Handles ML inference for facial recognition, image classification, and other enrichment tasks. It pulls the same Immich version tag as the main server and caches downloaded models under the model-cache volume.【F:immich_base/compose.yml†L33-L56】
ghcr.io/immich-app/immich-machine-learning:${IMMICH_VERSION:-release}backend_netmodel-cache:/cache/dev/dri for hardware accelerationalwayshwaccel.ml.yml if you need CUDA, ROCm, or other acceleration profiles.model-cache to avoid re-downloading large ML assets after rebuilds.