Title recognition

[neverlatetolearn0]

When I call the server mineru service using the vlm-http-client method, how can I configure it to achieve the configuration using title_aided_config in mineru.json

[dosubot[bot]]

To enable title recognition with vlm-http-client, set up your mineru.json like this:

{
  "llm-aided-config": {
    "title_aided": {
      "api_key": "your_api_key",
      "base_url": "https://dashscope.aliyuncs.com/compatible-mode/v1", // or your OpenAI-compatible endpoint
      "model": "qwen3-next-80b-a3b-instruct",
      "enable": true
    }
  }
}

You only need to configure this on the server side. MinerU loads mineru.json automatically (or from the path set by the MINERU_TOOLS_CONFIG_JSON environment variable) and applies the title_aided settings when you make API requests with backend=vlm-http-client and server_url pointing to your vLLM server (e.g., http://127.0.0.1:30000). No extra client-side parameters are needed—just ensure the config is correct and loaded by the server details & example.

If title_aided is enabled, MinerU will use your specified LLM to optimize and assign hierarchical levels to detected titles, resulting in multi-level Markdown output code logic.

Common issues: mineru.json not loaded (wrong path, missing env var), incorrect config structure, missing/invalid api_key/base_url/model, or unsupported fields like enable_thinking. If your config isn’t in the default location, set MINERU_TOOLS_CONFIG_JSON before starting the server config loading.

Version compatibility matters: recommended are MinerU 2.5.3+, mineru_vl_utils >=0.1.7, vllm >=0.10.2 version notes.

If you want to avoid automatic model downloads, manually download required models and set MINERU_MODEL_SOURCE=local and models-dir in mineru.json model advice.

Once configured, title recognition will work automatically when you use vlm-http-client.

To reply, just mention @dosu.

---

Docs are dead. Just use Dosu.