migrate to libprotobuf321

[ngam]

Checklist

• Used a personal fork of the feedstock to propose changes
• Bumped the build number (if the version is unchanged)
• Reset the build number to 0 (if the version changed)
• Re-rendered with the latest conda-smithy (Use the phrase @conda-forge-admin, please rerender in a comment in this PR for automated rerendering)
• Ensured the license file is being packaged.

[conda-forge-linter]

Hi! This is the friendly automated conda-forge-linting service.

I just wanted to let you know that I linted all conda-recipes in your PR (recipe) and found it was in an excellent condition.

[ngam]

@conda-forge-admin, please rerender

[ngam]

this wasn't solvable for some reason...

pytorch-cpu (0) Error: not solvable (bot CI job): main:

https://conda-forge.org/status/#other_migrations

[hmaarrfk]

@conda-forge-admin please rerender

[hmaarrfk]

Maybe this file needs to be updated for OSX https://github.com/ngam/pytorch-cpu-feedstock/blob/libprotobuf/recipe/conda_build_config.yaml#L1

[hmaarrfk]

@conda-forge-admin please rerender

[h-vetinari]

Maybe this file needs to be updated for OSX https://github.com/ngam/pytorch-cpu-feedstock/blob/libprotobuf/recipe/conda_build_config.yaml#L1

That's there since 8e4ce19. Ideally things should be fine if MACOSX_DEPLOYMENT_TARGET is not set, but 12.3 might not be able to compile for that target anymore.

I remember looking this up once... Based on https://developer.apple.com/support/xcode/, it looks we need to stay with xcode<14, otherwise we don't have support for deployment targets <11 anymore (or add __osx >=11)

[hmaarrfk]

i think lief-dev is now old, and just lief 11.2 or something is out.

[ngam]

Good catch. I hope we can still keep the sdk at 12.3 (that's how we enable MPS for osx)

[hmaarrfk]

This needs: asmjit/asmjit@c598476

[hmaarrfk]

maybe we should just downgrade the c ompiler.

[ngam]

@hmaarrfk no rush on this. I think it would be reasonable to get it done when the next minor pytorch is out. I have no idea where to start with debugging such a failure...

[hmaarrfk]

compilers often increase severity on "bad practices" with newer versions. I've linked to the patch that addressed this.

It seems that for now, the easiest might be to downgrade the compiler

[h-vetinari]

@hmaarrfk no rush on this. I think it would be reasonable to get it done when the next minor pytorch is out

Not merging this makes pytorch and all dependent packages run into the same kind of solvability problems we were discussing for tf. So while we shouldn't perhaps be rushing unduly, we should definitely get this in as soon as time allows, and not wait for the next version.

[ngam]

So while we shouldn't perhaps be rushing unduly, we should definitely get this in as soon as time allows, and not wait for the next version.

Yes, this is what I meant. I think the issue is resolved now though 😄

[ngam]

I can build the cuda ones unless you wanna do it yourself, hmaarrfk. Let me know

[hmaarrfk]

i don't have the compute power right now to build. so if you can build that would be great!

[ngam]

Ok, I will get this done late Tue or Wed

[ngam]

building now, surprisingly this takes significantly longer than the tensorflow builds in my setup... is there some sort of a caching mechanism we can activate? In the tensorflow builds, bazel automatically caches everything and I rely on that to accelerate the builds...

Also, @hmaarrfk, since you are the primary one here, how much longer do you want to keep 11.1 and 11.0 (as well as 10.2)? I am building them all (i.e. not editing this PR)

[hmaarrfk]

i think many downstream packages are still stuck on older cuda. I owuld like to keep the 11 series.

[hmaarrfk]

I'm probably duplicating effort. But I started build too.

[ngam]

I'm probably duplicating effort. But I started build too.

Well, at least build 11.0 and 10.2 pls. I am done with 11.2 and 11.1. I can upload now. I stopped my build. Hopefully you see this and focus on 10.2 and 11.0...

[ngam]

uploads: https://anaconda.org/ngam

sample log: linux_64_c_compiler_version10cuda_compiler_version11.2cudnn8cxx_compiler_version10numpy1.21python3.10.____cpython-log.txt

[hmaarrfk]

ok i've started with 10.2. historically it has 3 hours per build. So maybe 24 hours.

[ngam]

How are we doing on this? If things didn't complete, I am happy to build some of the remaining ones. I would say it takes on the order of 5 hours on my end (compared to your 24 hours), so...

[ngam]

I am going to continue building just in case. The builds will be uploaded in less than 5 hours (hopefully)

[hmaarrfk]

sorry about the delay, tbh, i had the builds, but ultimately more pressing fires happened :/ this is the main issue with CFEP03, it centralizes too much "power"

[hmaarrfk]

I'll wait for your builds, it makes the copying process more streamlined.

[ngam]

Okay, all uploaded: https://anaconda.org/ngam

[ngam]

In #137 (comment) I have one sample log; other logs exist, but I need to anonymize them and upload them. I can do that, let me know if you want to seem them. I used your automating script with only a small edit to make it work for Singularity.

[hmaarrfk]

thanks!

[ngam]

thanks!

Thank YOU. I know this is annoying especially for you, so thank you for doing this for the community all the time and for your patience.