Tabular Model Different Input versions


I have 5 different combinations of hydrological input parameters (for example one with only 5 parameters, one with 9 and so on). The target for each version is the same - a groundwater level observation. The input combinations with more parameters have less rows of observations than the versions with less parameters. My question is how should I compare the performance of these 5 different versions. Should I find some good learning rate and epoch number for one version and use the same configuration for the rest (like a baseline version) in order to evaluate the performance of each or should I try to optimize each and every single version differently and then compare them??

Thanks for the help!