Hunter Chasens

I couldn't get it running on Linux and a 7900xtx, tried both transformers and llamaCPP.

> A workaround that worked for me (Archlinux) is to use the systems `libgomp.so` instead of the included one. If you want to try it for yourself: > > 1....

@brucemacd Any update on this?

Making `cyrus-sasl-xoauth2-git` a dependency might be an issue. If you're not on Arch than you don't have access to the AUR. Without it, installing `cyrus-sasl-xoauth2-git ` requires building it from...

I definitely think OAuth is worth pursuing. I tested your fork on Fedora. Due to the difficulties setting up the dependencies, I found it much easier to just manually configure...

maybe by editing the "{{json .NetworkSettings.Networks }}"

Does this take into account task affinity (model affinity) for model types? Ollama requires time to load in new models, so it makes sense to send requests of the same...

> Ollama has to wait for the upstream llama.cpp backend ([ggerganov/llama.cpp#6849 (comment)](https://github.com/ggerganov/llama.cpp/issues/6849#issuecomment-2072860077)) to support it first. I just got added. Next Ollama release with a pull from the llamacpp mainline...

I'm seeing this with my 7900xtx

So I figured it out. When using ROCm it tries to select your first GPU which is your integrated graphics. There's not enough VRAM so you get a segmentation fault....