compose.x.speaches.nvidia.yml

services:
  speaches:
    image: ghcr.io/speaches-ai/speaches:${HARBOR_SPEACHES_VERSION}-cuda
    # Alternative to image - build in place
    # However, service's Dockerfile only compatible with ubuntu24.04, so
    # you won't be able to use significantly older CUDA versions
    # build:
    #   context: https://github.com/speaches-ai/speaches.git
    #   dockerfile: Dockerfile
    #   args:
    #     BASE_IMAGE: nvidia/cuda:12.6.3-cudnn-runtime-ubuntu24.04
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: all
              capabilities: [gpu]