Let's drop ollama-gemma2-2b and forbid GenAI models
Hey all, There's been some discussion about this via chat but no action has been taken. Some time ago, ollama-gemma2-2b was uploaded to [extra] which has caused some discussion for various reasons. I'm not writing this mail to address or open up the discussion for ethical or political concerns. This is about pragmatism. With the current state of things, I don't think we should have any LLMs or other GenAI models in official repos. I think it would be much more practical to offer downloaders such as huggingface-cli or ollama instead. GenAI models become out of date every other week and they tend to be rather large too. Downloaders allow users to download these models from specialized hosters at good speeds and to easily stay up to date. I don't think we're doing anyone a service by offering outdated GenAI models in the repos. This mail is explicitly NOT about other ML model types such as rnnoise. Those tend to be small and unchanging. They should stay in repos where it makes sense and if there's software that uses them (such as Mumble in this particular example). I propose this: - Drop ollama-gemma2-2b - Forbid other GenAI models from entering official repos for the time being What do you guys think? Cheers, Sven
Hi, On 13/11/2024 05:53, Sven-Hendrik Haase wrote:
Hey all,
- Drop ollama-gemma2-2b - Forbid other GenAI models from entering official repos for the time being
Agreed, as a side-note I would to also propose that we create a new RFC for uploading proprietary packages. We have been very lenient in the past and I am not saying this should change but we need to better document when X is allowed and how we publically document that we are allowed to re-distribute X. For example teamspeak is nicely documented [1] [1] https://gitlab.archlinux.org/archlinux/packaging/packages/teamspeak3/-/blob/... Greetings, Jelle
I completely agree. No notes. Campbell On November 12, 2024 11:53:48 PM EST, Sven-Hendrik Haase <svenstaro@archlinux.org> wrote:
Hey all,
There's been some discussion about this via chat but no action has been taken. Some time ago, ollama-gemma2-2b was uploaded to [extra] which has caused some discussion for various reasons.
I'm not writing this mail to address or open up the discussion for ethical or political concerns. This is about pragmatism. With the current state of things, I don't think we should have any LLMs or other GenAI models in official repos. I think it would be much more practical to offer downloaders such as huggingface-cli or ollama instead. GenAI models become out of date every other week and they tend to be rather large too. Downloaders allow users to download these models from specialized hosters at good speeds and to easily stay up to date.
I don't think we're doing anyone a service by offering outdated GenAI models in the repos.
This mail is explicitly NOT about other ML model types such as rnnoise. Those tend to be small and unchanging. They should stay in repos where it makes sense and if there's software that uses them (such as Mumble in this particular example).
I propose this:
- Drop ollama-gemma2-2b - Forbid other GenAI models from entering official repos for the time being
What do you guys think?
Cheers, Sven
+1
On 11/13/24 5:53 AM, Sven-Hendrik Haase wrote:
Hey all,
There's been some discussion about this via chat but no action has been taken. Some time ago, ollama-gemma2-2b was uploaded to [extra] which has caused some discussion for various reasons.
Thank you very much for assembling the information and bringing up the discussion here, much appreciated.
I propose this:
- Drop ollama-gemma2-2b - Forbid other GenAI models from entering official repos for the time being
What do you guys think?
I very much second this. The reasoning sounds fond and I believe this is a good default to agree on. Sincerely, Levente
On 13.11.24 05:53, Sven-Hendrik Haase wrote:
Hey all,
There's been some discussion about this via chat but no action has been taken. Some time ago, ollama-gemma2-2b was uploaded to [extra] which has caused some discussion for various reasons.
I'm not writing this mail to address or open up the discussion for ethical or political concerns. This is about pragmatism. With the current state of things, I don't think we should have any LLMs or other GenAI models in official repos. I think it would be much more practical to offer downloaders such as huggingface-cli or ollama instead. GenAI models become out of date every other week and they tend to be rather large too. Downloaders allow users to download these models from specialized hosters at good speeds and to easily stay up to date.
I don't think we're doing anyone a service by offering outdated GenAI models in the repos.
This mail is explicitly NOT about other ML model types such as rnnoise. Those tend to be small and unchanging. They should stay in repos where it makes sense and if there's software that uses them (such as Mumble in this particular example).
I propose this:
- Drop ollama-gemma2-2b - Forbid other GenAI models from entering official repos for the time being
What do you guys think?
Cheers, Sven
Thanks for the feedback everyone. Removed ollama-gemma2-2b from [extra].
participants (5)
-
Brett Cornwall
-
Campbell Jones
-
Jelle van der Waa
-
Levente Polyak
-
Sven-Hendrik Haase