Your local CUDA toolkit should be detected automatically. If that’s not the case, set the location via the env var e.g. to:
CUDA_HOME=/usr/local/cuda
in case you are using the default location.
The install log would then also show the detected CUDA toolkit version as well as its location.