Output of model_tz_xw in IV models if both treatment and IV are discrete

[EgorKraevTransferwise]

In case when treatment and IV are discrete, what does T*Z in the description of model_tz_xw refer to?

Firstly, what kind of component model do I need to supply in that case, a regressor or a classifier?

Secondly, what is the interpretation of that term for discrete treatment and instrument, in particular for multivalued treatments? Specifically in the case when the instrument is assigned randomly, it seems like it would make sense for that model to be a literal composition of model_t_xw and model_z_xw used in the same estimator (instead of fitting a separate one), but I can't see how to compose them correctly without understanding the meaning of the term.

Thanks a lot!