Missing nvcc?

[Ender1618]

I am on Ubuntu 18.04 and have installed pytorch via:
conda install pytorch torchvision torchaudio cudatoolkit=10.2 -c pytorch

It works and uses the systems GPU fine.

When running the init.sh script, it fails because it does not have access to nvcc.

When installing pytorch with cudatoolkit=10.2, this does not install nvcc.

The system already has the cuda toolkit installed under /usr/local/cuda-10.2.

But I can't get the pytorch installed via conda to use use the system cuda, it wants to use the the one installed by conda in the local conda environment.

How can can I get init.sh to work?