Can I download the machine learning files and host it on my pc?
Hi! I’m having an issue where the machine learning docker will refuse to start up (i’m guessing the computer i’m hosting immich on is too slow) and thus immich won’t start up either without it. Commenting it out in the yml fixes it and lets immich run, but i’m missing out on a huge part of why I wanted to get immich in the first place lol. Instead of it trying to download it from a remote server every time the docker starts up, is it possible to self-host it? Any help is appreciated, thank ya! (I’m very inexperienced with this sort of thing, so I may need to ask follow up questions!)
13 Replies
Side note, I tried leaving the docker on overnight to see if it would download and it was still on the same screen when I woke up, so unless it’s supposed to take forever to download, i’m stuck 😂
while someone else might guess what's going on. It might be good to include some logs 🙂
docker logs immich_machine_learning
those are the logs for the past hour and a half, i uncommented it in the yml/esv and let it run for a bit
Those logs look pretty normal apart from the looping. Could it be that the container is being killed because of running out of RAM?
it could be possible, i'm allocating basically everything to docker though so i have no clue why it's refusing to load lol
How much RAM do you have?
4 gigs on the one i’m running it off of. its not much but since i’m not using it for anything else i have it all it can spare allocated to the docker
Hmm, that might be just low enough that it will run out while starting up :/
Try setting
MACHINE_LEARNING_EAGER_STARTUP=false
, that should at least let it start up
Though you might still run into problems when you start running the ML jobsI have similar question: can i host ML container on my pc with gpu and other containers (non gpu intensive) on another machine like RPi4? So that i keep RPi4 running all the time, but fire up my pc once a day to finish processing ML jobs.
yes you can
GPU isn't supported yet, right?
But the rest is possible
Check out this thread, I'm doing same, running ML on a different machine.
https://discord.com/channels/979116623879368755/991643870523830382/1126919469617397880
this is actually very helpful, i didn’t realize you could offload the job to another pc! instead of trying to figure out how to get it working on my (relatively slow) immich setup, i’ll just send all the jobs to my main pc with 16 gigs of ram LOL thank you so much!