I am using the NVIDIA PhysicsNeMo framework to train a DeepONet model for a fluid dynamics problem. I have calculated the mean ($\mu$) and standard deviation ($\sigma$) for all input and output variables, and I am passing these normalization parameters using physicsnemo.sym.key.Key(name, scale=(,)) as in the example of darcy_AFNO. However, after instantiating the DeepONet model, checking the deeponet.input_scales and deeponet.output_scales properties shows that the model is defaulting to (0.0, 1.0) for inputs and None for outputs, instead of using my custom values.Why are the normalization parameters defined in my Key objects not being captured by the DeepONet instance? What is the correct way to pass these $\text{mean}$ and $\text{std}$ values during network instantiation to ensure the internal normalization layers are set up correctly? Should the normalization keys (C, Q, rho, mu, etc.) and their scale parameters be declared explicitly within the Hydra configuration file (e.g., in the `arch.deeponet` section) in addition to, or instead of, being passed via the Python `Key` objects? Thanks for anyone who could help.