Immich 2.5.2 Machine Learning Broken (OpenVINO / Intel)

[simplicite72]

I have searched the existing issues, both open and closed, to make sure this is not a duplicate report.

• Yes

The bug

Dear all,

I would like to report an issue that I noticed very recently (possibly less than a week ago), likely following the update to latest release of Immich (docker).

My Immich instance is configured to use Intel integrated GPU (Jasper Lake) via OpenVINO and has been working like a charm for several months flawlessly to accelerate facial recognition.

However, since recent update, facial recognition does not work anymore. I checked the log of the immich_machine_learning and I found error linked to OpenVINO.

I'm looking for other experiences and feedbacks regarding possibly similar issues.

The OS that Immich Server is running on

Debian 12

Version of Immich Server

v.2.5.2

Version of Immich Mobile App

v2.5.2

Platform with the issue

• Server
• Web
• Mobile

Device make and model

No response

Your docker-compose.yml content

services:
  immich-server:
    container_name: immich_server
    image: ghcr.io/immich-app/immich-server:release
     # extends:
    #   file: hwaccel.transcoding.yml
    #   service: cpu # set to one of [nvenc, quicksync, rkmpp, vaapi, vaapi-wsl] for accelerated 
    shm_size: '512m'
    command: ["start.sh", "immich", "--hostname", "0.0.0.0"]
    volumes:
      - ${UPLOAD_LOCATION}:/usr/src/app/upload
    env_file:
      - .env
    ports:
      - "${IMMICH_PORT}:2283"
    devices:
      - /dev/dri:/dev/dri
    depends_on:
      - redis
      - database
    restart: always
    networks:
      - big_bear_immich_network

  immich-microservices:
    container_name: immich_microservices
    image: ghcr.io/immich-app/immich-server:release
    shm_size: '512m'
    devices:
      - /dev/dri:/dev/dri
    env_file:
      - .env
    depends_on:
      - redis
      - database
    restart: always
    networks:
      - big_bear_immich_network

  immich-machine-learning:
    container_name: immich_machine_learning
    image: ghcr.io/immich-app/immich-machine-learning:release-openvino
    extends:
     file: hwaccel.ml.yml
     service: openvino
    volumes:
      - /DATA/AppData/big-bear-immich/model-cache:/cache
    environment:
      - DB_HOSTNAME=immich_postgres
      - DB_USERNAME=
      - DB_PASSWORD=
      - DB_DATABASE_NAME=immich
      - REDIS_HOSTNAME=immich_redis
    restart: always
    networks:
      - big_bear_immich_network

  redis:
    container_name: immich_redis
    image: redis:6.2-alpine
    restart: always
    networks:
      - big_bear_immich_network

  database:
    container_name: immich_postgres
    image: ghcr.io/immich-app/postgres:14-vectorchord0.4.3-pgvectors0.2.0
    env_file:
      - .env
    volumes:
      - /DATA/AppData/big-bear-immich/pgdata:/var/lib/postgresql/data
    restart: always
    networks:
      - big_bear_immich_network

networks:
  big_bear_immich_network:
    name: big-bear-immich_big_bear_immich_network
    external: true

Your .env content

N/A

Reproduction steps

I just noticed that no faces were recognized on newly uploaded photos

Relevant log output

026-02-02 15:16:07.894403429 [E:onnxruntime:, inference_session.cc:2544 operator()] Exception during initialization: /onnxruntime/onnxruntime/core/providers/openvino/backends/basic_backend.cc:135 onnxruntime::openvino_ep::BasicBackend::BasicBackend(std::unique_ptr<onnx::ModelProto>&, onnxruntime::openvino_ep::SessionContext&, const onnxruntime::openvino_ep::SubGraphContext&, onnxruntime::openvino_ep::SharedContext&, onnxruntime::openvino_ep::ptr_stream_t&) /onnxruntime/onnxruntime/core/providers/openvino/ov_interface.cc:24 auto onnxruntime::openvino_ep::OvExceptionBoundary(Func&&, std::format_string<_Args ...>&&, Args&& ...) [with Func = OVCore::CompileModel(std::shared_ptr<const ov::Model>&, std::string&, ov::AnyMap&, bool, const std::string&)::<lambda()>; Args = {const std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >&}; std::format_string<_Args ...> = std::basic_format_string<char, const std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >&>] [OpenVINO-EP] Exception while Loading Network for graph OpenVINOExecutionProvider_OpenVINO-EP-subgraph_1_0: Exception from src/inference/src/cpp/core.cpp:109:
Exception from src/inference/src/dev/plugin.cpp:53:
Exception from src/plugins/intel_gpu/src/runtime/ocl/ocl_engine.cpp:205:
[GPU] clCreateSubBuffer, error code: -30 CL_INVALID_VALUE

Additional information

Configurations for hardware-accelerated machine learning

If using Unraid or another platform that doesn't allow multiple Compose files,

you can inline the config for a backend by copying its contents

into the immich-machine-learning service in the docker-compose.yml file.

See https://immich.app/docs/features/ml-hardware-acceleration for info on usage.

services:
armnn:
devices:
- /dev/mali0:/dev/mali0
volumes:
- /lib/firmware/mali_csffw.bin:/lib/firmware/mali_csffw.bin:ro # Mali firmware for your chipset (not always required depending on the driver)
- /usr/lib/libmali.so:/usr/lib/libmali.so:ro # Mali driver for your chipset (always required)

rknn:
security_opt:
- systempaths=unconfined
- apparmor=unconfined
devices:
- /dev/dri:/dev/dri

cpu: {}

cuda:
deploy:
resources:
reservations:
devices:
- driver: nvidia
count: 1
capabilities:
- gpu

rocm:
group_add:
- video
devices:
- /dev/dri:/dev/dri
- /dev/kfd:/dev/kfd

openvino:
device_cgroup_rules:
- 'c 189:* rmw'
devices:
- /dev/dri:/dev/dri
volumes:
- /dev/bus/usb:/dev/bus/usb

openvino-wsl:
devices:
- /dev/dri:/dev/dri
- /dev/dxg:/dev/dxg
volumes:
- /dev/bus/usb:/dev/bus/usb
- /usr/lib/wsl:/usr/lib/wsl

[simplicite72]

I confirm this is an issue linked to the a recent relase of Immich.

I just tried with a previous version of immich_machine_learning (2.4.1 Openvino) with Immich-server 2.5.2

🟢 Face detection + recognition works now well with GPU (machine accelerated) immediately when pictures are uploaded
🟢 Recognised faces (persons) appear on Immich Web

immich-machine-learning: container_name: immich_machine_learning image: ghcr.io/immich-app/immich-machine-learning:v2.4.1-openvino extends: file: hwaccel.ml.yml service: openvino

[RDshift]

I am seeing similar behavior on unraid with the xe kernel driver manually enabled as well, except 2.4.1 also fails for me for GPU due to xe. Intel Arc A310.
Retested on both v2.5.3-all (with -openvino) and 2.5.3+2.4.1ML-openvino.

#
# WARNING: To install Immich, follow our guide: https://immich.app/docs/install/docker-compose
#
# Make sure to use the docker-compose.yml of the current release:
#
# https://github.com/immich-app/immich/releases/latest/download/docker-compose.yml
#
# The compose file on main may not be compatible with the latest release.

name: immich

services:
  immich_server:
    container_name: immich_server
    image: ghcr.io/immich-app/immich-server:${IMMICH_VERSION:-release}
    #extends: same in ml
    # file: hwaccel.transcoding.yml
    # file: hwaccel.ml.yml
      # service: vaapi
      # service: openvino
    devices:
      - /dev/dri:/dev/dri
    volumes:
      # Do not edit the next line. If you want to change the media storage location on your system, edit the value of UPLOAD_LOCATION in the .env file
      - ${UPLOAD_LOCATION}:/data
      - /etc/localtime:/etc/localtime:ro
#      - immich-generated:/usr/src/app/upload
    shm_size: '8g'
    environment:
      - REDIS_PASSWORD=${REDIS_PASSWORD}
      - LIBVA_DRIVER_NAME=iHD
#      - DB_SKIP_MIGRATIONS=true
    env_file:
      - .env
#    ports:
#      - '2283:2283'
# ADD THESE LINES (at the same indentation level as 'restart: always')
    networks:
#      unraid-br0:
#        ipv4_address: 192.168.1.104  # <-- CHOOSE A STATIC IP FOR IMMICH
      domain-internal-dmz:
      immich-net:
    depends_on:
      - redis
      - immich_postgres
    restart: always
    healthcheck:
      disable: false

  immich_machine_learning:
    container_name: immich_machine_learning
    # Explicitly using the OpenVINO image
    # image: ghcr.io/immich-app/immich-machine-learning:v2.4.1-openvino
    image: ghcr.io/immich-app/immich-machine-learning:${IMMICH_VERSION:-release}-openvino
    devices:
      - /dev/dri:/dev/dri
    volumes:
      - model-cache:/cache
    shm_size: '8g'
    environment:
      - REDIS_PASSWORD=${REDIS_PASSWORD}
      # Primary flag to tell Immich to look for the GPU
      - IMMICH_ML_DEVICE=openvino
      # Explicit OpenVINO device target (GPU.0 targets the first render device)
      - OPENVINO_DEVICE=GPU.0
      # Prevent OOM on 4GB Card
      - MACHINE_LEARNING_WORKERS=1
    env_file:
      - .env
    restart: always
    healthcheck:
      disable: false
    networks:
      - immich-net

  redis:
    container_name: redis
#    image: docker.io/valkey/valkey:8-bookworm@sha256:a137a2b60aca1a75130022d6bb96af423fefae4eb55faf395732db3544803280
#    image: valkey/valkey:alpine
    image: valkey/valkey:7-alpine
    environment:
      - REDIS_PASSWORD=${REDIS_PASSWORD}
    healthcheck:
      test: redis-cli -a $$REDIS_PASSWORD ping || exit 1
    restart: always
    volumes:
      - ${REDIS_LOCATION}:/usr/local/etc/valkey
      - valkey-data-v2:/data
    command: valkey-server --requirepass ${REDIS_PASSWORD} --save 60 1
    networks:
      - immich-net

  immich_postgres:
    container_name: immich_postgres
    #image: ghcr.io/immich-app/postgres:16-vectorchord0.4.3-pgvector0.8.1-pgvectors0.3.0
    image: ghcr.io/immich-app/postgres:16-vectorchord0.5.3-pgvector0.8.1
    environment:
      POSTGRES_PASSWORD: ${DB_PASSWORD}
      POSTGRES_USER: ${DB_USERNAME}
      POSTGRES_DB: ${DB_DATABASE_NAME}
      POSTGRES_INITDB_ARGS: '--data-checksums'
      DB_STORAGE_TYPE: 'SSD'
    volumes:
      - ${DB_DATA_LOCATION}:/var/lib/postgresql/data
    
    # Keep the memory limit
    shm_size: '10g'
    command: >-
     postgres 
      -c shared_preload_libraries='vchord.so' 
      -c search_path="$$user,public,vectors" 
      -c shared_buffers=35GB 
      -c maintenance_work_mem=16GB 
      -c work_mem=4GB 
      -c effective_cache_size=64GB 
      -c max_parallel_maintenance_workers=24 
      -c max_worker_processes=56 
      -c max_parallel_workers=24 
      -c random_page_cost=1.01
      -c effective_io_concurrency=500 
      -c wal_compression=on 
      -c max_wal_size=8GB 
      -c min_wal_size=4GB

    restart: always
    networks:
      - immich-net
    stop_grace_period: 2m
    
volumes:
  immich-generated:
    driver_opts:
      type: none
      o: bind
      device: /dev/shm/immich-generated
  model-cache:
    name: immich_model-cache
  pgdata:
    name: immich_pgdata
  valkey-data-v2:
    name: immich_valkey-data-v2
    
networks:
  immich-net:
    driver: bridge
    external: false
#  unraid-br0:
#    external: true
#    name: br0
  domain-internal-dmz:
    driver: bridge
    external: true

[joaquimBS]

Same here, 2.5.x broke ML capabilities. I am running 2.4.1 and now it works again. Synology DS423+.

[bleomycin]

Can also confirm the breaking change here. Thought I was losing my mind for a bit.

[gianmarcos96]

I have updated to 2.5.5 and still have the server error when i search something. My log is different, may someone can help me?

Nest] 25 - 02/06/2026, 8:16:39 PM WARN [Api:MachineLearningRepository~gfcq7p7w] Machine learning request to "http://immich-machine-learning:3003" failed: fetch failed [Nest] 25 - 02/06/2026, 8:16:39 PM ERROR [Api:ErrorInterceptor~gfcq7p7w] Unknown error: Error: Machine learning request '{"clip":{"textual":{"modelName":"ViT-L-16-SigLIP-256__webli","options":{"language":"it-null"}}}}' failed for all URLs Error: Machine learning request '{"clip":{"textual":{"modelName":"ViT-L-16-SigLIP-256__webli","options":{"language":"it-null"}}}}' failed for all URLs at MachineLearningRepository.predict (/usr/src/app/server/dist/repositories/machine-learning.repository.js:117:15) at process.processTicksAndRejections (node:internal/process/task_queues:103:5) at async MachineLearningRepository.encodeText (/usr/src/app/server/dist/repositories/machine-learning.repository.js:140:26) at async SearchService.searchSmart (/usr/src/app/server/dist/services/search.service.js:94:29)

[mikesnav]

Also experiencing this - was fine on 2.4.1 and broke when I moved up to 2.5.x

[Snuupy]

same but on rocm/gfx 9.00 / 5825U

ERROR [Api:ErrorInterceptor~iwgbq60k] Unknown error: Error: Machine learning request '{"clip":{"textual":{"modelName":"ViT-B-32__openai","options":{"language":"en-US"}}}}' failed for all URLs

edit: this is the issue #24734

[gianmarcos96]

Now I have updated to 2.5.6 and then i ran the process of machine learning for smart search. Now the search is working fine. I will check better at the end of the process

Update: after the process of smart search, now it works fine for me. 2.5.6 solved my issue

[MisterBolt]

I'm starting my adventure with Immich and i have the same error on proxmox with VM - Ubuntu 24.04.3 LTS.
Tested Immich ML - 2.4.1/2.5.5/2.5.6
I have Gemini Lake processor with legacy support for GPU drivers installed from the instruction here: https://github.com/intel/compute-runtime/releases/tag/24.35.30872.22

[gmcinalli]

Running v2.5.6, facial recognition still broken with OpenVINO.

[Tryfos]

Ok, this is VERY brutal, especially for newcomers to Immich :) With a humble N5095 here I've been struggling with all this for the past 3 days. Jasper Lake quirks (the forgotten CPU line by Intel), low power stuff for HEVC, VP9 etc., legacy opencl or not (it is legacy!), kernel parameters (guc), vainfo, clinfo, Proxmox LXC passthrough and a WHOLE lot of other things. Made and nuked almost 10 CTs in Proxmox :). Also: Trixie does not have opencl in its stable repositories. Changed to SID (in production Proxmox!), just to download the legacy opencl....In the end clinfo was perfect in my LXC container, gpu passthrough ok, hardware transcoding worked just fine. One thing was constantly failing: Hardware ML (exception during initialization, exception in ASGI application, exceptions in general and a whole lot of other extremely low-level debugging errors).
Chatgpt was screaming and insisting, N5095 cannot do ML reliably, give up! As soon as I read this thread, I said to myself "one last hail Mary" and I sincerely give up. I changed to ML image 2.4.1. as initially suggested above and LO AND BEHOLD, HA ML is magically working with N5095!!! All this time it's just a buggy 2.5.6 (or earlier) ML image!!! Extremely exhausting for a newcomer to deal with this, because it's the last thing to calculate...
(Of course I'm going to go ahead and try all the subsequent image versions until ML doesn't work again :)
Devs keep up the GREAT work you're doing, but for the love of God: "What has changed since ML 2.4.1 image"!!!!???? :)))))))

Edit: Proxmox 9 with Debian 13 LXC containers.
Edit2: Now I have to sanity check if (legacy) opencl is indeed needed to run HA ML or not or if it was the image all along....
Edit3: It's not needed, oh my God.....:)
Edit4: Tested with 2.5.5, 2.5.3 and 2.5.0, HA ML fails in all of them. So last known working HA ML image is indeed 2.4.1. The question now is how long we're going to be able to have different Immich and ML image versions without problems...