{!> ../../../docs_src/body_updates/tutorial002_py39.py!}
    ```

=== "Python 3.8+"

    ```Python hl_lines="35"
    {!> ../../../docs_src/body_updates/tutorial002.py!}
    ```

### Partial updates recap

In summary, to apply partial updates you would:

* (Optionally) use `PATCH` instead of `PUT`.
* Retrieve the stored data.
* Put that data in a Pydantic model.

You could also use it to generate code automatically, for clients that communicate with your API. For example, frontend, mobile or IoT applications.

## Recap, step by step

### Step 1: import `FastAPI`

```Python hl_lines="1"
{!../../../docs_src/first_steps/tutorial001.py!}
```

`FastAPI` is a Python class that provides all the functionality for your API.

// TODO(jayconrod): add method to resize the buffer. We'll need that when the
// mutator can increase input length. Only the coordinator will be able to
// do it, since we'll need to send a message to the worker telling it to

    ```Python hl_lines="1"
    async def needy_dependency(fresh_value: str = Depends(get_value, use_cache=False)):
        return {"fresh_value": fresh_value}
    ```

## Recap

Apart from all the fancy words used here, the **Dependency Injection** system is quite simple.

Just functions that look the same as the *path operation functions*.

But if you need to access variables and objects from the same **FastAPI** app, or you need to perform small background tasks (like sending an email notification), you can simply just use `BackgroundTasks`.

## Recap

      }
    }
    replicas.push_back(replica);
  }

  return success();
}

// Creates islands per replica from `tf_device.replicate` region and remap
// replicate results with new island outputs. A single island is created to
// forward control dependencies if there is a control dependency output from the
// replicate island. Devices are remapped from aliased devices to explicit

        "tax": 3.2
    }
}
```

instead of:

```JSON
{
    "name": "Foo",
    "description": "The pretender",
    "price": 42.0,
    "tax": 3.2
}
```

## Recap

You can add multiple body parameters to your *path operation function*, even though a request can only have a single body.

		sz := len(sv.fieldMap)
		ntvMap := reflect.MakeMapWithSize(typeDesc, sz)
		for name, val := range sv.fieldMap {
			refVal, err := val.Ref.ConvertToNative(elemType)
			if err != nil {
				return nil, err
			}
			ntvMap.SetMapIndex(reflect.ValueOf(name), reflect.ValueOf(refVal))
		}
		return ntvMap.Interface(), nil
	}

	if typeDesc.Kind() == reflect.Struct {
		ntvObjPtr := reflect.New(typeDesc)

    {!> ../../../docs_src/extra_models/tutorial005_py39.py!}
    ```

=== "Python 3.8+"

    ```Python hl_lines="1  8"
    {!> ../../../docs_src/extra_models/tutorial005.py!}
    ```

## Recap

Use multiple Pydantic models and inherit freely for each case.

  let constructor = "mlir::tf_saved_model::CreateDedupBoundInputBindingPass()";
}

def LowerGlobalsToMlProgramPass: Pass<"tf-saved-model-lower-globals-to-mlprogram", "mlir::ModuleOp"> {
  let summary = "Remove (and remap) function arguments that map to global tensors.";
  let description = [{
    This pass will lower AssignVariableOp and ReadVariableOp to their ml_program