Dear MarsSeed et al, The difference between tensorflow-rocm and tensorflow-amd-git is that tensorflow-rocm pulls from Google's official repo while tensorflow-amd-git builds from AMD's fork. Google's repo is behind AMD's fork. I don't know if this justifies having separate packages. I adopted tensorflow-amd-git six months ago because, at the time, tensorflow-rocm did not work on my not-officially-supported GPU. I can build and install that package without errors, but any code that actually runs on the GPU would cause a segfault or similar. However, tensorflow-amd-git worked fine. It is possible, now that we are at Tensorflow 2.13, that this is not an issue. I have no plans to test this due to the time commitment. I would not be opposed to deleting this package. My GPU, the RX 6750 XT (gfx1031) is not officially supported, and I think I might be the only person using tensorflow-amd-git. It's not a big deal to delete it. I'm not very active anymore. I hope this helps clarify the situation; I apologize if I have sent my remarks to the wrong place. Regards, Mark Peschel On 12/25/23, notify@aur.archlinux.org <notify@aur.archlinux.org> wrote:
MarsSeed [1] filed a deletion request for tensorflow-amd-git [2]:
Duplicte non-VCS package of AUR/tensorflow-rocm. The latter also conforms better to upstream fork's name.
[1] https://aur.archlinux.org/account/MarsSeed/ [2] https://aur.archlinux.org/pkgbase/tensorflow-amd-git/