Why does NHiTS need the target variable specified in the time_varying_unknown_reals attribute?

[TeodorChiaburu]

I was wondering why do I need to specify the target variable twice, when building a TimeSeriesDataset for a NHiTS? Once for the attribute 'target' and once for 'time_varying_unknown_reals'. If I don't specify it for the second attribute, I get a ValueError: [target_name] is not in list.

[KunzstBGR]

I discovered the same and would be interested in an answer. Is the N-HiTS architecture built in a way that the target has to be also an input (e.g. to predict rainfall I need historical rainfall values)?
Other algorithms such as TFT can predict a target without having this variable as an input.