- 08 12月, 2018 38 次提交
-
-
由 A. Unique TensorFlower 提交于
PiperOrigin-RevId: 224586292
-
由 Derek Murray 提交于
PiperOrigin-RevId: 224584699
-
由 Suharsh Sivakumar 提交于
PiperOrigin-RevId: 224581893
-
由 A. Unique TensorFlower 提交于
PiperOrigin-RevId: 224580438
-
由 A. Unique TensorFlower 提交于
PiperOrigin-RevId: 224578284
-
由 Scott Zhu 提交于
Also added test case to cover that. PiperOrigin-RevId: 224577299
-
由 A. Unique TensorFlower 提交于
but the implementation directly references google::protobuf. Fix it so that it agrees with its header file. PiperOrigin-RevId: 224576410
-
由 TensorFlower Gardener 提交于
PiperOrigin-RevId: 224576050
-
由 Alexandre Passos 提交于
Relies on being able to run the assert* test methods inside a py_func to run them inside the graph, so there's no need for self.evaluate or similar methods which create a graph/eager hybrid programming model. PiperOrigin-RevId: 224575790
-
由 Dan Moldovan 提交于
Use eager_py_func instead of the deprecated py_func. Add a bit of extra logic to account for the potential use of tf.print in Python 2. PiperOrigin-RevId: 224574979
-
由 Gaurav Jain 提交于
PiperOrigin-RevId: 224572844
-
由 A. Unique TensorFlower 提交于
PiperOrigin-RevId: 224571863
-
由 Jianwei Xie 提交于
PiperOrigin-RevId: 224570819
-
由 A. Unique TensorFlower 提交于
PiperOrigin-RevId: 224570187
-
由 Scott Zhu 提交于
PiperOrigin-RevId: 224567398
-
由 A. Unique TensorFlower 提交于
PiperOrigin-RevId: 224565798
-
由 A. Unique TensorFlower 提交于
PiperOrigin-RevId: 224565479
-
由 A. Unique TensorFlower 提交于
PiperOrigin-RevId: 224565203
-
由 Allen Lavoie 提交于
Pulls some utilities out of saver.py which are necessary to actually use it. The functional saver takes only SaveableObjects, so these are utilities for taking a list of whatever users pass in and converting them to those. One other code move for object-based checkpointing to avoid circular imports. Applications which need a SaverDef still use the old Saver. Serialization to SaverDef will be added to this saver in a followup. Does not actually wrap the new Saver's methods in @tf.function yet, since there are memory issues which need to be fixed first. PiperOrigin-RevId: 224561069
-
由 A. Unique TensorFlower 提交于
Fix NVPTX TargetMachine construction to use reloc/code model flags only if explicitly set. Fixes failures caused by r348585. PiperOrigin-RevId: 224560930
-
由 Igor Ganichev 提交于
Compiling a defun can be done with: @function.defun def f(x): return x + 1 with tf.device("XLA_GPU"): f(5) PiperOrigin-RevId: 224560868
-
由 Jared Duke 提交于
Strings are not compatible with ByteBuffer inputs/outputs. Make that clear in the documentation, also highlighting how they differ from primitive types. PiperOrigin-RevId: 224559052
-
由 A. Unique TensorFlower 提交于
PiperOrigin-RevId: 224548216
-
由 TensorFlower Gardener 提交于
PiperOrigin-RevId: 224547163
-
由 Gaurav Jain 提交于
- Add cond_v2/while_v2 dependencies - Add run_deprecated_v1 and run_v1_only where needed - Remove non-strict cond tests PiperOrigin-RevId: 224543529
-
由 A. Unique TensorFlower 提交于
PiperOrigin-RevId: 224543122
-
由 Alexandre Passos 提交于
PiperOrigin-RevId: 224541903
-
由 Yu-Cheng Ling 提交于
PiperOrigin-RevId: 224540375
-
由 Rachel Lim 提交于
PiperOrigin-RevId: 224535049
-
由 Mihai Maruseac 提交于
PiperOrigin-RevId: 224535013
-
由 Jianwei Xie 提交于
PiperOrigin-RevId: 224534209
-
由 Smit Hinsu 提交于
PiperOrigin-RevId: 224531965
-
由 Ayush Dubey 提交于
Since NCCL 2.1.2, NCCL uses cooperative group launch by default which results in deadlocks in the NcclManager test if number of ranks > number of GPUs. Setting NCCL_LAUNCH_MODE to PARALLEL, which was the default before NCCL 2.1.2, fixes this. PiperOrigin-RevId: 224531851
-
由 Scott Zhu 提交于
This heavily clear up the logic for save/load model since the weights for input bias is gone. The new implementation is mathematically same as the canonical LSTM. PiperOrigin-RevId: 224531752
-
由 Rohan Jain 提交于
to GPU while keeping everything in host memory. This is necessary for when we move to creating datasets on the go and for transformations such as copy_to_device we need to ferry the DatasetVariantWrapper. PiperOrigin-RevId: 224531292
-
由 Martin Wicke 提交于
Add default value for topdown aerg of io.gfile.walk. The default exists in Python's os.walk and in the v1 gfile too. PiperOrigin-RevId: 224530098
-
由 Eugene Brevdo 提交于
This provides a consistent API for users who have scalar "channels". PiperOrigin-RevId: 224528301
-
由 Brennan Saeta 提交于
#cleanup PiperOrigin-RevId: 224518435
-
- 07 12月, 2018 2 次提交
-
-
由 James Ring 提交于
PiperOrigin-RevId: 224513745
-
由 Gaurav Jain 提交于
In order to do so we introduce the concept of a FakeEagerSession which allows tests to run sess.run() with an empty feed_dict. PiperOrigin-RevId: 224495103
-