Building torch from source for old GPU

torch.compile
1

I need a 1.13 version of torch with compute capability 3.5 and cuda 10.2, since there is no prebuilt one, I’m doing it myself.
Have been using this guide (changing versions here and there) : Building PyTorch from source on Windows to work with an old GPU
I’m at last steps
performing command

python setup.py install --cmake

gives me

Could not find any of CMakeLists.txt, Makefile, setup.py, LICENSE, LICENSE.md, LICENSE.txt in D:\torch\pytorch\pytorch\third_party/ios-cmake                      
Did you run 'git submodule update --init --recursive --jobs 0'?

I did the submodule update already, it does nothing.

2

Those instructions seem out of date, specifically I’m not sure what the ccmake flag is supposed to do this in this case

Would suggest reading the official instructions pytorch/CONTRIBUTING.md at main · pytorch/pytorch · GitHub

1 Like
4

Had to start all over, but I’m getting new errors now.
Thank you.

Current problem - having VS2022, when requirements are from 2015 to 2019.