@paogorman comments:

One output that we have not included from the NN is the number concentration for liquid and ice cloud.

ZM handles this by just using a typical radius and deriving it from the mixing ratio tendencies from liquid and ice cloud. So I think we could follow suit with the following code:

ptend_loc%q(i,k,ixnumliq) = 3._r8 * max(0._r8, ptend%q(i,k,ixcldliq)) / (4._r8*3.14_r8* 8.e-6_r8**3*997._r8)
ptend_loc%q(i,k,ixnumice) = 3._r8 * max(0._r8, ptend%q(i,k,ixcldice)) / (4._r8*3.14_r8*25.e-6_r8**3*500._r8)

(Please see our google doc on the SAM-CAM interface for further discussion).
In any event it would be good to output cloud and liquid mixing ratios and number concentrations from your SCAM6 run and see how they behave.