I am trying to build the torchaudio extension on my university’s HPC, which provides PowerPC computing nodes for machine learning applications. So I have to build from source since no wheels for ppc64le are available, AFAIK.
I am also limited to PyTorch 1.6.0, because there is an optimized module available on the HPC and I would like to use it. I therefore am trying to build torchaudio v0.6.0 (according to the compatibility table in the torchaudio repo).
Here is the command I am running:
CC=gcc CXX=g++ BUILD_SOX=1 python setup.py install --user
When I do so, I first get an error about the third-party libs MAD and LAME and their config.guess, which cannot guess my platform. I was able to get around that by overwriting the respective config.guess files using an updated version taken from here.
After that, everything seems to build just fine, but when I import torchaudio, I get the warning:
UserWarning: torchaudio C++ extension is not available.
Can anyone tell me what this means exactly and how I can make the C++ extension available?