how to run with low version pytorch(1.12)-cuda(11.6) and high version gpu(sm89)

[j-cyoung]

I'm using rtx4090, supported by cuda-toolkit version >= 11.7. In a project I have to use torch1.12(compiled by cuda <= 11.6). I can use the torch 1.12 with internal cuda support, but I cannot compile the tiny-cuda-nn, because it shows that my current cuda version is not in compatibility with current torch vision.