Describe the bug

Dear all, I am computing dispersion correction for various inorganics.
I have installed the recent D3 and D4 ASE integrations from conda (dftd4 and simple-dftd3).
I find for a Scandium-Silicon-Iron alloy, differences between D3 and D4 forces of several eV/A.
I cannot identify myself, whether these are due to the method, or implementation.

To Reproduce

from dftd3.ase import DFTD3
from dftd4.ase import DFTD4
import ase.io

frame = ase.io.read("example.xyz")

calcd3 = DFTD3(method="r2scan", damping="d3bj")
calcd4 = DFTD4(method="r2scan")

frame.calc = calcd3
F_d3 = frame.get_forces()

frame.calc = calcd4
F_d4 = frame.get_forces()



print(F_d3-F_d4)

Expected behaviour
DFTD4 forces should not be several eV/A.

Additional context
.xyz file attached, needs to be renamed to .txt
example.xyz.txt