PiperOrigin-RevId: 164917891

PiperOrigin-RevId: 164916465

PiperOrigin-RevId: 164914552

PiperOrigin-RevId: 164911656

destination. It will reuse the same underlying buffer in those cases.

PiperOrigin-RevId: 164909906

PiperOrigin-RevId: 164902588

PiperOrigin-RevId: 164894867

PiperOrigin-RevId: 164894645

PiperOrigin-RevId: 164890915

Bugfix: number of input channels is not necessarily in the last dimension, after introduction of data_format param.

PiperOrigin-RevId: 164889729

Recover MonitoredSession when the Coordinator is requested to stop with one of the _PREEMPTION_ERRORS.

When SyncReplicasOptimizer is used, a preemption in the Coordinator may result in two cases:
Case 1) the session gets silently marked as complete
Case 2) the session gets stuck

This CL aims to solve and verify solutions for both of these problems. Fix 1 changes the should_stop logic. Fix 2 changes the CoordinatedSession.run() logic.

SyncReplicasOptimizer runs a separate set of threads using a Coordinator instance. Those threads do FIFOQueue.enqueue; the main thread does a blocking FIFOQueue.dequeue.

`sync_token_q` FIFOQueue is on parameter-servers. When one of the PS instances gets preempted, an AbortedError causes the Coordinator to stop via request_stop(ex). That by itself changes the state of MonitoredSession.should_stop() to True (Fix 1).

Results of the blocking Dequeue operation are sent to the chief worker via Recv. What happens next depends on the amount of tokens in `sync_token_q`. If there are enough for the next call to Dequeue to return, then the low-level "tf session run() call" returns. The next iteration of the `while not MonitoredSession.should_stop()` loop decides that the training is complete (Case 1).

If there are not enough tokens in `sync_token_q`, then the blocking Dequeue is going to keep waiting for them. This results in the graph execution getting stuck and the whole session getting garbage collected after 10 minutes (Case 2).

We decided to fix that by re-creating a session after it gets garbage collected (Fix 2). An alternative was to try to cancel the pending Dequeue operation, but it's not clear that it is the right thing to do and it is also not easy.

PiperOrigin-RevId: 164888390

END_PUBLIC
RELNOTES: n/a

BEGIN_PUBLIC
Automated g4 rollback of changelist 164825735

PiperOrigin-RevId: 164883478

PiperOrigin-RevId: 164881593

point accuracy is not high enough across all supported platforms.

PiperOrigin-RevId: 164879933

PiperOrigin-RevId: 164879220

PiperOrigin-RevId: 164867401

PiperOrigin-RevId: 164864147

PiperOrigin-RevId: 164862849

PiperOrigin-RevId: 164860564

Fixes an issue when 'CG' is used as an optimization method. For example, the CG method does not return the `nit` and `nfev` fields which leads to an error during logging. Test cases were also extended to cover additional optimization methods offered through SciPy.

PiperOrigin-RevId: 164849933

PiperOrigin-RevId: 164845473

PiperOrigin-RevId: 164831327

PiperOrigin-RevId: 164830580

This change allows more kinds of ops to be fused into the
loop.

Besides increasing spatial locality, fusing the ops can decrease
the code size by not generating an extra loop nest.

This change also adds two kinds of tests:
* Tests to make sure the fusion logic recognizes that fusion
can only occur the op acts elementwise on the operand.
* More tests in fusion_test to test how fused loops are lowered
with the newly added classes of ops.

PiperOrigin-RevId: 164825735

Android demo: revert calls to yuv -> rgb conversion methods so that Java fallback can be used if libtensorflow_demo.so method is not found. Resolves #12110

PiperOrigin-RevId: 164812287

RemoveEMA GT: transforms frozen graphs from the FakeQuantizeTraining GT making it compatible with the QuantizeNodes GT.

PiperOrigin-RevId: 164810566

PiperOrigin-RevId: 164805620

This change introduces an LLVMCompiler class, of which the
CPU and GPU compilers are subclasses. The LLVMCompiler
class provides the ability to inspect LLVM generated compiler
code by registering a callback. The callbacks can
be used to analyze IR before and after optimizations.

This also adds a simple test for the callback mechanism.

PiperOrigin-RevId: 164805348

PiperOrigin-RevId: 164804532

PiperOrigin-RevId: 164804406

PiperOrigin-RevId: 164803218

PiperOrigin-RevId: 164802741

Update Android Detect demo to use models exported using the Tensorflow Object Detection API.  Resolves #6738.

PiperOrigin-RevId: 164802542

This changes the `samples_per_second` parameter of the `encode_audio`
and `decode_audio` ops from an `Attr` to an `Input`, so that it can be
given arbitrary tensor values instead of only constants.

This change is important for use cases that want to use a single graph
to encode audio clips at arbitrary sample rates. (In particular, we want
to create a Python function that uses a long-running TensorFlow session
to encode audio; the sample rate cannot be known ahead of time, and we
don't want to have to reconstruct the graph on every call.)

PiperOrigin-RevId: 164799067

PiperOrigin-RevId: 164797105

slightly different semantics.

PiperOrigin-RevId: 164796436

PiperOrigin-RevId: 164794573

PiperOrigin-RevId: 164791375

Refactor Keras layers to rely on the core constraint implementation.

PiperOrigin-RevId: 164788653

PiperOrigin-RevId: 164787644