I am working on modeling a protein with two chains (heterodimer).
I have used two separate pdb files for each chain and oriented the two
template structures appropriately before the modeling.
Modeller is able to model both the chains, however not preserving the
orientation of both the chains.
If you are confident of this orientation, put the two chains in a single
PDB file, as is done in the example in the manual:
Modeller extracts internal distance restraints from each template
separately, so if your two chains are two separate templates, there will
be no inter-chain restraints.