On Sun, Jan 9, 2022 at 3:59 AM gevisz <gev...@gmail.com> wrote: > > вс, 21 нояб. 2021 г. в 17:12, Mark Knecht <markkne...@gmail.com>: > > > > Congrats! > > Thank you. However, it was not for long. On 30-12-2021 recompilation > of the same tensorflow-2.7.0 because of some changed dependencies > failed with the same f**ng "Bazel failed" error as before. > > So, I am currently going to degrade my Gentoo system to the state it > was in on 12-12-2021, when its last update was successful and froze it > forever. > > The problem is that I do not know how to do it but I am going to post > this question as a separate thread. (I am using webrsync method.) >
Sorry for the problems. I saw your other thread about downgrading Gentoo. I agree with the other responses you got there that Gentoo, in general, does not support standing still much less going backward. I will offer what will probably not be a popular comment but my opinion is Gentoo is exactly the wrong sort of distribution for doing work in tensorflow. With all of it's updates, limited testing of packages, extreme amounts of code building, and not being a distro that the official tensorflow folks even verify on, it's just too hard. (And a contributing factor to how I moved away in the beginning.) My thought is that you might create a 20.04 LTS Ubuntu VM (or possibly an LXC container) running whatever your desktop flavor of Gentoo is - I run Kubuntu - and just run tensorflow in the VM. You won't easily get GPU support unless you deal with passthrough, but the software will just work and you won't spend time dealing with building code which can be spent coding tensorflow. If you insist on running in Gentoo consider the LXD container running an older rev of Gentoo. (If you can find one) Get it working, if you can, and then never update it. lxc image list images: gentoo There are openrc and systemd versions available, but a Kubuntu stable container would more likely to 'just work' IMO. Good luck, Mark
