This behavior isn't going to be available with the `compile(cloning=...)` flag.

The tests had to updated for this change for a few reasons: 1) Cloning and no-cloning supports different features, so the expected thrown exceptions are different.  2) TFOptimizer-wrapped optimizers don't work due to internal bug issue # 130808953.

PiperOrigin-RevId: 258652546

Fixes some bugs in how the v2 loop handles batch callbacks when it runs out of data. Also makes it always recreate the validation data iterator between epochs even if validation_steps has been set, to be consistent with the behavior of the existing training loops.

PiperOrigin-RevId: 258652231

The experimental interface uses `cancellation.CancellationManager`:

```python
c_mgr = cancellation.CancellationManager()

@tf.function
def f(?):
  ?

cancelable_f = c_mgr.get_cancelable_function(f.get_concrete_function(?))

# Call a function that might run for a long time.
cancelable_f(?)

# Asynchronously:
c_mgr.start_cancel()
```

A subsequent change will add a publicly-accessible (probably experimental) API endpoint for `CancellationManager`.

PiperOrigin-RevId: 258648702

The side_input and conv_input_scale input tensors are not yet supported.

PiperOrigin-RevId: 258646677

This prevents deprecation warnings when using the public API.

PiperOrigin-RevId: 258645243

The trait is defined the MLIR core repository so tests should be there.

PiperOrigin-RevId: 258645091

PiperOrigin-RevId: 258640781

I only added the "run_with_all_model_types" annotation to one test so that the tests don't take too long to run.

PiperOrigin-RevId: 258640310

Disable the test_annotated_assign unit test because the annotation syntax is only recognized by Python 3.6 or later

PiperOrigin-RevId: 258638033

PiperOrigin-RevId: 258637728

* Replace computation/parameter
* Remove/replace scheduled computation/instruction

PiperOrigin-RevId: 258633717

PiperOrigin-RevId: 258630122

Add a `--nnapi_accelerator_name` flag to benchmark_model, allowing
explicit accelerator control on Android Q+.

PiperOrigin-RevId: 258626781

PiperOrigin-RevId: 258620905

PiperOrigin-RevId: 258618714

PiperOrigin-RevId: 258612058

Sparkfun Edge board with a camera.  Add raw2bmp script to convert images from
log file to viewable bitmaps.

PiperOrigin-RevId: 258608334

The recently added `FunctionLibraryDefinition::CopyFunctionDefFrom()` method performs a more efficient shallow copy.

PiperOrigin-RevId: 258604819

PiperOrigin-RevId: 258598105

PiperOrigin-RevId: 258597426

make_dataset_iterator -> experimental_distribute_dataset,
make_input_fn_iterator -> experimental_distribute_datasets_from_function,
unwrap -> experimental_local_results,
experimental_run -> experimental_run_v2

PiperOrigin-RevId: 258597234

PiperOrigin-RevId: 258596408

PiperOrigin-RevId: 258596337

PiperOrigin-RevId: 258591071

PiperOrigin-RevId: 258583855

PiperOrigin-RevId: 258582774

PiperOrigin-RevId: 258582149

PiperOrigin-RevId: 258582046

Update keras v2 optimizers to reuse coefficients which are shared across all updates, which reduces the total number of ops created by between 5% (for simple optimizers such as SGD and Adagrad) and 25% (for complicated optimizers such as Adam and NAdam). Separate copies are made for each device and dtype.

The effect of this change on run time is fairly minimal since Grappler is expected to consolidate most of these ops; however it does improve graph construction time.

PiperOrigin-RevId: 258581998

PiperOrigin-RevId: 258581066

PiperOrigin-RevId: 258580730

PiperOrigin-RevId: 258579478

The delegate is no longer experimental.

PiperOrigin-RevId: 258578637

PiperOrigin-RevId: 258578415

PiperOrigin-RevId: 258577336

PiperOrigin-RevId: 258575670

Make LinearOperatorDiag tape safe.  Also add a standard "check_tape_safe" method in the base test class.

PiperOrigin-RevId: 258570322

PiperOrigin-RevId: 258563058

PiperOrigin-RevId: 258562390

PiperOrigin-RevId: 258548017