Compilation errors trying to build a ramalama fedora image

[maxamillion]

hello all! I'm trying to build a ramalama container based on Fedora and using the Fedora ROCm libraries because those enable more gfx series of the APUs than the official ROCm packages (including one I happen to own and am wanting to test on)... however I'm getting a compilation error on something C/C++ that I lack the expertise to sort out and I was hoping someone might be able to help

I have my changes here: https://github.com/containers/ramalama/compare/main...maxamillion:ramalama:fedora-rocm?expand=1

build log here: https://maxamillion.fedorapeople.org/ramalama-build-out.txt

Thank you!

[ericcurtin]

Thanks for looking at this @maxamillion please open the failing PR with that change.

With copilot you can actually click a button to ask to diagnose the failing CI build, sometimes that saves a minute or two of a human parsing the issue!

[ericcurtin]

First of all you'll need this to be executed, it's not in your example Containerfile:

    rocm)
      common_flags+=("-DGGML_HIPBLAS=1")
      ;;

this currently is not building a ROCm based version of llama.cpp.

You can run the ubi9 image on Fedora for rocm.

But we can build a Fedora based version no biggie. But I would highly recommend comparing against how the UBI9 rocm image is built.

[ericcurtin]

The issue being ran into is from fmt from the kompute backend, that part will likely get skipped in a ROCm build (since ROCm is an alternative to kompute)

[maxamillion]

Thank you for the detailed response, I'll get that fixed up and test it. :)

[maxamillion]

that fixed it! PR here #596 ... thank you again!