services:
  msquant:
    build:
      context: ..
      dockerfile: docker/Dockerfile.gpu
    image: ghcr.io/MissionSquad/msquant:dev   # replace OWNER with your GitHub org/user
    container_name: msquant
    ports:
      - "8080:8080"
    environment:
      PYTORCH_CUDA_ALLOC_CONF: "expandable_segments:True,max_split_size_mb:64"
      HF_HOME: /workspace/hf
      HF_DATASETS_CACHE: /workspace/hf/datasets
      HF_HUB_ENABLE_HF_TRANSFER: "1"
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              device_ids: ["0","2"]     # adjust for your host
              capabilities: ["gpu"]
    volumes:
      - ../hf:/workspace/hf
      - ../out:/workspace/out
    runtime: nvidia