2026-04-26 23:22 UTC
It was the container itself after all! fffff I spent so much time trying to figure it out, but the official ollama container uses an older rocm version that doesn't work with the 9060xt that's in the server :blobcatgooglyholdingitsheadinitshands:
it's now working! Gemma4:26b is running on this. Note that Ollama connects through the frontend ingress node, which automatically routes it to the GPU node, the scheduler automatically makes sure the Ollama container is spun up on a machine with VRAM.
Replies (1)
-
@anthropy@mastodon.derg.nz 2026-08-05 21:06
I've been steadily adding more onto StackWyrm over the last few months, and it's gained many things like a built-in self-healing object storage, CI pipelines and docker/podman image registry, hardware monitoring and storage setup, and much more. right now I'm particularly giddy about getting an automated netboot system working, which can auto diagnose and even reimage servers with issues based on a custom ipxe image, fired off by the Problem Solution Actuator (PSA) system.