Undefined symbol at::native::_bmm_out_dtype_cuda

After searching through the source, I found that there are no function defined for this operation which leads to undefined symbol error

Could you share a minimal and executable code snippet to reproduce the undefined symbol error, please?

export CC="/usr/bin/gcc-12"
export CXX="/usr/bin/g++-12"
export DEBUG=1
export USE_DEBUG=1
export USE_CUDA=1
export USE_CUFILE=0
export BUILD_TEST=0
export USE_FLASH_ATTENTION=0
export MAX_JOBS=3


python setup.py develop

After the build, we simply have to run “import torch” which will trigger this error while loading the shared library

I wonder where is the function is defined in the codebase ?

Are you using a specific commit or just the latest main?

I think its a mistake from my side, it looks like one of my source files was empty. Really sorry for the trouble.