Hi,
First of all, thank you again for maintaining this excellent port of Immich for FreeBSD/Daemonless! It is highly appreciated.
I am running Immich on an older, but extremely reliable HP MicroServer N40L, which features an AMD Turion II Neo N40L processor. This CPU does not support the AVX (Advanced Vector Extensions) instruction set.
When running the current ghcr.io/daemonless/immich-ml:latest image, the machine learning container enters a crash loop, throwing SIGILL (Illegal Instruction) errors because the underlying libraries (ONNX Runtime / PyTorch) try to execute AVX instructions.
podman logs immich_ml | tail -50
[INFO] Booting worker with pid: 32878
[INFO] Control socket listening at /config/.gunicorn/gunicorn.ctl
[WARNING] Worker (pid:32878) was sent SIGILL!
[INFO] Booting worker with pid: 32901
[WARNING] Worker (pid:32901) was sent SIGILL!
Earlier version of ML was fine with my CPU but the new one is failing.
I know the new software now requires AVX (Advanced Vector Extensions) or AVX-512 instruction sets, whereas the older version did not.
Request
Would it be possible to add a -no-avx build/tag for the immich-ml image, similar to the official upstream Immich project? This would allow users with older x86-64 hardware (without AVX support) to utilize the machine learning features on FreeBSD.
Thank you for your time and help!
Hi,
First of all, thank you again for maintaining this excellent port of Immich for FreeBSD/Daemonless! It is highly appreciated.
I am running Immich on an older, but extremely reliable HP MicroServer N40L, which features an AMD Turion II Neo N40L processor. This CPU does not support the AVX (Advanced Vector Extensions) instruction set.
When running the current
ghcr.io/daemonless/immich-ml:latestimage, the machine learning container enters a crash loop, throwingSIGILL(Illegal Instruction) errors because the underlying libraries (ONNX Runtime / PyTorch) try to execute AVX instructions.podman logs immich_ml | tail -50
[INFO] Booting worker with pid: 32878
[INFO] Control socket listening at /config/.gunicorn/gunicorn.ctl
[WARNING] Worker (pid:32878) was sent SIGILL!
[INFO] Booting worker with pid: 32901
[WARNING] Worker (pid:32901) was sent SIGILL!
Earlier version of ML was fine with my CPU but the new one is failing.
I know the new software now requires AVX (Advanced Vector Extensions) or AVX-512 instruction sets, whereas the older version did not.
Request
Would it be possible to add a
-no-avxbuild/tag for theimmich-mlimage, similar to the official upstream Immich project? This would allow users with older x86-64 hardware (without AVX support) to utilize the machine learning features on FreeBSD.Thank you for your time and help!