Modeling cyclic peptides is something that’s currently being worked on, and in the future there likely will be established protocols and tools to more conveniently model them with Rosetta, but it’s not currently something that works “out of the box”.
The stopgap in the meantime is to impose constraints on the geometry between the N and C termini. Strong enough constraints will force a cyclic-like conformation. The downside is that too strong of constraints will stop conformational sampling. You may need to try ramping the constraints, or only applying constraints during certain phases of modeling (like minimization).