Demand forecasting TFT example - from_dataset() choice of data

I was hoping someone could clear up the use of from_dataset() in the demand forecasting example using TFT.

training = TimeSeriesDataSet(
    data[lambda x: x.date < training_cutoff],
    time_idx= ...,
    target= ...,
    # weight="weight",
    group_ids=[ ... ],
    max_encode_length=max_encode_length,
    max_prediction_length=max_prediction_length,
    static_categoricals=[ ... ],
    static_reals=[ ... ],
    time_varying_known_categoricals=[ ... ],
    time_varying_known_reals=[ ... ],
    time_varying_unknown_categoricals=[ ... ],
    time_varying_unknown_reals=[ ... ],
)

validation = TimeSeriesDataSet.from_dataset(training, data, min_prediction_idx=training.index.time.max() + 1, stop_randomization=True)

When creating the validation set using from_dataset() isn't it using the whole dataset that is used to create the training dataset and therefore leaking data into the validation set? Shouldn't the data used by the data after the training_cutoff be like sodata[lambda x: x.date > training_cutoff]?

Read the demand forecasting TFT example and the TimeSeriesDataSet.from_dataset() entry point first. Check how the training cutoff, full data argument, and min_prediction_idx interact, then determine whether the example leaks validation data and whether the example or its explanation needs correction.