Not sure if anyone else knows of a different common practice, but starting with a RANS run and then using that to calculate the integral length scale and the cell size of your mesh to resolve at least 80% of the TKE in LES is a good place to start. Aidan goes over it in this video: