Immich ML not working with rocm?
❓Question☑️Ready
My ML won't work. I have rocm 7.2 installed on the main machine, do you think I should downgrade to the same version as the docker image or that's not the problem? Thanks!
The container has access to the gpu vram...
My immich ML log is:
Memory access fault by GPU node-1 (Agent handle: 0x73b131cf3180) on address 0x73b175036000. Reason: Page not present or supervisor privilege.
[02/03/26 12:34:49] ERROR Worker (pid:1387) was sent code 134!
[02/03/26 12:34:49] INFO Booting worker with pid: 1426
[02/03/26 12:34:50] INFO Started server process [1426]
[02/03/26 12:34:50] INFO Waiting for application startup.
[02/03/26 12:34:50] INFO Created in-memory cache with unloading after 300s
of inactivity.
[02/03/26 12:34:50] INFO Initialized request thread pool with 16 threads.
[02/03/26 12:34:50] INFO Application startup complete.
[02/03/26 12:34:50] INFO Loading visual model
'ViT-SO400M-16-SigLIP2-384__webli' to memory
[02/03/26 12:34:50] INFO Setting execution providers to
['ROCMExecutionProvider', 'CPUExecutionProvider'],
in descending order of preference
Memory access fault by GPU node-1 (Agent handle: 0x73b12dcf3720) on address 0x73b147628000. Reason: Page not present or supervisor privilege.
The container has access to the gpu vram...
My immich ML log is:
Memory access fault by GPU node-1 (Agent handle: 0x73b131cf3180) on address 0x73b175036000. Reason: Page not present or supervisor privilege.
[02/03/26 12:34:49] ERROR Worker (pid:1387) was sent code 134!
[02/03/26 12:34:49] INFO Booting worker with pid: 1426
[02/03/26 12:34:50] INFO Started server process [1426]
[02/03/26 12:34:50] INFO Waiting for application startup.
[02/03/26 12:34:50] INFO Created in-memory cache with unloading after 300s
of inactivity.
[02/03/26 12:34:50] INFO Initialized request thread pool with 16 threads.
[02/03/26 12:34:50] INFO Application startup complete.
[02/03/26 12:34:50] INFO Loading visual model
'ViT-SO400M-16-SigLIP2-384__webli' to memory
[02/03/26 12:34:50] INFO Setting execution providers to
['ROCMExecutionProvider', 'CPUExecutionProvider'],
in descending order of preference
Memory access fault by GPU node-1 (Agent handle: 0x73b12dcf3720) on address 0x73b147628000. Reason: Page not present or supervisor privilege.
docker_compose.yml2.26KB
