// Creates parameters (placeholders) in the tracing `ctx` using the shape and
// dtype of `inputs`.
Status CreateParamsForInputs(AbstractContext* ctx,
                             absl::Span<AbstractTensorHandle* const> inputs,
                             std::vector<AbstractTensorHandle*>* params);

// A callable that takes tensor inputs and returns zero or more tensor outputs.
using Model = std::function<Status(AbstractContext*,

Status ReluModel(AbstractContext* ctx,
                 absl::Span<AbstractTensorHandle* const> inputs,
                 absl::Span<AbstractTensorHandle*> outputs) {
  return ops::Relu(ctx, inputs[0], &outputs[0], "Relu");
}

Status SparseSoftmaxCrossEntropyWithLogitsModel(
    AbstractContext* ctx, absl::Span<AbstractTensorHandle* const> inputs,
    absl::Span<AbstractTensorHandle*> outputs) {
  AbstractTensorHandle* loss;

}

void TF_AddInputList(TF_OperationDescription* desc, const TF_Output* inputs,
                     int num_inputs) {
  std::vector<NodeBuilder::NodeOut> input_list;
  input_list.reserve(num_inputs);
  for (int i = 0; i < num_inputs; ++i) {
    input_list.emplace_back(&inputs[i].oper->node, inputs[i].index);
  }
  desc->node_builder.Input(input_list);
}

  // inherit them from input)
  TF_Operation* control_inputs[100];
  int num_control_inputs = TF_OperationGetControlInputs(
      scalar3, control_inputs, TF_OperationNumControlInputs(scalar3));
  ASSERT_EQ(2, num_control_inputs);
  EXPECT_EQ(feed, control_inputs[0]);
  EXPECT_EQ(feed2, control_inputs[1]);

  num_control_inputs = TF_OperationGetControlInputs(
      feed3, control_inputs, TF_OperationNumControlInputs(feed3));

                 absl::Span<AbstractTensorHandle*> grad_inputs) override {
    // TODO(b/161805092): Support broadcasting.

    DCHECK(grad_outputs[0]);
    grad_inputs[0] = grad_outputs[0];
    grad_inputs[1] = grad_outputs[0];

    grad_inputs[0]->Ref();
    grad_inputs[1]->Ref();
    return absl::OkStatus();
  }
  ~AddGradientFunction() override {}
};

                         absl::Span<AbstractTensorHandle* const> inputs,
                         int input_index, bool use_function,
                         AbstractTensorHandle** numerical_grad) {
  vector<AbstractTensorHandle*> theta_inputs(inputs.size());
  for (int i{}; i < inputs.size(); ++i) {
    theta_inputs[i] = inputs[i];
  }

  AbstractTensorHandle* theta =
      theta_inputs[input_index];  // parameter we are grad checking

public class SynonymItem extends DictionaryItem {
    private final String[] inputs;

    private final String[] outputs;

    private String[] newInputs;

    private String[] newOutputs;

    public SynonymItem(final long id, final String[] inputs, final String[] outputs) {
        this.id = id;
        this.inputs = inputs;
        this.outputs = outputs;

        if (id == 0) {
            // create

    std::vector<TFE_TensorHandle*> device_inputs;
    device_inputs.reserve(inputs.size());
    for (int input_index = 0; input_index < inputs.size(); ++input_index) {
      // Parallel tensors are divided between operations by device.
      device_inputs.push_back(inputs[input_index]->tensor(device_index));
    }
    device_thread->StartExecute(
        context, operation_name, std::move(device_inputs), attributes,

   * propagate cancellation) and {@link #toString()}. To access the futures' <i>values</i>, {@code
   * AggregateFuture} attaches listeners that hold references to one or more inputs. And in the case
   * of {@link CombinedFuture}, the user-supplied callback usually has its own references to inputs.
   */
  /*
   * In certain circumstances, this field might theoretically not be visible to an afterDone() call

 *
 * <h3>Common ways to use</h3>
 *
 * <p>Getting a converter:
 *
 * <ul>
 *   <li>Use a provided converter implementation, such as {@link Enums#stringConverter}, {@link
 *       com.google.common.primitives.Ints#stringConverter Ints.stringConverter} or the {@linkplain
 *       #reverse reverse} views of these.
 *   <li>Convert between specific preset values using {@link