//   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);

// Return a tensor handle with given type, values and dimensions.

  TF_RETURN_IF_ERROR(ops::SparseSoftmaxCrossEntropyWithLogits(
      ctx, inputs[0], inputs[1], &loss, &backprop,
      "SparseSoftmaxCrossEntropyWithLogits"));
  // `gradient_checker` only works with model that returns only 1 tensor.
  // Although, `ops::SparseSoftmaxCrossEntropyWithLogits` returns 2 tensors, the
  // second tensor isn't needed for computing gradient so we could safely drop
  // it.
  outputs[0] = loss;

  SetViaProto(desc_, {});
  FinishAndVerify(desc_, {});
}

TEST(CAPI, SavedModel) {
  // Load the saved model.
  const string saved_model_dir = tensorflow::GetDataDependencyFilepath(
      tensorflow::io::JoinPath("tensorflow", "cc", "saved_model", "testdata",
                               "half_plus_two", "00000123"));
  TF_SessionOptions* opt = TF_NewSessionOptions();

#include "tensorflow/cc/framework/gradients.h"
#include "tensorflow/cc/framework/ops.h"
#include "tensorflow/cc/framework/scope_internal.h"
#include "tensorflow/cc/ops/while_loop.h"
#include "tensorflow/cc/saved_model/loader.h"
#include "tensorflow/core/distributed_runtime/server_lib.h"
#include "tensorflow/core/framework/logging.h"
#include "tensorflow/core/framework/op_gen_lib.h"

        ":kernels",
    ],
)

tf_cuda_cc_test(
    name = "c_api_test",
    size = "medium",
    srcs = ["c_api_test.cc"],
    data = [
        ":test_op1.so",
        "//tensorflow/cc/saved_model:saved_model_half_plus_two",
    ],
    linkopts = select({
        "//tensorflow:macos": ["-headerpad_max_install_names"],
        "//conditions:default": [],
    }),
    tags = [

    ```

    For a single/parameterized test e.g. `test_capture_variables` in
    `tensorflow/python/saved_model/load_test.py`:

    (Requires `python>=3.7`)

    ```bash
    bazel test ${flags} //tensorflow/python/saved_model:load_test --test_filter=*LoadTest.test_capture_variables*
    ```

    **Note:** You can add `--test_sharding_strategy=disabled` to the `flags` to

    *   `Model.fit_generator`, `Model.evaluate_generator`,
        `Model.predict_generator`, `Model.train_on_batch`,
        `Model.test_on_batch`, and `Model.predict_on_batch` methods now respect
        the `run_eagerly` property, and will correctly run using `tf.function`
        by default. Note that `Model.fit_generator`, `Model.evaluate_generator`,

                "//tensorflow/c:logging",
                "//tensorflow/c:ops",
                "//tensorflow/cc/saved_model:fingerprinting_impl",
                "//tensorflow/cc/saved_model:loader_lite_impl",
                "//tensorflow/cc/saved_model:metrics_impl",
                "//tensorflow/compiler/tf2tensorrt:op_converter_registry_impl",
                "//tensorflow/core/common_runtime:core_cpu_impl",

      RunModel(forward, ctx, inputs, model_outputs, use_function));
  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();