[Feature Request]: LMstudio rerank model isn't supported

[r3vs]

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).
• [x] 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.

RAGFlow workspace code commit ID

RAGFlow image version

v0.19.0-full

Other environment information

Windows 11 with Docker

Actual behavior

This is the log:

2025-06-07 00:45:35,419 ERROR 22

Fail to access model(text-embedding-qwen.qwen3-reranker-0.6b).The LmStudioRerank has not been implement

NoneType: None

2025-06-07 00:45:35,421 INFO 22 172.20.0.6 - - [07/Jun/2025 00:45:35] "POST /v1/llm/add_llm HTTP/1.1" 200 -

2025-06-07 00:45:37,097 ERROR 22

Fail to access model(text-embedding-qwen.qwen3-reranker-0.6b).The LmStudioRerank has not been implement

NoneType: None

2025-06-07 00:45:37,098 INFO 22 172.20.0.6 - - [07/Jun/2025 00:45:37] "POST /v1/llm/add_llm HTTP/1.1" 200 -

Expected behavior

The docs says that reranking models for LMstudio are supported.

Steps to reproduce

Additional information

No response

[dosubot[bot]]

Hi @r3vs! I'm Dosu and I’m helping the ragflow team.

The LMstudio rerank model is listed in the codebase, but its implementation is missing—the main method raises a "NotImplementedError" and isn't functional yet, which matches the error you're seeing in your logs. This means that, despite being recognized as a model type, LMstudio rerank is not actually supported in RAGFlow at this time see source.

There are also similar reports for other rerank models not being supported yet. As a workaround, some users have used LM_Studio to add a local reranker model or selected one from the FlagEmbedding repository see discussion.

I couldn't find documentation that explicitly states LMstudio rerank models are supported, so there may be a mismatch between the docs and the current implementation. If you have a specific doc reference, sharing it could help clarify the discrepancy.

To reply, just mention @dosu.

---

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

[ZhenhangTung]

@r3vs Noted

[joaorsbarbosa]

Any update on this? I got this error too, but the supported model list is contradicting this issue: https://ragflow.io/docs/dev/supported_models

[ZhenhangTung]

Any update on this? I got this error too, but the supported model list is contradicting this issue: https://ragflow.io/docs/dev/supported_models

@joaorsbarbosa which version are you using?

[joaorsbarbosa]

@joaorsbarbosa which version are you using?

v0.20.4 Although we have talked on Discord already! I can quote your reply here if you want, for reference sake.

[ZhenhangTung]

@joaorsbarbosa which version are you using?

v0.20.4 Although we have talked on Discord already! I can quote your reply here if you want, for reference sake.

@joaorsbarbosa Sure, if it’s not too much trouble for you 🙂.

[joaorsbarbosa]

For reference:

Haven't planned it yet. We're refactoring our pile parsing pipeline right now. A feature request is welcomed.

PS: Will submit that feature request in a bit! 😅

[Magicbook1108]

The LM Studio API endpoint follows the OpenAI-compatible format (/v1/chat/completions, /v1/completions, /v1/embeddings), which supports chat, vision, and embedding models.

However, LM Studio does not implement a dedicated /v1/rerank endpoint or reranking logic. Even if a reranker model is loaded, it is still handled as a chat model through the standard LLM pipeline. Using a rerank model via the chat interface is therefore conceptually misaligned—it asks a ranking model to generate text instead of computing relevance scores, leading to unstable or meaningless results.