training system size for dpgen process?

[KeouLee]

When the target system is too large, both exploration and QC step will be much slower. But if I choose a relatively small system with longer exploration time, will the exploration of configuration space suffice? I was wondering the general system size setting that will provide a good configuration space convergence. Thanks!

[KeouLee]

Say, will atom number more than 200 suffice?