[Bug] rerank private model docking failed

[yenaikafei]

Contact Information

yenaikafei@126.com

MaxKB Version

v2.1.2 (build at 2025-09-23T09:18, commit: c14ad2c)

Problem Description

用gpustack部署了llm embedding reranker模型 对接使用openai插件成功对接llm 和 embedding 因openai未提供reranker类型 按文档使用模型概况列表里支持reranker模型的供应商都试了一遍 看文档应该是都支持私有模型的 但是都没有成功对接
大部分都是提示model not found，也有Failed to get the model description, detail: Not Found

Steps to Reproduce

gpuStack模型试验场里测试是成功的，地址http://127.0.0.1/v1/rerank
对接使用局域网IP ，局域网IP /v1，局域网IP /v1/rerank，局域网IP /v1/reranker均未成功

The expected correct result

No response

Related log output

Additional Information

No response

[liqiang-fit2cloud]

openai接口不支持重排模型。
所有模型的支持见这里https://maxkb.cn/docs/v2/user_manual/model/model_summary/#3

[shaohuzhang1]

Bot detected the issue body's language is not English, translate it automatically. 👯👭🏻🧑‍🤝‍🧑👫🧑🏿‍🤝‍🧑🏻👩🏾‍🤝‍👨🏿👬🏿

---

The openai interface does not support rearranged models.
All model support is available here https://maxkb.cn/docs/v2/user_manual/model/model_summary/#3

[shaohuzhang1]

Bot detected the issue body's language is not English, translate it automatically. 👯👭🏻🧑‍🤝‍🧑👫🧑🏿‍🤝‍🧑🏻👩🏾‍🤝‍👨🏿👬🏿

---

Can't read Chinese? I have tried Alibaba Cloud Bailian SILICONFLOW Ollama Xorbits Inference, but none of them succeeded according to the documentation.

[yenaikafei]

openai接口不支持重排模型。 所有模型的支持见这里https://maxkb.cn/docs/v2/user_manual/model/model_summary/#3

列表里支持reranker模型的供应商都试了一遍 看文档应该是都支持私有模型的 但是都没有成功对接
阿里云百炼SILICONFLOW Ollama Xorbits Inference都试过

[shaohuzhang1]

Bot detected the issue body's language is not English, translate it automatically. 👯👭🏻🧑‍🤝‍🧑👫🧑🏿‍🤝‍🧑🏻👩🏾‍🤝‍👨🏿👬🏿

---

The openai interface does not support rearranged models. All model support is available here https://maxkb.cn/docs/v2/user_manual/model/model_summary/#3

I have tried all the suppliers in the list that support the reranker model. Looking at the documentation, it seems that they all support the private model, but none of them have successfully connected.
Tried Alibaba Cloud Bailian SILICONFLOW Ollama Xorbits Inference

[liqiang-fit2cloud]

openai接口不支持重排模型。 所有模型的支持见这里https://maxkb.cn/docs/v2/user_manual/model/model_summary/#3

列表里支持reranker模型的供应商都试了一遍 看文档应该是都支持私有模型的 但是都没有成功对接 阿里云百炼SILICONFLOW Ollama Xorbits Inference都试过

把填写的参数和报错信息都贴出来看看。

[shaohuzhang1]

Bot detected the issue body's language is not English, translate it automatically. 👯👭🏻🧑‍🤝‍🧑👫🧑🏿‍🤝‍🧑🏻👩🏾‍🤝‍👨🏿👬🏿

---

The openai interface does not support rearranged models. Support for all models is available here https://maxkb.cn/docs/v2/user_manual/model/model_summary/#3

I have tried all the suppliers in the list that support the reranker model. Looking at the documents, it seems that they all support private models, but they have not been successfully connected. I have tried Alibaba Cloud Bailian SILICONFLOW Ollama Xorbits Inference.

Post the filled in parameters and error messages for a look.

[yenaikafei]

openai接口不支持重排模型。 所有模型的支持见这里https://maxkb.cn/docs/v2/user_manual/model/model_summary/#3

列表里支持reranker模型的供应商都试了一遍 看文档应该是都支持私有模型的 但是都没有成功对接 阿里云百炼SILICONFLOW Ollama Xorbits Inference都试过

把填写的参数和报错信息都贴出来看看。

模型名称：bge-reranker-v2-m3
模型类型：重排模型
基础模型：BAAI/bge-reranker-v2-m3
URL：http://172.18.91.233/v1 （局域网IP）

报错：code:500
'认证失败，请检查参数是否正确：SiliconCloud API 请求失败: {"error":{"message":"Model not found","code":404,"type":"NotFound"}}'

[liqiang-fit2cloud]

既然用硅基流动的接口为什么URL不是硅基流动的API地址，而是局域网URL？
我这按正常配置是正常的https://maxkb.cn/docs/v2/img/model/siliconflow_rerank.png

[shaohuzhang1]

Bot detected the issue body's language is not English, translate it automatically. 👯👭🏻🧑‍🤝‍🧑👫🧑🏿‍🤝‍🧑🏻👩🏾‍🤝‍👨🏿👬🏿

---

Since the silicon-based mobile interface is used, why is the URL not the silicon-based mobile API address, but the LAN URL?
This is normal according to my normal configuration https://maxkb.cn/docs/v2/img/model/siliconflow_rerank.png

[yenaikafei]

既然用硅基流动的接口为什么URL不是硅基流动的API地址，而是局域网URL？ 我这按正常配置是正常的https://maxkb.cn/docs/v2/img/model/siliconflow_rerank.png

对接私有模型。。。不支持私有模型的吗？如果对接公网直接像阿里云百炼一样填个key就行了 还要填rul吗写死就行了啊？
embdding模型是可以对接私有模型的呀

对接参数：
模型名称：bge-reranker-v2-m3
模型类型：重排模型
基础模型：BAAI/bge-reranker-v2-m3
URL：http://172.18.91.233/v1 （局域网IP）

SILICONFLOW报错：
{
"code": 500,
"message": "认证失败，请检查参数是否正确：SiliconCloud API 请求失败: {"error":{"message":"Model not found","code":404,"type":"NotFound"}}",
"data": null
}

Ollama报错：
{
"code": 500,
"message": "API 域名无效",
"data": null
}

Xorbits Inference报错：
{
"code": 500,
"message": "认证失败，请检查参数是否正确：Failed to get the model description, detail: Not Found",
"data": null
}

[liqiang-fit2cloud]

1.每个模型供应商都有相应的sdk，因为每个供应商开放的接口不一样，不能用硅基流动的sdk去调比如阿里云、oopenai或者私有模型等其他供应商的接口，必然报错
2.对接私有模型只支持ollama，vllm，xinference，请选择相应的供应商
3.目前不支持gpustack等其他私有模型供应商