[Feature] Upgrade to ONNX Runtime 1.27.0 for official CUDA 13 support

[altheali533]

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

• Yes

The feature

I have searched the existing feature requests to make sure this is not a duplicate request.
Yes

The feature

ONNX Runtime v1.27.0 was just released and it's a significant milestone for CUDA 13 support:

• Support for CUDA 12 is being removed starting in 1.27.0
• CUDA 13 is now published as a stable package: onnxruntime-<os>-<arch>-gpu_cuda13-<version>
• onnxruntime-gpu 1.27.0 wheels for CUDA 13 are already live on PyPI

Release notes: https://github.com/microsoft/onnxruntime/releases/tag/v1.27.0

Previously, the blocker for Immich adding CUDA 13 support was that onnxruntime only had CUDA 13 as an unstable nightly build (no official stable release). That blocker is now resolved.

Immich's machine-learning/Dockerfile currently builds the CUDA image from nvidia/cuda:12.2.2-runtime-ubuntu22.04, which causes failures for users running newer NVIDIA drivers that only ship CUDA 13 (see #21910, #21911, and the arm64/DGX Spark discussion #10647). Many users are currently forced to either downgrade their host driver to a CUDA 12.x-compatible version or patch a custom Dockerfile themselves.

Describe the solution you'd like

Upgrade the machine-learning service to onnxruntime 1.27.0 and switch the -cuda image's base to a CUDA 13 runtime image (e.g. nvidia/cuda:13.x-runtime-ubuntu24.04 or similar), following the same approach as the recent ONNX Runtime bump PR (#23458).

This would:

• Fix CUDA execution provider failures for users on CUDA 13 / driver 580+ hosts
• Remove the need for manual LD_LIBRARY_PATH workarounds
• Enable native CUDA support for newer hardware (e.g. Blackwell-based GPUs, DGX Spark)

Describe alternatives you've considered

Currently working around this by pinning host NVIDIA driver/CUDA to a 12.x-compatible version, or building a custom machine-learning image from source. Neither is ideal for a "batteries included" self-hosted deployment.

Additional context

• [Feature Request] Support CUDA 13.0 microsoft/onnxruntime#26238 (original CUDA 13 feature request)
• Nightly packages for Python 3.14 or CUDA 13.x available for testing microsoft/onnxruntime#26547 (nightly CUDA 13.x packages announcement)
• immich-machine-learning CUDA fail #21910 / immich-machine-learning CUDA fail #21911 (CUDA fail on host CUDA 13)
• [Feature] Support for arm64 for CUDA (immich-machine-learning) #10647 (arm64 CUDA 13 / DGX Spark support request)

Platform

• Server
• Web
• Mobile