[Question]: Model(BAAI/bge-reranker-v2-m3@None) not authorized

[lizibin565758]

Self Checks

• [x] I have searched for existing issues search for existing issues, including closed ones.
• [x] I confirm that I am using English to submit this report (Language Policy).
• [ ] Non-english title submitions will be closed directly ( 非英文标题的提交将会被直接关闭 ) (Language Policy).
• [x] Please do not modify this template :) and fill in all the required fields.

Describe your problem

Hello, I am currently using version v0.22.0 deployed via Docker. My model, BAAI/bge-reranker-v2-m3, is a local model that has been successfully linked with HuggingFace. I am using the API to create a chat session, passing the following prompt: prompt.put("rerank_model", "BAAI/bge-reranker-v2-m3"); However, after creating the session and sending a question-answer request, I get the following error:

{ "code": 500, "message": "Model(BAAI/bge-reranker-v2-m3@None) not authorized", "data": { "answer": "*ERR: Model(BAAI/bge-reranker-v2-m3@None) not authorized", "reference": [] } }

I did not modify the rerank option, and it was left blank. However, there is a checkbox with an "X" inside it, which I believe indicates that the rerank model has been selected, even though it is not displayed. I can also see from the F12 console that the model is indeed selected.

Could you please help me resolve this issue? Currently, I am unable to perform RAG-based knowledge base question answering.

[dosubot[bot]]

Hi @lizibin565758! I'm Dosu and I’m helping the ragflow team.

The error happens because RAGFlow requires rerank models to be explicitly registered and authorized for your tenant, even if they are visible in the UI. For HuggingFace models, you must add the rerank model via the "Add Model" or equivalent UI/API, making sure to select "HuggingFace" as the provider/factory. Internally, the system appends a suffix (like ___HuggingFace) to the model name during registration. If you only specify BAAI/bge-reranker-v2-m3 without registering it this way, the backend cannot authorize it and returns the error you see.

To fix this, re-add the rerank model using the UI or API, select "HuggingFace" as the provider, and ensure the model is enabled. After this, the rerank model should appear as available and authorized, and you can reference it in prompts as BAAI/bge-reranker-v2-m3 (the system will map it to the registered model with the correct suffix) see details.

If you still see the error, double-check that the model is enabled and has a valid API key/config for HuggingFace, and that your tenant has access to it see backend logic.

To reply, just mention @dosu.

---

^{How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other}    

[lizibin565758]

I can see that the rerank model has already been registered in my model list. So, what value should I pass in order to correctly create the chat session? prompt.put("rerank_model", ""); – What value should I pass here?

I tried the following:

BAAI/bge-reranker-v2-m3 – I was able to create the chat session, but during the conversation, I encountered the following error:

{ "code": 500, "message": "Model(BAAI/bge-reranker-v2-m3@None) not authorized", "data": { "answer": "ERR: Model(BAAI/bge-reranker-v2-m3@None) not authorized", "reference": [] } }

BAAI/bge-reranker-v2-m3___HuggingFace@HuggingFace – When I used this value to create the chat session, I got the following error:

{ "code": 102, "message": "rerank_model BAAI/bge-reranker-v2-m3___HuggingFace@HuggingFace doesn't exist" }

BAAI/bge-reranker-v2-m3@HuggingFace – When I used this value to create the chat session, I got the following error:

{ "code": 102, "message": "rerank_model BAAI/bge-reranker-v2-m3@HuggingFace doesn't exist" }

Can you please tell me what value I should actually use here?

[dosubot[bot]]

You need to set prompt.put("rerank_model", "BAAI/bge-reranker-v2-m3")—just the base model name, without any suffix or @HuggingFace. This is the value the backend expects and matches against its internal list of allowed rerank models and your tenant's registered models.

If you still get a "not authorized" error, it means the model is registered but not authorized for your tenant. Double-check that the model is enabled and authorized in your model list. If it appears in the UI but isn't working, try disabling and re-enabling it, or re-registering it with the HuggingFace provider selected. The backend will handle the internal suffixing automatically, but the API parameter should be the plain model name: BAAI/bge-reranker-v2-m3 (reference).

To reply, just mention @dosu.

---

^{How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other}    

[Magicbook1108]

@lizibin565758 bge-reranker-v2-m3 is freely available on SiliconFlow. Is there a specific reason you prefer running it locally?

[OneKe]

@Magicbook1108 I'm encountering the same issue: when trying to replace the built-in BAAI/bge-large-zh-v1.5 of the old version with the one from SiliconFlow, an error occurs.

[buua436]

@OneKe Thanks for reporting — could you share the exact error message/logs? To help us pinpoint the cause.

[OneKe]

@OneKe Thanks for reporting — could you share the exact error message/logs? To help us pinpoint the cause.

#11515 @buua436

[buua436]

@OneKe If possible, could you share the logs from the RAGFlow container so we can pinpoint the issue?

You can run: docker logs -f docker-ragflow-cpu-1

[OneKe]

@buua436

[buua436]

@OneKe Could you please help confirm whether this knowledge base is a team-shared one, and that you are not the creator of it? In our reproduction, we found a scenario that can trigger the issue: if another team member changes the embedding model for the KB, but the KB creator account does not have that embedding model configured/authorized, the problem may occur. Please confirm if this matches your case (shared KB + non-creator + embedding changed by others), so we can narrow down the root cause and provide the proper fix or guidance.

[OneKe]

@buua436 I confirm that this is the knowledge base I created myself.

[Magicbook1108]

Try the model below in siliconflow, it's free.