Storyteller on Unraid transcribes on CPU instead of the Nvidia GPU despite `--runtime=nvidia`

I moved my instance to Unraid to take advantage of GPU acceleration and added --runtime='nvidia' to the container's Extra Parameters, but transcription never touches the GPU. There is no GPU utilisation while a book is aligning, and the logs show no errors at all.

Resolution: Hardware acceleration lives in a separate image, and the Unraid community template still points at the plain non-accelerated repository, so the container just silently falls back to CPU when it cannot find a GPU. Change the container's Repository field to the CUDA image, e.g. registry.gitlab.com/storyteller-platform/storyteller:latest-cuda-12.9.0 or registry.gitlab.com/storyteller-platform/storyteller:latest-cuda-13.1.0, choosing by card generation -- CUDA 12.9 for older cards such as Pascal/GTX 10xx and Quadro P-series, 13.1 for anything newer than the 2000 series. Keep --runtime=nvidia in Extra Parameters and add two container variables: NVIDIA_VISIBLE_DEVICES set to your GPU's id (the GPU-xxxxxxxx UUID) and NVIDIA_DRIVER_CAPABILITIES set to all. --device=/dev/dri is not required for Nvidia cards. Also work through the GPU configuration page in the Storyteller docs; this combination was confirmed working on both a Quadro P2000 and a GTX 1080.