Nvcc fatal : Value 'sm_86' is not defined for option 'gpu-architecture'

Troubleshooting
5

Is at a desired behavior that setting target = 'cuda -arch=sm_80' doesn’t really change the cuda arch when compiling the model?