etol, the parameter you changed is indeed unitless and is equal to Delta_E/Etot,

Delta_E being the energy variation

between two iterations and Etot the total energy of the system. You might want to
decrease slightly this number for your
next dataset.

That being said I checked some of your files, and it seems that the distance
between the partial dislocations is
significantly larger in Ag than in Al. It also seems that your dissociation length
(distance between the partials)
is significantly larger than I expected in Al. I wonder if it is related to the
small size of the crystal (the image
forces from the free surfaces can modify the equilibrium distance between the
partial dislocations) or if it is because
of the very short relaxation time.
Variation of dissociation idstance

You might still be able to differentiate the two materials from the diffraction
pattern.
For the screw dislocation, it doesn't have time to form the constriction node which
is the true equilibrium
configuration but I guess it is not critical as long as we see a clear difference
between Ag and Al (which seems to
be the case).

If you introduce the Frank loop you will already get a stacking fault in the
unrelaxed configuration
(you can choose if its an intrinsic (2 consecutives hcp planes) or extrinsic (1 hcp
/ 1 fcc / 1 hcp) stacking fault).
Just to explain you the difference between the 2: a single partial dislocation
always create an intrinsic stacking fault
when gliding on its slip plane. An extrinsic stacking fault is created when 2
partial dislocation glide on two
consecutives {1 1 1} planes (it's also called a nanotwin).

Then you can relax your crystal, your stacking fault should remain in your crystal
during the relaxation.

Put the (111) reflection, train with >1 reflection.

Test experimental dataset.