ci : fix Docker workflow

[ggerganov]

I just disabled the workflow since it has been crashing for a while now:

[KerfuffleV2]

System.IO.IOException: No space left on device : '/home/runner/runners/2.310.2/_diag/Worker_20231014-105256-utc.log'

It's just that the disk is full, right? Or maybe just isn't allocated large enough to run the current tasks. (At least all the ones I saw were just No space left on device errors.)

[extradosages]

This might be relevant: docker/build-push-action#968.
Here is a suggested solution: actions/runner#1807 (comment).

[samm81]

I thought I'd give this a shot, but couldn't manage to replicate the issue. after forking, the "Publish Docker image" workflow runs fine in my repo. I tried reverting back to 11bff29 since that was the commit of the most recent failure before the workflow was manually disabled, but even that succeeded 🤷

I went ahead and added the jlumbroso/free-disk-space action to the workflow since it seems straightforward. test runs in my repo indicate that it frees up 34gb on the github test runners (turns out they only have 24gb remaining free by default of the 80gb total disk they get) & that it doesn't interfere with building and publishing the image.

pr #4150

[rhuddleston]

@ggerganov thoughts on that PR?

[ggerganov]

Should I try to merge #4150 and re-enable the docker action?

[rhuddleston]

Yea let's try it

[ggerganov]

Queued: https://github.com/ggerganov/llama.cpp/actions/runs/7293206278

[rhuddleston]

nice looks like it's worked!

[rhuddleston]

On another note I see there is full-cuda-afefa319f1f59b002dfa0d1ef407a2c74bd9770b and full-cuda tags but should also have full-cuda-b1678 tag

[rhuddleston]

#4584 here's my attempt at this

[ggerganov]

How to fix this: