[Question]: Why can't I successfully invoke the rerank model deployed by vllm when using the API at /api/v1/retrieval?

[lovekarri]

Self Checks

• I have searched for existing issues search for existing issues, including closed ones.
• 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).
• Please do not modify this template :) and fill in all the required fields.

Describe your problem

my rerank model was successfully deployed with vllm,and I have successfully use this rerank model to test callback with my knowledge base

but when I try to use api , the request was wrong with this response:
{ "code": 100, "data": null, "message": "LookupError('Model(vllm/bge-reranker-v2-m3) not authorized')" }

this is my api code:
try: response = requests.post( url="http://192.168.191.101:8092/api/v1/retrieval", headers={ "Authorization": "Bearer ragflow-kzNmJhYWE4MmRlODeXZjBhN2IyMGE2Nj", "content-Type": "application/json" }, json={ "question": "特种设备检验检测过程中发现设备存在严重事故隐患时该如何处理？", "dataset_ids": ["9ec38706301b11f08abf0a66212dace0"], "top_k": 3, "page": 1, "page_size": 3, "keyword": False, "highlight": False, "similarity_threshold": 0.5, "vector_similarity_weight": 0.7, "rerank_id": "vllm/bge-reranker-v2-m3", }, timeout=120, ) response.raise_for_status() except requests.exceptions.RequestException as e:

how can I use this rerank model?

[lovekarri]

this is my api code

try:
    response = requests.post(
        url="http://192.168.191.101:8092/api/v1/retrieval",
        headers={
            "Authorization": "Bearer ragflow-kzNmJhYWE4MmRlODExZjBhN2IyMGE2Nj",
            "content-Type": "application/json"
        },
        json={
            "question": "特种设备检验检测过程中发现设备存在严重事故隐患时该如何处理？",
            "dataset_ids": ["9ec38706301b11f08abf0a66212dace0"],
            "top_k": 3,
            "page": 1,
            "page_size": 3,
            "keyword": False,
            "highlight": False,
            "similarity_threshold": 0.5,
            "vector_similarity_weight": 0.7,
            "rerank_id": "vllm/bge-reranker-v2-m3",
        },
        timeout=120,
    )
    response.raise_for_status()
except requests.exceptions.RequestException as e:

[dosubot]

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

The error LookupError('Model(vllm/bge-reranker-v2-m3) not authorized') means that the rerank_id you provided is not recognized or authorized by RAGFlow. This usually happens if the rerank_id is not in the list of valid rerank models configured for your deployment. Known valid rerank models include BAAI/bge-reranker-v2-m3 and maidalun1020/bce-reranker-base_v1—these are the IDs you should use in your API request, not the vllm/ prefix or custom names unless your admin has explicitly registered them in the backend configuration [source].

To resolve this:

• Check the "Rerank model" dropdown in the Retrieval Testing page of the UI. Only the options listed there are valid rerank_id values for the API [source].
• If you are using a custom or locally deployed model, ensure it is registered and authorized by the administrator in the knowledge base settings.
• Try using the rerank_id BAAI/bge-reranker-v2-m3 in your API call instead of vllm/bge-reranker-v2-m3.
• If your deployment uses VLLM and you need a special format, try bge-m3@VLLM or bge-m3___VLLM as the rerank_id, as this has resolved similar issues for others [source].

Also, make sure your API key is valid and has the necessary permissions. The /api/v1/retrieval endpoint requires a valid API key in the Authorization header as Bearer <YOUR_API_KEY>. If the key is invalid or missing, you will get an authentication error [source].

If you continue to see authorization errors, confirm with your admin that the rerank model is set up and authorized for your knowledge base. If you are using a local or custom model, it must be registered in the backend configuration to be accessible via the API.

To reply, just mention my name @dosu.

---

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

[lovekarri]

@dosu
yes, I'm sure the name of vllm/bge-reranker-v2-m3 was valid rerank model name, it's in the "Rerank model" dropdown in the Retrieval Testing page of the UI. the image above showed this name in the Retrieval Testing page, and it's active.
my API key in the Authorization header as Bearer <YOUR_API_KEY> is active too.
If you need more information, please let me know.

[dosubot]

Based on the backend configuration and code, the rerank model ID vllm/bge-reranker-v2-m3 is not recognized as a valid rerank model by the RAGFlow API, even if it appears in the UI dropdown. The backend only accepts exact string matches for model names defined in its configuration files, and currently, only BAAI/bge-reranker-v2-m3 (and its "Pro" variant) are recognized as valid rerank models for this purpose. There is no aliasing or mapping that allows vllm/bge-reranker-v2-m3 to be accepted by the API, so using this ID will always result in a "Model not authorized" error, regardless of your API key status or UI appearance [source].

To resolve this, you need to ensure that the rerank model is registered in the backend configuration with the exact name you intend to use in the API. If you want to use a VLLM-deployed model, you may need to register it under a supported name (such as BAAI/bge-reranker-v2-m3) or adjust your API call to use a name that matches the backend configuration. If your deployment requires a special format, try using bge-m3@VLLM or bge-m3___VLLM, as this has worked for embedding models in similar cases [source].

If the model is not listed in the backend configuration, you will need administrator access to add or authorize it. Simply seeing the model in the UI dropdown does not guarantee backend authorization for API use [source].

If you need more details on how to register or authorize a custom rerank model in the backend, let me know your deployment details and I can help you further.

To reply, just mention my name @dosu.

---

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

[lovekarri]

finally resloved this problem.
you mustbe change the key.
the key 'rerank_id' is effective when you use Xinference,but if use vllm, the key must change to 'rerank_model'.

try:
    response = requests.post(
        url="http://192.168.191.101:8092/api/v1/retrieval",
        headers={
           ...
        },
        json={
           ...
            "rerank_model": "vllm/bge-reranker-v2-m3",
        },
        timeout=120,
    )
    response.raise_for_status()
except requests.exceptions.RequestException as e: