| Commit message (Collapse) | Author | Age |
|
|
|
| |
Change: 144763407
|
|
|
|
|
|
|
|
| |
Also add functions to test whether we're running on Linux or macOS, and convert
every check over to them.
Fixes #6869
Change: 144761341
|
|
|
|
| |
Change: 144760356
|
|
|
|
|
|
| |
as that state can be cleared.
Change: 144760207
|
|
|
|
| |
Change: 144755322
|
|
|
|
| |
Change: 144754803
|
|
|
|
|
| |
Fixes #6833
Change: 144752893
|
|
|
|
|
|
| |
for `(non_)trainable_weights`.
Change: 144752883
|
|
|
|
| |
Change: 144752664
|
|
|
|
|
|
|
| |
This extra parameter got introduced in a previous commit as part
of some performance improvements. But because it is unused, it
should be reverted back.
Change: 144749662
|
|
|
|
| |
Change: 144749626
|
|
|
|
|
|
|
|
|
|
|
|
| |
This patch:
- Updates // comments to ///. I manually reverted some comments that shouldn't be docs (e.g. TODOs), but may have missed some.
- Indents code blocks so they get formatted as such in the docs.
- Removes /* */ comments from example code since it messes up Doxygen.
- Puts a space between {{ and }} since it messes up devsite.
- Adds some // START_SKIP_DOXYGEN and // END_SKIP_DOXYGEN comments for functions that aren't part of the public API (incomplete)
This will likely require further small fixups, but this gets something to be generated.
Change: 144749351
|
|
|
|
| |
Change: 144749245
|
|
|
|
|
|
|
| |
Make LoggingTensorHook to print tensors in the order they were given [if list].
Added formatter option, to support custom string formatting.
Added numpy printing options configuration to tweak precision and summarization.
Change: 144748847
|
|
|
|
| |
Change: 144747557
|
|
|
|
| |
Change: 144746814
|
|
|
|
|
| |
No implementations are yet provided for these operations.
Change: 144743665
|
|
|
|
|
|
|
| |
Now, at least for the public APIs that return port::Status, they can grab the
port::Status that the implementation would like to return and use its
additional information in reporting to the user.
Change: 144741667
|
|
|
|
| |
Change: 144739269
|
|
|
|
| |
Change: 144738033
|
|
|
|
|
| |
This method is only used in other `HloInstruction` creator methods.
Change: 144737654
|
|
|
|
| |
Change: 144737436
|
|
|
|
|
| |
provided. This is probably a better default to have.
Change: 144737122
|
|
|
|
| |
Change: 144734632
|
|
|
|
| |
Change: 144733955
|
|
|
|
|
|
|
|
| |
Still hidden from tf.contrib.seq2seq namespace for rapid iteration.
Not for general use until the API has stabilized a bit.
Barebones documentation for now.
Change: 144733714
|
|
|
|
|
|
| |
ambiguity on broadcast rules.
Change: 144733601
|
|
|
|
| |
Change: 144733155
|
|
|
|
| |
Change: 144729490
|
|
|
|
|
| |
Similarly for other "dynamic" Ops.
Change: 144728885
|
|
|
|
|
|
| |
to optimizer name checking)
Change: 144727719
|
|
|
|
| |
Change: 144726940
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The previous implementation of resize_bilinear operation is a straight
forward, naive implementation. Unfortunately, it leaves a significant amount
of room for improvement.
This change improves the performance of the operation by avoiding unnecessary
recomputation (for where the operation was CPU bound), and avoids redundant
memory references (for where the operation was memory bound).
In my benchmark (loosely based on inception image pipeline), this implementation
change improves speed between 1.5-2x.
Additionally, to ensure correctness, I've preserved the old implementation of
resize_bilinear, and added a number of tests that initalize a random image,
and ensure that the outputs are identical.
Change: 144726607
|
|
|
|
|
|
| |
commonly used in example models.
Change: 144721588
|
|
|
|
|
|
|
|
|
|
|
| |
Based on instrumentation of the imagenet/inception_train typical input &
output image sizes, we have devised a micro benchmark to verify the performance
of the resize_bilinear op.
To compile and run:
- bazel build -c opt --copt=-mavx //tensorflow/python/image_ops_test
- ./bazel-bin/tensorflow/python/image_ops_test --benchmarks=ResizeBilinear
Change: 144718335
|
|
|
|
|
|
|
|
|
| |
generators are stateful.
Prevents Tensorflow from constant-folding _XlaLaunch ops containing random-number ops. Fixes Github issue: #6854
Enable RandomStandardNormal for the XLA CPU backend.
Change: 144716342
|
|
|
|
| |
Change: 144713722
|
|
|
|
| |
Change: 144689500
|
|
|
|
| |
Change: 144675800
|
|
|
|
| |
Change: 144673014
|
|
|
|
| |
Change: 144669682
|
|
|
|
| |
Change: 144656658
|
|
|
|
| |
Change: 144648190
|
|
|
|
|
|
|
| |
The newly-added GrpcDebugWrapperSesion and the existing DumpingDebugWrapperSession now share a same superclass, namely the newly-created NonInteractiveDebugWrapperSession in python/debug/wrappers/framework.py.
GrpcDebugWrapperSession is not yet exposed in the tensorflow.python.debug namespace, for rapid iteration reasons. It'll be made public once its API has stabilized together with grpc_debug_server.
Change: 144647632
|
|
|
|
| |
Change: 144623381
|
|
|
|
|
| |
(can be used to have larger batch sizes for smaller input sequences.)
Change: 144621786
|
|
|
|
| |
Change: 144609556
|
|
|
|
|
|
| |
fixes to it.
Change: 144588235
|
|
|
|
| |
Change: 144584357
|
|
|
|
| |
Change: 144575750
|