Talk:Ollama

Where are the LLM stored?

When Ollama is installed and being used, you can download an LLM but where is the LLM stored? Colper (talk) 14:19, 1 May 2025 (UTC)

They're stored in /usr/share/ollama/.ollama/models, and the location can be changed with the environment variable OLLAMA_MODELS (e.g. by adding the environment variable to the service unit). See this: https://github.com/ollama/ollama/blob/main/docs/faq.md#where-are-models-stored Sindastra (talk) 14:48, 1 May 2025 (UTC)

They are stored under /var/lib/ollama, but in the current latest 0.10.1-1 ollama, there is a symlink from /usr/share/ollama to /var/lib/ollama. Yxz (talk) 10:09, 3 August 2025 (UTC)

failed to listen: listen tcp :11435: bind: address already in use

seems to be a popular question with lots of different answers - which is the recommended remedy, if e.g. https://github.com/bernardo-bruning/ollama-copilot is to be run? Thanks! Ng0177 (talk) 06:29, 18 July 2025 (UTC)

You need to either configure a different port for ollama or find out what is using the port and change its configuration instead. — Lahwaacz (talk) 11:14, 19 July 2025 (UTC)

LLM-assisted changes

This has been highlighted in Talk:Llama.cpp - ArchWiki, but I feel like articles about LLMs are fated to be poisoned by LLMs.

TL;DR:

• Don't use LLMs to update/generate ArchWiki articles, at least don't touch mine.
• If you can't afford a search engine and/or are too lazy to think with your brain and type with your hands like the authentic American craftsmanship, don't touch an ArchWiki article.

Thanks for your cooperation.

SH3NG1UN (talk) 08:15, 4 February 2026 (UTC)

Hi @SH3NG1UN

Since you reverted my changes, I guess this comment is in part addressed to me.

Do you think I used an LLM to edit this article? If so, what makes you think so? I did not. I just spent 10 minutes reading the last version and carefully amending it to (as the description says):

- add a missing info about launching the server directly, not through the systemd service

- simplify some paragraphs that seemed a little heavy + add a reference to the official Ollama docs for the troubleshooting AMD GPU part.

You do not need to be aggressive with "too lazy/your brain/your hands" language. Instead, please justify why this looks like an LLM-generated edit to you and why you consider it to be deleted.

Also, this is a wiki article, it belongs to every arch user even though you might have created it or added the most part of the info. Therefore, "Don't touch mine" sounds quite inappropriate. Taste20 (talk) 09:31, 4 February 2026 (UTC)

If so, I do apologize for mistakenly thinking you used an LLM, but you really need to put more considerations into your ArchWiki contributions.

When I was expanding this page to what it looks like it is today, I took inspirations from the article about Docker, since Ollama essentially is the Docker for LLMs. The outline is in fact stolen from that article. I don't feel shame to steal, because it's often better to maintain consistent styles with other articles.

Regarding your contribution in general, it was effectively a rewrite, and we won't do that. This will break the original logic chain the author has baked into it. In fact, I may confess I personally think your version is even harder to read than mine. Too many concepts are squeezed together without a clean distinction, and I have noticed several places where Ollama is misspelled as ollama.

Besides, technically speaking, a Docker daemon can also be launched without service manager, but why don't people do that anyway? Because daemons are simply not designed to be executed directly from regular shell sessions, both Ollama and Docker. So we don't write that on ArchWiki.

Regarding the "mine" problem, it's true that the article belongs to everyone, but each contribution does have its owner. The "mine" refers more to the contribution, not the article in general, and since I was the first guy attempted to fill in the details, the duty to maintain it is legally mine.

PS: LLM pollution is a real issue and is happening right now on ArchWiki. Every precautions must be taken to maintain the quality of this Wiki in general. Sorry again for overreacting on this one, though. SH3NG1UN (talk) 10:03, 4 February 2026 (UTC)

Thanks for your apology.

OK, I understand that the serve command would be a really unusual way to launch the Ollama server (apart from developing the tool itself), and therefore is not worth mentioning here.

Apart from this section of the contributing rules, which only states that large rewrites should be discussed beforehand, I could not find anything about not doing small rewrites.

Regarding my "ROCm troubleshooting" part edit, I still think that linking to this Ollama docs section is appropriate, since it provides a list of supported LLVM targets and is explicit about which GPU is supported out of the box. Taste20 (talk) 10:30, 4 February 2026 (UTC)

Linking supported ROCm targets is great, but this is beyond Ollama's scope. Let's consult AMD's own requirements instead: https://rocm.docs.amd.com/projects/install-on-linux/en/latest/reference/system-requirements.html

Feel free to contribute again, and sorry again for blaming you. SH3NG1UN (talk) 10:42, 4 February 2026 (UTC)

It's also ArchWiki:Contributing#Do not make complex edits at once which was not followed properly in your rewrite. — Lahwaacz (talk) 13:29, 7 February 2026 (UTC)

Users and groups section

I was troubleshooting Ollama with ROCm and at one point switched to the install script on ollama.com. I noticed it creates the user ollama and adds it to the render and video groups, then adds my user to the ollama group. The official Arch package simply creates the user ollama and adds no groups. I was having an issue with both the Arch package and also still with the ollama.com-installed binaries, whenever I tried to run a model the service would fail with open /dev/null: permission denied and this would not be resolved until the ollama user was added to the users group.

I'd suggest adding a "Users and groups" section to this page documenting that the ollama user may need to be a part of users, render, and video. Bhoffman (talk) 17:26, 23 June 2026 (UTC)

Then why does everything work fine on my machine without these