But that type can itself be another Pydantic model.

So, you can declare deeply nested JSON "objects" with specific attribute names, types and validations.

All that, arbitrarily nested.

### Define a submodel

For example, we can define an `Image` model:

//// tab | Python 3.10+

```Python hl_lines="7-9"
{!> ../../docs_src/body_nested_models/tutorial004_py310.py!}
```

////

                name=response_name,
                type_=self.response_model,
                mode="serialization",
            )
            # Create a clone of the field, so that a Pydantic submodel is not returned
            # as is just because it's an instance of a subclass of a more limited class
            # e.g. UserInDB (containing hashed_password) could be a subclass of User

      for (Range<Integer> range2 : RANGES) {
        for (Range<Integer> range3 : RANGES) {
          Map<Integer, Integer> model = Maps.newHashMap();
          putModel(model, range1, 1);
          putModel(model, range2, 2);
          putModel(model, range3, 3);
          RangeMap<Integer, Integer> test = TreeRangeMap.create();
          test.put(range1, 1);
          test.put(range2, 2);
          test.put(range3, 3);

* Das Gleiche gilt für `app/routers/users.py`, es ist ein weiteres Submodul: `app.routers.users`.
* Es gibt auch ein Unterverzeichnis `app/internal/` mit einer weiteren Datei `__init__.py`, es handelt sich also um ein weiteres „Python-Subpackage“: `app.internal`.
* Und die Datei `app/internal/admin.py` ist ein weiteres Submodul: `app.internal.admin`.

  return ops::MatMul(ctx, inputs[0], inputs[1], &outputs[0],
                     /*transpose_a=*/false,
                     /*transpose_b=*/false, "MatMul");
}

absl::Status MulModel(AbstractContext* ctx,
                      absl::Span<AbstractTensorHandle* const> inputs,
                      absl::Span<AbstractTensorHandle*> outputs) {
  return ops::Mul(ctx, inputs[0], inputs[1], &outputs[0], "Mul");
}

    ASSERT_EQ(errors::OK, s.code()) << s.message();
    x.reset(x_raw);
  }

  std::vector<AbstractTensorHandle*> outputs(1);
  absl::Status s = RunModel(RecordOperationWithNullGradientFunctionModel,
                            ctx.get(), {x.get()}, absl::MakeSpan(outputs),
                            /*use_function=*/!std::get<2>(GetParam()));

                            absl::Span<AbstractTensorHandle*> outputs,
                            bool use_function) {
  AbstractTensorHandle* model_outputs[1];

  // Run the model.
  TF_RETURN_IF_ERROR(
      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));

        * When a `response_model` is declared, the same `response_model` type declaration won't be used as is, it will be "cloned" to create an new one (a cloned Pydantic `Field` with all the submodels cloned as well).