Possible enhancement to support deterministic ML container images?

[RyzeNGrind]

I have come across this https://github.com/replicate/cog and I see they have this open issue re: reproducible builds.

And as I type this issue, it seems I have a reply there pointing to: Cognix A Cog container but for Nix. Perhaps we can combine work efforts or converge at a mutual goal to get production ready deterministic ML model container images?

When working with ComfyUI or InvokeAI, sometimes you want to be able to test working with models in a reproducible manner and this may be an opportunity to expand the Nixified-AI flake to include some potentially useful projects as buildinputs?

Happy to hear what others have to say on this as well.

[max-privatevoid]

I must ask (again): Why involve containers in this?