How to use the dbnd.pipeline function in dbnd

@pipeline
def t_pipeline(p_str="from_pipeline_ctor"):
    # type: (str) -> str
    a = t_A(p_str)
    return t_B(a)

@pipeline
def my_xcom_pipeline(p_date=None):
    # type: ( datetime ) -> Tuple[str, str]
    corrected_date = date_correction(p_date=p_date)
    string_date = convert_to_string(p_date=corrected_date)
    parsed_string = my_second_pipeline(p_date=corrected_date)
    return parsed_string, string_date

@pipeline
def deep_wide_pipe_int(width=25, depth=25, pause=0):
    # type: (int, int, int) -> int
    return large_pipe_int(width=width, depth=depth, pause=pause)

@pipeline
def my_pipeline_search(x_range=3):
    results = {}
    for x in range(x_range):
        t2, t3 = my_pipeline(p_str=str(x))
        results[x] = t3
    return results

@pipeline
def experiment_concat():
    # if type is specified, we can mix between strings and task outputs
    l = ["aa", gen_token("bbb")]
    return concat_tokens(l)

@pipeline(result=("model", "validation"))
def train_model_for_customer(
    task_target_date,
    data: pd.DataFrame = None,
    alpha: float = 1.0,
    l1_ratio: float = 0.5,
    period=datetime.timedelta(days=7),
    selected_features: List[str] = None,
):
    if data is None:
        partners = fetch_partner_data(task_target_date=task_target_date, period=period)
        data = calculate_features(
            selected_features=selected_features, raw_data=partners
        )
    training_set, test_set, validation_set = split_data(raw_data=data)

    model = train_model(

@pipeline
def ingest_partner_b(task_target_date):
    raw_data = data_source(name="b", task_target_date=task_target_date)
    return clean_data_spark(raw_data=raw_data)

@pipeline
def fetch_data(task_target_date, period=datetime.timedelta(days=7)):
    all_data = []
    for d in period_dates(task_target_date, period):
        data = fetch_wine_quality(task_target_date=d)
        all_data.append(data)

    return data_combine(all_data, sort=True)

@pipeline(result=("model", "validation", "serving"))
def tran_model_and_package():
    model, validation = train_model_for_customer()
    serving = package_as_docker(model=model)
    return model, validation, serving

@pipeline(result=("model", "validation", "serving"))
def predict_wine_quality_package():
    model, validation = predict_wine_quality()
    serving = package_as_docker(model=model)
    return model, validation, serving