| Commit message (Collapse) | Author | Age |
|
|
|
|
|
|
|
|
|
|
| |
https://github.com/tensorflow/community/pull/16.
In addition to the changes in the doc, I made the following updates (these changes make sense to me and I didn't notice them when compiling the doc):
* deprecate saved_model.builder.SavedModelBuilder - replaced with saved_model.SavedModelBuilder
* deprecate python_io.tf_record_iterator - replaced with io.tf_record_iterator
* deprecate python_io.TFRecordWriter - replaced with io.TFRecordWriter
* move reduce_join to tf.string
PiperOrigin-RevId: 215253944
|
|
|
|
| |
PiperOrigin-RevId: 214824023
|
|
|
|
| |
PiperOrigin-RevId: 212582822
|
|
|
|
|
|
| |
saved_model/signature_def_utils
PiperOrigin-RevId: 211858972
|
|
|
|
|
|
| |
into saved_models.utils.get_tensor_from_tensor_info
PiperOrigin-RevId: 210417762
|
|
|
|
|
|
| |
self.test_session() has been deprecated in cl/208545396 as its behavior confuses readers of the test. Moving to self.session() instead.
PiperOrigin-RevId: 209696110
|
|
|
|
| |
PiperOrigin-RevId: 208094190
|
|
|
|
|
|
| |
to/from checkpoint. Currently, the model topology is still loaded from json (placed under saved_model/assets). Later, we will load from saved_model.pb.
PiperOrigin-RevId: 206412614
|
|
|
|
| |
PiperOrigin-RevId: 206048542
|
|
|
|
|
|
| |
we remove duplicated code paths by mapping legacy_init_op into main_op in the SavedModelBuilder, and we deprecate the legacy_init_op arg. Note that the loader will still look for both, so old SavedModels will still load without trouble.
PiperOrigin-RevId: 206026743
|
|
|
|
| |
PiperOrigin-RevId: 205875586
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This change makes Estimator.train() support v2 summaries (tf.contrib.summary.*) out-of-the-box, to match the support for v1 summaries. Estimator.train() will now handle the boilerplate necessary to initialize a file writer and enable summary writing every N steps, and will ensure that its own automatically exported summaries (for loss and global_step/sec) get written to the same underlying events file.
As part of this change, tf.train.SummarySaverHook, tf.train.CheckpointSaverHook, tf.train.StepCounterHook, and tf.train.ProfilerHook have also been adapted to write summaries using the v2 summary system (via a compatibility layer), instead of using FileWriterCache.
A couple additional smaller changes are:
- the 'session' parameter to FileWriter() can now be a callable returning a tf.Session instance.
- the introduction of tf.contrib.summary.record_summaries_if() which takes a boolean tensor for direct control of tf.contrib.summary.should_record_summaries().
- EstimatorSpec.train_op, besides a tf.Operation, is now allowed to be any Tensor-equivalent object rather than just a tf.Tensor.
PiperOrigin-RevId: 205843986
|
|
|
|
| |
PiperOrigin-RevId: 205746329
|
|
|
|
|
|
| |
we remove duplicated code paths by mapping legacy_init_op into main_op in the SavedModelBuilder, and we deprecate the legacy_init_op arg. Note that the loader will still look for both, so old SavedModels will still load without trouble.
PiperOrigin-RevId: 205728344
|
|
|
|
| |
PiperOrigin-RevId: 205466000
|
|
|
|
| |
PiperOrigin-RevId: 205449314
|
|
|
|
|
|
| |
variable restoration steps.
PiperOrigin-RevId: 201268712
|
|
|
|
| |
PiperOrigin-RevId: 200802842
|
|
|
|
|
|
| |
variable restoration steps.
PiperOrigin-RevId: 200747752
|
|
|
|
| |
PiperOrigin-RevId: 200525639
|
|
|
|
| |
PiperOrigin-RevId: 200467580
|
|
|
|
|
|
|
|
| |
tf.saved_model.loader.load
This change explicitly declares import_scope as a kwarg for tf.saved_model.loader.load. Previously, tf.saved_model.loader.load implicitly accepted import_scope and passed it through to import_meta_graph through **saver_kwargs.
PiperOrigin-RevId: 200249417
|
|
|
|
|
|
| |
in Estimator model functions through to the Builder when saving.
PiperOrigin-RevId: 199546645
|
|
|
|
|
|
|
| |
Revert #18413. Too many internal test failures due to the name scope change caused by this change.
Revert #18192. Cannot use re2::StringPiece internally. Need alternative for set call. Will pull and clean this up in a separate change.
PiperOrigin-RevId: 197991247
|
|
|
|
|
|
| |
point to distinct files.
PiperOrigin-RevId: 197835288
|
|
|
|
|
|
| |
This is in preparation for removing the _USE_C_API toggle altogether.
PiperOrigin-RevId: 196920481
|
|
|
|
|
|
| |
Estimator with two new methods, available in tf.contrib: export_all_saved_models and export_saved_model_for_mode.
PiperOrigin-RevId: 195485922
|
|
|
|
| |
PiperOrigin-RevId: 194274698
|
|
|
|
| |
PiperOrigin-RevId: 190878279
|
|
|
|
| |
PiperOrigin-RevId: 187883080
|
|
|
|
|
|
|
|
|
| |
It turns out that the original Python code modifies the graph_def argument
to add default attr values. I'm not sure if the behavior is covered by
our API guarantees since it's not documented, but let's keep the behavior
consistent for now.
PiperOrigin-RevId: 185193037
|
|
|
|
|
|
|
|
|
|
|
| |
with C API.
The test originally altered the Python version of the op registry,
which is not reflected in the C API. This changes the test to alter
the serialized node def instead of the op def, and renames the test to
testInconsistentConsumerDefaultAttrs.
PiperOrigin-RevId: 185067838
|
|
|
|
| |
PiperOrigin-RevId: 184897758
|
|
|
|
| |
PiperOrigin-RevId: 184708277
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This change makes the test_util.with_c_api decorator call
reset_default_graph() after enabling or disabling the C API instead of
creating a new Graph. This makes it more robust to tests that call
reset_default_graph(), which requires that the current default graph
isn't nested (which the C API-enabled Graph previously was).
In addition, enables the C API with saved_model_test.py (which
required the above change). A few tests still need further changes,
which I'll post in subsequent patches.
PiperOrigin-RevId: 183739148
|
|
|
|
| |
PiperOrigin-RevId: 183467186
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Following APIs now accept an additional argument (`strip_default_attrs`) to
enable/disable (default:disabled) stripping of default valued attributes in a NodeDef:
o meta_graph: export_meta_graph, create_meta_graph.
o saver: Saver.save, Saver.export_meta_graph.
o builder: SavedModelBuilder.add_meta_graph,
SavedModelBuilder.add_meta_graph_and_variables.
o estimator: Estimator.export_savedmodel.
Related changes:
o Pywrap C++ AreAttrValuesEqual to compare two AttrValue instances.
This allows for a single/canonical way of comparing AttrValues in C++/Python.
o Add a utility method to meta_graph.py to get the node def by name in a graph def.
o Update SavedModelBuilder documentation on relevance of strip_default_attrs flag.
PiperOrigin-RevId: 180619001
|
|
|
|
|
|
| |
tensorflow/python/saved_model.
PiperOrigin-RevId: 179599527
|
|
|
|
| |
PiperOrigin-RevId: 179309732
|
|
|
|
|
|
| |
the implicit device.
PiperOrigin-RevId: 176544698
|
|
|
|
|
|
|
|
| |
* Log report of which signatures are produced and which TF Serving APIs are targeted.
* Improve docstrings for signature_def builders, explaining the TF Serving API constraints.
* Accept a single Tensor as a prediction output (which will be named 'output').
PiperOrigin-RevId: 172831366
|
|
|
|
|
|
|
|
| |
Previously, when the user restores graph from one meta graph generated from
saved_model and then re-generates another saved model, the re-generated model
will be invalid because it will contain duplicate legacy_init_ops.
PiperOrigin-RevId: 171099152
|
|
|
|
| |
PiperOrigin-RevId: 170376578
|
|
|
|
|
|
| |
This can be thought of as a smoothed version of a ReLU. On Imagenet, alpha=0.3 gives 0.6-1% improvement in validation accuracy compared to ReLU, by reducing the generalization gap.
PiperOrigin-RevId: 170376244
|
|
|
|
| |
PiperOrigin-RevId: 170376578
|
|
|
|
|
|
| |
and shapes.
PiperOrigin-RevId: 168820997
|
|
|
|
| |
PiperOrigin-RevId: 165530192
|
|
|
|
|
|
| |
way of getting a SparseTensor of a particular shape.
PiperOrigin-RevId: 165322708
|
|
|
|
| |
PiperOrigin-RevId: 164791375
|
|
|
|
| |
PiperOrigin-RevId: 161242660
|