[14:14:22] hello! [14:14:48] It is Tuesday, my day-o-meetings! [16:07:19] o/ [16:20:48] been reading thru the ML observability book that elukey linked last week -- it makes a pretty strong case for including explainability/interpretability as part of your metrics [16:23:09] in theory, kserve could help use tools like LIME, SHAP, anchor-text, etc in a plug-n-play manner inside our inference services [16:24:08] which could help us expose stuff like feature importance, drift detection etc [16:27:25] IIRC there was some exploratory work done by the scoring platform team using LIME on some of the ORES models [16:32:32] ah yeah found the repo: https://github.com/adamwight/ores-lime [17:30:21] That could be really interesting, because then we could get those analytics into our analytics system and wikidata for the model cards [19:32:28] niiiiice i think i finally got the editquality isvc image upgraded to kserve v0.7.0 [19:35:18] doin a quick test to make sure, fingers crossed :) [19:52:47] Update on the bird name, the Branding department is going to help us out with branding everything, which could be cool [19:53:11] Apparently nobody trusts me to name things, which is fair [19:55:14] LOL [19:56:17] idk LiftWing is a pretty good one [19:56:31] but i do understand a need for cohesion across projects :) [19:56:51] wasn't there another name prior to LiftWing? [20:00:18] Heavy Wing [20:00:27] but people thought it sounded too much like a bomber [20:00:30] which, fair [20:58:05] ahh derp i forgot to update model.py to import kserve instead of kfserving on both editquality and the articlequality images [20:58:09] fixing [21:26:19] oof scratch that still need some more work on the editquality kserve upgrade :( [21:26:40] running into issues with cjk tokenization w/ some of the revscoring deps [21:29:00] it seems the tweepy package has change the way it parses a corpus (corpiii??)