[10:43:45] elukey: it is likely the sklearn dep can be dropped entirely, as it's only really used in CI and examples of that project (fairseq). I can change my for of fairseq accordingly. [10:44:20] (I know you're out today/tomorrow, so that message is both for you-on-Monday and making sure I don't forget) [10:44:45] We likely don't need to maintain a fairseq fork anyway, as I only needed it for debugging. [10:45:21] Unless of course we want to be on the safe side, since the code we use is on a non-main branch and not really "released" in any sense [10:58:49] klausman: o/ (I was checking the internetz anyway) - I used SKLEARN_ALLOW_DEPRECATED_SKLEARN_PACKAGE_INSTALL=True and it worked, but we can definitely change the name of the pkg in out setup.py and make it work. I am a little worried about our "fork" going stale compared to the main branch of fairseq, maybe this is something to discuss at the AWS meeting as well (if it didn't happen this morning) [10:59:13] it's in-progress :) [11:04:32] ack :) [11:04:58] So the current state of fairseq is making even Meta sad. Everybody is using it in different ways and things are very messy. So Guillaume is/has been working on a successor, fairseq2, that will be both easier to use and cleare in what code should be where. He says he'll probably have something to share early next year. [11:05:33] With that in mind, I think we can stay on my fork-of-a-branch thing (maybe move it to Gitlab) for now. [11:07:12] yes let's move to gitlab, better to have all deps in the WMF court [11:07:21] agreed [11:07:47] I'll make a change to my GH fork to drop sklearn for testing. We can do the move-to-GL stuff next week. [11:08:50] to have less variables, let's just change the sktlearn dep to the new name [11:09:00] sure [11:09:04] thanks :) [11:10:48] Do you also want me to change the assorted imports and calls? [11:14:16] eh, done&done [11:42:16] ack :) [11:42:23] the new image that I have locally is around 7.43GB [11:42:30] not sure if I left something though [11:42:39] (previously 11.XGB) [11:44:32] is it still ubuntu-based? [11:45:06] also, get off IRC, it's a holiday! :-P [11:45:33] yep it is still using the original cuda-ubuntu image [11:46:11] https://github.com/elukey/stopes/blob/aws_publish/misc/aws_publish/dockerfiles/Dockerfile has been updated a lot, lemme know if it is ok [11:46:17] on Monday I'll deploy to staging :) [11:46:19] o/ [11:59:44] Roger! [12:42:20] * klausman lunch [12:51:24] (03PS1) 10AikoChou: Create a test folder and add lua scripts for wrk [machinelearning/liftwing/inference-services] - 10https://gerrit.wikimedia.org/r/866360 (https://phabricator.wikimedia.org/T323613) [12:53:50] (03PS2) 10AikoChou: Create a test folder and add lua scripts for wrk [machinelearning/liftwing/inference-services] - 10https://gerrit.wikimedia.org/r/866360 (https://phabricator.wikimedia.org/T323613) [15:38:51] (03PS3) 10AikoChou: revertrisk: fix mwapi session host headers [machinelearning/liftwing/inference-services] - 10https://gerrit.wikimedia.org/r/865713 (https://phabricator.wikimedia.org/T323023)