*What* didn't compile? The gpu library?

No problem here with 10.1 on Ubuntu 18.04 LTS. I'm using the
git checkout from around May/15 but always patch the
following files before the compilation:

for my needs with
     CUDA_ARCH = -arch=sm_75

(because of my RTX-20xx accelerators.)

Then, the compilation is performed from
*within* the 'gpu' directory:

   cd mylammps/lib/gpu/
   make -f Makefile.linux.single

After the compilation ended successful (which may take quite a while),
you'll find some new files in the gpu directory (ls -latr):
  - libgpu.a (the gpu link library)
  - Makefile.lammps (build-information for creating the gpu-executable)
  - nvc_get_devices (run it and see if LAMMPS is able to detect your gpu)

Later, you may "activate" built-in gpu support for your current LAMMPS
executable by doing:
  make yes-gpu

before the last build step from the mylammps/src directoy;
this tells the build process to include the gpu-library
you *already built without error* before in the gpu directory
as described above.