Flatpak - Playground Services not using GPU acceleration

[grantomation]

Bug description

Using flatpak, the pod created for an AI service does not pick up GPU acceleration. It only uses CPU inference.

Operating system

Linux - Fedora 43

Installation Method

from Podman-Desktop extension page

Version

next (development version)

Steps to reproduce

• I am running a fresh fedora 43.

• I have an NVIDIA GPU with the drivers installed correctly from rpmfusion, I also installed container runtime and everything from the podmnan command line is working as I would expect it should. I can run a container that gives me the output of nvidia-smi.

• I have downloaded podman desktop 1.23.1 from flathub.

• I installed the AI Lab extension from the gui.

• I have enable GPU support in the podman settings.

• Using flatseal I can see that podman desktop has access to the GPU.

• When I create a service it does not recognise that it should be GPU accelerated (it says CPU inference)

• The container is created quay.io/ramalama/ramalama-llama-server@sha256:9560fdb4f0bf4f44fddc4b1d8066b3e65d233c1673607e0029b78ebc812f3e5a

• The logs of the container show
  podman logs be9bb138cc02
  ggml_vulkan: No devices found.
  build: 1 (b0d5299) with cc (GCC) 15.2.1 20250808 (Red Hat 15.2.1-1) for x86_64-redhat-linux

Relevant log output

podman logs be9bb138cc02
ggml_vulkan: No devices found.
build: 1 (b0d5299) with cc (GCC) 15.2.1 20250808 (Red Hat 15.2.1-1) for x86_64-redhat-linux

Additional context

No response