<@jflory7:fedora.im>
16:30:03
!startmeeting Fedora AI/ML SIG - 2026-07-16
<@meetbot:fedora.im>
16:30:04
Meeting started at 2026-07-16 16:30:03 UTC
<@meetbot:fedora.im>
16:30:04
The Meeting name is 'Fedora AI/ML SIG - 2026-07-16'
<@jflory7:fedora.im>
16:30:09
!topic Roll call & hello's
<@jflory7:fedora.im>
16:30:11
!hi
<@zodbot:fedora.im>
16:30:12
Justin Wheeler: Justin Wheeler (jflory7) - he / him / his
<@jflory7:fedora.im>
16:30:36
!group members ai-ml-sig
<@zodbot:fedora.im>
16:30:38
Members of ai-ml-sig: bstinson, Christiano Anderson, David Duncan, Davide Cavalca, dhellmann, Jason Montleon, Jonathan Steffan, kaitlynabdo, Mohammadreza Hendiani, markmc, rdevasth, rishi, Michel Lind, Tim Flink, thunderbirdtr, trix, Alexander Lent
<@kph1992:matrix.org>
16:30:58
hello. first time joining here.
<@trix:fedora.im>
16:31:02
!hi
<@zodbot:fedora.im>
16:31:03
Tom Rix: Tom Rix (trix)
<@gordonmessmer:fedora.im>
16:31:25
!hi
<@zodbot:fedora.im>
16:31:26
Gordon Messmer: Gordon Messmer (gordonmessmer)
<@davdunc:fedora.im>
16:31:28
Kevin Hannon: me too.
<@jflory7:fedora.im>
16:31:33
Hey there Kevin Hannon, welcome! Nice to have you here. You can use `!hi` for zodbot to look up your username in FAS, and it is our convention for showing up at a meeting 🙂
<@kph1992:matrix.org>
16:31:42
!hi
<@jflory7:fedora.im>
16:31:43
davdunc: Oh hey, yeah, I haven't seen you here yet, have I? 😄
<@zodbot:fedora.im>
16:31:44
No Fedora Accounts users have the @kph1992:matrix.org Matrix Account defined
<@davdunc:fedora.im>
16:31:56
not in a long time.
<@tflink:fedora.im>
16:32:05
!hi
<@zodbot:fedora.im>
16:32:05
tflink: Tim Flink (tflink)
<@jflory7:fedora.im>
16:32:06
Kevin Hannon: Oh heheh, I forgot! You need to add your Matrix ID to https://accounts.fedoraproject.org for zodbot to "recognize" you.
<@davdunc:fedora.im>
16:32:08
things have changed since then.
<@jflory7:fedora.im>
16:32:33
davdunc: Obviously! I saw your LinkedIn post and attended the FRCL call. I'm excited to have you in any capacity we can have you 😄
<@jflory7:fedora.im>
16:32:47
As a reminder for new and old folks, today's meeting agenda can be found here:
<@jflory7:fedora.im>
16:32:49
!link https://board.net/p/fedora-aiml-sig-meeting-agenda
<@davdunc:fedora.im>
16:32:56
appreciate that. I'll be more Fedora forward from now on. :D
<@jflory7:fedora.im>
16:33:03
We currently have four topics triaged.
<@kph1992:matrix.org>
16:33:15
!hi
<@jflory7:fedora.im>
16:33:18
Anyone else have a last-minute thing they want to throw out there?
<@zodbot:fedora.im>
16:33:18
No Fedora Accounts users have the @kph1992:matrix.org Matrix Account defined
<@trix:fedora.im>
16:33:29
do we have quorum ?
<@kph1992:matrix.org>
16:33:40
okay. I did add to my matrix account but maybe it takes time to sync
<@jflory7:fedora.im>
16:33:48
Kevin Hannon, davdunc: Oh, yeah! And since you are both new, please introduce yourselves! Tell us a sentence or two about you, why you are here / what you are curious about, etc.
<@kph1992:matrix.org>
16:33:54
okay. I did add to my matrix to my fedora account but maybe it takes time to sync
<@davdunc:fedora.im>
16:34:09
yes.
<@jflory7:fedora.im>
16:34:19
Tom Rix: I think so. You, me, tflink, Gordon Messmer, and we have Kevin Hannon and davdunc here as newcomers.
<@jflory7:fedora.im>
16:34:52
I'll wait for Kevin and David to post a short intro, and then we can get into the agenda topics promptly
<@kph1992:matrix.org>
16:34:57
I'm Kevin and I mostly work on Kubernetes. I've been playing around with llama.cpp and pi. I'm interested in learning more on helping out here as I like fedora.
<@davdunc:fedora.im>
16:35:22
I am David Duncan, I work on the Cloud SIG primarily, but have packaged much software and I am very interested in building images that are specific to cloud hardware and workloads.
<@jflory7:fedora.im>
16:35:35
Kevin Hannon: Hah, it is almost like you pre-empted Tom Rix's two topics for today 🙂
<@jflory7:fedora.im>
16:36:11
Welcome to both of you, Kevin Hannon and davdunc! After the meeting, definitely check out our ticket tracker and drop a comment or ask questions in #ai-ml:fedoraproject.org if anything sticks out to you.
<@jflory7:fedora.im>
16:36:14
With that, I think we can proceed
<@jflory7:fedora.im>
16:36:34
!topic llama.cpp F46 (@trix)
<@jflory7:fedora.im>
16:36:38
!info @trix is looking for people interested in maintaining new llama.cpp backends for F46, with priority on vulkan to keep vulkan support in ollama. CI for the unstable backend would also help.
<@jflory7:fedora.im>
16:36:45
!halp Maintainers needed for new llama.cpp backends, priority on vulkan. Also looking for CI setup help.
<@trix:fedora.im>
16:36:55
ok. some forward looking llama.cpp will need some help in f46.
<@jflory7:fedora.im>
16:37:20
Tom Rix: Just curious, have you already posted to devel@ or anywhere else about the maintainer help?
<@jflory7:fedora.im>
16:37:49
I was wondering if there is any sense of urgency, if is this blocking F46 inclusion, or more of a general concern? Happy to help amplify the call if that would help
<@jflory7:fedora.im>
16:37:59
I was wondering if there is any sense of urgency, if this is blocking F46 inclusion, or more of a general concern? Happy to help amplify the call if that would help
<@trix:fedora.im>
16:38:21
it is an outreach to folks that don't see their favorite backend on the list of backends.
<@trix:fedora.im>
16:38:30
or their favorite cpu.
<@trix:fedora.im>
16:38:59
keep in mind, i work at AMD, so most things are only really work on AMD things.
<@kph1992:matrix.org>
16:39:01
would there be a package like llama-cpp-vulcan, llama-cpp-cpu?
<@jflory7:fedora.im>
16:39:08
I think that is fair 🙂
<@trix:fedora.im>
16:39:22
ex/ i have no powerpc hw, nor nvidia, nor intel gpu.
<@trix:fedora.im>
16:39:29
or s390..
<@kph1992:matrix.org>
16:39:31
as of now its llama-cpp only works on amd gpu
<@tflink:fedora.im>
16:39:42
That is one way to do things and how debian/ubuntu do it IIRC
<@jflory7:fedora.im>
16:40:27
Is there a standard way of doing this, and if not, should there be in Fedora?
<@trix:fedora.im>
16:40:30
so ex/ testing for F45 will be AMD forcused unless some other want to help. this will cover some common case liek cpu only.
<@trix:fedora.im>
16:40:36
but not well.
<@trix:fedora.im>
16:41:07
so poke me offline if you have time and interest.
<@kph1992:matrix.org>
16:41:33
I'd ❤️ to have GPUs but as of now I am mostly doing cpu testing. Vulcan / BLAS / CPU I have familarity with building. But I won't be able to help with Vulcan on GPUs.
<@davdunc:fedora.im>
16:42:33
Tom Rix: is there functional testing that can be handled by just running AMD instances (no GPU)?
<@tflink:fedora.im>
16:42:42
which part? I think it's usually left up to the packagers
<@davdunc:fedora.im>
16:42:48
b/c those are available in the QE
<@jflory7:fedora.im>
16:43:05
!info Anyone interested in packaging new backends for llama.cpp should reach out to Tom Rix for more info.
<@jflory7:fedora.im>
16:43:12
!meetingname fedora-ai-ml-sig
<@meetbot:fedora.im>
16:43:13
The Meeting Name is now fedora-ai-ml-sig
<@jflory7:fedora.im>
16:43:20
(forgot to do that to fix the grouping with past meetings)
<@kph1992:matrix.org>
16:44:06
Tom Rix: do you have a spec filed defined somewhere so I could see how that is done for llama.cpp
<@jflory7:fedora.im>
16:44:20
I am thinking of the sub-packages for different hardware enablement packages/drivers. I might be out of my league a bit here, but I guess the thought I was thinking lately was whether the Fedora Packaging Guidelines should say much or anything about AI-related packages. I am not sure, but feels like something this group would be the right group to do
<@trix:fedora.im>
16:44:48
cpu llm is extreemly slow. in theory yes, something could be done, in practice even a dinky apu is better.
<@davdunc:fedora.im>
16:44:56
packager and QE can be distinct.
<@jflory7:fedora.im>
16:45:19
I guess the topic of FPG is not strictly on-topic for this llama.cpp for F46 topic, but it is adjacent I guess to getting more people onboarded and doing consistent standards for AI-related packages, if a need exists for that
<@tflink:fedora.im>
16:46:14
I'm unclear on what AI-related packages would have to do with this unless we're using an agent to do the packaging work. The idea of having multiple backends for one setup isn't unique to AI
<@tflink:fedora.im>
16:46:31
but we are diving down a rat hole a bit
<@trix:fedora.im>
16:46:40
there is overlap maybe, a ci is needed i think to keep llama.cpp actually stable. ex/ we had issue with vulkan not working, but now maybe it is..
<@trix:fedora.im>
16:47:07
fedora and rhel have never done a good job on multiple gpu backends.
<@trix:fedora.im>
16:47:39
yes, rathole. we can move on.
<@jflory7:fedora.im>
16:48:24
!idea Perhaps later, revisiting the idea of additions to Fedora Packaging Guidelines for AI. Will revisit at a later time.
<@jflory7:fedora.im>
16:48:33
!topic pi-coding-agent in F45/rawhide (@trix)
<@jflory7:fedora.im>
16:48:37
!info pi-coding-agent is now in F45/rawhide. @trix is looking for people interested in maintaining and testing non-x86 arches (priority on aarch64), and for folks to hook up other local LLMs and add to the wiki (priority on llama.cpp or lemonade).
<@jflory7:fedora.im>
16:48:43
!halp Testers and maintainers needed for pi-coding-agent on non-x86 arches, especially aarch64. Also looking for contributors to integrate other local LLMs.
<@trix:fedora.im>
16:48:45
an fyi, pi-coding-agent is in fedora now. so we have at least some agentic coding going in f45
<@jflory7:fedora.im>
16:48:52
Tom Rix: Over to you again
<@zodbot:fedora.im>
16:49:00
jflory7 has already given cookies to trix during the F44 timeframe
<@trix:fedora.im>
16:49:57
this opens up folks who want to add their favorite agent. atm i am look at how to add skill packages to pi.
<@trix:fedora.im>
16:50:03
that is f46 things.
<@trix:fedora.im>
16:50:26
thanks Gordon Messmer for the help reviewing
<@jflory7:fedora.im>
16:50:47
Gordon Messmer++
<@zodbot:fedora.im>
16:50:48
jflory7 gave a cookie to gordonmessmer. They now have 2 cookies, 1 of which were obtained in the Fedora 44 release cycle
<@trix:fedora.im>
16:51:09
other fyi. lemonade-server landed, folks could give that a try.
<@jflory7:fedora.im>
16:51:21
Very curious about that. Skills packages meaning what though, like Agent Skills with `agr`/`lola` or something else?
<@trix:fedora.im>
16:51:32
i think that works with npu which i don't cover.
<@tflink:fedora.im>
16:51:37
rawhide only, right? are there plans to release it on older Fedoras?
<@trix:fedora.im>
16:52:06
we can poke the maintainer.
<@trix:fedora.im>
16:52:39
a problem we have is blazing fast ai vs old releases. i don't really look back.
<@jflory7:fedora.im>
16:53:06
I wonder if Hummingbird makes this any better or easier or not
<@kph1992:matrix.org>
16:53:11
I play around with Pi and llama.cpp so I'm happy to at least write something in wiki. I can't use the llama-cpp package from fedora but I can at least test out rawhide pi-coding-agent.
<@trix:fedora.im>
16:53:34
here is skill i am looking at now . https://pi.dev/packages/@gaodes/pi-graphify
<@zodbot:fedora.im>
16:53:36
No Fedora Accounts users have the @kph1992:matrix.org Matrix Account defined
<@tflink:fedora.im>
16:53:43
out of curiosity, why can't you use the package?
<@jflory7:fedora.im>
16:54:03
Kevin Hannon: Docs are always useful if you are willing to write them. BTW, what is your FAS username?
<@trix:fedora.im>
16:54:54
graphify may make grok-ing big code bases like linux kernel doable.
<@kph1992:matrix.org>
16:54:59
tflink: atm its amd gpu only and won't run on cpu. We would need other rpm packages for llama-cpp for cpu or vulcan. But I don't think that impacts docs at all.
<@jflory7:fedora.im>
16:55:01
Whoa, this looks cool, I'd love to try this with data dumps of the 2024-2026 Flock post-event surveys 😛
<@jflory7:fedora.im>
16:55:21
That would also be super cool
<@tflink:fedora.im>
16:56:06
it works on cpu, doesn't it? it just downloads all of rocm with it
<@jflory7:fedora.im>
16:56:09
Kevin Hannon: Would you be willing to take an action item to take a first go at writing some `pi-coding-agent` docs in the AI/ML SIG Fedora Wiki page?
<@trix:fedora.im>
16:56:18
llama.cpp has cpu backends for x86 and aarch64
<@kph1992:matrix.org>
16:57:02
hmm. when I tried it I got a failure because it couldn't find a gpu..
<@trix:fedora.im>
16:57:21
https://fedoraproject.org/wiki/SIGs/AI-ML#pi-coding-agent
<@jflory7:fedora.im>
16:57:36
OK, not a first go 🙂
<@jflory7:fedora.im>
16:57:41
!link https://fedoraproject.org/wiki/SIGs/AI-ML#pi-coding-agent
<@trix:fedora.im>
16:58:01
i have the hookup to ollama. i am looking for other folks hook ups
<@trix:fedora.im>
16:58:21
open a bz and poke me.
<@kph1992:matrix.org>
16:58:47
Sure Justin Wheeler . I think my FAM is kannon92
<@jflory7:fedora.im>
16:59:06
kannon92++
<@zodbot:fedora.im>
16:59:08
jflory7 gave a cookie to kannon92. They now have 1 cookie, 1 of which was obtained in the Fedora 44 release cycle
<@trix:fedora.im>
16:59:36
those are my fyi's .. we can move on if we want.
<@jflory7:fedora.im>
16:59:38
!action @kannon92 Review and edit the existing `pi-coding-agent` docs on the AI/ML SIG Fedora Wiki page, and coordinate in #ai-ml:fedoraproject.org on changes as needed.
<@jflory7:fedora.im>
16:59:42
Tom Rix: Super, thanks.
<@jflory7:fedora.im>
17:00:11
Kevin Hannon: Thanks also for volunteering to help out. You can always ping in #ai-ml:fedoraproject.org if you have any questions. For example, I think I need to help you out with permissions so you can actually edit the wiki, unless you are in other FAS groups already
<@jflory7:fedora.im>
17:00:23
Anyways, we can move on, and discuss more in open floor or the team chat room
<@jflory7:fedora.im>
17:00:39
!topic #31: Establish a Skills Reviewers sub-team for the skills-library (@jwheel)
<@kph1992:matrix.org>
17:00:39
I edited it last week on a minor issue so I think I have permissions
<@jflory7:fedora.im>
17:00:43
!link https://forge.fedoraproject.org/ai-ml/tickets/issues/31
<@jflory7:fedora.im>
17:00:53
OK, then perfect! Never mind then 😄
<@jflory7:fedora.im>
17:01:09
!info Since we discussed #31 two weeks ago, five people raised their hands to join: @cverna, @davdunc, @lbarczio, @cybette, and @ttomecek. @lbarczio also volunteered to draft a human reviewer guide and proposed packaging-focused skills like scratch build fixing and patch backporting. @cverna opened a CI/linting discussion in `ai-ml/skills-library#13`.
<@jflory7:fedora.im>
17:01:20
I would like to call a vote to create the skills-reviewers team on Forgejo, seed it with these six members (myself included), and post an open call on Matrix and Discourse for anyone else who wants to join. The team would get write access to `ai-ml/skills-library` with a lightweight review process. To start, at least one reviewer approval before merge. Does anyone object?
<@jflory7:fedora.im>
17:03:32
davdunc: I realize you are the only one present of all those people 🙂
<@davdunc:fedora.im>
17:03:36
:)
<@davdunc:fedora.im>
17:03:42
I am all for it.
<@jflory7:fedora.im>
17:03:57
The Europeans are likely having dinner now 😛
<@davdunc:fedora.im>
17:04:03
let me know how I can build the support for my friends.
<@davdunc:fedora.im>
17:04:29
I'll propose what I can.
<@jflory7:fedora.im>
17:04:30
Anyone else have any input? I am curious to align this with what you are exploring Tom Rix, but maybe that is a little more wide than just Fedora specifically
<@jflory7:fedora.im>
17:05:16
davdunc: I think if we can support @lbarczio on drafting some reviewer guidelines, this is a great first step. Otherwise, either come up with some new skills yourself, or help us review anything that pops up in the meantime. I think a lot of this, we are simply going to figure out as we go about what works well or not.
<@trix:fedora.im>
17:05:41
i am not sure. there is A LOT of energy with folks doing skills.
<@jflory7:fedora.im>
17:05:55
Generally, or in Fedora?
<@trix:fedora.im>
17:06:20
generally.
<@davdunc:fedora.im>
17:06:30
Justin Wheeler: I'll start there.
<@jflory7:fedora.im>
17:06:56
Well, as you are exploring, if there are overlaps with anything you are seeing and building common skills for Fedora contributions, perhaps with any AI dev work in Fedora, it could be interesting.
<@jflory7:fedora.im>
17:07:16
But otherwise, if no objections and since the ticket is open for a while, I am going to take this as assent to move forward and proceed with the plan
<@jflory7:fedora.im>
17:08:07
!agreed We will create the skills-reviewers team on Forgejo, seed it with @jflory7, @cverna, @davdunc, @lbarczio, @cybette, and @ttomecek, and post an open call on Matrix and Discourse for anyone else who wants to join. The team would get write access to `ai-ml/skills-library` with a lightweight review process. To start, at least one reviewer approval before merge.
<@trix:fedora.im>
17:08:27
here is what i mean. https://bugzilla.redhat.com/show_bug.cgi?id=2468413
<@jflory7:fedora.im>
17:08:41
!action @jflory7 Create skills-reviewer team, then close issue #31. Continue planning/working on Agent Skills in the `ai-ml/skills-library` repo.
<@trix:fedora.im>
17:09:09
likely pi has something similar, i just need some time to look.
<@jflory7:fedora.im>
17:09:15
Oh wow. This is something new for me to dig into 👀
<@jflory7:fedora.im>
17:09:23
!link https://bugzilla.redhat.com/show_bug.cgi?id=2468413
<@jflory7:fedora.im>
17:09:35
I guess that's all I have then for this topic.
<@trix:fedora.im>
17:09:58
clawhub can query a repo of skills, you can see there are rpmbuild related skill already.
<@jflory7:fedora.im>
17:10:24
I think I'll have to look at this, even though I have been avoiding OpenClaw in the priority of "AI things I need to spend time learning"
<@trix:fedora.im>
17:10:25
so we don't need to reinvent the wheel, its open source.
<@jflory7:fedora.im>
17:10:57
I have been looking into `agr` a lot lately, especially since it is already packaged in Fedora
<@jflory7:fedora.im>
17:10:59
!link https://github.com/computerlovetech/agr
<@jflory7:fedora.im>
17:11:07
Surely it is one of you, I guess? 🙂
<@davdunc:fedora.im>
17:11:27
Tom Rix: what you are describing is similar to what we do with packages like cloud-init where we make our own modifications to support the config locations specific to a fedora installation.
<@jflory7:fedora.im>
17:11:30
Oh, it is Ankur Sinha (FranciscoD@fedora) (he/him/his)!
<@jflory7:fedora.im>
17:12:44
We are getting into the last 18 minutes. I'm going to go ahead and press on.
<@jflory7:fedora.im>
17:13:01
Anyone curious about Fedora Agent Skills, come chat in #ai-ml:fedoraproject.org 🙂
<@jflory7:fedora.im>
17:13:07
!topic #34: Scope gpu01 access to ai-packagers-sig and open up ai-ml-sig membership (@jwheel)
<@jflory7:fedora.im>
17:13:12
!info I have edited #34 significantly since the last meeting. The title and body are different from what was there before. Please re-read if you looked at it earlier.
<@jflory7:fedora.im>
17:13:34
I am not looking for a decision today. I would appreciate feedback, especially from tflink, either here or async on the ticket or in our Matrix room. I also understand if people need time to read through the updated text first
<@tflink:fedora.im>
17:14:12
both Tom Rix and I commented on the ticket saying that we think it's too soon to be having too many conversations about this
<@trix:fedora.im>
17:14:19
my feedback i left in the ticket, it is too early to do this.
<@tflink:fedora.im>
17:14:25
there are just too many moving parts right now
<@jflory7:fedora.im>
17:14:31
Oh hah, I had to refresh
<@jflory7:fedora.im>
17:14:34
I didn't see the comments yet
<@jflory7:fedora.im>
17:15:26
OK, that is fair. But we did get a ticket like #33 with a general request for access. The reporter ended up closing it on their own, but I think we should have a standard idea of how to respond or answer questions about this access.
<@trix:fedora.im>
17:16:17
for that one, i reminded the requestor that the use is for ci.
<@jflory7:fedora.im>
17:16:24
It seems like the idea of GPU access/infra being available has been pinned onto the SIG, but that might because anything else is unclear. I want to make sure that if the machines are intended for CI purposes in the F45 release cycle, we are making that clear for everyone to avoid misunderstandings. I also don't think there is anything controversial at all about earmarking the existing infra for CI
<@jflory7:fedora.im>
17:16:48
Yep, but I think this is not clear. I am not sure of the best way to make it more clear until there are fewer moving parts
<@jflory7:fedora.im>
17:16:53
I know a lot of this has to do with Dell…
<@davdunc:fedora.im>
17:17:02
maybe we need to define our requests for hardware based on requests/features?
<@tflink:fedora.im>
17:17:29
I'm still not sure what all can and can't be said about what is going on
<@trix:fedora.im>
17:17:49
i do not think the basics of owns the machine are clear or even if the machine is a loaner or not.
<@jflory7:fedora.im>
17:17:53
In the long-term, this would be the ideal. The challenge we have now is, there is a lot of stuff in flight, some hardware was brought in with specific intentions for use, and we are stuck waiting for Dell to give us hardware we need to mount the GPUs that Fedora Infrastructure currently possesses
<@jflory7:fedora.im>
17:18:16
Sure. I don't want to put anyone on the spot, and there is no need to give all the context here and now
<@trix:fedora.im>
17:18:37
it would be good to at least let the ci team have a say, since ci.
<@tflink:fedora.im>
17:18:47
it seems like we should know more soon, though. hopefully in the next few weeks
<@jflory7:fedora.im>
17:18:51
We could simply say, the hardware is in a transitionary/early state as we figure out how it all fits into Fedora Infra and whether we can make it available for non-CI use cases in the future?
<@jflory7:fedora.im>
17:19:07
Just so people don't keep coming with ideas or plans about how to use the GPU 🙂
<@tflink:fedora.im>
17:19:14
that makes sense and is a decent summary
<@davdunc:fedora.im>
17:19:28
tflink: I always default to "less corporate-sided information" unless I am the second person to say it.
<@jflory7:fedora.im>
17:19:40
Yes, true. I forget that Fedora Quality and Fedora CI are technically different teams/people.
<@jflory7:fedora.im>
17:19:59
Luckily, Cristian Le is a pretty easy person to track down, so maybe we can ask him for input in the team meeting room later
<@jflory7:fedora.im>
17:20:06
Err, just team room
<@jflory7:fedora.im>
17:20:08
Not team meeting room
<@gordonmessmer:fedora.im>
17:20:26
I think the GPU system(s) need an acceptable use policy regardless of how jobs are run. "CI" is just a mechanism to launch batch jobs. The batch jobs can do pretty much anything. That is to say that "CI" is more of a mechanism than a purpose, unless we provide a definition that includes acceptable purposes.
<@jflory7:fedora.im>
17:21:00
If you feel confident that clarity is imminent, we can wait out and use the "transitionary/exploratory" explanation for anything that comes up before then. But if it gets into >30 days, I think we should write this down in Docs somewhere so it is explicit
<@trix:fedora.im>
17:21:31
this is also about trust. if the machine is in a high area of trust. then random folks shouldnt use it.
<@jflory7:fedora.im>
17:21:47
We could discuss what the logistics of what this looks like, even in the abstract. For example, I am REALLY curious about how we might consider scheduling and resource sharing. It reminds me of work I did years back in HPC and it was one of the things that always fascinated me
<@trix:fedora.im>
17:21:51
i include myself as a random person.
<@jflory7:fedora.im>
17:22:16
Yes, true. This is why I made the pitch for `ai-packagers-sig` as the FAS group to link to hardware access. It also makes adding people to `ai-ml-sig` as members less of a privileged thing
<@jflory7:fedora.im>
17:22:45
We should have a couple rings of trust where people can contribute and do things. We don't want to give someone free reign to GPUs on Day 0, but figuring out how someone can work up to that trust is important
<@tflink:fedora.im>
17:22:59
I can only go on what I've been told but I'm hoping that it's imminent
<@trix:fedora.im>
17:22:59
my ai test is an agent that breaks into networks.. let's not do that.
<@jflory7:fedora.im>
17:23:36
Then let's put a pin in writing anything officially until the August 13th meeting, perhaps?
<@tflink:fedora.im>
17:23:57
sure, hopefully things will be more clear by then
<@jflory7:fedora.im>
17:24:06
And we can "verbally" agree to this "transitionary/exploration phase" as the reason for why GPU access is not more open than it is now
<@jflory7:fedora.im>
17:24:16
Do ^^both of these things sound fine as short-term resolutions?
<@jflory7:fedora.im>
17:24:43
I agree with Gordon Messmer that an acceptable use policy is likely something for us to work on, and we could begin sketching out what that looks like even without hardware
<@tflink:fedora.im>
17:24:44
well, that and general confusion but we can probably leave it as the exploration phase part
<@jflory7:fedora.im>
17:24:46
But I think that is a new ticket
<@jflory7:fedora.im>
17:24:57
The wild, wild, west
<@trix:fedora.im>
17:25:22
no. take open access off the table.
<@jflory7:fedora.im>
17:25:49
I don't think any part of our infra is open access beyond read-only Infrastructure Apprentice roles
<@jflory7:fedora.im>
17:26:02
I think the acceptable use is how we define what access and by whom is appropriate
<@jflory7:fedora.im>
17:26:11
But this is a long-term conversation, and we are running against a clock 🙂
<@trix:fedora.im>
17:26:36
come back when things are clearer.
<@jflory7:fedora.im>
17:26:44
Perhaps we can use ticket #34 to track the specifics of this hardware we already have, and I could open a new ticket to discuss an AUP generally for Fedora GPUs, assuming we pass this to Fedora Infrastructure at some later time
<@gordonmessmer:fedora.im>
17:26:49
Lightning storm in Seattle just took down my main Internet connection.
<@tflink:fedora.im>
17:27:00
there have been packager-checkout-able systems in the past for more exotic hardware like s390 and ppc so it's not like this would be completely new ground
<@jflory7:fedora.im>
17:27:11
Oof. Take care! 🙏
<@trix:fedora.im>
17:27:16
hinting we have 3 minutes left..
<@jflory7:fedora.im>
17:27:26
This is a new area to me, I'd love to know better at what ground has already been tread
<@jflory7:fedora.im>
17:27:41
Yes, indeed. Going to minute-ize the discussion, then we can start wrapping up
<@jflory7:fedora.im>
17:29:14
!agreed In the short-term, as far as existing/available GPU infrastructure is concerned, the immediate focus for the available hardware is for CI purposes in the F45 release cycle. We commit to following up on the August 13th SIG meeting if there is no new clarity to the hardware access to write something officially. For now, we agree for any requests that come up in the meantime, we will close as `wontfix`, explaining that the GPU hardware is in a "transitionary/exploration" phase while we discover what acceptable use looks like for the Fedora community.
<@jflory7:fedora.im>
17:29:21
I think that covers it mostly
<@jflory7:fedora.im>
17:29:51
!idea We will need to think up an Acceptable Use Policy for GPUs, generally however. This is not dependent on having any actual hardware yet. This should be a new ticket.
<@jflory7:fedora.im>
17:30:03
!action @jflory7 Open a new ticket for Acceptable Use Policy discussion, likely targeting Q4
<@jflory7:fedora.im>
17:30:07
!topic Open floor
<@jflory7:fedora.im>
17:30:13
OK, we are right at the meeting end time 🙂
<@jflory7:fedora.im>
17:30:20
Anything urgent or timely? Or shall we wrap?
<@tflink:fedora.im>
17:30:31
nothing from me
<@jflory7:fedora.im>
17:30:31
!action @jflory7 Post a meeting summary of today's meeting on https://discussion.fedoraproject.org
<@jflory7:fedora.im>
17:30:46
Alrighty then, let's do it. Thanks folks for being here and joining in!
<@jflory7:fedora.im>
17:30:57
Kevin Hannon, davdunc: And thanks for joining us!
<@jflory7:fedora.im>
17:31:00
See y'all next time.
<@jflory7:fedora.im>
17:31:02
!endmeeting