memcpy(theta_data.data(), TF_TensorData(theta_tensor),
         TF_TensorByteSize(theta_tensor));

  // Initialize space for the numerical gradient.
  vector<float> dtheta_approx(num_elems);

  // Get theta shape and store in theta_dims.
  int num_dims = TF_NumDims(theta_tensor);
  vector<int64_t> theta_dims(num_dims);
  GetDims(theta_tensor, theta_dims.data());

  // Initialize auxilary data structures.

};

struct TF_Library {
  void* lib_handle;
  TF_Buffer op_list;
};

struct TF_Graph {
  TF_Graph();

  mutable tensorflow::mutex mu;
  tensorflow::Graph graph TF_GUARDED_BY(mu);

  // Runs shape inference.
  tensorflow::ShapeRefiner refiner TF_GUARDED_BY(mu);

  // Maps from name of an operation to the Node* in 'graph'.
  std::unordered_map<tensorflow::string, tensorflow::Node*> name_map
      TF_GUARDED_BY(mu);

  // Computes the rank of the tensor handle.
  //
  // Shapes are specified via callbacks because retrieving the shape of a tensor
  // is a blocking operation for async eager; custom devices should avoid
  // retrieving shapes of tensors they wrap until the custom device tensor's
  // shape is explicitly requested where possible.
  int (*num_dims)(void* data, TF_Status* status);

  // Computes the axis length at `dim_index`.

            EndpointPair.unordered(1, 1),
            EndpointPair.unordered(1, 3))
        .inOrder();
  }

  /**
   * Populates the graph with nodes and edges in a star shape with node `1` in the middle.
   *
   * <p>Note that the edges are added in a shuffled order to properly test the effect of the
   * insertion order.
   */
  private void populateTShapedGraph() {
    putEdge(2, 1);

            EndpointPair.unordered(1, 1),
            EndpointPair.unordered(1, 3))
        .inOrder();
  }

  /**
   * Populates the graph with nodes and edges in a star shape with node `1` in the middle.
   *
   * <p>Note that the edges are added in a shuffled order to properly test the effect of the
   * insertion order.
   */
  private void populateTShapedGraph() {
    putEdge(2, 1);

    // .device of shape is GPU since the op is executed on GPU
    device_name = TFE_TensorHandleDeviceName(retvals[0], status.get());
    ASSERT_EQ(TF_OK, TF_GetCode(status.get())) << TF_Message(status.get());
    ASSERT_TRUE(absl::StrContains(device_name, "GPU:0")) << device_name;

    // .backing_device of shape is CPU since the tensor is backed by CPU
    backing_device_name =

It is just a function that can take all the same parameters that a *path operation function* can take:

{* ../../docs_src/dependencies/tutorial001_an_py310.py hl[8:9] *}

That's it.

**2 lines**.

And it has the same shape and structure that all your *path operation functions* have.

You can think of it as a *path operation function* without the "decorator" (without the `@app.get("/some-path")`).

And it can return anything you want.

And Swagger UI has supported this particular `examples` field for a while. So, you can use it to **show** different **examples in the docs UI**.

The shape of this OpenAPI-specific field `examples` is a `dict` with **multiple examples** (instead of a `list`), each with extra information that will be added to **OpenAPI** too.

This schema definition includes your API paths, the possible parameters they take, etc.

#### Data "schema"

The term "schema" might also refer to the shape of some data, like a JSON content.

In that case, it would mean the JSON attributes, and data types they have, etc.

#### OpenAPI and JSON Schema

                                                           TF_DeleteStatus);

  TF_Tensor* tensor_shape = Int32Tensor({37, 1});
  TF_Operation* shape = Const(tensor_shape, func_graph.get(), s.get(), "shape");
  TF_Operation* random =
      RandomUniform(shape, TF_FLOAT, func_graph.get(), s.get());

  TF_Output outputs[] = {{random, 0}};
  *func = TF_GraphToFunction(func_graph.get(), name,