Multiple Input Models or Input Normalization?

I’m new to ML and TensorFlow, so I’m just coming up to speed with both. I haven’t seen any good TensorFlow JS examples of multiple input models, but I assume this is a thing?

I’m trying to understand two basic concepts:

  1. How should I be organizing my feature input data? Is it best to create one input model or to divide the data into logical sections?

  2. The examples use a very generic methodology for normalizing inputs. Input normalization is required because the Tensor has no idea about the range of the values it’s going to be seeing, right? Should you normalize each Tensor’s inputs individually in cases where the measures are each varied? i.e. One Tensor’s values may be 0 - 1, but the next Tensor’s values may be -10000 - 10000.

Thanks for helping with the newbie questions.