host: 0.0.0.0
  port: 8001

# takes about 1:30 minutes to change the cluster to RUNNING mode on EC2  g5.2xlarge
applications:
- import_path: ray.serve.llm:build_openai_app
  name: Multimodal LLMs
  route_prefix: "/"
  args:
    llm_configs:
      - model_loading_config:
          model_id: deepseek-ai/DeepSeek-OCR
          model_source: deepseek-ai/DeepSeek-OCR # Qwen/Qwen2.5-VL-7B-Instruct
        engine_kwargs:
          max_model_len: 8192 # 12288   # safe for 24GB VRAM
          # dtype: bfloat16
          gpu_memory_utilization: 0.9
  ray_actor_options:
    resources: {"gpu": 1}  # Use default GPU resource```

### Issue Severity

None