[04:20:05] 10Lift-Wing, 10Machine-Learning-Team, 10I18n, 10NewFunctionality-Worktype, 10Patch-For-Review: Create a language detection service in LiftWing - https://phabricator.wikimedia.org/T340507 (10santhosh) > @santhosh is this model service going to be used by an existing service? To start with, the frontend o... [06:11:58] (03CR) 10Santhosh: "I see that CI is failing to build the fasttext wheel. But that is surprising since debian bullseye, python3.9 is sufficient for building i" [machinelearning/liftwing/inference-services] - 10https://gerrit.wikimedia.org/r/932828 (https://phabricator.wikimedia.org/T340507) (owner: 10Santhosh) [07:32:09] 10Machine-Learning-Team, 10Gerrit-Privilege-Requests: Grant ML Team members +2 rights to the recommendation-api repository - https://phabricator.wikimedia.org/T340531 (10kevinbazira) @Aklapper, thank you for the ping. I have linked my SUL staff account and it's now visible on my Phabricator profile. [07:45:51] 10Machine-Learning-Team, 10Gerrit-Privilege-Requests: Grant ML Team members +2 rights to the recommendation-api repository - https://phabricator.wikimedia.org/T340531 (10kevinbazira) 05Resolved→03Open @thcipriani thank you for helping with this. It looks like we were added to the recommendation-api [[ http... [08:23:57] (03CR) 10Elukey: Add language identification service (031 comment) [machinelearning/liftwing/inference-services] - 10https://gerrit.wikimedia.org/r/932828 (https://phabricator.wikimedia.org/T340507) (owner: 10Santhosh) [08:27:06] (03CR) 10Elukey: Add language identification service (031 comment) [machinelearning/liftwing/inference-services] - 10https://gerrit.wikimedia.org/r/932828 (https://phabricator.wikimedia.org/T340507) (owner: 10Santhosh) [08:29:30] (03CR) 10Ilias Sarantopoulos: Add language identification service (031 comment) [machinelearning/liftwing/inference-services] - 10https://gerrit.wikimedia.org/r/932828 (https://phabricator.wikimedia.org/T340507) (owner: 10Santhosh) [08:31:02] elukey: o/ regarding the above I think fasttext needs python3-dev debian package.. [08:31:08] I m testing it now [08:31:12] and g++ ofc [08:42:45] going afk for 20-30' [08:57:00] (03PS7) 10Elukey: Add language identification service [machinelearning/liftwing/inference-services] - 10https://gerrit.wikimedia.org/r/932828 (https://phabricator.wikimedia.org/T340507) (owner: 10Santhosh) [08:57:12] (03PS8) 10Elukey: Add language identification service [machinelearning/liftwing/inference-services] - 10https://gerrit.wikimedia.org/r/932828 (https://phabricator.wikimedia.org/T340507) (owner: 10Santhosh) [09:00:22] (03PS9) 10Elukey: Add language identification service [machinelearning/liftwing/inference-services] - 10https://gerrit.wikimedia.org/r/932828 (https://phabricator.wikimedia.org/T340507) (owner: 10Santhosh) [09:09:30] (03CR) 10CI reject: [V: 04-1] Add language identification service [machinelearning/liftwing/inference-services] - 10https://gerrit.wikimedia.org/r/932828 (https://phabricator.wikimedia.org/T340507) (owner: 10Santhosh) [09:15:38] (03CR) 10Elukey: Add language identification service (031 comment) [machinelearning/liftwing/inference-services] - 10https://gerrit.wikimedia.org/r/932828 (https://phabricator.wikimedia.org/T340507) (owner: 10Santhosh) [09:19:19] (03PS10) 10Ilias Sarantopoulos: Add language identification service [machinelearning/liftwing/inference-services] - 10https://gerrit.wikimedia.org/r/932828 (https://phabricator.wikimedia.org/T340507) (owner: 10Santhosh) [09:22:34] (03CR) 10Ilias Sarantopoulos: Add language identification service (031 comment) [machinelearning/liftwing/inference-services] - 10https://gerrit.wikimedia.org/r/932828 (https://phabricator.wikimedia.org/T340507) (owner: 10Santhosh) [09:24:04] (03CR) 10Elukey: Add language identification service (031 comment) [machinelearning/liftwing/inference-services] - 10https://gerrit.wikimedia.org/r/932828 (https://phabricator.wikimedia.org/T340507) (owner: 10Santhosh) [09:24:52] isaranto: pinged releng but they don't know exactly the issue, do you think it is the tox.ini ? [09:24:57] the error msg is very weird [09:25:53] (03CR) 10CI reject: [V: 04-1] Add language identification service [machinelearning/liftwing/inference-services] - 10https://gerrit.wikimedia.org/r/932828 (https://phabricator.wikimedia.org/T340507) (owner: 10Santhosh) [09:26:48] isaranto: do we need to also upload the tox.ini file to the PR? [09:27:25] this error is because the tox ini is missing but also entrypoint is not run correctly [09:27:49] yes I confirm, releng told me that the weird name is the generated docker image [09:28:04] very clear issue reporting... [09:28:06] anyway [09:29:24] isaranto: you have also to upload the actual tox.ini file no? [09:29:32] to be able to copy it [09:30:06] I have copied it. we are using the same tox.ini for all files [09:30:10] *for all images [09:30:47] ahhh right I forgot about that bit [09:33:23] w8 im debugging it locally, cause after that i got another error [09:33:24] argh [09:36:16] (03CR) 10Elukey: [C: 03+2] ores-legacy: redirect root to docs [machinelearning/liftwing/inference-services] - 10https://gerrit.wikimedia.org/r/933970 (owner: 10Ilias Sarantopoulos) [09:36:43] going to build/deploy --^ [09:37:32] (03Merged) 10jenkins-bot: ores-legacy: redirect root to docs [machinelearning/liftwing/inference-services] - 10https://gerrit.wikimedia.org/r/933970 (owner: 10Ilias Sarantopoulos) [09:38:29] Hello. Why all ORES models returns 404 error? [09:38:29] https://api.wikimedia.org/service/lw/inference/v1/models/enwiki-damaging:predict [09:39:04] Iluvatar: hi!!! [09:39:33] hi :) [09:39:55] Iluvatar: we should probably limit the HTTP GETs, since the endpoint needs POSTs to work [09:40:31] Iluvatar: https://api.wikimedia.org/wiki/Lift_Wing_API/Reference/Get_revscoring_damaging_prediction [09:40:52] e.g. curl -s "https://api.wikimedia.org/service/lw/inference/v1/models/enwiki-damaging:predict" -X POST -d '{ "rev_id": 2453354 }' would work [09:45:49] Iluvatar: (trying to match usernames) - have I opened a github issue to a project that you maintain? [09:46:56] anyway, all feedback about Lift Wing is really welcome, thanks! [09:47:28] elukey: I mean to ask: doe we have Prom/icinga monitoring of the o-legacy endpoint? [09:47:43] meant* [09:48:01] klausman: we have the health checks, but currently they are not set in production mode IIRC [09:48:10] ack, thanks. [09:49:18] klausman: my turn :) In https://github.com/SWViewer/swviewer-service/issues/1 I noticed that it was mentioned OAuth 1 vs 2, but I am not 100% clear what the Meta form that we link in our pages do [09:49:35] elukey, yes, it’s me. [09:49:35] Ok, thanks, but eswikiquote and eswikibooks still return 404 error. [09:49:38] there is indeed a "Client version" value to set, by default 1.0 [09:50:05] Iluvatar: you rock thanks for the patience :) [09:50:55] Iluvatar: do you mind to share what are you using to call eswiki{quote,books} ? [09:50:58] so we can repro [09:51:15] klausman: maybe we should add more infos to the page, re oauth 1 vs 2 [09:51:28] elukey: yep [09:52:00] https://api.wikimedia.org/wiki/Lift_Wing_API/Reference/Get_revscoring_damaging_prediction : python sample code. Just replace enwiki to eswikibooks [09:53:14] elukey: the API GW docs on wikitech explicitly mention that they'r OA2 tokens, so I'll add that detail on the LW usage page [09:53:31] Iluvatar: one caveat is that you'll need to use "eswikibookswiki" [09:53:45] it is redundant I know [09:53:55] (see URI parameters) [09:55:23] klausman: sure sure, what I mean is if it is really a oauth 1 token or not (the indefinite expire one) [09:55:34] I am ignorant/confused about what meta offers :D [09:55:47] Mh, I see your point. [09:58:24] deployed the redirect to /docs-> https://ores-legacy.wikimedia.org/ [10:02:26] > 12:53 it is redundant I know [10:02:28] yes, always in the past (streams, DB replicas) wiki were named: Wikipedia — lang + wiki (ru wiki). Others project: lang + domain (eswikibooks). Except wikidata (wikidatawiki). But here eswikibookswiki. I didn't notice that, sorry. [10:07:51] FYI, ml-etcd2001 will briefly go down for reboot [10:12:36] (03PS1) 10Elukey: ores-legacy: improve banner in /docs [machinelearning/liftwing/inference-services] - 10https://gerrit.wikimedia.org/r/934284 [10:12:43] moritzm: ack! [10:13:55] folks I tried to improve the docs page for ores-legacy in --^ [10:14:00] comments etc.. are welcom [10:14:02] *welcome [10:14:07] IIUC we can use markdown in there [10:15:57] elukey: the non-expiring token I used was an OA2 owner token. I don't know if API GW supports non-expiring (or long-lifetime) OA2 tokens. I have asked in the relevant place on Slack. [10:16:39] If OA2 is a hard requirement, we should probably add (a link to) example code to have an app request token renewals in a simple way, to lower the barrier of entry. [10:18:58] sure sure, I think that the right point that it was made in the github issue is that expiring tokens may require too much code for a simple bot [10:19:16] but if there is a way to have non expiring oauth 2.0 tokens we are good [10:19:23] Agreed. I think a token like the one I used might be the way to go [10:19:56] https://api.wikimedia.org/wiki/Special:AppManagement Here, you can click on "Create" and make a Personal token. [10:20:44] Those do not expire, but you also shouldn't just give it away for obvius reasons. If you make an app for others to use, you need to use a client token, with the associated expiry and extra code. [10:21:06] https://www.mediawiki.org/wiki/OAuth/For_Developers has some more details [10:21:18] 10Machine-Learning-Team, 10Continuous-Integration-Infrastructure, 10Release-Engineering-Team: Python torch fills disk of CI Jenkins instances - https://phabricator.wikimedia.org/T338317 (10elukey) @hashar is there a clean up command that we (ML SREs) can run to help the clean up while we get bigger partition... [10:21:53] But that documentation still links to MEta, whereas the API portal might be the New Hotness, so to speak [10:23:41] (03CR) 10Ilias Sarantopoulos: [C: 03+1] "nice touch!" [machinelearning/liftwing/inference-services] - 10https://gerrit.wikimedia.org/r/934284 (owner: 10Elukey) [10:27:47] elukey: added some links to our Usage page. Once we have a reply re API Portal vs. Meta on Slack, I'll clean it up some more. [10:27:48] klausman: so, to undestand, tokens generated from https://meta.wikimedia.org/wiki/Special:OAuthConsumerRegistration/propose/oauth2 have an expiry? [10:28:20] AIUI, yes [10:28:45] okok [10:29:05] I am more and more convinced that we should simply avoid oauth tokens [10:29:20] The API GW auth flows are in a bit of a transitionary phase (towards API portal), so thinks point into several directions at once [10:29:30] things* [10:29:32] it is a big entry barrier and it doesn't give to us a lot of benefits, except a base rate limit [10:29:55] Question is how we identify users then, if not by some sort of token. [10:30:16] but what do you need the user identification for? [10:30:29] especially if you can't do anything about it :D [10:30:30] Being able to block haywire or malicious use [10:30:44] this is already possible with combinations of ip address/uas/etc.. [10:31:02] like we do for requestctl [10:31:05] Also, we would have a similar problem as we have now with ORES in that we don't know who uses a service [10:31:17] we do know, via User Agents [10:31:22] If we ever wnated to turn down an IS [10:31:32] and we have a WMF policy that if people don't use them we can drop the traffic [10:31:47] I was not aware of said policy [10:32:00] (03PS11) 10Ilias Sarantopoulos: Add language identification service [machinelearning/liftwing/inference-services] - 10https://gerrit.wikimedia.org/r/932828 (https://phabricator.wikimedia.org/T340507) (owner: 10Santhosh) [10:32:02] I have to find the link but people are expected to use a good UA [10:32:10] Then that wfm. [10:32:17] I'll try to come up with an alternative proposal [10:32:25] this is getting really frustrating [10:32:31] going afk for lunch! [10:32:35] auth tokens do add quite some complexity in both docs as well as code on both client and server side. [10:32:47] Lunch sounds like a splendid idea. [10:38:16] (03CR) 10CI reject: [V: 04-1] Add language identification service [machinelearning/liftwing/inference-services] - 10https://gerrit.wikimedia.org/r/932828 (https://phabricator.wikimedia.org/T340507) (owner: 10Santhosh) [11:19:19] ml-etcd2003 will briefly go down for a reboot as well [11:26:27] Going for lunch! [12:32:43] ml-etcd2002 will briefly go down for a reboot as well [12:37:40] (03CR) 10Elukey: [C: 03+2] ores-legacy: improve banner in /docs [machinelearning/liftwing/inference-services] - 10https://gerrit.wikimedia.org/r/934284 (owner: 10Elukey) [12:39:06] (03Merged) 10jenkins-bot: ores-legacy: improve banner in /docs [machinelearning/liftwing/inference-services] - 10https://gerrit.wikimedia.org/r/934284 (owner: 10Elukey) [13:16:21] (03PS12) 10Ilias Sarantopoulos: Add language identification service [machinelearning/liftwing/inference-services] - 10https://gerrit.wikimedia.org/r/932828 (https://phabricator.wikimedia.org/T340507) (owner: 10Santhosh) [13:17:55] (03PS13) 10Ilias Sarantopoulos: Add language identification service [machinelearning/liftwing/inference-services] - 10https://gerrit.wikimedia.org/r/932828 (https://phabricator.wikimedia.org/T340507) (owner: 10Santhosh) [13:22:34] (03CR) 10CI reject: [V: 04-1] Add language identification service [machinelearning/liftwing/inference-services] - 10https://gerrit.wikimedia.org/r/932828 (https://phabricator.wikimedia.org/T340507) (owner: 10Santhosh) [13:26:46] isaranto: do you mind to check ores-legacy? [13:26:51] how does it look like? [13:27:08] I see some old cached content (hit front on the Marseille's varnishes I think) [13:27:57] sure, I am checking it. where do u see the cached content? [13:28:13] I mean which content and how do u verify it? [13:29:58] also I am going to remove the root endpoint from the swagger UI, doesn't serve anything over there [13:30:07] I don't see the redirect anymore [13:30:18] and I don't see the new banner visualized [13:30:28] but I am still wondering if my set up is messed up or not [13:30:48] klausman: o/ I merged https://gerrit.wikimedia.org/r/c/operations/deployment-charts/+/933060. could you deploy changeprop when you have a moment? thanks :) [13:31:20] everything works for me, redirect +new banner. I'm a bit lost, how could I verify if there is cached content? [13:31:37] aiko_: will do [13:31:56] isaranto: ack thanks, you can check the response HTTP headers [13:32:08] some of them are cache-related [13:32:17] and they tell you the cp nodes that generated the page [13:32:21] a ok thanks [13:36:56] I'm getting this a miss and a hit in x-cache when calling root ("cp3064 miss, cp3058 hit/1") [13:38:34] isaranto: I was about to ask, you go through esams [13:38:40] and I am going through marseille [13:38:51] even on my phone the content is old [13:39:10] 10Machine-Learning-Team, 10Continuous-Integration-Infrastructure, 10Release-Engineering-Team: Python torch fills disk of CI Jenkins instances - https://phabricator.wikimedia.org/T338317 (10hashar) I more or less fire fought some of them by heading to the instance and issuing `sudo docker buildx prune --force... [13:41:29] 10Machine-Learning-Team, 10Continuous-Integration-Infrastructure, 10Release-Engineering-Team: Python torch fills disk of CI Jenkins instances - https://phabricator.wikimedia.org/T338317 (10hashar) >>! In T338317#8971453, @isarantopoulos wrote: > There isn't any cache in the image since only the specific file... [13:41:33] (03PS14) 10Ilias Sarantopoulos: Add language identification service [machinelearning/liftwing/inference-services] - 10https://gerrit.wikimedia.org/r/932828 (https://phabricator.wikimedia.org/T340507) (owner: 10Santhosh) [13:43:27] ok I purged ores-legacy's content, and now I see the redirect, but not the new banner [13:43:55] I go through marseille for ssh but for web I had no idea [13:44:11] I do get it on my phone as well from different network [13:44:16] yeah we use GeoDNS on our Auth DNS servers [13:44:43] Greece is probably routed to Amsterdam [13:45:03] (03CR) 10Ilias Sarantopoulos: Add language identification service (031 comment) [machinelearning/liftwing/inference-services] - 10https://gerrit.wikimedia.org/r/932828 (https://phabricator.wikimedia.org/T340507) (owner: 10Santhosh) [13:45:50] 10Machine-Learning-Team, 10API Platform, 10Anti-Harassment, 10Cloud-Services, and 18 others: Migrate PipelineLib repos to GitLab - https://phabricator.wikimedia.org/T332953 (10JArguello-WMF) [13:49:25] (03CR) 10CI reject: [V: 04-1] Add language identification service [machinelearning/liftwing/inference-services] - 10https://gerrit.wikimedia.org/r/932828 (https://phabricator.wikimedia.org/T340507) (owner: 10Santhosh) [13:51:42] elukey: I solved the issue with the langid test docker image. it is now failing because it is expecting the readability pipeline which exists in aiko's patch (and that one is failing because it expects the langid one). Shall we merge one of them so the other succeeds? [13:51:58] (03PS7) 10Ilias Sarantopoulos: readability: add readability model server [machinelearning/liftwing/inference-services] - 10https://gerrit.wikimedia.org/r/931987 (https://phabricator.wikimedia.org/T334182) (owner: 10AikoChou) [13:52:35] isaranto: nice! Yes I think we can do it [13:52:43] both are ready to be merged. Aiko tried to merge the readability one but it failed... [13:53:02] I'm going to try to merge the readability first [13:53:32] (03CR) 10Ilias Sarantopoulos: [V: 03+2 C: 03+2] readability: add readability model server [machinelearning/liftwing/inference-services] - 10https://gerrit.wikimedia.org/r/931987 (https://phabricator.wikimedia.org/T334182) (owner: 10AikoChou) [13:54:16] ok done. had to bypass jenkins on this one. [13:54:16] ok so if I edit /etc/hosts to point ores-legacy to esams I get the right page [13:56:45] (03PS15) 10Ilias Sarantopoulos: Add language identification service [machinelearning/liftwing/inference-services] - 10https://gerrit.wikimedia.org/r/932828 (https://phabricator.wikimedia.org/T340507) (owner: 10Santhosh) [13:57:05] we probably need to bypass the varnish cache for the moment, ores doesn't do it [13:59:38] (03CR) 10Ilias Sarantopoulos: "I merged the readability patch and rebased (that one also had a CI pipeline which was failing on this patch)." [machinelearning/liftwing/inference-services] - 10https://gerrit.wikimedia.org/r/932828 (https://phabricator.wikimedia.org/T340507) (owner: 10Santhosh) [14:00:56] aiko_: I merged the readability patch. new images are being generated, so it is ready to be deployed [14:02:24] https://gerrit.wikimedia.org/r/c/operations/puppet/+/934336/ [14:02:42] sigh I noticed that the markdown stuff didn't work in the descriptio [14:03:39] but https://fastapi.tiangolo.com/tutorial/metadata/ says the opposite [14:07:15] yeah this is strange, lemme check locally [14:09:12] 10Machine-Learning-Team, 10Continuous-Integration-Infrastructure, 10Release-Engineering-Team: Python torch fills disk of CI Jenkins instances - https://phabricator.wikimedia.org/T338317 (10hashar) I reopened it merely to investigate whether maybe pytorch layer could be shrunk somehow but that does not seem t... [14:09:19] 10Machine-Learning-Team, 10Continuous-Integration-Infrastructure, 10Release-Engineering-Team: Python torch fills disk of CI Jenkins instances - https://phabricator.wikimedia.org/T338317 (10hashar) 05Open→03Resolved [14:11:12] 10Machine-Learning-Team, 10Gerrit-Privilege-Requests: Grant ML Team members +2 rights to the recommendation-api repository - https://phabricator.wikimedia.org/T340531 (10thcipriani) 05Open→03Resolved @kevinbazira bah, sorry about that! Should be fixed now. Removed from `mediawiki-services-recommendation-ap... [14:11:16] nice CI passes! [14:14:28] What language code should I send to agnostic model? for example, simplewiki (??) or hywwiki. "hyw" is “unsupported lang”). [14:16:07] Iluvatar: have you tried with other examples? If you find some language that are not supported we can ping the Research team and ask (they own the model) [14:17:34] klausman: I've set "pass" as caching for ores-legacy, since before turning it on we'd need to properly set our caching headers/strategy [14:17:53] ack! [14:19:38] I think need to request exact schema of "project <> language". Otherwise, there will be a lot of such questions. [14:24:05] definitely yes [14:24:17] maybe not all languages are supported [14:24:40] what were the ones that didn't work? hyw/simple only? Or more? [14:27:27] At the moment, the problem is only with these. [14:29:06] and be-x-old: time-out error :). But… [14:29:07] > 20:12 I will fix this issue asap. [14:30:24] ah yes yes that is a bug IIUC [14:33:24] +so (sowiki) “unsupported lang” [14:34:17] do you have a specific curl command to paste so we can repro/report? [14:39:36] Simple python code: https://pastebin.com/hU2nMETM [14:41:34] (Ideally, the wiki <> lang mapping should be done on backend. Not on clients.) [14:46:07] ack thanks! [14:49:40] (03CR) 10Elukey: Add language identification service (033 comments) [machinelearning/liftwing/inference-services] - 10https://gerrit.wikimedia.org/r/932828 (https://phabricator.wikimedia.org/T340507) (owner: 10Santhosh) [14:57:45] (03PS1) 10Ilias Sarantopoulos: ores-legacy: fix markdown in docs and remove root [machinelearning/liftwing/inference-services] - 10https://gerrit.wikimedia.org/r/934343 (https://phabricator.wikimedia.org/T330414) [15:00:46] 10Machine-Learning-Team, 10Patch-For-Review: Create ORES migration endpoint (ORES/Liftwing translation) - https://phabricator.wikimedia.org/T330414 (10isarantopoulos) {F37123192 width = 100% } [15:02:12] elukey: found the issue with markdown. now it looks like this https://phabricator.wikimedia.org/T330414#8976683 [15:02:12] because we had a tab before each sentence it was putting eveything in a box thus not rendering any markdown inside it. So it is an issue with markdown in general (TIL!) [15:22:12] isaranto: wow nice! [15:22:56] (03CR) 10Elukey: [C: 03+1] ores-legacy: fix markdown in docs and remove root [machinelearning/liftwing/inference-services] - 10https://gerrit.wikimedia.org/r/934343 (https://phabricator.wikimedia.org/T330414) (owner: 10Ilias Sarantopoulos) [15:26:24] (03CR) 10Elukey: [C: 03+2] ores-legacy: fix markdown in docs and remove root [machinelearning/liftwing/inference-services] - 10https://gerrit.wikimedia.org/r/934343 (https://phabricator.wikimedia.org/T330414) (owner: 10Ilias Sarantopoulos) [15:27:45] (03Merged) 10jenkins-bot: ores-legacy: fix markdown in docs and remove root [machinelearning/liftwing/inference-services] - 10https://gerrit.wikimedia.org/r/934343 (https://phabricator.wikimedia.org/T330414) (owner: 10Ilias Sarantopoulos) [15:34:12] +taywiki (unsupported lang: tay) [15:42:09] Iluvatar: o/ here is the list of supported lang: https://phabricator.wikimedia.org/P49491 [15:46:45] thanks! [15:47:03] aiko_: o/ I am wondering if we could have a special value or call to return that list [15:47:09] maybe something for the explainer? [15:49:35] https://ores-legacy.wikimedia.org/docs updated, thanks Ilias! [15:49:48] cc: chrisalbon_ --^ [15:51:59] 10Machine-Learning-Team, 10Patch-For-Review: Create ORES migration endpoint (ORES/Liftwing translation) - https://phabricator.wikimedia.org/T330414 (10elukey) We added https://ores-legacy.wikimedia.org/ and the idea is to eventually add a CNAME ores.wikimedia.org -> ores-legacy.wikimedia.org, that contains a b... [15:52:40] awesome [15:55:10] We probably want a list of supported languages, but the model card is the best place for them [15:55:41] chrisalbon_: yes definitely, but from some users' comments it would be nice to have it on an API, so they don't need to check the docs :) [15:58:44] If we did the change wouldn't we have to make it avaliable for all models on Lift Wing for consistency? [15:59:38] for revscoring definitely not [16:00:01] for the rest it shouldn't be a big deal, IIRC the model binaries have the list of supported models [16:00:07] or this is what Research wanted to do [16:00:15] so it would be just exposing a dict basically [16:00:47] (03PS16) 10Ilias Sarantopoulos: Add language identification service [machinelearning/liftwing/inference-services] - 10https://gerrit.wikimedia.org/r/932828 (https://phabricator.wikimedia.org/T340507) (owner: 10Santhosh) [16:01:57] (03CR) 10Ilias Sarantopoulos: Add language identification service (032 comments) [machinelearning/liftwing/inference-services] - 10https://gerrit.wikimedia.org/r/932828 (https://phabricator.wikimedia.org/T340507) (owner: 10Santhosh) [16:04:01] (03CR) 10Elukey: Add language identification service (032 comments) [machinelearning/liftwing/inference-services] - 10https://gerrit.wikimedia.org/r/932828 (https://phabricator.wikimedia.org/T340507) (owner: 10Santhosh) [16:04:43] going afk for today folks! [16:04:44] o/ [16:06:07] \o [16:06:30] aiko_: I am currently deployin the changeprop change in staging/codfw. If nothing explodes, I will do eqiad in 15m or so [16:06:40] having a post parameter e.g. "supported_languages" would be ok. and the information about that could be in the inference services repo so it can even be shared among some model servers [16:06:46] ciao Luca! [16:19:40] bye luca! [16:24:43] changeprop in eqiad pushed, keeping an eye on things [16:25:15] i'm logging off as well. I didn't touch the llm load tests again as I got caught up with the rest of stuff. more tomorrow! [16:25:20] \o [16:25:22] have a nice evening/day ! [16:25:32] 10Machine-Learning-Team: The nsfw model hangs in predict() after moving to Kserve 0.10 - https://phabricator.wikimedia.org/T331416 (10Htriedman) @elukey Feel free to remove it from Lift Wing for the moment! Thanks for letting me know. [16:28:10] revertrisk-multilingual: IndexError: pop index out of range (some edits, for example 39814738 ukwiki). Sample code: https://pastebin.com/BBc3rxHq [16:29:03] (03PS17) 10Ilias Sarantopoulos: Add language identification service [machinelearning/liftwing/inference-services] - 10https://gerrit.wikimedia.org/r/932828 (https://phabricator.wikimedia.org/T340507) (owner: 10Santhosh) [20:31:11] 10Machine-Learning-Team, 10Foundational Technology Requests: Content Translation Recommendations API - https://phabricator.wikimedia.org/T293648 (10leila) @elukey the short answer is that we don't have an agreement, yet, and we should. : ) can you please provide a high level list of items that need maintenanc... [21:48:46] 10Machine-Learning-Team, 10Data-Engineering, 10Edit-Review-Improvements-Integrated-Filters, 10Event-Platform Value Stream, and 2 others: Integration of Revert Risk Scores to Recent Changes as a filter - https://phabricator.wikimedia.org/T329071 (10JArguello-WMF)