version: "3" services: Roboflow-GPU-1: image: roboflow/roboflow-inference-server-trt restart: always environment: - CUDA_VISIBLE_DEVICES=0 #A2 - They have device IDs - MODEL_CACHE_DIR=/tmp/cache volumes: - shared-volume:/tmp/cache deploy: resources: reservations: devices: - driver: nvidia device_ids: ['0'] capabilities: [gpu] Roboflow-GPU-2: image: roboflow/roboflow-inference-server-trt restart: always environment: - CUDA_VISIBLE_DEVICES=0 #A2 - They have device IDs - MODEL_CACHE_DIR=/tmp/cache volumes: - shared-volume:/tmp/cache deploy: resources: reservations: devices: - driver: nvidia device_ids: ['0'] capabilities: [gpu] Load-Balancer: image: lb restart: always environment: - CUDA_VISIBLE_DEVICES=0 - MODEL_CACHE_DIR=/tmp/cache ports: - 9001:80 volumes: shared-volume: