[08:19:03] 10Machine-Learning-Team, 10Add-Link, 10Growth-Team: Investigate `UnicodeEncodeError` thrown by Add-A-Link training pipeline for fywiki model - https://phabricator.wikimedia.org/T325521 (10kevinbazira) Discussed this issue with @MGerlach and he advised that it could be caused by the [[ https://github.com/wiki... [09:31:56] (03PS5) 10Elukey: 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) (owner: 10AikoChou) [09:32:06] (03CR) 10Elukey: [V: 03+2] 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) (owner: 10AikoChou) [10:15:00] (03PS1) 10Elukey: Remove nvidia-ml-py3 from requirements.txt files [machinelearning/liftwing/inference-services] - 10https://gerrit.wikimedia.org/r/869729 [10:15:30] going to test some builds without nvidia-ml-py3 [10:15:34] no idea why it is in our deps [10:15:39] but it is not open-source [10:17:39] going afk for bit (need to check a co-working space here in my town) [11:27:53] (03CR) 10Elukey: "Tested the revscoring model server (but not the others) with Docker. CI built all images so I am pretty sure that the nvidia package is no" [machinelearning/liftwing/inference-services] - 10https://gerrit.wikimedia.org/r/869729 (owner: 10Elukey) [11:34:23] (03CR) 10Ilias Sarantopoulos: [C: 03+1] "I also don't see a reason why this would needed!" [machinelearning/liftwing/inference-services] - 10https://gerrit.wikimedia.org/r/869729 (owner: 10Elukey) [11:34:50] * elukey lunch! [11:44:51] elukey: o/ how to know if a package is open-source or not? [11:54:53] aiko: o/ I checked LICENSE at the bottom of https://pypi.org/project/nvidia-ml-py/ [11:55:40] also "homepage" links to Nvidia's main page, they don't have a github page etc.. [11:56:00] in general all things that nvidia ships are binary only (usually, we'll see in the future) [13:05:04] elukey: I see, thanks! [13:10:47] 10Machine-Learning-Team, 10Add-Link, 10Growth-Team: Investigate `UnicodeEncodeError` thrown by Add-A-Link training pipeline for fywiki model - https://phabricator.wikimedia.org/T325521 (10kevinbazira) Looked at the wikipedia2vec repo and found `UnicodeEncodeError` issues: https://github.com/wikipedia2vec/wik... [13:44:46] 10Machine-Learning-Team: Automate publishing python packages to PyPI - https://phabricator.wikimedia.org/T325561 (10isarantopoulos) This is an example action https://github.com/wikimedia/drafttopic/pull/67 that will push to PyPI As I think about it we have 3 options: - publish a package whenever we merge a... [13:49:19] 10Machine-Learning-Team, 10Research: Update torch's settings in the Knowledge Integrity repo - https://phabricator.wikimedia.org/T325349 (10achou) We rebuilt our docker image with Luca's knowledge integrity fork repository which removed torch dependency. When we tested the image in ml-sandbox, we got ` Messag... [13:57:22] 10Machine-Learning-Team, 10Add-Link, 10Growth-Team: Investigate `UnicodeEncodeError` thrown by Add-A-Link training pipeline for fywiki model - https://phabricator.wikimedia.org/T325521 (10kevinbazira) While looking into wikipedia2vec and where to fix the `UnicodeEncodeError` issue, I noticed it happens at th... [14:09:42] (03CR) 10Elukey: [C: 03+2] Remove nvidia-ml-py3 from requirements.txt files [machinelearning/liftwing/inference-services] - 10https://gerrit.wikimedia.org/r/869729 (owner: 10Elukey) [14:15:25] (03Merged) 10jenkins-bot: Remove nvidia-ml-py3 from requirements.txt files [machinelearning/liftwing/inference-services] - 10https://gerrit.wikimedia.org/r/869729 (owner: 10Elukey) [14:25:06] 10Machine-Learning-Team, 10Patch-For-Review: Test revscoring model servers on Lift Wing - https://phabricator.wikimedia.org/T323624 (10isarantopoulos) The above plots show that we can enable MP for editquality models if we see fit it makes them much more stable and keeps latency low even in the 99th percentile [14:55:34] elukey: I found that even if we remove nvidia-ml-py from requirement.txt, we still install it because it seems a dep for kserve. [14:56:05] elukey: I tried to run pip install kserve==0.8.0 in a virtualenv, and it also installed nvidia-ml-py [14:56:25] aiko: ufff thanks :( [14:57:35] it is not directly in https://github.com/kserve/kserve/blob/release-0.8/python/kserve/requirements.txt [14:57:45] but it must be a dep [14:58:17] maybe a dep of a dep :( [14:59:13] aiko: if you have the venv already set, can you install pipdeptree and see what dep brings in the nvidia one? [14:59:49] elukey: ok let me try [15:01:21] gpustat==1.0.0 [15:02:44] aiko: meeting, let's see who brings in gpustat in kserve later [15:04:12] okkki wait a sec [15:16:45] 10Machine-Learning-Team, 10artificial-intelligence, 10Bad-Words-Detection-System, 10revscoring: Add language support for Esperanto (eo) - https://phabricator.wikimedia.org/T325577 (10calbon) a:03calbon [15:30:24] 10Machine-Learning-Team, 10Add-Link, 10Growth-Team: Investigate `UnicodeEncodeError` thrown by Add-A-Link training pipeline for fywiki model - https://phabricator.wikimedia.org/T325521 (10calbon) a:03kevinbazira [15:31:56] 10Machine-Learning-Team, 10artificial-intelligence, 10Bad-Words-Detection-System, 10revscoring: Add language support for Serbo-Croatian - https://phabricator.wikimedia.org/T325483 (10calbon) a:03calbon [15:37:31] 10Machine-Learning-Team, 10Research-Backlog, 10Section-Level-Image-Suggestions, 10Structured-Data-Backlog: Productionize section alignment model training - https://phabricator.wikimedia.org/T325316 (10calbon) a:03calbon [15:38:42] aiko: I started to put together a Python package I'm calling mwconstants that has a bunch of common constants/utils for working with wikitext etc. The idea is that the constants should be mostly static things like all the Wikipedia languages or namespace prefixes we use in our ML model code but are a pain to update everywhere as they slowly go out of date. and the utils are basic functions that could be reused between models like mapping [15:38:43] links to namespaces. one thing i haven't decided is whether to keep this super simple in terms of dependencies or add wikitext functions that would depend on mwparserfromhell too (e.g., stripping syntax from wikitext). not urgent but would love to collaborate if you think this would be useful for making the liftwing models more maintainable. repo: https://gitlab.wikimedia.org/repos/research/mwconstants [15:41:05] 10Machine-Learning-Team: Enrich revertrisk image tag with model's package version - https://phabricator.wikimedia.org/T325295 (10calbon) a:03AikoChou [15:43:11] 10Machine-Learning-Team: Create a pre-commit hook for inference-services repo - https://phabricator.wikimedia.org/T325198 (10calbon) a:03isarantopoulos [15:57:57] 10Lift-Wing, 10Machine-Learning-Team: Test batch prediction for revert-risk model - https://phabricator.wikimedia.org/T323023 (10calbon) a:05achou→03isarantopoulos [16:00:35] * elukey little break [16:05:21] 10Machine-Learning-Team: Upgrade python from 3.7 to 3.9 in docker images - https://phabricator.wikimedia.org/T325657 (10isarantopoulos) [16:07:44] btw does anyone use PAWS? https://wikitech.wikimedia.org/wiki/PAWS [16:33:51] isaranto: I don't use it much, but I know it is often used in community and wikimedia's contributors. [17:26:33] * elukey afk!