Currently, the sanity check step runs a few example inputs using LAMMPS Python interface.
By default, these just run the standard CPU implementation of the packages.
Many of LAMMPS potentials have accelerated variants that use packages such as Kokkos, Intel, OpenMP, etc which can greatly speed up the execution.

Many applications leverage Kokkos to run large-scale simulations on GPUs.
Some easyconfigs to build LAMMPS with Kokkos (and CUDA) are available, but not properly tested.
To test this, one would just need to change the command used to launch the simulations, e.g. by specifying to use the kokkos suffix of the pair styles, as well as the number of GPUs (and of threads, using OpenMP threading at the same time):
https://docs.lammps.org/Speed_kokkos.html#running-lammps-with-the-kokkos-package