Meeting started at 2026-08-13 16:31:05 UTC

The Meeting name is 'fedora-ai-ml-sig'

!hi

tflink: Tim Flink (tflink)

At least zodbot is finally back 😛

!hi

didn't know it went away

Carol Chen: Carol Chen (cybette) - she / her / hers

FreeIPA woes have been plaguing infra for the last couple of weeks

yeah, I've had plenty of fun wiht that. it's at least better than it was now

slow rolling on the oh-hum stuff..

ok, let's get this party started

err..

Tom Rix: this one is yours

noo. my topic is rocm 7.14

i can talk about llama.cpp if folks want.

!hi

I think llama.cpp was July 16th

rocm 7.14 has been in holding for a while, with the branch of F45, the wheels have started again on rawhide content.

Also, hey Carol Chen Jeremy Newton 👋

oh, I'm looking at the wrong day

my bad

once rocm 7.14 settles, i'll be updating llama.cpp etc.

my ai packages

so maybe in 2 weeks we can talk about those.

newer ROCm is coming out soon??

!hi

that will land in the 'preview' rocm slot and live in a copr for a while. we are expecting it will be a major update so we will have to see. the compat-rocm bits we did for F45 will will lean on if needed.

any rocm questions ?

moving on.

this is for awareness.. i have been playing around with pi-coding-agent in Fedora to do fedora work. here is a sample https://github.com/trixirt/fedora-pi-prompts/blob/main/documentation/fedora-release-notes.xml

sorry, distracted

This looks really neat

I wonder if Petr Bokoc has seen anything like that already before for release notes 👀

it has been useful in running a prompt over a directory full of my packages and getting it to automagically clean stuff up.

there is a real balance of giving pi enough information and not asking it to do too much.

i'll be adding more as i run into problems with my day-to-day.

the take away is fedora is cabable of day-to-day ai tasks.

do you think that different backing models would make much of a difference in the output?

go explore and have fun.

i am using qwen3.6:35b the biggest ollama has for F45. i'll be trying out others as i get back to updating ollama.

picking the model is really a personal choice, depends on hw

and some models aren't reasoning or have tools support.

yeah, it was more musing on my part. I suspect that the output wouldn't be completely different when using models of a similar cailber

yes. also i am using beefy strixhalo system.

for the best results, get some good hardward and don't fight with models

any other topics before we get to the one that will likely take the rest of the time we have today?

What is the RPM package name for Pi? I was trying to find it

pi-coding-agent

Thanks

the workflow is 'pi < your-prompt'

Ah, in F45, not F44, right?

correct.

pi landed in F45 as well as openclaw

Nice

a bit on the next topic. has everyone _read_ the proposal ?

this was filed earlier today

Justin Wheeler: this is yours

Sure. We were already talking about this in #ai-ml:fedoraproject.org and I also see some new comments on the ticket

same suspects so far, I think. mostly me and Tom Rix

yes, so point of order, is it productive to discuss this when everyone else on the call is catching up ?

If we want to keep a big-tent approach for AI Use, then we can fit it in here, and then I can propose a vision/mission statement to close out #27 and start our Fedora Docs site

One goal I'd like to help out with is getting content from the Wiki into a docs site, for instance

If we have a big-tent approach, I can make different Antora modules for each of the existing sub-SIGs so there is a clear home for the content, and then one for AI Use

I brought forward the proposal for splitting because I figured that would be the easiest approach, but in hindsight, maybe not

I'm not saying that we have to keep everything big tent style. I am, however, annoyed that you've chosen to make a proposal like this in lieu of other discussion

but I suspect that has come across rather clearly in the last hour or so

for me (as a relatively newcomer) I prefer the big-tent approach with clear signposts of sub-groups depending on one's interest. I think there's overlap in the interest areas as well, people package the tools they want to use.

Justin Wheeler: what are you hoping to get out of the discussion in meeting today?

I would be fine to close this as `wontfix` and begin working on #27 (i.e., a vision/mission that fits the big-tent scope) and #29 (i.e., improved welcome page in Fedora Docs)

I am curious if Gordon Messmer has any thoughts here too

I could probably get those two tickets into PRs before my PTO for the second half of August

1: there are probably good reasons to split out sub-SIG groups for access controls, but at the current volume it's probably best to keep a broad unified sig to host comms

is that not what we are already doing?

I didn't realize that #27 was still open

yeah. we doing that now.

What I mean is: I think splitting the SIG, per se, is probably not the best approach

Oh yeah, probably worth mentioning the part about the EU Cyber Resilience Act. I am not sure what this will all look like in Fedora, but I could see it being a requirement for Fedora to have clear docs/lists of package maintainers for a given set of packages. My understanding is that if the EU regulatory authority goes to Red Hat to ask questions about security issues, this information would need to be provided to the EU regulatory authority in some capacity. So, I think making sure that the `ai-ml-sig` group does not assign any package commit rights is sufficient to check whatever box is necessary for this compliance work.

related: one of the things I'm seeing elsewhere right now is that the owner of packages and the owners of their dependencies aren't always communicating enough, and more broadly, packagers and the developers of the software they package aren't communicating enough, so I tend to think narrow scopes are overall more harm than good

I think if we decide on a big-tent approach, the closing criteria for #27 would be a Fedora Docs page with our vision/mission clearly spelled out, so there is not ambiguity

for at least the third time - THE ai-ml-sig FAS GROUP DOES NOT HAVE ANY PACKAGING PERMISSIONS

Yes, got it now, just adding that part since it came up before

don't we already have clear lists of people with write access to packages via FAS?

Thanks, this explains some of the concerns about structure.

In part yes, but I guess Infra/Releng will need to adapt some things for the reporting parts later. I am not a policy expert but I am seeing some of the CRA work happen adjacent to where I am at Red Hat

this is general fedora problem, not a specific us problem,

That's right

do you, though? this is the second time you've asserted this as "fact" since you first said you understood it

Yes, I am expressing a -1 on the idea of "splitting" the SIG, per se

i'm sure fesco will figure somethign out

Yeah, FESCo will be involved in part

OK, I will write it down this time 🙂

Consensus seems to be pretty uniform for no split, so let's keep to that

2: as far as proposals vs other forms of discussion: I think proposals are the right way to start discussions that are non-trivial.

And I will have some clear things to work on after this meeting

You just can't offer a suggestion that is detailed and justifies the specific changes or actions proposed otherwise

I can assign myself some `!action` items, but I feel like we have an `!agreed` on what to do for this ticket

My 2¢: I don't dislike the proposals per se, it was more the feeling of being blindsided.

For example, I certainly didn't see the split proposal coming out of the earlier discussions about onboarding and roles.

Indeed, I had thought we had resolved against splitting the group at that time.

I don't have an issue with proposals in general and I'm not saying that we shouldn't have more of them. I have an issue with _this_ proposal

That being said, I do realize that I have a track record of opposing proposals as of late but I hope that isn't conflated with opposing all proposals or opposing the idea that anyone could make a proposal

I will get those actions done before EOD tomorrow, or earlier, and folks can discuss across August while I am out. When I'm back, I will circle back on any feedback or change requests

(But other folks have already said that, more or less.)

FWIW, I really think getting this down in writing will help, and I am excited to spend time on the Fedora Docs site

Starting with the small stuff will get us all familiar with the workflow and Antora

Later down the line, I'd love to talk about getting content out of the Wiki and the Docs site so it is better organized. But that will come later and we will talk about it first before it happens

I have an action item from the Fedora Docs Team to find a new home for the Quick Docs AI content, some of which might be entirely outdated/irrelevant at this point

I might work on importing or dropping that content in early September too, when I'm back

For inspiration, I really like the site that the Neuro SIG put together that describes what they've tested and which packages are available, and why.

please let people know if there are docs that are unclear - one of the issues that was sourced for the genesis of this proposal was a mistake on a wiki page that took me about a minute to fix

That might be able to replace the sort of ad hoc list we have now

Also, I am sorry for blindsiding y'all and for the frustration. It really wasn't my intention, and FWIW, I am glad we had this conversation. Now that I am moving from the FCA role, I will be putting more actual work time into things like our docs and onboarding guidance, instead of just talking about it…

Ack 👍️

I can use that for inspiration

I can write an `!agreed` and we could move to open floor, unless there are other things around this proposal we want to discuss?

on a somewhat related note - making a big proposal ~90 minutes before the meeting can make participation difficult for folks who can't prioritize work time to read it

Yeah that was on me, I meant to open it sooner but I know how it probably looked

for future long proposals, please get them posted at least a day before the meeting and preferably a week or so before so everyone gets a chance to read the proposal and know if they should show up for the meeting to discuss

In either case, there seems to be unanimous consent by those here that splitting is not the path we are doing

for now, at least

I will do better on this in the future.

yes. when i asked for 'point of order' i was asking that question in this meeting, folks blew past that.

let's try to do better..

bygones.

the operating ethos since we revived the various ai-ml SIGs has been to keep process and overhead at a minimum until it's needed. if changes need to be made, they need to be made

anything else on this today? especially from folks who caught up during discussion?

I'm set

No objections or updates here for the remainder.

Glad I could catch up with the group after a few months.

if there's nothing else on this topic, we can move on to open floor

Welcome back 🙂

Justin Wheeler: I volunteered you there but I assume you don't mind

Not at all 👍️

anyhow, moving on to

any other items for discussion today?

in the ~5 minutes we have left

I guess just keep an eye out for two PRs from me before the end of the week on the Docs repo. My goal is to get them up before I go on PTO, but there won't be a rush to review them quickly, which should give plenty of time to read and discuss

Nothing else from me

let's close out the meeting, then

o///

Thanks for chairing tflink!

thanks for coming everyone