wrong linux/amd64 docker image

[lilunxm12]

0.1.4-proxy: Pulling from codelion/optillm no matching manifest for linux/amd64 in the manifest list entries

I encountered this error when pulling and it turns out there's no linux/amd64 but only unknown/unknown in the registry

I've checked https://github.com/codelion/optillm/blob/main/.github/workflows/publish.yml but couldn't find what's wrong

[codelion]

It looks like the GH Action runner is getting out of space - https://github.com/codelion/optillm/actions/runs/13032992652

Are you able to build the container locally?

[lilunxm12]

It looks like the GH Action runner is getting out of space - https://github.com/codelion/optillm/actions/runs/13032992652

Are you able to build the container locally?

just finished building without error.

And it's really huge for proxy. I see torch and cudnn as pypi dependency, do we really need that for proxy only image?

[codelion]

We don't need those for proxy, I suspect it is sneaking in transitively via some other dependency here - https://github.com/codelion/optillm/blob/main/requirements_proxy_only.txt I will review and fix it in next release.

[lilunxm12]

We don't need those for proxy, I suspect it is sneaking in transitively via some other dependency here - https://github.com/codelion/optillm/blob/main/requirements_proxy_only.txt I will review and fix it in next release.

Oh, my bad, I did not use Dockerfile.proxy_only, I'll try it again now

[lilunxm12]

That's the result, I think it's still a bit too large for a proxy

[codelion]

Managed to reduce the size to 487.25, none of the plugins or local inference would work in the proxy but the main approaches would all work.