Re: uploading pytorch-cuda to experimental soon
While it is very difficult to build pytorch-cuda, I think I can try debomatic-amd64
in the future, which seems to be the only choice that can pull non-free
dependency, and does not expire, nor cost me a fortune. But I'm not sure whether it
can finish the build without timeout, as it has got merely a (IIRC) Xeon E5-2697v3 [1].
Anyway... the next time I struggle at the amd64 build... I'll simply give it up.
I hesitate on toggling "XS-Autobuild: yes" because I'm still confused by cudnn EULA
in the case if the debian server downloads and installs cudnn for the build.
Anyone present in the keyring is welcome to contribute the pytorch-cuda build
when the pytorch-cuda in the archive is out of sync with the cpu version.
[1] The cpu build on this service takes roughly 2~3 hours. The cuda build will
only take several times longer.
On Sat, 2023-03-25 at 17:34 -0400, M. Zhou wrote:
> I'm preparing the upload of pytorch-cuda to experimental. The last
> two steps are waiting for the build, and testing it with my nvidia gpu.
> I don't know how much time it will take because I struggle at finding
> a proper amd64 machine to build it.
Reply to: