Again, I'm not familiar with SI units for these models, but
an overlap of 75% is far too large. An overlap of a percent
or two is all you should see for hard granular particles.
So either your particles are too soft or your velocities are too big,
or your timestep is too big. The granular model in
LAMMPS is not for deformable particles, except in the
limit of small overlaps (e.g. the Hertzian or Hookean model).
Steve