Trying to resolve the compilation error:

error: more than one conversion function from "__nv_bfloat16" to "uint8_t"

. The error seems to related to the architecture of the gpu as mentioned here: https://github.com/NVIDIA/cutlass/issues/4

But it does not seem to resolve the problem.

Relevant compiler flag:

  -- CUDA supported arches: 7.0;7.5;8.0;8.6;8.9;9.0
  -- CUDA target arches: 86-real
  -- CMake Version: 3.30.3
  -- CUTLASS 3.5.1

What could be the problem?

I wish I could provide a small example to reproduce the problem but it does not seem to be applicable in this case.