Wikimedia Technical Committee/Minutes/2020-08-26

Present: Daniel K, Tim S, Giuseppe L, Timo T, Dan A, Roan K.

New link service

https://www.mediawiki.org/wiki/Growth/Personalized_first_day/Structured_tasks#Prioritizing_%22add_a_link%22
Guiseppe: Showing suggested links for editors using ML. How do we store ML models in production, update them?
Dan: isn’t this the purview of Chris’s ML infrastructure team?
Giuseppe: long-term, yes, but not directly their responsibility right now. The general architectural issue is: we want a sustainable way to generate ML models, and ship them to applications using them. This process should be streamlined - the current pattern of slapping the models in git-lfs is not sustainable long-term.
Roan: explains about the project, a collaboration with Research to suggest links to add to articles. Docs/write-ups are emerging (and more focused on UX)
Daniel: Create an RFC?

https://phabricator.wikimedia.org/T260714
Daniel: Should it go on last call?
Tim: no urgency to closing this. The review they’re asking for hasn’t been done yet.
Timo: They’ve done a fair bit of research and outreach offline prior to the RFC. They’re looking for explicit approval from would-be users of their new API for parser extensions. Once that completes, they should ask for Last Call.

https://phabricator.wikimedia.org/T260330
Tim: New initiative being started by Platform Engineering. Will turn into an RFC.
Tim: Affects every shell execution, can be used in small/medium wiki instances that aren’t hosted by WMF, seems cross-cutting, affects over 30 extensions.