llamacpp cache access denied in llamaswap:cpu

[DIGist]

Updated my llamaswap:cpu container to try out the new Qwen3-vl support.

ghcr.io/mostlygeek/llama-swap cpu ee1d1321e80b 18 hours ago 130MB
ghcr.io/mostlygeek/llama-swap c4db55692ff6 7 days ago 130MB

I was getting error messages in llamaswap that it couldn't load the models even though they had all worked previously.

I got into the llamaswap container with "docker exec -it harbor.llamaswap /bin/bash" and found I was unable to access anything under /root

I have altered my compose.llamaswap.yaml and llamaswap/config.yaml to use this path instead and the models are working again as a temporary measure: /nroot/.cache/llama.cpp

[av]

Thank you for the report!

I just checked, indeed, latest llamaswap image now uses app user by default instead of the root, so previous mounts were invalid. I've updated Harbor's llamaswap.yml file and it'll be included in the next release