effectively parallelize checkpointing and training for workloads like TPUEstimator, where the weights are only updated after a number of device iterations. PiperOrigin-RevId: 214670991
* Support dynamic LR for Keras optimizer by setting the global Keras session.Gravatar A. Unique TensorFlower2018-09-25
| | | | PiperOrigin-RevId: 214532827
* Fix documentationGravatar A. Unique TensorFlower2018-09-25
| | | | PiperOrigin-RevId: 214499034
* Add outputs and target cross replica concat, so each core sees the same ↵Gravatar Jianwei Xie2018-09-25
| | | | | | output and targets and produces the same loss and metrics. PiperOrigin-RevId: 214494877
* Optimize Resnet-50 infeed pipeline by adding sharding and cache.Gravatar A. Unique TensorFlower2018-09-25
| | | | PiperOrigin-RevId: 214489904
* Fix keras_support.tpu_model example usages.Gravatar Ruoxin Sang2018-09-25
| | | | PiperOrigin-RevId: 214476713
* [TF:XLA] Introduce CollectivePermute op.Gravatar Youlong Cheng2018-09-24
| | | | PiperOrigin-RevId: 214373714
* Upgrade cloud tpu profiler to 1.11.Gravatar A. Unique TensorFlower2018-09-24
| | | | PiperOrigin-RevId: 214359786
* [TPU] Change the TPU DeviceAssignment class to use a flatter (replica, ↵Gravatar Peter Hawkins2018-09-21
| | | | | | | | | | logical core) indexing scheme for cores. Previously the DeviceAssignment class mixed both a general concept (a mapping from (replica, logical core) to physical TPU core) and a specific instantiation of that concept, by imposing a particular 3D grid structure on the logical core numbers. This was excessive ? while the physical core numbers have a particular structure, there is no need to impose any particular structure on the logical core numbers. This change simplifies the DeviceAssignment scheme, changing it so logical cores within a replica are numbered sequentially without any particular semantics. PiperOrigin-RevId: 213984629
* Support 16 ways model parallelism.Gravatar Youlong Cheng2018-09-20
| | | | PiperOrigin-RevId: 213913013
* Moving tpu_embedding_config.proto to tpu_embedding_configuration.proto, ↵Gravatar Daryl Ng2018-09-20
| | | | | | refactoring it, adding several new fields and an EmbeddingOutputLayout message to provide experimental support for controlling the embedding output. PiperOrigin-RevId: 213849572
* Fix the error message thrown when running eval on podGravatar Sourabh Bajaj2018-09-19
| | | | PiperOrigin-RevId: 213730668
* Add xla.compile(), a low-level API that compiles graph with XLA.Gravatar Yanan Cao2018-09-18
| | | | PiperOrigin-RevId: 213574904
* Add Keras TPU support for the new metrics.Gravatar Pavithra Vijay2018-09-17
| | | | PiperOrigin-RevId: 213378552
* Add type checking at the beginning of tpu.shard().Gravatar Yunxing Dai2018-09-17
| | | | | | Otherwise a message like "TypeError: Tensor objects are only iterable when eager execution is enabled. To iterate over this tensor use tf.map_fn." will be thrown, which is confusing. PiperOrigin-RevId: 213371676
* Add support for predicting models with learning_phase.Gravatar Jianwei Xie2018-09-17
| | | | PiperOrigin-RevId: 213327633
generated by cgit v1.2.3 (git 2.39.1) at 2024-06-13 12:43:41 +0000