[Feature] Dockermods for streamlining ML containers

[mritunjaymusale]

I have searched the existing feature requests, both open and closed, to make sure this is not a duplicate request.

• Yes

The feature

Hi,
I recently found out about docker mods that linux server images support .
https://github.com/linuxserver/docker-mods
Through this I was able to use CPU's iGPU for transcoding for jellyfin https://mods.linuxserver.io/?mod=jellyfin
I was wonder if it's possible for you guys to implement a similar mechanism where a one line of code in your main compose file enables gpu utlization without having to mess with a second docker file. This way you guys can have reproducible builds with proper gpu utlization that can be helpful for the ML containers that immich uses especially on AMD CPUs where ROCM stack is hit or miss depending upon the iGPU.

Platform

• Server
• Web
• Mobile

[bo0tzz]

I don't entirely follow what you mean? We already have a single Dockerfile, with multiple targets for the various accelerators.