Hi. I am struggling to compile many pre-build packages for windows, which are compiled with different cuda versions and torch versions.
example:
- package_name-package_version-torch1.8.0-cu102-cp36-cp36m-win_amd64.whl
- package_name-package_version-torch1.8.0-cu110-cp36-cp36m-win_amd64.whl
- package_name-package_version-torch1.8.0-cu111-cp36-cp36m-win_amd64.whl
I’m not sure I understand the issue correctly, so could you explain what “compile pre-built packages” would mean? Are you having trouble installing the Windows pip wheels or building wheels from source?
Thanks for your reply. I develop a pakcage based on pytorch, which is supported in Windows. I want to release my pre-build packages (wheel) with differenct cuda and pytorch. But I don’t want to manually compile it. And I want to compile in docker but torch don’t provide image for windows (Docker Hub).
Since you want to build your own custom wheels you could take a look at the pytorch/builder repository where the conda binaries and pip wheels are built.
1 Like
Thanks for your reply, I will have a try.