An easy thing you can do is start with
set.mm and just start deleting large swaths of the database until you have what you want. Obviously you can always draw a line and delete anything after a certain point, and many theorems are independent of their neighbors so you can also trim it down more selectively.
(This is how
iset.mm was born.)
You can
also
be more precise about this by using the "trace axioms" command (or an external script, it's not that hard) to find all the dependents of a chosen set of theorems.