[Bug] Information is lost after output tokens exceeds the limit

[Nan93]

Contact Information

No response

MaxKB Version

1.9.1

Problem Description

我使用内部vllm部署的模型，虽然不清楚最大输出tokens是多少，但使用maxkb和cherry-studio通常在输出1000个tokens左右就会中断不再输出。我认为可能是vllm配置有问题，正在找部署人员排查。

我的问题是，maxkb对于这种中断输出，没有办法记录：

1. 在应用的对话历史中无法找到这次对话
2. 当中断后，我点击停止，再输入继续时，提示我不清楚要继续什么。
3. 对话页面一旦刷新，无法回答的那次输入输出也会消失。

这给我带来了很大的困扰。我想知道是否可能会改变？

Steps to Reproduce

我的maxkb的应用是基础应用，模型的tokens被设置为最大上限100000，没有配置任何知识库。

我使用“请输入N遍以下代码”来进行测试，直到超出模型输出tokens上限就会复现。

或许可以部署个模型使其输出tokens为最小来尝试复现。

The expected correct result

No response

Related log output

Additional Information

No response

[shaohuzhang1]

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

---

Contact Information

No response

MaxKB Version

1.9.1

Problem Description

I deployed the model using the internal vllm, and although it is not clear what the maximum output tokens is, using maxkb and cherry-studio usually breaks or no longer outputs when outputting about 1000 tokens. I think there may be a problem with the vllm configuration and I am looking for deployment personnel to troubleshoot.

My problem is that maxkb has no way to record for such interrupt output:

1. This conversation cannot be found in the dialog history of the application
2. When I click Stop and then enter Continue, it prompts me that I don’t know what to continue.
3. Once the dialogue page is refreshed, the input and output that cannot be answered will also disappear.

This caused me a lot of trouble. I wonder if it might change?

Steps to Reproduce

My maxkb application is the basic application, the model's tokens is set to the maximum maximum of 100,000, and no knowledge base is configured.

I used "Please enter the following code N times" to test it until the upper limit of model output tokens will be reproduced.

Perhaps a model can be deployed to minimize its output tokens to try reproducing.

The expected correct result

No response

Related log output

Additional Information

No response

[Shenguobin0102]

你好，感谢反馈，我们会在后续版本进行修复。

[shaohuzhang1]

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

---

Hello, thanks for the feedback, we will fix it in subsequent versions.