[Bug]: GraphRAG doesn't get generated - 400 Please use a valid role: user, model.

[marcfon]

Is there an existing issue for the same bug?

• I have checked the existing issues.

Branch name

main

Commit ID

--

Other environment information

* dev version

KB settings
* chunk method: knowledge graph
* embedding model: Gemini (tried others as well)

Actual behavior

Traceback (most recent call last):
  File "/ragflow/graphrag/graph_extractor.py", line 128, in __call__
    result, token_count = self._process_document(text, prompt_variables)
  File "/ragflow/graphrag/graph_extractor.py", line 177, in _process_document
    if response.find("**ERROR**") >=0: raise Exception(response)
Exception: **ERROR**: 400 Please use a valid role: user, model.
**ERROR**: contents must not be empty```

### Expected behavior

_No response_

### Steps to reproduce

```Markdown
Add a plain text document to the KB and parse.

Additional information

No response

[KevinHuSh]

Pull the dev version of docker. It's fixed.

[marcfon]

Pull the dev version of docker. It's fixed.

Thanks for the response Kevin. The error still exists for me in the latest dev version.

I followed the upgrade instructions from here. With the difference that docker compose up ragflow -d doesn't work for me (no configuration file provided: not found). I need to cd ~/ragflow/docker and run docker compose up -d.

[marcfon]

@KevinHuSh just pulled the latest dev version today and checked again. Still getting the same error when processing a file.

[marcfon]

@KevinHuSh I think I have found the (root) issue which still exists in the latest dev version.

Inside Model Providers > System Model Settings > Chat model, when I select gpt-4o-mini as the Chat model everything works fine, selecting something else for instance gemini-1.5-flash-lastest it throws the 400 Please use a valid role: user, model. error.

[KevinHuSh]

Could you submit an issue for gemini-1.5-flash-lastest?

[marcfon]

@KevinHuSh #2720