src/sagemaker/serve/model_server/torchserve/inference.py [27:68]:
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
    shared_libs_path = Path(model_dir + "/shared_libs")

    if shared_libs_path.exists():
        # before importing, place dynamic linked libraries in shared lib path
        shutil.copytree(shared_libs_path, "/lib", dirs_exist_ok=True)

    serve_path = Path(__file__).parent.joinpath("serve.pkl")
    mlflow_flavor = _get_mlflow_flavor()
    with open(str(serve_path), mode="rb") as file:
        global inference_spec, native_model, schema_builder
        obj = cloudpickle.load(file)
        if mlflow_flavor is not None:
            # TODO: Add warning if it's pyfunc flavor since it will need to enforce schema
            schema_builder = obj
            loaded_model = _load_mlflow_model(deployment_flavor=mlflow_flavor, model_dir=model_dir)
            return loaded_model if callable(loaded_model) else loaded_model.predict
        elif isinstance(obj[0], InferenceSpec):
            inference_spec, schema_builder = obj
        elif isinstance(obj[0], str) and obj[0] == "xgboost":
            model_class_name = os.getenv("MODEL_CLASS_NAME")
            model_save_path = Path(__file__).parent.joinpath("model.json")
            native_model = load_xgboost_from_json(
                model_save_path=str(model_save_path), class_name=model_class_name
            )
            schema_builder = obj[1]
        else:
            native_model, schema_builder = obj
    if native_model:
        framework, _ = _detect_framework_and_version(
            model_base=str(_get_model_base(model=native_model))
        )
        if framework == "pytorch":
            native_model.eval()
        return native_model if callable(native_model) else native_model.predict
    elif inference_spec:
        return partial(inference_spec.invoke, model=inference_spec.load(model_dir))


def input_fn(input_data, content_type):
    """Placeholder docstring"""
    try:
        if hasattr(schema_builder, "custom_input_translator"):
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -



src/sagemaker/serve/model_server/torchserve/xgboost_inference.py [31:71]:
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
    shared_libs_path = Path(model_dir + "/shared_libs")

    if shared_libs_path.exists():
        # before importing, place dynamic linked libraries in shared lib path
        shutil.copytree(shared_libs_path, "/lib", dirs_exist_ok=True)

    serve_path = Path(__file__).parent.joinpath("serve.pkl")
    mlflow_flavor = _get_mlflow_flavor()
    with open(str(serve_path), mode="rb") as file:
        global inference_spec, native_model, schema_builder
        obj = cloudpickle.load(file)
        if mlflow_flavor is not None:
            schema_builder = obj
            loaded_model = _load_mlflow_model(deployment_flavor=mlflow_flavor, model_dir=model_dir)
            return loaded_model if callable(loaded_model) else loaded_model.predict
        elif isinstance(obj[0], InferenceSpec):
            inference_spec, schema_builder = obj
        elif isinstance(obj[0], str) and obj[0] == "xgboost":
            model_class_name = os.getenv("MODEL_CLASS_NAME")
            model_save_path = Path(__file__).parent.joinpath("model.json")
            native_model = load_xgboost_from_json(
                model_save_path=str(model_save_path), class_name=model_class_name
            )
            schema_builder = obj[1]
        else:
            native_model, schema_builder = obj
    if native_model:
        framework, _ = _detect_framework_and_version(
            model_base=str(_get_model_base(model=native_model))
        )
        if framework == "pytorch":
            native_model.eval()
        return native_model if callable(native_model) else native_model.predict
    elif inference_spec:
        return partial(inference_spec.invoke, model=inference_spec.load(model_dir))


def input_fn(input_data, content_type):
    """Placeholder docstring"""
    try:
        if hasattr(schema_builder, "custom_input_translator"):
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -