I’m trying to reproduce an algorithm using qubits for the CV context using GKP states in StrawberryFields.
In this article Photonic quantum computers  PennyLane Demos, a universal set of qubit analog gates (T, CNOT, ControlZ, Hadamard, PauliX, PauliZ) is presented.
For my problem I need two things:

Gates analogous to X and Z rotations.

That the parameters of these rotations are trainable.
For me it is clear how to derive the rotations from the universal set, for example, for a specific rotation theta.
But it gets messy when I consider putting a theta as a trainable parameter.
Am I missing something or is it really messed up?
Thank you very much.