{
  "runtime": {
    "llama_server": "E:\\ollama\\模型\\llama-b8642-bin-win-cuda-13.1-x64\\llama-server.exe",
    "ollama_url": "http://127.0.0.1:11434",
    "host": "127.0.0.1",
    "port": 8787,
    "default_model": "apex-qwen3",
    "model_dir": "E:\\model",
    "runtime_dir": "E:\\llm-backend\\runtime",
    "cache_dir": "E:\\llm-backend\\cache",
    "n_gpu_layers": "auto",
    "ctx_size": 32768,
    "batch_size": 2048,
    "ubatch_size": 512,
    "flash_attn": "auto",
    "fit_vram": true,
    "auto_load": false,
    "api_key": null,
    "max_loaded_models": 1,
    "keep_alive_seconds": 300
  },
  "models": [
    {
      "id": "apex-qwen3",
      "kind": "transformers",
      "path": "E:\\ollama\\Apex_AI-catmodel",
      "enabled": true,
      "description": "本机 Qwen3 safetensors 模型",
      "options": {
        "torch_dtype": "float16",
        "device": "cuda",
        "device_map": "auto",
        "load_in_4bit": false,
        "attn_implementation": "sdpa"
      }
    },
    {
      "id": "gguf-local",
      "kind": "gguf",
      "path": "E:\\model\\example.gguf",
      "enabled": false,
      "description": "将 path 改成真实 GGUF 文件后启用",
      "options": {}
    },
    {
      "id": "onnx-local",
      "kind": "onnx",
      "path": "E:\\model\\example.onnx",
      "enabled": false,
      "description": "ONNX Runtime/Optimum 扩展入口",
      "options": {}
    },
    {
      "id": "mistral-nemo-ollama",
      "kind": "ollama",
      "remote_model": "mistral-nemo-12b:latest",
      "enabled": false,
      "description": "兼容本机 Ollama 模型仓库；不复制权重"
    },
    {
      "id": "deepseek-cloud",
      "kind": "cloud",
      "cloud_profile": "deepseek",
      "remote_model": "deepseek-chat",
      "enabled": false,
      "description": "OpenAI 兼容云端模型"
    }
  ],
  "cloud_profiles": [
    {
      "id": "deepseek",
      "base_url": "https://api.deepseek.com",
      "api_key_env": "DEEPSEEK_API_KEY",
      "default_model": "deepseek-chat",
      "timeout": 120
    }
  ]
}