| Commit message (Collapse) | Author | Age |
|
|
|
|
|
|
| |
Add --config=android_arm and --config=android_arm64 Bazel configs
to add options needed for Android Bazel builds.
PiperOrigin-RevId: 173340664
|
|
|
|
|
|
| |
Add tf.contrib.data.make_saveable_from_iterator(iterator) that builds a SaveableObject for an iterator so it can be saved/restored using tf.Saver.
PiperOrigin-RevId: 173340191
|
|
|
|
|
|
| |
With this method users have access to a raw Session while getting the benefit of recoverable behavior of MonitoredSession.
PiperOrigin-RevId: 173334319
|
|
|
|
| |
PiperOrigin-RevId: 173333650
|
|
|
|
|
|
|
| |
Also disables all functional layers until a non-default store is
implemented.
PiperOrigin-RevId: 173333446
|
|
|
|
|
|
| |
The 'filter' variable wasn't initialized for OIHW filter format.
PiperOrigin-RevId: 173327533
|
|
|
|
| |
PiperOrigin-RevId: 173324359
|
|
|
|
| |
PiperOrigin-RevId: 173324238
|
|
|
|
| |
PiperOrigin-RevId: 173324074
|
|
|
|
|
|
|
|
|
|
|
| |
* Get graph execution working, by adding init_variables() and
supporting graph mode in evaluate_on_dataset().
* Use track_*() instead of add_*() to match Network.
* Fill in several doc strings.
* Detect metric added to two different Evaluators.
* Return prefix along with metrics from metrics property.
PiperOrigin-RevId: 173320585
|
|
|
|
| |
PiperOrigin-RevId: 173320547
|
|
|
|
|
|
|
| |
grpc and grpc::internal namespaces while versions of gRPC
are in transition
PiperOrigin-RevId: 173317900
|
|
|
|
|
|
| |
presence of input Tensors.
PiperOrigin-RevId: 173314020
|
|
|
|
| |
PiperOrigin-RevId: 173313459
|
|
|
|
| |
PiperOrigin-RevId: 173312134
|
|
|
|
| |
PiperOrigin-RevId: 173308794
|
|
|
|
| |
PiperOrigin-RevId: 173307212
|
|
|
|
|
|
|
|
|
|
| |
We want variables with eager execution to have object semantics instead of name semantics and this is a small step in that direction.
This means that the functional style layer invocations (tf.layers.dense() etc.)
will NOT work when eager execution is enabled. Instead, use of the object-oriented
layers is advised.
PiperOrigin-RevId: 173306447
|
|
|
|
|
|
|
| |
Fix collocate_gradients for initial losses.
Remove NcclBroadcast gradient test for now. The generated AddN to accumulate the broadcast outputs before passing it to the gradient function is CPU only and cannot be collocated with NcclBroadcast on the GPU.
PiperOrigin-RevId: 173306409
|
|
|
|
|
|
|
| |
e.g.:
http://ci.tensorflow.org/job/nightly-matrix-linux-gpu-clang/159/console
PiperOrigin-RevId: 173305545
|
|
|
|
|
|
|
|
| |
* Avoid situation where variables are created in the wrong graph.
* Add an init_variables() method that returns an op that will initialize
any non-initialized variables.
PiperOrigin-RevId: 173302832
|
|
|
|
| |
PiperOrigin-RevId: 173299853
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
171084886 had to be rolled back twice due to various open source build issues.
I'm trying again, now that I think I've addressed all the pertinent issues.
Original CL description:
Don't use dlsym to resolve symbols in the CPU JIT
Instead of resolving symbols via dlsym when JITting for the CPU backend, use a
registry based mechanism. This lets us kill off the --export_dynamic hack that
we used to need for CustomCall on the CPU backend.
PiperOrigin-RevId: 173277862
|
|
|
|
|
|
|
|
| |
* Switch GPU implementation of Determinant to use the more numerically stable kernel as well.
* Change behavior for Determinant on matrices with (numerically) infinite determinants to match the behavior of numpy.linalg.det: Return inf for matrix with infinite determinant.
* Misc. cleanup in code working around missing support for complex in the NVCC compiler.
PiperOrigin-RevId: 173277377
|
|
|
|
|
|
| |
BeamSearchDecoder.
PiperOrigin-RevId: 173273139
|
|
|
|
| |
PiperOrigin-RevId: 173272637
|
|
|
|
| |
PiperOrigin-RevId: 173266690
|
|
|
|
| |
PiperOrigin-RevId: 173263867
|
|
|
|
|
|
|
|
|
|
|
|
| |
Bug is at:
#5804
Fix is to add a unique identifier to each temp file name. The id is unique to
the process. Multiple processes could still have a conflict, though even there
the odds do go down somewhat with this fix.
PiperOrigin-RevId: 173261202
|
|
|
|
| |
PiperOrigin-RevId: 173260210
|
|
|
|
| |
PiperOrigin-RevId: 173214110
|
|
|
|
|
|
|
| |
Create "add()" method (analogous to Keras') now that add/track_layer()
is inappropriate to use.
PiperOrigin-RevId: 173213945
|
|
|
|
| |
PiperOrigin-RevId: 173213868
|
|
|
|
| |
PiperOrigin-RevId: 173208878
|
|
|
|
| |
PiperOrigin-RevId: 173207602
|
|
|
|
|
|
| |
tensor have more than 2^32 elements.
PiperOrigin-RevId: 173203430
|
|
|
|
| |
PiperOrigin-RevId: 173194115
|
|
|
|
| |
PiperOrigin-RevId: 173193748
|
|
|
|
|
|
| |
random adjustments, it is possible to improve the model generalization.
PiperOrigin-RevId: 173192179
|
|
|
|
| |
PiperOrigin-RevId: 173188867
|
|
|
|
|
|
| |
components).
PiperOrigin-RevId: 173188044
|
|
|
|
| |
PiperOrigin-RevId: 173184568
|
|
|
|
| |
PiperOrigin-RevId: 173178780
|
|
|
|
| |
PiperOrigin-RevId: 173176850
|
|
|
|
| |
PiperOrigin-RevId: 173172916
|
|
|
|
|
|
| |
arguments.
PiperOrigin-RevId: 173172808
|
|
|
|
| |
PiperOrigin-RevId: 173172604
|
|
|
|
| |
PiperOrigin-RevId: 173172018
|
|
|
|
| |
PiperOrigin-RevId: 173171871
|
|
|
|
| |
PiperOrigin-RevId: 173170394
|