What return and combine layers for single head models?

[Dheeru66k]

I am looking at your documentation of your model parameters like return and combine layers for MTLRefinenet+mobilenet v2 .
https://github.com/DrSleep/DenseTorch/blob/dev/docs/Models.md
You have mentioned as below

But if train a single head like "segm" alone or "depth" alone, what should be the return and combine layers for mobilenet+lwrefinenet. Looking for your support.

I also have other issue, when i load my model trained checkpoints, its says some keys are missing, should i also have to load optimizer state_dict?