# One GPU: vLLM serves the weights, openjev-server exposes the decision API on :3000.
services:
  vllm:
    image: vllm/vllm-openai:v0.29.0
    command: >
      --model openjev/openjev-FP8 --served-model-name qwen --port 8000 --enable-prefix-caching --max-model-len 16384
      --gpu-memory-utilization 0.90 --limit-mm-per-prompt '{"image":1}' --trust-remote-code --max-num-seqs 64
      --max-logprobs 64 --gdn-prefill-backend triton --quantization fp8
    deploy: { resources: { reservations: { devices: [{ driver: nvidia, count: 1, capabilities: [gpu] }] } } }
    volumes: ["~/.cache/huggingface:/root/.cache/huggingface"]
  openjev:
    build: .
    depends_on: [vllm]
    ports: ["3000:3000"]
    environment:
      OPENJEV_TOKEN: ${OPENJEV_TOKEN:-}