Try the same using Modeller - Does your template have all the necessary chains that are to be modelled? If yes then Modeller should work fine.

HHpred is designed for ternary structure prediction so it won't model the dimeric form. Either you know an homologous complex then you model the two chains with HHpred and superpose the chains on the homologous template. If you know the type symmetry in the case of homodimeric structure, you can apply a symmetry operator. If you don't have any knowledge about the complex, you need to use a docking server like SwarmDock, Haddock or Pydock.

Modeller won't help either, actually HHpred as far as I know uses modeller in its pipeline to build the structure with templates it finds with HHsearch or HHblits.

Agree with Raphael.

To look for dimeric templates: try searching HHpred using a single chain sequence for homologous proteins, then look at the top hits in RCSB (advanced search -> put in pdb codes, print out info in customized table format). You can then filter the table for pdbs that have multiple chains of the same homologous protein and physically look at them to see if they are dimers. of course, it would help if you have some data about the dimeric state of your protein (e.g. what residues are on the dimer interface) so you can choose the most appropriate template.