Encountered issues run the pipeline in agent tap when using different parser

[jacksonasl-07]

When i try to run a pipeline in agent tap, there are some issues when choose to use different parser (already install the minerU and docling):
MinerU: [ERROR][MinerU] Process failed with exit code -9
Naive: ValueError('Parser_0: PDF VLM language does not support empty value.')
Docling: [ERROR]Model(Docling@None) not authorized
How can i fix these issues?

[dosubot[bot]]

Hi @Yeon1007! I'm Dosu and I’m helping the ragflow team.

Here’s how to address each parser error:

MinerU: [ERROR][MinerU] Process failed with exit code -9
Exit code -9 means the MinerU process was killed, usually due to running out of memory or hitting resource limits. Try reducing the chunk task page size, processing one document at a time, and ensuring your server has enough RAM. If you’re running multiple tasks in parallel, lower the concurrency settings (e.g., set MAX_CONCURRENT_CHUNK_BUILDERS=1). Also, make sure you’re using compatible versions of MinerU and RAGFlow, and check for missing fields like 'bbox' in your PDF blocks if you’ve customized the code. More details: source, source.

Naive: ValueError('Parser_0: PDF VLM language does not support empty value.')
This error happens when the parser expects a non-empty language field for certain VLM-based PDF parsers. If you’re using a parser that requires a language (not DeepDoc, Plain Text, MinerU, or TCADP), make sure to specify the lang parameter in your pipeline config. For the naive parser, this usually means the PDF content is empty or not supported (e.g., image-only PDFs). Try using a stronger parser (like Docling or MinerU) for ingestion, or pre-process the PDF to extract text before uploading. See: source.

Docling: [ERROR]Model(Docling@None) not authorized
This usually means the Docling model is not properly installed or authorized. Make sure you have the docling Python package installed and set the environment variable USE_DOCLING=true. If you’re running in Docker, update your .env file and rebuild/restart the container. If you still see authorization errors, check if Docling requires additional credentials or configuration in your environment. Details: source.

For all parsers, after changing parser settings or updating environment variables, delete and re-upload your PDFs for the new parser to take effect. If issues persist, check backend logs and confirm all required services are running.

To reply, just mention @dosu.

---

Docs are dead. Just use Dosu.

[Ganesh-403]

@jacksonasl-07 Each of these errors seems to point to a different part of the parser configuration rather than a single root cause.

1. MinerU: Process failed with exit code -9

On Linux, exit code -9 usually means the process was terminated by SIGKILL, which is commonly caused by the OS killing it due to running out of memory or another resource limit.

If you're using MinerU, make sure:

• The MinerU service is correctly configured (MINERU_APISERVER or local backend).
• The server has sufficient RAM for the PDF being processed.
• Try processing a smaller PDF or reducing concurrent parsing jobs to see if the issue persists. RAGFlow supports MinerU as a remote parser service and documents the required environment variables for configuring it. :contentReference[oaicite:0]{index=0}

2. Naive: PDF VLM language does not support empty value

This one is a little different.

The Naive parser itself is intended for plain-text PDFs and skips OCR/layout analysis. However, if a VLM-based parser or PDF pipeline is selected, RAGFlow requires a default VLM (image-to-text model) to be configured. I suspect the pipeline is attempting to use VLM functionality, but the language/model configuration is incomplete, resulting in the empty-language error. :contentReference[oaicite:1]{index=1}

Could you check whether you've configured a default VLM under Model Providers → Set default models?

3. Docling: Model(Docling@None) not authorized

From what I can tell, this usually indicates that RAGFlow cannot locate a usable Docling backend.

There are two supported approaches:

• Local Docling (USE_DOCLING=true with the required package installed), or
• Docling Serve by setting:

DOCLING_SERVER_URL=http://<docling-server>:5001

If DOCLING_SERVER_URL is not configured, RAGFlow falls back to the local installation. If neither is available, you'll likely see authorization or initialization errors similar to the one you're getting. :contentReference[oaicite:2]{index=2}

A few questions that would help narrow this down:

• Which RAGFlow version are you running?
• Are you using Docker or a source installation?
• Did you enable USE_DOCLING=true (or configure DOCLING_SERVER_URL) before starting the containers?
• Does the same PDF parse successfully with the default DeepDoc parser?

The answers to those questions will help determine whether this is a parser configuration issue, a resource issue (MinerU), or a regression.

If this solves your problem, feel free to mark it as the accepted answer so others can find it easily.