Meeting started at 2026-05-21 16:30:35 UTC

too fast :D

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

!hi

!hi

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

Clément Verna: Clement Verna (cverna) - he / him / his

Hey

give a minute or 2 for people to join in

!hi

Alexander Lent: Alexander Lent (xanderlent) - he / him / his

alright, we'll start with Clément Verna 's topic since they have to leave shortly

!hi

!hi

I found my way 🙂

Justin Wheeler: Justin Wheeler (jflory7) - he / him / his

Looks similar to something Justin Wheeler has done as well. Clément Verna feel free to elaborate

Sounds good. And this one sounds fun!

And not so stressful 😛

Thanks Carol Chen , basically I would like to start creating skills for using Agents while contributing to Fedora.

Clément Verna: My only comment to your ask is, maybe we build a library of skills in the org, instead of a single repo with many skills? Or is there a particularly good reason to use a mono-repo?

It would be nice to have git repo where we can gather these and have folks contribute

On Tuesday, I realized the thing I put together, effectively, is like a skill, and I changed the repo name to make that more clear.

Oh, interesting, Element Web has prettier web previews now

Justin Wheeler: Didn't think too much about it, the mono-repo I guess let you make it easier to install these skills with your agent of choice

Clément Verna: If there is a technical reason for a mono-repo, no objections from me. But I wonder if we could repurpose the `skill-meeting-secretary` repo into the mono-repo, and it could be the first skill

I put this out there because I was experimenting, I have a Gemini Gem that I use it for, but I'd love to see how others could collaborate, fork, modify, etc.

(i.e., I have no qualms to someone painting up the repo with new commits)

Example of skills we use in CoreOS https://github.com/coreos/ai-helpers/tree/main/skills

(i.e., I have no objections to someone painting up the repo with new commits)

do these work on openclaw ?

This feels like the kind of thing we _totally_ should be leaning into, and it is way more practical for helping other folks use AI/ML tools in Fedora or to work on Fedora

I'm wondering what others think

Tom Rix: should be it is just .md files

…I might have just made this tag 😛

I don't have a strong opinion, just that have 1 place to gather everything (at least at the start) might make it easier to contribute and use

is it worth adding a 'Skills' section to the wiki ?

+1 for making it easy to discover skills and install

https://fedoraproject.org/wiki/SIGs/AI-ML#Training

Would you be willing to take an action item to repurpose my Meeting Secretary skill repo into a mono-repo for many kinds of skills, generally?

Tom Rix: Definitely worth adding, but we might need to figure out (1) what our intention is with maintaining/creating Fedora skills, and (2) whether we should lean into Fedora Docs long-term over the Wiki

Sounds good to me

jflory7 gave a cookie to cverna. They now have 204 cookies, 1 of which were obtained in the Fedora 44 release cycle

Clément Verna: What is a good deadline for you here?

Let's say, for a PR to reorg the repo

Do you have weekly meetings? If so happy to report back next week

and what name should we use?

we have meeting every 2 weeks

Also, I renamed the repo… hope I am not jumping the gun… https://forge.fedoraproject.org/ai-ml/skills-library

Then even better :)

Oh, huh, some Matrix federation lag… lots of messages appeared at once for me

One thing, that is a little bit painful while working with Agents and Fedora is the anti-bot protection. It makes it hard for agents to fetch the doc pages for example

Or read the code on forge.fp.o

I use Claude Code + Forgejo API token

Works like a charm and fixes the rate-limit

A skill for Forgejo API would be neat 😛

Great sounds like something that should go in the README of that repo

Cool… well, anything else folks want to bring up here?

Clément Verna: Thanks for bringing the topic here and for willing to work on this here 🙌

yes, thanks Clément Verna ! are we good to move on to next topic?

+1 from me

No problem, looking forward to this. I think this can help a lot of contributors

Tom Rix: take it away!

Thanks, all. It's late here so I need to drop (family time). Thanks for having me

Several thing from me. we have text to speech package python-piper-tts now.

the training side is an RFE in the wiki, i am breaking down the required packages.

there is one in the wiki that coould use a review, there will be a lot. i'll talk about training more next meeting.

got a link to the page tracking that? ( don't have that handy

https://fedoraproject.org/wiki/SIGs/AI-ML#Training

#link https://fedoraproject.org/wiki/SIGs/AI-ML#Training

the big thing i think we get is 'lighting'

Thanks for linking up with the #a11y-wg:fedoraproject.org folks. I saw you jump in over there

IRC commands are ingrained in our brains forever 😄

and a lot of basic TTS libs.

I am working on a bunch of those (was going to provide an update once you were finished)

i would really like to showcase ai helping with accessiblity. so any help is appriciated.

I am working on executorch and a gstreamer ML framework that sits on top of that, executorch requires torch audio/vision etc and I have a bunch of those already building although I am seeing an issue on x86 around ROCm that causes build issues that I might need Tom Rix 's assistance on

It will have support for a bunch of NPUs that we now support in F-44 plus vulkan

poke me about that. i have been ollama-ing mostly these days. as new versions of ollama roll out newer agents 'ollama launch <something>' pop out.

we landed openclaw. i would like to also land pi

I need to get back to ollama on aarch64 to do some PRs but I've not checked if I'm still blocked from contributing

https://bugzilla.redhat.com/show_bug.cgi?id=2464801

pi ties into emacs coding agent.

i'd like to get emacs and any other editor we care about llm aware.

vim i think i solved with documentation.

i would hope folks with there own favorite things could also contribute packages.

Can we get the LLMs to answer the vi vs emacs question and have a disucssion amoung themselves? 😂

i have done that.

emacs's answer was the usual 'both sides are right'

any other editors ?

I'm happy to assist with aa64 enablement and maintenance as needed

atm i am looking at somethign qt and the new hermes agent.

Tom Rix: nano is the default editior in Fedora

both you and ai can't get out of vim

does nano have a pugin thing like vim,emacs ?

Good point, would be nice to have something working with the default text editor

no idea, I've never used it, was just mentioning

i am hoping the die hard nano person speaks up..

🦗🦗

I think it is a nice-to-have for `nano`

I used pico quite a bit way back when.... but not so much nano

i'd like folks tools in fedora to be ai useful as possible.

vscode would be the other one but I'm guessing most people use that as a flatpak?

i looked a vscode a long time ago. we do not have a vscode native rpm do we ?

i got stuck on 1. have a vscode rpm.

i wasn't willing to do that.

I doubt we have anything native hence why I questioned against a flatpak

I don't think so, just the one from Microsoft I think

Not sure how most Fedora users install VSCode but I imagine a lot of people take it straight from Microsoft

ms thing did not build cleanly. so i ran away.

there are a lot of those things.

It could be a `good first issue` for the person who really wants to scratch that itch later. I think it is wise and sensible to not get stretched out too thin, and focus on the tools where we have the interest and passion

so that's my update. and thats a good lead into the 'stakeholder' topic.

you were braver than me, I never ever approached 🤣

Thanks Tom Rix , let's move on to ...

so the stakeholder topic.

Tom Rix: You start, I can add on top after you?

this is for Tom Rix and also Justin Wheeler

(and all of us here I suppose)

that's not a first issue at all, it would scare the most hardy of contributors away

i want folks to know this meeting is a great place to talk about AI things to be proposed to the public.

OK, not a _first_ issue, but I think you get what I mean. 🙂 Someone who wants to scratch the itch can come and put their hat in the ring. I digress!

well hello fellow fedora peeps

^^context for those who haven't followed it yet

Sathush Rajamanthri: Hi there 👋

this was imo very badly presented to the public, i got a look well after it went out.

so if folks want to try out an idea before it goes live, please bring it up here and in the matrix channel.

i do not think we have the people here to talk about the specific proposal, so we can skip that till it is a topic ?

I have some adds, with some caveats since people aren't here

(Well, not deeply about the specific proposal, but sort of)

Tom Rix: Finished? Didn't want to cut you off

go crzy

It was handled badly in a lot of ways, there were key stakeholders that would be impacted (kernel team) that hadn't heard of it until I asked their opinion on it and were shocked, that's not the way to get buyin

So, obviously, it is not fair for us to _really_ discuss this while Gordon is on PTO. I hope we could bring it up more in detail next time on specific feedback and objections, and also give Gordon a place to "home" this work. We will have to postpone that conversation until he is here.

But this whole thing got me thinking…

And it tied into the topic I brought up weeks or a month or two back, I forget, on understanding the scope of this SIG

There is a short-term thing, which we can suggest to Gordon, and a long-term thing, which we could work on as a group

I think the short-term thing is to pitch to Gordon that the AI/ML SIG be a formal stakeholder in the planned work. Initiative or no Initiative. This feels like the right place to collaborate, and like Tom Rix said, I also want this meeting to be a great place for people to talk about things before going super public.

There were a number of process hiccups (or worse) that happened in this, some of which, I know I am accountable for… I know that with Aoife back, we are thinking through some of this on the Fedora Council side, at least

We could discuss this idea next week with Gordon, once he is back.

The long-term thing is about how we promote ourselves

So, we _should_ be more of a landing place for AI stuff, but I think that message is not clear. In the same way that it was not to me, many weeks ago before I started showing up here

The long-term idea is collaborating, as a SIG, on a longer-term project to get the docs site cleaned up, and move more things out of the Wiki.

The Docs site exists, but it is pretty lacking. I set it up, without a plan or commitment in place on how we use it

But maybe now is a nice time?

Anyways, I think that's all I got for now… curious for thoughts.

Anyways, I think that's all I got for now… curious for thoughts. (Both on the short-term and/or long-term parts.)

maybe we need to hook up some docs ? i add a bit to the wiki this last week.

but also i don't want to be the only one writing the docs.

There's a ticket addressing the docs issue, I'd be happy to contribute to it but with feedback / consensus from the SIG members

there are a lot of BIG things that die without hooking in the day-to-day little things.

this is imo another failure of the ai desktop poposal. the idea is too big that normal folks don't know how to make there thing a little better.

I think there is an initial thing of what we want to publish in the docs, and what makes sense to move from the Wiki versus letting bitrot take its course

I am heavily involved in the Fedora Docs end of things, so I am happy to help out with a reviewer/PR workflow for us

A "Getting involved" page would be excellent, and it would be the perfect place to articulate what Tom Rix said at the outset, to make it obvious and clear that this SIG can be a launching/starting point for big ideas

i would s/big ideas/your idea/

I think another page in the AI/ML SIG docs could even potentially be about the AI Developer Desktop, Initiative or not. I think this fits into the short-term thing of this SIG being a stakeholder on the effort/work in general

I will be honest, I am not peeking at the Wiki too often because (1) the hierarchy of multiple SIGs/WGs is still a bit confusing to me, and (2) it seems mostly focused on packaging, which is great, but not my wheelhouse directly

So, I am wondering, could we action some folks to migrate Wiki pages to AsciiDoc?

I would raise my hand to be a AsciiDoc/Antora reviewer

I would raise my hand to be an AsciiDoc/Antora reviewer

packaging is how these complicated things get enabled in fedora. otherwise its something else. so it is necessary ex/ to get some idea of what things we want to land.

wikis are much easier to edit though, so for things that are WIP or constantly changing, it's easier than putting them in docs. so maybe we need clear links from docs to the wiki depending on the topic

I could potentially work on this, maybe tag-teaming it with Carol or anyone else who wants to help out

we do not have many people working on packaging.

For sure!! Not devaluing the work at all, it is important, I just mean I am less familiar with what folks are maintaining in the Wiki today, content-wise

We _should_ be promoting that work

that limits what ai things are in fedora.

Carol Chen: It is a great question. The Wiki, today, is really meant to be more like a scratchpad for ephemeral things which we use and forget about

I'm not sure if we have regular, ephemeral content like this or not

Tom Rix: Do you think any of the Wiki content focused on packaging could fit into a "packaging" Antora module?

We could potentially even set up a CODEOWNERS thing on Forgejo, so folks who are doing the packaging work are automatically tagged for content changes on packaging stuff

here's were i expose my ignorance, i have no idea what that is.

I could initialize the Antora module, so where to put stuff is more clear

I think a lot of the packaging stuff is transitory stuff that needs to be done

Antora module is basically a directory for specific content, within a larger component (i.e. site)

It is more sophisticated than a directory, but there's only 4 minutes left in this meeting 😄

not so much stuff that goes into docs

Peter Robinson: OK, good to know. I am really leaning on y'all to tell me what makes sense for docs vs. wiki. I just think we should definitely lean into using our docs site more

if we are getting to the end.. how about move docs to a topic for next meeting ?

For Tom basically Antora is the tools that builds the rendered docs

we can sort out the details outside of the meeting. Justin Wheeler count me in to assist you in this (I'm not that knowledgeable about the docs part but I'll follow your lead!)

WFM. This means, next Thursday, we should chat with Gordon about the Initiative proposal, and follow-up on how we organize our content between docs and wiki

Maybe you and I can start on something for the "getting involved" page and pitch it to the team!

2 weeks.

✋️

Oh, right! Two weeks is doable for me and Carol to get a draft of some getting involved docs

Peter Robinson: Go for it

I have a quick topic that I touched on briefly above

New `!topic` or same?

I am working on a bunch of AI bits for the edge. Executorch and the gstreamer ML stack is one of it

I am aiming for a F-45 feature

oops sorry I was opening the floor up for Peter Robinson

we have 80ish devices we support with fully open NPU stack, arm (the company) is sponsoring a bunch of work here

my plan is a bunch of stuff like audio (alexa ish eventually), HomeAssistant and various other style use cases

What is the F45 feature look like? Packaging executorch/gstreamer ML, or something more?

it'll be a fully open stack so it should be nice and not so controversial side of AI

that plus some usable demo apps (still TBD)

piper-tts is in homeassistent. yeah.

arm (the company) has a bunch of internal demos I'm working with the team to get open sourced

so that's just a heads up, can discuss more in future meetings, just wanted to get awareness :)

(slightly off topic, I like how Justin reacts with 💪 everytime arm is mentioned)

Also it'll run on RPi5 and thhere's an announcement there due next week for Fedora ;-)

Arm also makes the kelidiai CPU acceleration library, which is on my "need to package" list, let me know if you want even a partial spec.

I have a partial spec, I need to work with that team to support it as a dynamic library, it's currently only supported statically linked into projects

oh lol, I wasn't even thinking about that, but yeah 🤣

Ahhh, yes, I think that was the blocker. Thank you! It should speed up PyTorch, OpenCV, etc

Peter Robinson: Mind writing an `!info` to summarize it for the logs? Then, I reckon we yield the channel since we are 8 minutes over time

yup. It's on my list to reach out to the team for that

I would guess Carol Chen is ready to sleep soon 😄

wished you'd mentioned that earlier ;-)

Better late than never 😄 It's a good update though!

Nice to see ARM stepping up there, and also sponsoring Flock ❤️

(Just need to get out of paperwork hell for that 😛)

Peter Robinson++ Carol Chen++

!hi

I'm a bit slow with the summary, sorry!

jflory7 gave a cookie to pbrobinson. They now have 119 cookies, 1 of which were obtained in the Fedora 44 release cycle

Conan Kudo 🫩: Neal Gompa (ngompa) - he / him / his

jflory7 has already given cookies to pbrobinson during the F44 timeframe

jflory7 has already given cookies to cybette during the F44 timeframe

All good!

I think we can go ahead and wrap here, then?

Made it just in time for the ending 😄

yep, thanks everyone! I will add the logs and stuff to discussion forum

perfect!

Can you remove the bit about rpi5, it doesn't fit with the NPU bits

Carol Chen: Would you mind (tomorrow) posting the summary to Discussion?

Oh awesome, yay thanks!

that was a separate point

No `!undo` in new Meetbot, but could append `!info`

I have no idea how meetbot works

Lazy but it works 😛

there are no undos implemented yet :(

someone really should do that

Carol Chen: Thanks for chairing today!

will sort it out in the discussion post. ending meeting now. thanks everyone!

Anyone want to do it next time?

👀

(We can move to #ai-ml:fedoraproject.org

I'll be at PyData London in 2 weeks so someone else has to chair

(We can move to #ai-ml:fedoraproject.org)

yeah we'll sort it out there

tflink: should be back.