#metabrainz

YoMo12: yes, sure. if its an exact match though, probably wait till tomorrow because the mapper will likely catch up.

how does poetry handle the case where required libs use dependency pinning and disagree on the exact version?

outsidecontext you can just specify the deps in pyproject.toml without poetry: https://github.com/metabrainz/liblistenbrainz/b...

rdswift[m]: i think that's a good idea, especially adding chatlogs. web searchable chatlogs would be nice.

luc

meh, still getting use to nick completion on matrix

It depends on the degree of disagreement I'd say. If you have for example `<1` and `>1` (so there is not common version), it just fails

lucifer (IRC): Wolfgang responded and suggested that we might start simple and build a network that determined BMP or no BMP.

if there is at least one common match, it will be picked.

gcrkrause: so it essentially ignores the pinning for dependencies and only considers the version ranges?

mayhem[m]: hmm, not sure what that means. it sounds like classfication but that's not exactly what we are doing here.

if you install the package without poetry (eg with any other frontend) the lock will be ignored anyways

BPM > 60 or < 60 would make more sense.

I guess then not commiting the lock file and have it git ignored is the way to go.

so as soon as someone runs `pip install liblistenbrainz` there will be no lock file taken into account

one thing I want to emphasize in this alg is that there are songs that have no BPM (classical, ambient). The alg needs to do a good job of classifying those first, then it can deteremine BPM. that is what he is suggesting.

ah okay, that makes sense.

do we have a dateset of audio files?

outsidecontext But if we ignore everything that poetry introduces as feature, why using it? I mean the dependencies are already managed, they are specified in pyproject.toml and can be updated by renovate if desired

I'd argue its not desired to bump ranges for a lib

so in the end we only want to change the dependencies manually

oh duh. naw, don't worry about it

we will need to think more about [off] in the context of matrix

<11YoMo12> Even different artists and albums that are currently not mapping will be mapped?

if we add the chatlogs thing, we can disable history on matrix.

i mean we have chatlogs anyway so we could disable it anytime.

yomo12: in theory yes.

oh, yeah. we can just keep the chatlogs on IRC. good point.

lucifer: i just checked the message written below the issue LB-1459 about integration test in Websockets Backend.

LB-1459: Add integration tests for websockets backend https://tickets.metabrainz.org/browse/LB-1459

in the longer term, i guess a matrix based BrainzBot makes sense.

agreed.

liblistenbrainz has exactly one dependency, I therefore tend to decide to not use poetry until we really need it

GauravGupta993: hi!

Hi!

do you have the server setup locally?

gcrkrause: I agree

gcrkrause (IRC): then we'll need it for the content-resolver. it has a thicket of requirements.

I'm going to be working towards release versions of troi/content-resolver soon, so that will be great.

GauravGupta993: so for the tests, you would need to add the websockets container from this file https://github.com/metabrainz/listenbrainz-serv... to this file: https://github.com/metabrainz/listenbrainz-serv...

I just need to stop finding bugs and dreaming up new features that I want to implement instantly

Although I'd argue you can still get aways without poetry if you have a ton of requirements. I don't know, at the beginning I was a huge fan of poetry, and it indeed solves a lot of issues for Funkwhale being a application, however there are two major issues. The lock files are not used with other frontends and you cannot customize the build

process

GauravGupta993: then submit a listen to API, see https://github.com/metabrainz/listenbrainz-serv... for examples.

I just verified the liblistenbrainz wheel contains the tests, which is something we have to fix

finally ensure that the listen comes out from the websocket.

lucifer: It will mainly confirm what we know already: doable and more memory-efficient.

hi, I'm unable to see a login form on the login end point at the local host

lucifer: I’d really like to discuss it again once we switched to SolrCloud 9 (which also involve moving deployment to ansible).

GauravGupta993: that part you will have to figure out. use a python socketio client (add the dependency to requirements_dev.txt).

yvanzo: awesome, that sounds great to me. thanks!

yukteax: do you get any error on the webpage or the docker logs?

lucifer: Understood, Thanks for instructions. Till now I have read some articles about how to make integration tests and some information about socketio and checked other already written integration tests in repo. I will start working on issue now.

sounds great, feel free to ask for help if you get stuck.

Ok

[musicbrainz-server] 14mwiencek opened pull request #3150 (03master…event-art-archive): CAA-84: Event Art Archive https://github.com/metabrainz/musicbrainz-serve...

o/lucifer https://usercontent.irccloud-cdn.com/file/Bo30W...

no errors, but no form either, instead it opens this pop up

which doesn't let me log-in

kind of confusing, not sure if this is intended behavior

yukteax: oh looks like the MUSICBRAINZ_CLIENT_ID or MUSICBRAINZ_CLIENT_SECRET is missing from your config.py

thats' odd, I'll reverify if stuff is in order

also, make sure you copied the config.py.sample file to create a new config.py file in the same directory (instead of modifying the existing file)

works now, looks like I exchanged the client id w/ secret. my bad

ah makes sense.

this is a browser behavior that happens if you return a 401 or 403 on a login page so not much we can do to make this clearer.

but i guess we can mention it in server setup docs.

Can we not have a similar ai/ml project as lb radio? Ik we have gone through this before. But I would want to know everything about an artist if I write a prompt without having to click through the UI, mayhem[m] . What is stopping us from thinking in this direction? Just curious nothing more

Can’t we just fine tune an existing LLM over the MusicBrainz, ListenBrainz, CB db to achieve this?

resources. motivation, #fuckai. training data.

what exactly does "fine tune an existing LLM over the MusicBrainz, ListenBrainz, CB" mean?

you can't just "AI" something. that's what the AI bros do in order to get $M.

Btw

we in reality need to take concrete features we wish to implement (that we can't implement in a simpler way) and then see if we have the data with which to train an AI.

that said, what data do you want to use for making something that already exists, is almost complete and soon will be scalable?

For LB radio, I think it does make sense to have a better UI. Maybe a LLM to parse prompt to json or like html select options.

It will sell well if it's the former fwiw.

we need to improve the parser, yes. but there is no AI needed.

akshaaatt (IRC): read the documentation on how to write a query, please.

remember, this is not an end-user too. this is intended for developers, so that we learn how to design features for the website that a nice to use.

fair point yes.

artist:(linking park)

is the query you want.

I think the only valid suggestions of ML application I heard today is to BPM and mapping. the rest is useless.

I wonder if someone builds a wrapper around LB radio and an api call to ChatGPT and spits out Spotify playlists, can it get millions in vc funding? 🤔🤔

what are you waiting for lucifer?

but, I bet VCs would fall for it.

Not good at lying especially not with a straight face lol

Best thing I read today

🤣🤣

hard to argue

though the tyres still pollute

mayhem[m]: I am thinking if we can use the LLM to spit out the exact information for us without having to touch the GUI at all. If our fine tuned model understands our schema, and the relevant actions present, we could replace the gui entirely. The user could prompt to retrieve the relevant data. We could even use it to fix the data present on MB

yes, sounds nice. who is going to build it? with whose resources?

It shouldn’t be that hard technically because models like LLAMA are already open sourced. We need to think about just fine tuning it with our data which needs to be properly documented. Sure it requires effort and compute but could be a one time thing

I am not commenting from the perspective of saying that we should do it

But it is definitely something worth it and will bring us a lot of traction because the market is headed that way

do you think having a tool built with AI vs one with postgres will have more market traction, given that they are equal?

users dont give a fuck about how you made something. us geeks, yes. but dont project geek values on your proposed target market.

I don’t think the postgres one is equal. The AI one can infer what I want. The postgres one will always be deterministic

not that the postgres one is deterministic. I worked hard on it being NOT that.

very liberal application of random() lol

have you ever done a machine learning project, akshaaatt (IRC) ?

Small projects. But I have worked on applied ml

methinks you understimate the level of effort required for building data sets for ML.

the BPM dataset is going to be a bear. I need to have 10k tracks for which I am sure of the BPM. how I am going to do that, is still out for determination.

doing that for LB-radio. 🤯

True

zas: I found a bug in the scan function for the content resolver. If you hit ^C during a scan, it will NEVER complete the scan. The following scan will find nothing new.

https://arxiv.org/pdf/2307.16789.pdf mayhem[m]

the team that wrote that is TWICE the size of our team. what are you trying to tell me, akshaaatt (IRC) ?

Hahaha true

mayhem[m]: fair enough re removing that UX project But I'm curious what's meant by not having the underlying data? Maybe I've underestimated the difficulty of grabbing social media links from MB

link to ticket/idea aerozol?

lucifer: https://tickets.metabrainz.org/browse/MBS-13444

MBS-13444: ShareBrainz - MB/LB smart links service

Just lost power at home. Hang on

Im aware I was stretching it by putting it into the gsoc page though, but it was so empty, so I thought why not 😁

But really a LB project anyway tbh

*not

Check behind the couch mayhem

yes, we have this data easily available. might have been a misunderstanding.

Also, holy crap there's a lot of messages since yesterday. Usually I try skim everything but maybe not today... Nice to see it bustling!

Ok thanks lucifer, wasn't sure if I'd missed something

the UI heavy part can be an issue.

For sure, I think it's a good idea to leave it off gsoc if there's other things to work on

how/where would you show these cards?

Exactly as in the mock-up - it is it's own page (or even a new project website). Have you seen artists sharing links to link.tree and similar sites? They are very popular. But artists make them, they are not crowd sources like us

yup i have seen those.

I think we would have an edge. But yeah it is a (almost entirely?) UI project

i see

no UI projects?

the UI in this case looks very simple (as in even i can build it) so i do wonder if an exception can be made but then these always look simple in the start and then end up being "difficult"

jasje: yup, no projects involving significant ui/ux work.

One thing I thought was nice is that it's definitely less complicated than our other tickets, in that someone can make something from 'scratch' without the usual 5 million links to MB/LBetc code

Ok, I see the difference. The forum post is more ambitious than aerozol's idea

yes of course

its fairly isolated

Spot on Lucifer. Maybe if the right person came along but yeah UI is not why GSoC students come to us

I was going to ask that MB revamp could come up as Gsoc proj but you ppl don't put it. Why is that?

didn't*

'project: revamp all of mb'