CI maintenance: Update checkout, setup-python and upload-artifact to latest versions

[EwoutH]

A bit of CI maintenance, which updates the following actions to their latest versions:

• c1ca357: actions/checkout to v4
• 1d5dc54 actions/setup-python to v5
• 417be61: actions/upload-artifact to v4

For each, the main change is an update from Node 16 LTS to Node 20 LTS.

If preferred, I can add a Dependabot configuration which opens these PRs automatically in the future (in another PR).

[phymbert]

Thanks, have you tested on your fork that it works ? If yes please send links

[EwoutH]

@phymbert here are the runs on my fork: EwoutH#1

[phymbert]

@phymbert here are the runs on my fork: EwoutH#1

https://github.com/EwoutH/llama.cpp/actions/workflows/build.yml is empty

[EwoutH]

Right, missed that one, triggered it here: https://github.com/EwoutH/llama.cpp/actions/runs/8539946210

[phymbert]

I am prudent because the way the repo is cloned has an impact on the build-info.o object.

[EwoutH]

I understand.

All the runs have passed successfully.

[phymbert]

Thanks for having looking at this, please wait for @ggerganov approval

[EwoutH]

Thanks for merging!

Would a Dependabot configuration be useful, which opens such PRs automatically in the future?

[phymbert]

@ggerganov @EwoutH I guess it blocks master as now artifact name are duplicated:
https://github.com/ggerganov/llama.cpp/actions/runs/8551108117/job/23429429188

I suggest to revert and wait for a solution

[EwoutH]

Give me a sec, looking into it: https://github.com/actions/upload-artifact/blob/main/docs/MIGRATION.md#migration

If I understand correctly it just requires naming artifacts differently.

[EwoutH]

Basically there are two options:

1. Keeping the single artifact artifact, with all different ZIP files (AvX, AVX2, Arm, etc.) in that single file. This requires adding a merge step at the end of the run.
2. Upload each generated ZIP as it’s own artifact (so we would get artifact_avx, artifact_avx2, etc.

What’s preferred?

[phymbert]

Let's keep it simple, one artifact per job run

[ggerganov]

I don't have a preference and I'm not sure what is useful for most people. We can do the one artifact per job as suggested and if we see it is not optimal for some reason we can update later

[phymbert]

I doubt people are using it. Previously, it was merged:
https://github.com/ggerganov/llama.cpp/actions/runs/8541354320

But IMHO, separating per backend is easier than downloading a 1GB file. As you say, we can improve later on if it hurts for some reason.

[EwoutH]

#6482 should solve the artifact naming conflict. It will upload multiple smaller artifacts instead of one big one, which I think is more useful for most users and developers.