I have not yet ported the SDP relaxation to PowerModelsAnnex. However, the SOC relaxation there uses a similar variable space, so an extension of that model would be a good place to start.
SDP model is the easiest to implement. It is very similar to most of the papers you would find about that model. The sparse version omits some entires in the
W matrix and is hence more complex. But is also about 10 times more scalable.
In my experience, 100 buses is around the limit for the basic
SDP model and about 1000 buses for the sparse version. In both cases you will want to use the MOSEK SDP solver. The open-source solvers are lagging far behind in terms of reliability compared to MOSEK.
FWIW, I generally do not use the SDP model in my work due to solver reliability challenges. Making SDP OPF formulation reliable in practice seems to be an open research topic.
I hope that helps!