python312Packages.tensorflow: use tensorflow-bin by default #345658
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
GaetanLepage
force-pushed
the
tensorflow
branch
from
184346b to
e80cf0a
Compare
ofborg
bot
added
8.has: clean-up
8.has: package (new)
|
|
cc @NixOS/cuda-maintainers, #342112 (where this was discussed some) FWIW, I have a local work‐in‐progress branch to bump the TensorFlow version and I think it looks feasible. However I’m not a TensorFlow expert and have no intrinsic motivation to maintain it so I can’t promise I will see it through to completion. This seems like a reasonable fallback. |
|
The main blocker for this PR is the failure of |
ofborg
bot
added
the
11.by: package-maintainer
|
I can see how this patch is tempting given the crude tools we have, but I think having |
Thanks for the recommendation. |
GaetanLepage
force-pushed
the
tensorflow
branch
from
e9d3c97 to
804211c
Compare
|
No, sorry for being obscure. Of course I acknowledge I couldn't block this change as an emergency measure based only on nonsensical purist considerations... maybe in future in situations like this we agree on a deadline for merging? @zeuner I'm sooo happy to hear this:) |
|
If we satisfy TensorFlow dependencies with our binary package – if it is our blessed TensorFlow – then it seems like it would be confusing and bad for |
|
Indeed, after the creation of this PR, @zeuner has been resuming his effort on #272838. |
It's a valid point, but this already happens in many places, e.g. with Side note: if we introduced an |
|
I think there’s a difference between overriding the default version for some packages, and overriding the variant used for everything, as we would be doing here. This is just a redirecting of the alias |
Thanks
Well that's where we diverge: I haven't been considering |
|
After discussion on Matrix, I think we should move forward with this for now. I hope @zeuner’s work bears fruit and we can have an up‐to‐date TensorFlow source build for 25.05, but as it stands, the outdated TensorFlow source build is keeping alive an end‐of-life Bazel, a CUDA that was scheduled for removal in #342112, and an LLVM version that would otherwise likely be fairly easy to drop. I’m watching #272838 with interest, but the version it would update to would still reference compilers that are causing fuss now. I would prefer to merge this and mark the source build as broken so that we can drop the old CUDAs and compilers early in the cycle and benefit from the resulting clean‐up, with the hopes of restoring and switching back to the source build on an up‐to‐date version during the 25.05 cycle, rather than leaving the situation unresolved until we potentially have to rush that before release, having spent effort on keeping the old compilers going before then. That said, of course, I’d appreciate @zeuner’s input since they’re the one doing the vital work here :) |
13 tasks
SomeoneSerge
approved these changes
Nov 21, 2024
wegank
added
the
12.approvals: 1
|
|
Oh let's just bite the bullet already. I hope one day we can afford to have a distro provisioned entirely and provably from vcs-tracked source, with bootstrap chains that can be inspected and comprehended by simple human beings; one that can be forked, built, and cached in a sustainable way by single individuals without an expensive build farm; one where all software can be guaranteed to only talk to external services or leak fingerprints when requested by the user, one that won't be rendered hopelessly useless if nuclear war or a natural disaster happens to our internet infrastructure, and all of that dreamy dreamy stuff, but right now many of our maintainers aren't properly funded, and we have to have priorities and accept the embarrassment |
|
Thanks @GaetanLepage, thanks all |
|
We'll now see more of this: #358973 |
This was referenced Nov 26, 2024
To me, That being said, I'm not too familiar with the necessities the 24.11 release comes with. So, since I'm limited in resources and thus cannot provide a sufficiently tested and polished (up to 2.17 are in the queue) source build of a newer release this month, there might be no better option at this point as to stick with a binary download without control over instruction sets etc. Just hoping this won't lead to dependencies important to the subsequent TF source builds being removed from |
Thanks once again for investing your time on this @zeuner ! Using the binary download was the best short-term solution available to avoid having too many packages broken. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Things done
It has been several months that our tensorflow build has been stuck on version 2.13.0 (released Jul. 5 2023).
Also, tensorflow 2.13.0 is not supported on python 3.12, which makes all its dependents incompatible too.
To this date, python 3.11 is still part of our pair of supported versions in nixpkgs, but this will not last forever.
As no one seems to have both the time and skills to tackle this tricky update (#WeLoveBazel), I propose to use our up-to-date binary package in the meantime.
cc @SomeoneSerge @zeuner @mweinelt @abbradar
nix.conf? (See Nix manual)sandbox = relaxedsandbox = truenix-shell -p nixpkgs-review --run "nixpkgs-review rev HEAD". Note: all changes have to be committed, also see nixpkgs-review usage./result/bin/)Add a 👍 reaction to pull requests you find important.