Page MenuHomePhabricator

Request creation of research-collaborations-api VPS project
Closed, ResolvedPublic

Description

Project Name: research-collaborations-api

Wikitech Usernames of requestors: @Miriam,@Isaac,@diego,

Purpose: Run experimental APIs using complex ML models, developed in collaboration with external academics, students and interns.

Brief description: The research team collaborates with external researchers our the world, to create services and APIs that run on top Machine Learning models. We have a bottle on , due the lack of access of external collaborators to servers where they can test and deploy those models. Toolforge is not enough, due the size of some of those models. We use packages like Fasttext, SemBert, and Glove. Some of the ongoing tasks that will benefit from this project will be: T263646, T263860, T273213, T260564, just to mention some examples.

How soon you are hoping this can be fulfilled: as soon as possible

Event Timeline

Any ideas how to avoid the "umbrella project" problem here? This sounds like a project to create a sandbox for hosting multiple instances each with a different ownership, purpose, and lifecycle. That convenience for the project members typically comes with a tradeoff of lower accountability when the cloud-services-team needs to contact the instance maintainer especially when asking to purge unused instances or comply with an operating system upgrade mandate.

Hi @bd808, I get your point. I can take the responsibility on keeping track of all these instances, and be the point of contact with you.

Moreover, given that we work with several collaborators within the year, we are completely aware of the limited resources (we have done some cleanups of existing projects with @Isaac) , so it is very unlikely we will left instances abandoned. Each collaboration has a PoC within the team, and if anybody within the team request more space for this project, that person will need to check/justify what they are doing with other instances.

Hi @bd808, I get your point. I can take the responsibility on keeping track of all these instances, and be the point of contact with you.

Moreover, given that we work with several collaborators within the year, we are completely aware of the limited resources (we have done some cleanups of existing projects with @Isaac) , so it is very unlikely we will left instances abandoned. Each collaboration has a PoC within the team, and if anybody within the team request more space for this project, that person will need to check/justify what they are doing with other instances.

Thanks for these thoughts @diego. Keeping some public mapping of instances to "owners" and purpose, like https://wikitech.wikimedia.org/wiki/Nova_Resource:Reading-web-staging is doing, would be a nice practice for a project like this to adopt as well.

This has been approved in our 2021-02-17 meeting.

Project research-collaborations-api has been created.