as of a few minutes ago, the latest stable version of LAMMPS is 5 Jun 2019.
please note, that you are trying to run a system with about 100 atoms.
since LAMMPS users domain decomposition for MPI parallelization, there is little value in running with more than 2-4 MPI ranks. using 24 is most certainly overkill and will lead to problems. this is a known “feature” of the code in USER-REAXC, that is assumes, that the number of atoms per sub-domain does not change much. but with so few atoms and so many MPI ranks in comparison, the relative change becomes very large. also, it is questionable, if LAMMPS actually becomes faster by using more than 8 MPI ranks for this system. you might try a combination of MPI and OpenMP (3 OpenMP threads per MPI rank for a total of 24).