Commit message (Collapse) | Author | Age | |
---|---|---|---|
* | Move equal_graph_def.h/cc from core/graph to core/util | 2017-03-29 | |
| | | | | | | This fixes build dependencies for a future patch (specifically allows referencing equal_graph_def.h from core/framework) Change: 151668115 | ||
* | Improves the way to determine the train_steps_per_iteration for ↵ | 2017-03-29 | |
| | | | | | | continuous_train_and_eval. Change: 151666217 | ||
* | [TF:XLA:GPU] Fix flaky convolution tests by increasing test tolerance. | 2017-03-29 | |
| | | | | Change: 151662713 | ||
* | Move some asserts and other code from contrib.framework to core. | 2017-03-29 | |
| | | | | Change: 151657164 | ||
* | Make quantize training rewriter add variables to save and restore ops if a ↵ | 2017-03-29 | |
| | | | | | | save op exists. This allows checkpointing of variables created by the rewriter. Change: 151656461 | ||
* | Improve error message when converting unconvertable lists of objects to Tensor. | 2017-03-29 | |
| | | | | | | | | | | | At the moment, we don't know how to convert a list of objects such as TensorShape or Dimension to a Tensor, or a list of objects where type promotion may be slightly ambiguous (e.g., [1.0, Dimension(1)]). Before this change, the code would dump an unhelp error message when trying to interpret these objects using compat.as_bytes(), so instead we catch the error and raise a slightly more informative error that suggests where the original list comes from to the user. Change: 151653212 | ||
* | Add a SessionRunHook to manage synchronization of the row and column sweeps ↵ | 2017-03-29 | |
| | | | | | | in the WALS solver. Change: 151652662 | ||
* | Unregister the CUPTI callbacks if needed when the gpu tracer is deleted to ↵ | 2017-03-29 | |
| | | | | | | prevent them from accessing freed memory. Change: 151650677 | ||
* | [TF contrib seq2seq] Cleanup normalized attention documentation and API | 2017-03-29 | |
| | | | | | | (removes the unnecessary r offset; since with softmax attention it makes no difference) Change: 151636533 | ||
* | Add support for `input_layer_min_slice_size` to `DNNLinearCombinedEstimator`, | 2017-03-29 | |
| | | | | | | and make `DNNLinearCombinedClassifier` and `DNNLinearCombinedRegressor` subclasses of `DNNLinearCombinedEstimator`. Change: 151635758 | ||
* | Ensure tf.layers._Layer objects are not used across multiple graphs. | 2017-03-29 | |
| | | | | | This wreaks havoc with tracking variables and other internal state and just isn't supported. Change: 151633894 | ||
* | matmul_with_broadcast added to linear_operator_util. This is like matmul, but | 2017-03-29 | |
| | | | | | | | will work for non-matching batch dimensions. This is achieved by broadcasting, by adding a Tensor of zeros to arguments that cannot be determined to already be of the broadcast shape. Change: 151631491 | ||
* | Use pip2 instead of `pip` for installing pip packages for python2. | 2017-03-29 | |
| | | | | Change: 151630285 | ||
* | Fusion uses the same cost model for all backends when it comes to | 2017-03-29 | |
| | | | | | rematerializing for fusion. Let subtypes override the default cost model. Change: 151626001 | ||
* | tfdbg: address thread sanitizer warning | 2017-03-29 | |
| | | | | Change: 151619218 | ||
* | Use a constant initialization for Phased LSTM test. | 2017-03-29 | |
| | | | | Change: 151616172 | ||
* | Add test for assertion error when trying to decode jpegs at 16 bit. | 2017-03-29 | |
| | | | | Change: 151613583 | ||
* | Include PhasedLSTMCell to the tf.contrib.rnn module. | 2017-03-29 | |
| | | | | Change: 151607738 | ||
* | Add plugin assets for the Embedding Projector and expose the API via ↵ | 2017-03-29 | |
| | | | | | | tf.contrib.tensorboard.plugins.projector Change: 151603470 | ||
* | Handle optional collections parameter correctly in tf.summary.text. | 2017-03-29 | |
| | | | | Change: 151602957 | ||
* | Remove unnecessary extraneous local variable. | 2017-03-29 | |
| | | | | | We can re-use the status local variable. Change: 151602342 | ||
* | Remove superfluous dependencies. | 2017-03-29 | |
| | | | | Change: 151600516 | ||
* | Seed the random number generator with a constant before evaluating its quality. | 2017-03-29 | |
| | | | | Change: 151596861 | ||
* | Disable input forwarding for tensors fed from python or returned from py_func. | 2017-03-29 | |
| | | | | Change: 151591184 | ||
* | Comment update for scattered_embedding_column. | 2017-03-29 | |
| | | | | Change: 151590509 | ||
* | Fix transpose benchmark's throughput computation. | 2017-03-29 | |
| | | | | Change: 151589798 | ||
* | Internal change. | 2017-03-29 | |
| | | | | Change: 151587999 | ||
* | Set the visibility of the bigquery_reader_ops target to public. | 2017-03-29 | |
| | | | | Change: 151586548 | ||
* | Consolidate the session_handle delete session runs. | 2017-03-29 | |
| | | | | | Experiments on MNIST shows about 5x speed up: 2000+us -> ~400us. Change: 151585909 | ||
* | [TF contrib seq2seq] Update AttentionWrapper behavior: | 2017-03-29 | |
| | | | | | | * New variable_scope in the wrapper's call. * Dense layers no longer use bias (following the papers) Change: 151583457 | ||
* | Make `ReferenceResolver` serializable. | 2017-03-29 | |
| | | | | Change: 151583070 | ||
* | Pass in optional dtype to image decoder. Default behavior is identical. | 2017-03-29 | |
| | | | | Change: 151582088 | ||
* | py_func no longer copies return values (attempt #3) | 2017-03-29 | |
| | | | | Change: 151578180 | ||
* | Add some basic info logging to Supervisor to make debugging slightly easier. | 2017-03-29 | |
| | | | | Change: 151567906 | ||
* | [XLA] Move kPad from GpuElementalIrEmitter::MakeElementGenerator to ↵ | 2017-03-29 | |
| | | | | | | | | | ElementalIrEmitter::MakeElementGenerator There is nothing GPU specific in GpuElementalIrEmitter::MakeElementGenerator for kPad. Move it into the base implementation so that all subcalses have it as an implementation. Change: 151564674 | ||
* | Improve Phased LSTM. | 2017-03-29 | |
| | | | | | Allow time to be of type float64. Change: 151547044 | ||
* | Fixed some formatting issues in Installing TF on Linux and Installing TF on ↵ | 2017-03-28 | |
| | | | | | | Mac OS. Change: 151538676 | ||
* | [XLA] Change TODO(leary) to TODO(b/...) | 2017-03-28 | |
| | | | | Change: 151537962 | ||
* | Remove unnecessary memory allocation. | 2017-03-28 | |
| | | | | | | Minor cleanup to remove unnecessary pointer allocation of the cancellation manager. Change: 151528507 | ||
* | Set the model_dir in run_config in contrib.Estimator if None | 2017-03-28 | |
| | | | | Change: 151527479 | ||
* | Decrease volume of spammy logs. | 2017-03-28 | |
| | | | | | Remove unused _monitored_train. Change: 151526705 | ||
* | Fix incorrect inclusion of tensor_forest all_ops. | 2017-03-28 | |
| | | | | Change: 151524146 | ||
* | Patching a typo in the install mac slave and install pip packages with relation | 2017-03-28 | |
| | | | | | to the markdown install command. Change: 151522575 | ||
* | Autogenerated Change: Change TensorBoard TAG to 51 | 2017-03-28 | |
| | | | | Change: 151519658 | ||
* | [XLA] Break fix for compiler scoping error in OSS compiler. | 2017-03-28 | |
| | | | | Change: 151518190 | ||
* | Fix shape handling errir in convolutional layer. | 2017-03-28 | |
| | | | | | | Indexing TensorShape returns a Dimension object, and the result cannot be compared directly to None. Change: 151517447 | ||
* | Do not update given dictionary in feature-column public utils. | 2017-03-28 | |
| | | | | Change: 151516918 | ||
* | Support for domain buckets, e.g gs://bucketname.domain-example.com/... | 2017-03-28 | |
| | | | | Change: 151516660 | ||
* | Add tf.summary.text to the TensorFlow API. | 2017-03-28 | |
| | | | | | tf.summary.text writes scalar string tensors into summaries, for visualization in the TensorBoard text dashboard. Markdown is supported. Change: 151511881 | ||
* | Update TensorBoard README. | 2017-03-28 | |
| | | | | Change: 151511272 |