The Open Web UI project initially started out as a Ollama frontend. To run local models, one needs GPU support and Cloudron currently does not have GPU support. Over time, Open Web UI has shifed to be a frontend for OpenAI API compatible services . It is also now possible to configure Ollama in another computer which has a GPU and then point Open Web UI to use that Ollama.

I guess we can mark this as stable since it's still useful while we figure out GPU support in Cloudron.

@vladimir-d In the future, I believe we should let the app handle most of it's configuration, environment-wise.

Variables such as RAG_EMBEDDING_MODEL, etc. are all set via the application UI and saved in the config.json file.

Now, for the initial issue of this topic, I've solved it by resetting the embedding model engine.

@LoudLemur said in ETA for GPU support? Could we contribute to help it along?:

@girish said in ETA for GPU support? Could we contribute to help it along?:

I think the hard part is the GPU support in docker is varied.

From Arya:

"As of 2023, GPU support in Docker, particularly for AI applications, has made significant strides but still faces challenges. The main issue is that Docker was originally designed for CPU-based applications and struggled to efficiently utilize GPU resources. ...

Thanks for that info, I wasn't aware of that challenge but it certainly makes sense.

The Open WebUI installation page talks about running the Docker image with GPU support but doesn't mention those problems: https://docs.openwebui.com/getting-started/

@firmansi not needed. apps are packaged as containers that have the right dependencies

LiteLLM is getting fairly mainstream imo (although things are moving very fast) and does more stuff than just accessing other models (load balancing, logging, spending checks…) so have a LiteLLM process started besides OpenWebUI would be useful.
In the meantime, I've followed this specific comment (https://github.com/open-webui/open-webui/issues/3288#issuecomment-2219524566) from the thread and I was able to get the anthropic/claude models with just configuration of openwebui, so that's already a bit of a win.

@Jack-613 this is a Cloudron related forum, so if you are not using OpenWebUI on Cloudron, then you should probably ask in the upstream project instead.

😖

Ah ok. I was under the impression that ollama was listening just on the localhost inside this app. If its not, this would work then. I'll test.

@eddowding said in Which models know most about Doughnut Economics?:

Does it matter much if you can add docs to them and have them answer intelligently?

I tried that yesterday. It didn't help in the slightest.

Answering my own question - yes - follow this: https://github.com/open-webui/open-webui/issues/3288

@taowang maybe worth reporting upstream? I think config.py is source code and not a configurable file. Even in their docker it's not stashed.

@shrey said in Using Mistral API seems broken on cloudron:

@nebulon Thanks!
Indeed, that was it.

RAM was set at 2GB. Whereas, it seems to require a min. of 5.5GB to function.

That's right, and if you can provide it with at least 8-16GB of RAM you will see a huge difference. These things are meant to consume large resources for now, but it's getting better.

I already have a volume, but I will create one more and mount it!

@JOduMonT yeah, I think you are on the old package. Please start another instance. In the latest package, we have fixed the NLTK download directory.

@girish There is no problem. We lost all the content of our conversations on OpenWebUI, but we prefer that it happens now rather than in six months and after six months of use. More stability, we will always be for it and it is better that it is done at the beginning as now.

This is what was updated in Cloudron

image.png