What to tell a cmake process about CUDA locations

I am attempting to build a package (LAMMPS, from Sandia Labs) that has a CUDA option. When I run cmake, I get errors like this:

Could NOT find CUDA (missing: CUDA_INCLUDE_DIRS CUDA_CUDART_LIBRARY) (found version “11.5”)

Since there are multiple versions in the tree beginning with /opt/nvidia/hpc_sdk/Linux_x86_64, it’s not clear how to set variables like this. It SEEMS like the build package assumes that everything is located in /usr/local/cuda. I have such a tree as well, but it seems outdated.

Is there a correct way to set this up?