What are the current options for disabling services like ML?
I thought of
1. setting the job concurrency to 0 (not allowed)
2. pausing the job(s): only works if you upload some pictures and are fast enough to click the UI, there is no option to pause while the queue is empty. This can become a problem, because if the system starts lagging due to too many jobs, it's hard to click the button in the UI... Also, pausing the ML job does not free the memory, which is my main concern.
3. removing the ml container completely: Are there any side effects apart from log messages complaining? is this the right way to do it? Since ideally I would like to only disable the clip service but keep the face recognition, this is not very practical.
4. set an environment flag: This is currently only documented for the reverse geocoding, is it supported for ml stuff as well?
I'd gladly update the docs with a paragraph on this if you could tell me the best practise. To me it seems that there is not a clear idea of where to keep and document settings currently. Some things have env vars, which are basically only referred to in the docker-compose installation documentation, i.e. hard to find for others, and some others are only reachable via web UI (like job concurrency).
2 Replies
I think https://immich.app/docs/FAQ#how-to-disable-machine-learning-and-typesense is what you're looking for
FAQ | Immich
What is the difference between the cloud icon on the mobile app?
Thanks! That is in part what I am looking for. However, it would be cool to turn off only some ML models and prevent them from being loaded into RAM. But I understand that this does not seem possible yet.