Portal:Toolforge/Admin/Monthly meeting/2025-07-15

Jul 15, 2025 | Toolforge monthly meeting

Attendees (shuffled)

DC: Not yet started looking into it. We need to upgrade Kyverno before. I tested locally with the old version and it works but I don’t trust it, the docs say it’s not compatible with k8s 1.30. I have yet to test anything related to the Kyverno upgrade.

SKS: I created a draft of the scoring framework in my personal meta.wikimedia space. I also updated the Phab task T368600 with a link to it. I will wait for comments and eventually publish it to wikitech. I’m currently having problems with logging in to Wikitech.

DC: We added scheduled jobs, and now you can check the current status of a deployment from a CI using the deploy token. I also added a gitlab example yaml that people can reuse to deploy their tools.
DC: Most of the tools we have should be able to use the new push-to-deploy system, for pre-built images we need to think about it. Tools that just run a command could also benefit from push-to-deploy.
DC: You can also change the configuration by modifying a file in the repository, without using SSH.
DC: a few people have already started playing with push-to-deploy, including Taavi, Lucas, Musikanimal and others.

Taavi is working on Loki, got it deployed it to tools and toolsbeta, gathering logs from tools though there’s no way for users to get the logs out yet, working on a patch for jobs-api

OSI is currently not listing any LLM as meeting their open source requirements, so we should probably not allow running them on Cloud VPS at the moment (with or without GPUs)
Things might change in the future, and e.g. Isaac in the task above wrote "my guess is that the underlying multilingual BERT model used by revert-risk does meet OSAID requirements as it seems to be trained just on Wikimedia dumps"
We could collect user requests for useful features/projects that would require GPUs, and consider adding GPUs when we have one or more good use cases