[08:20:25] o/ [08:20:42] I have just tested on stat1004 the http_cache for revscoring, it seems working nicely :) [08:20:45] I'll update the task [08:26:29] 10Lift-Wing, 10Machine-Learning-Team (Active Tasks), 10Patch-For-Review: Test async preprocess on kserve - https://phabricator.wikimedia.org/T309623 (10elukey) Another experiment that I made in these days, namely adding a simple HTTP cache to revscoring: https://github.com/elukey/revscoring/commit/aaa8a59c6f... [08:26:35] added all the infos in --^ [08:29:34] in theory the scalability of revscoring kserve pods should be way better with Aiko's AsyncSession + http_cache in revscoring [08:30:00] and the code update are not invasive, and backward-compatible with any clients using them [10:28:48] * elukey lunch [11:01:49] (03CR) 10AikoChou: [C: 03+1] "Thank you Luca for hacking this. It's great we can test locally with Docker!" [machinelearning/liftwing/inference-services] - 10https://gerrit.wikimedia.org/r/808247 (https://phabricator.wikimedia.org/T301878) (owner: 10Elukey) [11:52:15] 10Machine-Learning-Team, 10ORES, 10Growth-Team, 10PageCuration: Detect hoaxes - https://phabricator.wikimedia.org/T312841 (10Atsme) Can you add me to the group of editors? Apparently, no one can post there without first being added. [11:59:42] 10Machine-Learning-Team, 10ORES, 10Growth-Team, 10PageCuration: Detect hoaxes - https://phabricator.wikimedia.org/T312841 (10Novem_Linguae) @Atsme, good morning, hope you're doing well. > Apparently, no one can post there without first being added. Where is "there"? [12:08:57] 10Machine-Learning-Team, 10ORES, 10Growth-Team, 10PageCuration: Detect hoaxes - https://phabricator.wikimedia.org/T312841 (10Atsme) Good morning, Novem! @Isaac - see calbon's comment above. [13:26:29] (03CR) 10Elukey: [C: 03+2] editquality: add support for revision-score events [machinelearning/liftwing/inference-services] - 10https://gerrit.wikimedia.org/r/808247 (https://phabricator.wikimedia.org/T301878) (owner: 10Elukey) [13:49:14] mmm I don't see anything building the new image after --^ [13:55:52] seems to be stuck in ci queues? https://integration.wikimedia.org/zuul/ [13:59:43] taavi: yeah I was checking it as well, I didn't get if it was stuck (so something was kicked off) or not [14:04:22] the build was kicked off https://integration.wikimedia.org/ci/job/inference-services-pipeline-editquality-publish/17/ [14:04:24] goood [14:16:01] filed the deployment-chart change to test the event generation code! [14:16:02] https://gerrit.wikimedia.org/r/c/operations/deployment-charts/+/813633 [14:21:59] * elukey afk for a bit! [14:22:01] need to go to the vet [15:40:49] Morning all! [16:04:30] morning :) [16:08:09] deployed the event change in staging for enwiki, it doesn't work but it returns an error msg :) [16:08:42] progress! [16:09:07] yep! Not entirely sure why the event gets rejected by eventgate [16:09:27] (these are difficult things to test via Docker or ML-Sanbox) [16:10:09] as far as I got for the ml-collab thread, people would like to have a way to deploy models to cloud vps instances, but kserve may be really overkill [16:11:37] Yeah, I think kserve is overkill. Honestly what they need is a path for dockerizing a model in a flask app and putting it on cloud VPS [16:12:24] It is worthwhile for us to figure out eventually because I'd like a guide for volunteers to deploy their own models on toolforge [16:12:52] For example, a model we don't feel comfortable hosting, a community member can still get it up on toolforge. [16:12:59] But that is more like a detailed tutorial [16:13:08] not a system change like a kserve instance [16:16:00] yeah [16:18:08] but even for people trying to deploy to liftwing, in theory there is little gain in testing something as far as in ml-sanbox etc... [16:18:39] they can definitely test locally with Docker with more ease, and we can take care of the k8s config [16:18:58] (this is why I started testing/hacking with local Docker) [16:22:19] going afk, see you tomorrow folks! [16:22:37] ml-sandbox and all of Lift Wing shouldn't be accessed by other teams, the ML team has to be the ones to deploy for folks, otherwise we introduce a huge host of problems in the short term. [16:22:37] In the long term, it should be a goal of train wing that models created by folks in kubeflow can be deployed without us, but that is a longterm goal [16:22:43] night luca! [18:01:45] 10Machine-Learning-Team, 10ORES, 10Growth-Team, 10PageCuration: Detect hoaxes - https://phabricator.wikimedia.org/T312841 (10Isaac) Adding some context to make sure I understand the request correctly: [[https://www.mediawiki.org/wiki/Page_Curation|Page Curation]] is a tool that was built by Wikimedia Found... [18:10:05] (03CR) 10Jforrester: [C: 03+2] Bump minimum required version for upgrade to 1.31 [extensions/ORES] - 10https://gerrit.wikimedia.org/r/804664 (owner: 10Umherirrender) [18:14:04] (03Merged) 10jenkins-bot: Bump minimum required version for upgrade to 1.31 [extensions/ORES] - 10https://gerrit.wikimedia.org/r/804664 (owner: 10Umherirrender) [18:14:14] (03CR) 10Jforrester: [C: 03+2] Convert to abstract schema [extensions/ORES] - 10https://gerrit.wikimedia.org/r/757989 (https://phabricator.wikimedia.org/T268566) (owner: 10Umherirrender) [18:14:41] 10Machine-Learning-Team, 10MediaWiki-extensions-ORES, 10Patch-For-Review, 10Platform Team Initiatives (Abstract Schema): Convert ORES to AbstractSchema - https://phabricator.wikimedia.org/T268566 (10Jdforrester-WMF) 05Open→03Resolved [18:19:32] (03Merged) 10jenkins-bot: Convert to abstract schema [extensions/ORES] - 10https://gerrit.wikimedia.org/r/757989 (https://phabricator.wikimedia.org/T268566) (owner: 10Umherirrender) [21:12:17] 10Machine-Learning-Team, 10ORES, 10Growth-Team, 10PageCuration: Detect hoaxes - https://phabricator.wikimedia.org/T312841 (10Atsme) Thank you, Isaac. Your response is encouraging. @Bluerasberry was able to present the issue with greater clarity than I could. My vision is that such a tool would not be too...