# NVIDIA GPU overlay: whisper.cpp with CUDA offload instead of the CPU build.
#
#   docker compose -f docker-compose.yml -f docker-compose.gpu.yml up -d --build
#
# Needs the NVIDIA Container Toolkit on the host (`nvidia-ctk`), which is what
# makes `deploy.resources.reservations.devices` work. Check it first:
#
#   docker run --rm --gpus all nvidia/cuda:12.6.3-base-ubuntu22.04 nvidia-smi
#
# This builds a SECOND whisper.cpp (the runtime-cuda target) and pulls a CUDA
# base image — expect several GB and a long first build. The CPU image is
# untouched and remains the default; nothing here is required to run an archive.
#
# AMD/ROCm and Apple Metal are not covered: whisper.cpp supports both, but each
# needs its own base image and device plumbing. Build a variant of the
# `whisper-cpu` stage with the right -DGGML_* flag if you need one.

services:
  editor:
    build:
      target: runtime-cuda
    image: archilyzer:${ARCHILYZER_TAG:-local}-cuda
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: all
              capabilities: [gpu]

  # The other three run the same image so `docker compose ... up` never rebuilds
  # a second variant behind your back. None of them touches the GPU.
  site:
    build:
      target: runtime-cuda
    image: archilyzer:${ARCHILYZER_TAG:-local}-cuda
  homepage:
    build:
      target: runtime-cuda
    image: archilyzer:${ARCHILYZER_TAG:-local}-cuda
  umtool:
    build:
      target: runtime-cuda
    image: archilyzer:${ARCHILYZER_TAG:-local}-cuda