[Bug] After qwen2 asked the question, the question has been answered, but the source is not displayed, and there is always a "stop answering" button, and it ends after a few minutes.

[xiaoxiao261258]

Contact Information

No response

MaxKB Version

v1.8.0

Problem Description

提出问题后，会先正常回答问题，回答完之后，无任何响应，同时会一直有“停止回答”的字样。过了几分钟后，才会显示“知识来源”，“引用分段”这些信息。如下是一个示例，同时，去除了敏感信息。跟论坛上的如下问题一样。
https://bbs.fit2cloud.com/t/topic/6228

Q： 审计模型

A：根据您的问题:审计模型，为您在知识库中匹配到以下结果

XXXXXXXXXXXXXXXXXX

**知识来源 引用分段1

XXXX

消耗 tokens: 1655 耗时: 270.19s**

Steps to Reproduce

模型选择 vLLM，使用的是本地部署的Qwen2模型。

The expected correct result

问题正常回答后能及时结束。

Related log output

No response

Additional Information

No response

[baixin513]

感谢反馈，目前我们的测试环境上还无法重现，麻烦在模型管理中，截图看一下添加模型的信息，保密信息请马赛克。

[shaohuzhang1]

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

---

Thanks for the feedback. Currently, we cannot reproduce it in our tests. The trouble is in the model management. Take a screenshot to see the information of the added model. Please mosaic the confidential information.

[xiaoxiao261258]

尝试使用openai的供应商添加,看看是否有你这个情况

感谢反馈，目前我们的测试环境上还无法重现，麻烦在模型管理中，截图看一下添加模型的信息，保密信息请马赛克。
非常感谢，但是公司安保限制，无法贴图到github，我文字描述下。
选择的vLLM模型，权限不管选私有还是公用，都有这个问题。模型类型为“大语言模型”，基础模型填的是“Qwen2-7B-
Instruct”，API域名填的是其他服务器上部署的模型的地址，格式为“http://XXX.XXX.XXX.XXX:8000/v1”, API Key 填的是“EMPTY”

[shaohuzhang1]

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

---

Thanks for the feedback. Currently, it cannot be reproduced in our test environment. The trouble is in model management. Take a screenshot to see the information of the added model. Please mosaic the confidential information.

Thank you very much, but due to company security restrictions, I cannot post the image to github. I will describe it in text.

The selected vLLM model has this problem regardless of whether the permissions are private or public. The model type is "Large Language Model", the basic model is filled in "Qwen2-7B-Instruct", and the API domain name is filled in with the address of the model deployed on other servers, in the format of "http://XXX.XXX.XXX.XXX :8000/v1", the API Key is "EMPTY"

[xiaoxiao261258]

尝试使用openai的供应商添加,看看是否有你这个情况

感谢反馈，目前我们的测试环境上还无法重现，麻烦在模型管理中，截图看一下添加模型的信息，保密信息请马赛克。
非常感谢，但是公司安保限制，无法贴图到github，我文字描述下。
选择的vLLM模型，权限不管选私有还是公用，都有这个问题。模型类型为“大语言模型”，基础模型填的是“Qwen2-7B-
Instruct”，API域名填的是其他服务器上部署的模型的地址，格式为“http://XXX.XXX.XXX.XXX:8000/v1”, API Key 填的是“EMPTY”

我设置的是一个简单配置，没有用高级编排。
修改成了open ai添加模型，结果如下：
匹配不到知识库的情况，会很快结束（我设置了无引用知识库分段时，指定回答内容）。
但是一旦匹配到了知识库，还是会存在这个问题。不过等待的时间似乎缩短了一点，选vLLM时，都在4分钟出头，现在是2分钟出头的样子。

[shaohuzhang1]

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

---

Try using openai's supplier addition to see if this is the case for you.

Thanks for the feedback. Currently, it cannot be reproduced in our test environment. The trouble is in the model management. Take a screenshot to see the information of the added model. Please mosaic the confidential information.
Thank you very much, but due to company security restrictions, I cannot post the image to github. I will describe it in text.
The selected vLLM model has this problem regardless of whether the permissions are private or public. The model type is "Large Language Model", and the basic model is "Qwen2-7B-
Instruct", the API domain name is filled in with the address of the model deployed on other servers, the format is "[http://XXX.XXX.XXX.XXX:8000/v1"](http://XXX.XXX.XXX .XXX:8000/v1%E2%80%9D), the API Key is "EMPTY"

If the knowledge base cannot be matched, it will end quickly (when I set up a non-referenced knowledge base segment, I specified the answer content).
But once the knowledge base is matched, this problem will still exist.

[baixin513]

在v1.9版本修改过回答完成后仍然显示停止回答的问题，您可以等v1.9版本发布后在试一下。

[shaohuzhang1]

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

---

After the v1.9 version has modified the answer, it still displays the question that has stopped answering. You can wait for the v1.9 version to be released and try it again.

[xiaoxiao261258]

在v1.9版本修改过回答完成后仍然显示停止回答的问题，您可以等v1.9版本发布后在试一下。

非常感谢~

[shaohuzhang1]

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

---

After the v1.9 version has modified the answer, the question still shows that the answer has been stopped. You can wait for the v1.9 version to be released and try it again.

Thank you very much~

[xiaoxiao261258]

在v1.9版本修改过回答完成后仍然显示停止回答的问题，您可以等v1.9版本发布后在试一下。

release notes里面看了下，没有找到类似的问题。
同时，刚才升级了版本，仍然存在该问题。

[shaohuzhang1]

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

---

After the v1.9 version has modified the answer, it still displays the question that has stopped answering. You can wait for the v1.9 version to be released and try it again.

I looked through the release notes and couldn't find any similar issues.
At the same time, I just upgraded the version and the problem still exists.

[baixin513]

那可能还是与你们的模型环境、maxkb的部署环境等因素有关系。我们这边无法复现。

[shaohuzhang1]

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

---

That may still be related to your model environment, maxkb deployment environment and other factors. We can't reproduce it here.

[nxiao77]

@xiaoxiao261258 Your docker should have internet connection，then this issue maybe disappear. Have a try.