Been installing llama.cpp using aur and have ccache installed. Well for some reason one of the dependencies has some system lib incompatibly when it uses the cached object, you'd think running ccache -c would fix it, but no had to manually delete the cache folder

Hasn't llama been ripping out and rewriting all the low level code in their project?

Whats going on with that?

@Wisecrack I'm not too familiar with the internals I just know it works. They did remove the multi-model stuff for a while, but have re-introduced it recently to support the mmoproj file type (thought I haven't seen how to use it with the wrapping libraries)

@BordedDev "mmoproj"

Not familiar.

I think the group behind llama is planning on some sort of commercial offering though, thats what it looks like.

@Wisecrack That's just the file format for models which support multiple things like vision. It's an additional file to the actual model

It wouldn't surprise me

@BordedDev just has the same feel as every other open project that denied going commercial right before it announced it was going commercial or obtaining a venture partner.

@Wisecrack it definitely has a weird organizational structure, they also keep spinning up new cli tools instead of documenting and integrating into their core library. For multimodal stuff the main maintainer didn't want to do it because "there should be more people with core knowledge", something weird is going on

@BordedDev While I think he's right that there should be more people with core knowledge (and I can't really place why), it also sounds distinctly like pseudo-sincerity or rationalization, but I don't know enough about the maintainers to say whether they're being genuine or not.

@Wisecrack That was the same feel I had, kinda sour grapes that people didn't take his original way of doing the multi-modal stuff (he also made the guff format afaik which is a very efficient format for models)