| Commit message (Collapse) | Author | Age |
|
|
|
|
|
| |
only works for processes running on CPU's only.
PiperOrigin-RevId: 170725482
|
|
|
|
|
|
| |
conversion from non-integer types to uint16_t.
PiperOrigin-RevId: 170717628
|
|
|
|
| |
PiperOrigin-RevId: 170716623
|
|
|
|
| |
PiperOrigin-RevId: 170710055
|
|
|
|
| |
PiperOrigin-RevId: 170705212
|
|
|
|
| |
PiperOrigin-RevId: 170696315
|
|
|
|
|
|
| |
AlgebraicSimplifier.
PiperOrigin-RevId: 170695891
|
|
|
|
| |
PiperOrigin-RevId: 170696315
|
|
|
|
|
|
| |
AlgebraicSimplifier.
PiperOrigin-RevId: 170695891
|
|
|
|
|
|
|
| |
This affects whether we generate movaps (with hard 16-byte alignment) or
movups.
PiperOrigin-RevId: 170687148
|
|
|
|
|
|
|
|
| |
VariableScope ignores the reuse argument when eager execution is enabled
and treats it as AUTO_REUSE. So the caller doesn't have to explicitly do
so.
PiperOrigin-RevId: 170672112
|
|
|
|
| |
PiperOrigin-RevId: 170671787
|
|
|
|
|
|
|
|
|
|
| |
inverse_stft.
Also add tests for stft/inverse_stft when the shape/rank of the inputs are unknown.
Fixes GitHub Issue #13363.
PiperOrigin-RevId: 170662530
|
|
|
|
|
|
| |
operations fail.
PiperOrigin-RevId: 170661089
|
|
|
|
|
|
| |
constant.
PiperOrigin-RevId: 170637740
|
|
|
|
| |
PiperOrigin-RevId: 170637630
|
|
|
|
| |
PiperOrigin-RevId: 170617321
|
|
|
|
| |
PiperOrigin-RevId: 170607818
|
|
|
|
| |
PiperOrigin-RevId: 170604029
|
|
|
|
| |
PiperOrigin-RevId: 170597778
|
|
|
|
| |
PiperOrigin-RevId: 170595295
|
|
|
|
| |
PiperOrigin-RevId: 170594836
|
|
|
|
| |
PiperOrigin-RevId: 170584354
|
|
|
|
| |
PiperOrigin-RevId: 170572688
|
|
|
|
|
|
|
| |
of the setting of centered. centered only affects the offset relative to the
image.
PiperOrigin-RevId: 170558824
|
|
|
|
|
|
|
|
|
|
|
|
| |
This way users of the class don't have to remember to capture each one manually to avoid premature deallocation and memory races for asynchronous op kernels.
* Add simple tests that run multiple ops concurrently for linalg ops that use CudaSolver.
* Put a lock around the calls to cusolverDn*getrs and cusolverDn*gesvd, which appear not to be threadsafe.
* Misc. cleanup in linalg GPU kernels.
I ran all the related tests 1000 times without failure. Before this change, tests for matrix_solve and svd would fail or hang occasionally.
PiperOrigin-RevId: 170557380
|
|
|
|
|
|
| |
lower-triangular with positive diagonal.
PiperOrigin-RevId: 170554998
|
|
|
|
|
|
|
|
| |
BeamSearchDecoder.
Fixes #9832, #12569.
PiperOrigin-RevId: 170553460
|
|
|
|
| |
PiperOrigin-RevId: 170548275
|
|
|
|
|
|
| |
Previous changes export the model in accordance with the known export strategies when train_and_evaluate runs in the distributed mode. This change adds a similar support for the local mode.
PiperOrigin-RevId: 170546015
|
|
|
|
| |
PiperOrigin-RevId: 170540520
|
|
|
|
|
|
|
| |
A DynamicUpdateSlice where the update shape is the same as the output
shape and the input indices are all 0 is equal to its update.
PiperOrigin-RevId: 170539478
|
|
|
|
| |
PiperOrigin-RevId: 170537829
|
|
|
|
| |
PiperOrigin-RevId: 170537570
|
|
|
|
|
|
|
| |
Network has not been part of TensorFlow's public API for any release.
Users should use keras.Model instead for now.
PiperOrigin-RevId: 170534633
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Like HloComputation::instructions(), HloModule::computations() used to
return a list of unique_ptrs. But this is an implementation detail that
shouldn't be leaked into the public API.
This patch also adds HloModule::MakeNonFusionComputations(), because
many of the callers of computations() went on to filter out all the
fusion computations.
It would be possible to implement MakeNonFusionComputations() "in place"
using a filtering iterator, but I don't think it's necessary -- we never
have *that* many computations, and since many callers go on to copy the
list of non-fusion computations, making it unconditionally a copy is
simpler and avoids a footgun.
PiperOrigin-RevId: 170529051
|
|
|
|
| |
PiperOrigin-RevId: 170527085
|
|
|
|
| |
PiperOrigin-RevId: 170526899
|
|
|
|
| |
PiperOrigin-RevId: 170525148
|
|
|
|
| |
PiperOrigin-RevId: 170524684
|
|
|
|
| |
PiperOrigin-RevId: 170524461
|
|
|
|
|
|
|
| |
The interleave transformation has block_length=1 as a default value. This
change keeps sloppy_interleave and interleave in sync.
PiperOrigin-RevId: 170523435
|
|
|
|
| |
PiperOrigin-RevId: 170522593
|
|
|
|
| |
PiperOrigin-RevId: 170522553
|
|
|
|
|
|
| |
Added gradient to tf.mod
PiperOrigin-RevId: 170522376
|
|
|
|
|
|
|
| |
From @rhaertel80:
There are cases where it is useful to have externally defined keys and have these keys passed through from the input to the output. As an example, consider a batch prediction service: The service simply runs inference on the users graph and returns the results. Keys are essential because there is no order guarantee on the outputs so they need to be rejoined to the inputs via keys or transclusion of the inputs in the outputs.
PiperOrigin-RevId: 170521852
|
|
|
|
| |
PiperOrigin-RevId: 170517511
|
|
|
|
|
|
| |
_replace does not run validations. Here we provide a new 'replace' which does the validations.
PiperOrigin-RevId: 170516477
|
|
|
|
| |
PiperOrigin-RevId: 170515496
|
|
|
|
| |
PiperOrigin-RevId: 170508628
|