docker-compose NVIDIA-Nemotron-3-Super-120B-A12B-NVFP4

docker-compose.yml

services:
  vllm:
    image: vllm/vllm-openai:v0.17.1-cu130
    restart: unless-stopped
    container_name: vllm-nemotron
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: all
              capabilities: [gpu]
    environment:
      - NVIDIA_VISIBLE_DEVICES=all
      - VLLM_WORKER_MULTIPROC_METHOD=spawn
    volumes:
      - /home/mg/.cache/huggingface:/root/.cache/huggingface
    ports:
      - "8000:8000"
    ipc: host
    command: >
      --model nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-NVFP4
      --port 8000
      --trust-remote-code