The rerank_id parameter is not functioning properly

[maxiaomeng121-star]

[retriever-http] HTTP POST http://127.0.0.1:9380/api/v1/retrieval
[retriever-http] headers: {"Authorization": "Bearer ragf...8Tis", "Content-Type": "application/json"}
[retriever-http] payload: {"question": "test", "dataset_ids": ["c54559cad7ac11f098d31dbfcf0"], "page": 1, "page_size": 15, "top_k": 25, "vector_similarity_weight": 0.32, "rerank_id": "BAAI/bge-reranker-v2-m3@SILICONFLOW", "keyword": false, "highlight": false}

[retriever-http] HTTP POST http://127.0.0.1:9380/api/v1/retrieval
[retriever-http] headers: {"Authorization": "Bearer ragf...8Tis", "Content-Type": "application/json"}
[retriever-http] payload: {"question": "test", "dataset_ids": ["c54559cad7ac11f098d31dbfcf0"], "page": 1, "page_size": 15, "top_k": 25, "vector_similarity_weight": 0.32, "rerank_id": "BAAI/bge-reranker-v2-m3", "keyword": false, "highlight": false}

When I call the retrieval interface and pass the rerank_id parameter, the rerank model is not actually invoked. I have already configured it correctly.How can I troubleshoot where the issue might be?

[dosubot[bot]]

The most common reason rerank_id doesn't trigger the rerank model is that the value must exactly match the model name registered and authorized for your tenant, including any required provider suffix (like @SILICONFLOW, @VLLM, or ___VLLM for vllm deployments). Even a small mismatch or missing suffix will cause the reranker to be skipped, sometimes without an error message—so double-check the exact string as shown in the UI's "Retrieval Testing" > "Rerank model" dropdown. There is no public API to list valid rerank_id values, so the UI is the source of truth for what works in API calls (reference).

If you're using vllm, try using the key rerank_model instead of rerank_id, and use the vendor suffix format (e.g., bge-m3___VLLM) (reference).

Check backend logs for any "Model(xxx) not authorized" errors when you make the API call. If you see no errors, compare your API request payload with a working UI request to spot hidden differences. If you recently added or changed the rerank model, make sure there are no stale or conflicting records in your backend configuration or tenant_llm table, and restart RAGFlow if you updated config files (reference).

If the issue persists, capturing and comparing the full request payloads from both SDK and UI can help identify hidden differences or errors (reference).

To reply, just mention @dosu.

---

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

   

[Magicbook1108]

Rerank id should be like this, model_name@factory_name

[maxiaomeng121-star]

[retriever-http] HTTP POST http://127.0.0.1:9380/api/v1/retrieval
[retriever-http] headers: {"Authorization": "Bearer ragf...8Tis", "Content-Type": "application/json"}
[retriever-http] payload: {"question": "test", "dataset_ids": ["c54559cad7ac11f098d31dbfcf0"], "page": 1, "page_size": 15, "top_k": 25, "vector_similarity_weight": 0.32, "rerank_id": "BAAI/bge-reranker-v2-m3@SILICONFLOW", "keyword": false, "highlight": false}

Still not working: (

[Magicbook1108]

Any logs ?