[17:10:06] hey chrisalbon, how are you? [17:34:59] chtnnh going well, yourself? [17:35:16] not too bad! [17:36:37] I was wondering whom and when should I speak to someone on the team regarding the future of the ML platform here at WMF as the NSFW media classifier is now an approved outreachy project and has started getting initial contributions [17:38:37] chtnnh: I'd be a bit curious about that, specifically in T260068 i'm working up a method to classify queries to commonswiki as NSFW based on classifying the returned media [17:38:38] T260068: Classify completion candidate image results - https://phabricator.wikimedia.org/T260068 [17:39:11] looking at the task you linked ebernhardson [17:40:56] I see that the task you are referring to is quite new. The one I am referring to (T214201) has been around for a while and I decided to pursue it as an outreachy project [17:40:57] T214201: Implement NSFW image classifier using Open NSFW - https://phabricator.wikimedia.org/T214201 [17:41:30] chtnnh: right, i'm saying i'd be very interested in your project, perhaps most relevant is the last line of my comment "In a more ideal world where there is a trivial way for analytics jobs to query api's that transform page titles into is_safe predictions that could be a great simplifier, but in a quick evaluation there are significant roadblocks to making that happen." [17:41:42] i suspect your project is filling that gap? Or could i encourage to change it to fill that gap? :) [17:42:53] the project would certainly benefit from more guidance, the current scope of the project is to build a model to classify media (image, video) as SFW or NSFW [17:43:14] We are trying to achieve upload time classification of these media [17:43:39] interesting, as long as it's stored somewhere that's perfect. I imagine it would also go through the backlog eventually? [17:45:23] that is ideal [17:45:36] but I will have to research more on the technical feasibility of that aspect [17:46:25] currently we are just focusing on the ML aspects of the problem, storing the classification and building a service around wikimedia infrastructure can be a future project [17:46:41] as the internship timeline for outreachy only allows the student to do so much [17:47:30] right, as an intership if this includes building models i suspect it won't get much further than that initially. Building out the system that connects mediawiki, your models (via heavywing?), and our various storage patterns is a whole different skill set [17:47:44] (although there is overlap at times) [17:49:36] i can't help much with model building, but if you get to the other parts and have questions about mw/prod feel free to reach out [17:50:57] great [17:51:28] hopefully heavy wing will support this model and we can move forward with deploying it [17:52:06] why dont you add yourself as a subscriber to the task I linked and maybe we can share updates across phab if i dont catch you around irc [17:52:15] sure [17:52:31] I owe you all the heavy wingdraft design doc. It is currently being reviewed by some WMF but after that I'll dump it on wikitech [17:52:58] sounds great chrisalbon , any eta? [17:53:18] Probably Tuesday, since monday is a US holiday [17:55:01] alright! [21:28:39] 10ORES, 10Machine Learning Platform, 10Okapi, 10Operations, and 3 others: ORES redis: max number of clients reached... - https://phabricator.wikimedia.org/T263910 (10calbon) {F32378655} Suspiciously similar just happened to ores on eqiad. I reset the uwsgi service on all ores100x boxes and will monitor. [21:32:16] 10ORES, 10Machine Learning Platform (Current): Script to restart ORES uwsgi service - https://phabricator.wikimedia.org/T264121 (10calbon) 05Resolved→03Open [21:33:26] 10ORES, 10Machine Learning Platform (Current): Script to restart ORES uwsgi service - https://phabricator.wikimedia.org/T264121 (10calbon) Due to {T263910} continuing to be a problem (I suspect), I am bringing back this task. Cumin isn't an option because it is SRE-specific. I'll get on coding it up in fabric.