//
// This is intended to be used when a peer failure is detected.
TF_CAPI_EXPORT extern void TFE_AbortCollectiveOps(TFE_Context* ctx,
                                                  TF_Status* status);

// Checks the health of collective ops peers. Explicit health check is needed in
// multi worker collective ops to detect failures in the cluster.  If a peer is
// down, collective ops may hang.

          echo "TITLE=$(git log -1 ${{ github.event.inputs.git_commit }} --format="%s")" >> "$GITHUB_OUTPUT"
    - name: Create Pull Request with changes
      uses: peter-evans/create-pull-request@2b011faafdcbc9ceb11414d64d0573f37c774b04 # v4.2.3
      with:

} TFE_ContextDevicePlacementPolicy;
// LINT.ThenChange(//tensorflow/c/eager/immediate_execution_context.h)

// Sets the default execution mode (sync/async). Note that this can be
// overridden per thread using TFE_ContextSetExecutorForThread.
TF_CAPI_EXPORT extern void TFE_ContextOptionsSetAsync(TFE_ContextOptions*,
                                                      unsigned char enable);

         * It has an added advantage since you can you easily switch to different hardware accelerators (cpu, gpu, tpu) as per the task.
         * All you need is a good internet connection and you are all set.
      * Try to build TF from sources by changing CPU optimization flags.
   
   *Please let us know if this helps.*
   
windows_comment: >

you want to debug a kernel, you can compile specific files with `-g` using the
`--per_file_copt` bazel option. For example, if you want to debug the Identity
op, which are in files starting with `identity_op`, you can run

```bash
bazel build --config=dbg --per_file_copt=+tensorflow/core/kernels/identity_op.*@-g //tensorflow/tools/pip_package:build_pip_package
```

    // Note: we hold onto mu_ for the rest of this function. In practice, this
    // is okay, as stat requests are typically fast, and concurrent requests are
    // often for the same file. Future work can split this up into one lock per
    // key if this proves to be a significant performance bottleneck.
    absl::MutexLock lock(&mu_);
    if (LookupLocked(key, value)) {
      return TF_SetStatus(status, TF_OK, "");
    }

* <SIMILAR TO ABOVE SECTION, BUT FOR OTHER IMPORTANT CHANGES / BUG FIXES>
* <IF A CHANGE CLOSES A GITHUB ISSUE, IT SHOULD BE DOCUMENTED HERE>
* <NOTES SHOULD BE GROUPED PER AREA>

* `tf.lite`
    * Quantization for `FullyConnected` layer is switched from per-tensor to
      per-channel scales for dynamic range quantization use case (`float32`
      inputs / outputs and `int8` weights). The change enables new quantization

  // once unless the tape is persistent) and produces the gradient of the target
  // tensors with respect to the source tensors. The output gradients are used
  // if not empty and not null. The result is populated with one tensor per
  // target element.
  Status ComputeGradient(
      AbstractContext* ctx, absl::Span<AbstractTensorHandle* const> targets,
      absl::Span<AbstractTensorHandle* const> sources,

            # copybara:uncomment_begin(google-only)
            # "-Wl,--undefined-version",
            # copybara:uncomment_end(google-only)
        ],
    }),
    linkstatic = 1,
    per_os_targets = True,
    roots = [
                "//tensorflow/c/experimental/filesystem:filesystem_interface",
                "//tensorflow/c/experimental/stream_executor:stream_executor",

} TF_Filesystem_Option_Type;

typedef struct TF_Filesystem_Option {
  char* name;                         // null terminated, owned
  char* description;                  // null terminated, owned
  int per_file;                       // bool actually, but bool is not a C type
  TF_Filesystem_Option_Value* value;  // owned
} TF_Filesystem_Option;

/// SECTION 2. Function tables for functionality provided by plugins