[Bug] 服务端不能正确响应模型禁用,禁用后依然支持执行任意模型请求
Opened this issue · 1 comments
📦 部署方式
官方安装包
📌 软件版本
v2.15.8
💻 系统环境
Ubuntu
📌 系统版本
20.04.6
🌐 浏览器
Edge
📌 浏览器版本
131.0.2903.70
🐛 问题描述
我的服务使用 github 源码编译并运行在自己的阿里云服务器上,代码 commit: a2adfbb
服务使用自己的 OpenAI token。
我限制了使用的 API 版本, 比如不允许使用 gpt-4
.
在新的浏览器上访问我的网站, 确实不显示 gpt-4
了, 显示如下:
但如果我在之前访问过的浏览器上访问我的网站时, 我发现我不允许的选项全都在, 并且可以正常使用:
服务器后台似乎缺乏验证,使用者可以随意选择想使用的 API.
📷 复现步骤
- 服务端正常发布页面, 不做任何 api 限制
- 网页使用者登录并使用
gpt-4
问几个问题 - 服务端修改并重新发布,禁用部分对话模型, 比如
gpt-4
- 网页使用者继续登录并找到原来的问题, 继续使用
gpt-4
追问 - 此时网页能继续使用
gpt-4
作答, 不符合预期。
🚦 期望结果
服务器报错, 提示此模型被禁用, 或者不支持此模型,要求切换。
📝 补充信息
No response
Bot detected the issue body's language is not English, translate it automatically.
Title: [Bug]
📦 Deployment method
Official installation package
📌 Software version
v2.15.8
💻 System environment
Ubuntu
📌 System version
20.04.6
🌐 Browser
Edge
📌 Browser version
131.0.2903.70
🐛 Problem description
My service is compiled using github source code and runs on my own Alibaba Cloud server. Code commit: a2adfbb
The service uses its own OpenAI token.
I have restricted the API version used, for example gpt-4
is not allowed.
When I visit my website on a new browser, gpt-4
is indeed no longer displayed. It is displayed as follows:
But if I visit my website on the browser I visited before, I find that all the options I'm not allowed are there and can be used normally:
The server background seems to be verified, and users can choose the API they want to use.
📷 Steps to reproduce
- The server publishes the page normally without any API restrictions.
- Web users log in and use
gpt-4
to ask a few questions - Modify and re-publish the server and disable some dialogue models, such as
gpt-4
- The web user continues to log in and finds the original question, and continues to use
gpt-4
to ask the question - At this time, the web page can continue to use
gpt-4
to answer questions, which is not as expected.
🚦 Expected results
The server reports an error, indicating that this model is disabled or does not support this model and requires switching.
📝 Supplementary information
No response