tape.RecordOperation(inputs, {exp_output}, gradient_function.release());
  TF_RETURN_IF_ERROR(tape.ComputeGradient(ctx,
                                          /*targets*/ {exp_output},
                                          /*sources=*/inputs,
                                          /*output_gradients=*/{},
                                          /*result=*/outputs));
  exp_output->Unref();
  return absl::OkStatus();
}

    TF_RETURN_IF_ERROR(tape.ComputeGradient(ctx, /*targets=*/temp_outputs,
                                            /*sources=*/inputs,
                                            /*output_gradients=*/{}, outputs));
    for (auto temp_output : temp_outputs) {
      temp_output->Unref();
    }
    return absl::OkStatus();
  };
}

}  // namespace internal

    }
    inputs_.push_back(t->output_);
    *output = tensorflow::down_cast<TracingTensorHandle*>(outputs[0]);
    return absl::OkStatus();
  }

  Status Finalize(OutputList* outputs, AbstractFunction** f) override {
    std::vector<TF_Output> graph_outputs;
    graph_outputs.reserve(outputs->outputs.size());
    for (auto* abstract_output : outputs->outputs) {

      - opened
      - synchronize
jobs:
  changes:
    runs-on: ubuntu-latest
    # Required permissions
    permissions:
      pull-requests: read
    # Set job outputs to values from filter step
    outputs:
      docs: ${{ steps.filter.outputs.docs }}
    steps:
    - uses: actions/checkout@v4
    # For pull requests it's not necessary to checkout the code but for master it is
    - uses: dorny/paths-filter@v3

        absl::StrCat("ZerosLike", ToId(t)).c_str()));
  }
  TF_RETURN_IF_ERROR(op->AddInput(t));
  int num_outputs = 1;
  std::vector<AbstractTensorHandle*> outputs(num_outputs);
  TF_RETURN_IF_ERROR(
      op->Execute(absl::Span<AbstractTensorHandle*>(outputs), &num_outputs));
  *result = outputs[0];
  return absl::OkStatus();
}
}  // namespace

Status GradientRegistry::Register(

                      absl::Span<AbstractTensorHandle*> outputs) {
  std::vector<AbstractTensorHandle*> temp_outputs(2);
  TF_RETURN_IF_ERROR(
      ops::IdentityN(ctx, inputs, absl::MakeSpan(temp_outputs), "IdentityN"));
  // Although, `ops::IdentityN` returns 2 tensors, the first tensor isn't needed
  // for computing gradient so we could safely drop it.
  outputs[0] = temp_outputs[1];
  temp_outputs[0]->Unref();
  return absl::OkStatus();
}

  AbstractTensorHandlePtr model_out(model_outputs[0]);

  TF_Tensor* model_out_tensor;
  TF_RETURN_IF_ERROR(GetValue(model_out.get(), &model_out_tensor));
  int num_dims_out = TF_NumDims(model_out_tensor);
  TF_DeleteTensor(model_out_tensor);

  // If the output is a scalar, then return the scalar output
  if (num_dims_out == 0) {
    outputs[0] = model_out.release();

  }

  TFE_Execute(op_.get(), unwrapped_results.data(), &real_num_outputs, status);
  if (TF_GetCode(status) != TF_OK) {
    cancellation_manager_->StartCancel();
    return;
  }
  unwrapped_results.resize(real_num_outputs);
  outputs->reserve(real_num_outputs);
  for (TFE_TensorHandle* unwrapped_result : unwrapped_results) {
    outputs->emplace_back(unwrapped_result);
  }
}

                absl::Span<AbstractTensorHandle*> outputs) {
  return ops::AddV2(ctx, inputs[0], inputs[1], &outputs[0], "Add");
}

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

Status SqrtModel(AbstractContext* ctx,

// being equivalent to the following python code.
//
// if use_function:
//   outputs = tf.function(model)(inputs)
// else:
//   outputs = model(inputs)
Status RunModel(Model model, AbstractContext* ctx,
                absl::Span<AbstractTensorHandle* const> inputs,
                absl::Span<AbstractTensorHandle*> outputs, bool use_function);

Status BuildImmediateExecutionContext(bool use_tfrt, AbstractContext** ctx);