SOS! Errors! Can't even start building TorchAudio from source? Why!

audio
1

I have a Kepler sm_30 graphic card, in order to use CUDA, I need to build PyTorch from source.

So I did. I installed CUDA 10.2 and cuDNN 7.6.5 on my Windows 7 Professional 64 bit OS, installed CMake, Ninja in the conda environment, along with Microsoft Visual Studio Community 2019.

Then I successfully have PyTorch 1.10.0 wheel built. Awesome!

Under the same conda environment, I then successfully built TorchVision 0.11.0 wheel. Great!

Finally, when I tried to build TorchAudio, errors occurred, and the build process couldn’t even start.

Below are the detailed screenshot of the build process. What can I do? Please please please help!

TorchAudio Error_Page_1
TorchAudio Error_Page_1961×1243 272 KB

TorchAudio Error_Page_2
TorchAudio Error_Page_2961×1243 238 KB

TorchAudio Error_Page_3
TorchAudio Error_Page_3961×1243 442 KB

2

The error messages say you have issues with C++ / C compilers. C compiler is unknown.

Try Ubuntu first - Pytorch is well tested on it.

BTW, this discussion Category is for torch.compile and PyTorch 2.0 related compiler issues, not for Pytorch building issues.

3

Thanks for helping move this post from “compile” to the “audio” category.

Well, I still need some expert opinions here, as I only use Windows, and never use Linux or Ubuntu.

I am a musician, not a tech CEO.

Has anyone in this forum ever tried to build a torchaudio wheel on the Windows platform?

4

Need to enable MSVC tool chain so that cl command is recognized.

For MSVC 2019 it’s

call "C:\Program Files (x86)\Microsoft Visual Studio\2019\Community\VC\Auxiliary\Build\vcvarsall.bat" x64

See Building on Windows — Torchaudio 2.0.1 documentation