[07:58:28] o/ [09:11:35] https://github.com/kserve/kserve/blob/release-0.7/python/kserve/kserve/kfmodel.py#L63 [09:11:42] postprocess is not awaited in 0.7.... [09:11:46] * elukey cries in a corner [09:12:53] interesting, it is not awaited in 0.8 and the upcoming 0.9 too [09:15:48] mmmm maybe post-process doesn't need to be a coroutine [09:15:56] since we await the http call in it [09:16:39] so probably it needs to be [09:17:05] yep [09:21:47] ok trying to move everything to predict() [09:22:15] it is relatively easy to test a kserve model using only Docker afaics [09:22:25] if there is no doc we should write one [11:02:59] (03PS14) 10Elukey: editquality: add support for revision-score events [machinelearning/liftwing/inference-services] - 10https://gerrit.wikimedia.org/r/808247 (https://phabricator.wikimedia.org/T301878) [11:03:39] (03CR) 10Elukey: editquality: add support for revision-score events (031 comment) [machinelearning/liftwing/inference-services] - 10https://gerrit.wikimedia.org/r/808247 (https://phabricator.wikimedia.org/T301878) (owner: 10Elukey) [11:05:15] (03CR) 10Elukey: "Aiko: I had to move the code from postprocess to predict, since kfserver.py doesn't await postprocess but only preprocess and predict (sam" [machinelearning/liftwing/inference-services] - 10https://gerrit.wikimedia.org/r/808247 (https://phabricator.wikimedia.org/T301878) (owner: 10Elukey) [11:05:49] * elukey lunch! [11:30:10] 10Machine-Learning-Team, 10ORES, 10Growth-Team, 10PageCuration: Detect hoaxes - https://phabricator.wikimedia.org/T312841 (10Novem_Linguae) [11:36:40] 10Machine-Learning-Team, 10ORES, 10Growth-Team, 10PageCuration: Possible ORES outage for PageCuration tags "vandalism", "spam", "attack" - https://phabricator.wikimedia.org/T312843 (10Novem_Linguae) [13:08:43] (03CR) 10Elukey: "Tested with Docker like the following:" [machinelearning/liftwing/inference-services] - 10https://gerrit.wikimedia.org/r/808247 (https://phabricator.wikimedia.org/T301878) (owner: 10Elukey) [13:09:05] I have tested --^ with Docker and two containers using the bridge networking, really simple and easy [13:09:09] I'll write some docs about it! [13:12:15] 10Machine-Learning-Team, 10ORES, 10Growth-Team, 10PageCuration: Possible ORES outage for PageCuration tags "vandalism", "spam", "attack" - https://phabricator.wikimedia.org/T312843 (10elukey) Hi! From the ORES infrastructure point of view there is nothing in our metrics that indicates a problem. It would b... [13:12:31] (probably also useful for the folks trying to use the ml-sandbox for the inspiration week) [14:22:53] After two days of looking at the uwsgi code, I can safely say that I hace C :) [14:22:55] hate* [14:28:06] klausman: I tried, on ores1001, to comment the `logformat` directive and I don't see anymore the empty UDP frames [14:28:23] but I don't see the logs on logstash either, so maybe the standard format is not right [14:28:44] one thing that I noticed is that on netbox nodes we have the same uwsgi config (except for the logformat) and it works fine afaict [14:28:51] Do you see nonempty UDP frames? [14:29:07] yes you can check now on ores1001 (it is depooled) [14:29:22] Hmm. I had meant to do a deeper dive into the format syntax. [14:29:36] Something in the escape/quoting logic there just seemed... off [14:30:07] Still, the add'l logging should not change any mistakes in the original logging format. [14:31:37] Morning all [14:32:17] morning! [14:32:32] klausman: I suspect that it may also be related to the ordering of the directives in the .ini file [14:32:50] it should not count in theory, but maybe there is some weird corner case in the logging code [14:32:51] There is that. [14:33:03] What a mess. I can undertsand why the project is abandonware [14:33:05] but on the netbox nodes we have the same config, that works [14:38:04] Are they Buster as well? [14:42:20] 10Machine-Learning-Team, 10ORES, 10Growth-Team, 10PageCuration: Possible ORES outage for PageCuration tags "vandalism", "spam", "attack" - https://phabricator.wikimedia.org/T312843 (10calbon) Hi Novem_Linguae! ORES and the RevScoring models look okay to me. It is serving predictions for all categories (van... [15:00:39] klausman: yep [15:01:03] I may have to give that a differential look tomorrow, then [15:16:00] cleared our and pooled ores1001 [15:27:26] I started https://wikitech.wikimedia.org/wiki/Machine_Learning/LiftWing/KServe [15:27:35] there is a quick way to use Docker to run a model server [15:44:46] added also the Example 2 in https://wikitech.wikimedia.org/wiki/Machine_Learning/LiftWing/KServe#Run_KServe_locally_via_Docker [15:45:01] Very nice! [15:46:12] also passed everything to the slack channel in which people are hacking [15:46:33] it seems an easier way compared to the ML-Sandbox, that has also the complexity of K8s [15:48:53] going afk for today, have a nice rest of the day folks! [15:48:58] \o [16:05:41] night elukey! [16:36:45] I'm heading out as well. later, everyone! [16:43:04] Night klausman! [17:09:27] 10Machine-Learning-Team, 10ORES, 10Growth-Team, 10PageCuration: Detect hoaxes - https://phabricator.wikimedia.org/T312841 (10calbon) Hey Novem_Linguae! Let me give it some though, but off the top of my head it won't be easy. For example, what exactly is a hoax compared to other vandalism? Is there a detect... [17:32:52] 10Machine-Learning-Team, 10ORES, 10Growth-Team, 10PageCuration: Detect hoaxes - https://phabricator.wikimedia.org/T312841 (10Atsme) Hi, Calbon - regarding the hoax detection, if a stub or article is published in main space with no sources, we can draftify, but hoaxes are usually sourced. Is it possible to... [21:29:43] 10Machine-Learning-Team, 10ORES, 10Growth-Team, 10PageCuration: Detect hoaxes - https://phabricator.wikimedia.org/T312841 (10calbon) @Isaac would be in a better place than me to determine the viability of that approach. I think the knowledge extraction on the cited source would be pretty difficult. For exa...