[Feature] Add knobs to limit CPU usage on ML container

[jbtrystram]

I have searched the existing feature requests, both open and closed, to make sure this is not a duplicate request.

• Yes

The feature

I use my workstation to run the ML container, as it has a dedicated GPU. It work nicely, but i switched to an AMD gpu recently and the rocm container use the GPU and 100% of the CPU. Which prevents me to use my machine at the same time, as it becomes too slow.
It would be nice to add some threads limits as an env variable, as i see the app uses everything available :

juil. 07 17:03:36 device-129.home immich_ml[168095]: [07/07/25 15:03:36] INFO     Starting gunicorn 23.0.0                           
juil. 07 17:03:36 device-129.home immich_ml[168095]: [07/07/25 15:03:36] INFO     Listening at: http://[::]:3003 (4)                 
juil. 07 17:03:36 device-129.home immich_ml[168095]: [07/07/25 15:03:36] INFO     Using worker: immich_ml.config.CustomUvicornWorker 
juil. 07 17:03:36 device-129.home immich_ml[168095]: [07/07/25 15:03:36] INFO     Booting worker with pid: 5                         
juil. 07 17:03:38 device-129.home immich_ml[168095]: [07/07/25 15:03:38] INFO     Started server process [5]                         
juil. 07 17:03:38 device-129.home immich_ml[168095]: [07/07/25 15:03:38] INFO     Waiting for application startup.                   
juil. 07 17:03:38 device-129.home immich_ml[168095]: [07/07/25 15:03:38] INFO     Created in-memory cache with unloading after 300s  
juil. 07 17:03:38 device-129.home immich_ml[168095]:                              of inactivity.                                     
juil. 07 17:03:38 device-129.home immich_ml[168095]: [07/07/25 15:03:38] INFO     Initialized request thread pool with 20 threads.   
juil. 07 17:03:38 device-129.home immich_ml[168095]: [07/07/25 15:03:38] INFO     Application startup complete.                      
juil. 07 17:03:46 device-129.home immich-ml[168064]: 8cb1e03741d01ecbd94f26c4132500659fcd296c679eb54ffd90a0bbe9f4f62b
juil. 07 17:03:46 device-129.home systemd[2432]: Started immich-ml.service.
juil. 07 17:04:14 device-129.home immich_ml[168095]: [07/07/25 15:04:14] INFO     Loading detection model 'buffalo_l' to memory      
juil. 07 17:04:14 device-129.home immich_ml[168095]: [07/07/25 15:04:14] INFO     Setting execution providers to                     
juil. 07 17:04:14 device-129.home immich_ml[168095]:                              ['ROCMExecutionProvider', 'CPUExecutionProvider'], 
juil. 07 17:04:14 device-129.home immich_ml[168095]:                              in descending order of preference                  
juil. 07 17:04:25 device-129.home immich_ml[168095]: [07/07/25 15:04:25] INFO     Loading recognition model 'buffalo_l' to memory    
juil. 07 17:04:25 device-129.home immich_ml[168095]: [07/07/25 15:04:25] INFO     Setting execution providers to                     
juil. 07 17:04:25 device-129.home immich_ml[168095]:                              ['ROCMExecutionProvider', 'CPUExecutionProvider'], 
juil. 07 17:04:25 device-129.home immich_ml[168095]:                              in descending order of preference                  
juil. 07 18:02:35 bazzite immich_ml[168095]: [07/07/25 16:02:35] INFO     Shutting down due to inactivity.                   
juil. 07 18:02:35 bazzite immich_ml[168095]: [07/07/25 16:02:35] INFO     Shutting down                                      
juil. 07 18:02:35 bazzite immich_ml[168095]: [07/07/25 16:02:35] INFO     Waiting for application shutdown.                  
juil. 07 18:02:35 bazzite immich_ml[168095]: [07/07/25 16:02:35] INFO     Application shutdown complete.                     
juil. 07 18:02:35 bazzite immich_ml[168095]: [07/07/25 16:02:35] INFO     Finished server process [5]                        
juil. 07 18:02:35 bazzite immich_ml[168095]: [07/07/25 16:02:35] ERROR    Worker (pid:5) was sent SIGINT!                    
juil. 07 18:02:35 bazzite immich_ml[168095]: [07/07/25 16:02:35] INFO     Booting worker with pid: 222                       
juil. 07 18:02:36 bazzite immich_ml[168095]: [07/07/25 16:02:36] INFO     Started server process [222]                       
juil. 07 18:02:36 bazzite immich_ml[168095]: [07/07/25 16:02:36] INFO     Waiting for application startup.                   
juil. 07 18:02:36 bazzite immich_ml[168095]: [07/07/25 16:02:36] INFO     Created in-memory cache with unloading after 300s  
juil. 07 18:02:36 bazzite immich_ml[168095]:                              of inactivity.                                     
juil. 07 18:02:36 bazzite immich_ml[168095]: [07/07/25 16:02:36] INFO     Initialized request thread pool with 20 threads.   
juil. 07 18:02:36 bazzite immich_ml[168095]: [07/07/25 16:02:36] INFO     Application startup complete.         

Platform

• Server
• Web
• Mobile

[bo0tzz]

There are already env vars to tune this, see https://immich.app/docs/install/environment-variables#machine-learning.

[jbtrystram]

Nice, thanks !