ChatGPTNextWeb
[Bug] 硅基流动的 DeepSeek R1 / V3 模型 通过环境变量配置 API地址与Key 时 使用模型会返回 503 错误，但通过设置页面手动配置可正常使用
📦 部署方式
Vercel
📌 软件版本
commit add9ca2
💻 系统环境
Windows
📌 系统版本
Windows 10
🌐 浏览器
Edge
📌 浏览器版本
132.0.2957.140
🐛 问题描述
在环境变量中填写了API地址和Key后（SILICONFLOW_API_KEY、SILICONFLOW_URL），选定硅基流动的 DeepSeek R1 / V3 模型并发送消息时均返回 503错误，具体响应如下：
{
"code": 50603,
"message": "System is too busy now. Please try again later.",
"data": null
}
提示服务器繁忙。
而手动在设置页面填写相同的API地址和Key却能正常使用。（使用硅基流动文档的格式）
📷 复现步骤
- 在vercel的项目管理中填写
SILICONFLOW_API_KEY与SILICONFLOW_URL - 重新部署项目
- 打开网站，选中硅基流动的 DeepSeek R1 / V3 模型，发送消息。
- API返回上述响应
🚦 期望结果
在vercel的环境变量中填写硅基流动的API地址和Key后能够正常使用其 DeepSeek R1 / V3 模型，不再需要手动在设置页面填写
📝 补充信息
似乎只有满血版的会这样，其他的蒸馏模型正常
Bot detected the issue body's language is not English, translate it automatically.
Title: [Bug] DeepSeek R1/V3 model with silicon-based flow When configuring API addresses and Keys through environment variables Using the model will return a 503 error, but it can be used manually through the settings page
same, guessing SiliconFlow has set up some limitations for Vercel, it works well and faster on Netlify.
Bot detected the issue body's language is not English, translate it automatically.
Same gentleman
Bot detected the issue body's language is not English, translate it automatically.
Same
Bot detected the issue body's language is not English, translate it automatically.
Hello, in the browser developer tools, can you confirm that the keys used in the two modes are the same? You can also try the customized version https://nextchat-sf.vercel.app
Bot detected the issue body's language is not English, translate it automatically.
Hello, in the browser developer tools, can you confirm that the keys used in the two modes are the same. You can also try [Customized Version] (https://nextchat-sf.vercel.app). Click the lower left corner to automatically enter the key. More convenient
Can this customized version be used directly by fork?
为什么我git pull了最新版本，填写好.env中的SILICONFLOW_API_KEY、SILICONFLOW_URL 再 yarn build & yarn start，并没有在模型里找到siliconflow的DeepSeek R1 / V3？只有deepseek官方的。
而且我尝试添加CUSTOM_MODELS=+deepseek-ai/DeepSeek-R1@SiliconFlow=DeepSeek-R1,+deepseek-ai/DeepSeek-V3@SiliconFlow=DeepSeek-V3，还是走的BASE_URL和OPENAI_API_KEY，是我姿势不对吗
Bot detected the issue body's language is not English, translate it automatically.
Why did I pull the latest version and fill in SILICONFLOW_API_KEY, SILICONFLOW_URL in .env and yarn build & yarn start, and did not find the DeepSeek R1/V3 of siliconflow in the model? Only deepseek official ones.
And I tried adding CUSTOM_MODELS=+deepseek-ai/DeepSeek-R1@SiliconFlow=DeepSeek-R1,+deepseek-ai/DeepSeek-V3@SiliconFlow=DeepSeek-V3, or go Base Url and OPENAI_API_KEY are my posture wrong?
为什么我git pull了最新版本，填写好.env中的SILICONFLOW_API_KEY、SILICONFLOW_URL 再 yarn build & yarn start，并没有在模型里找到siliconflow的DeepSeek R1 / V3？只有deepseek官方的。
而且我尝试添加CUSTOM_MODELS=+deepseek-ai/DeepSeek-R1@SiliconFlow=DeepSeek-R1,+deepseek-ai/DeepSeek-V3@SiliconFlow=DeepSeek-V3，还是走的BASE_URL和OPENAI_API_KEY，是我姿势不对吗
Bot detected the issue body's language is not English, translate it automatically.
Hello, in the browser developer tools, can you confirm that the buttons used in the two modes are the same? You can also try [customized version] (https://nextchat-sf.vercel.app). Click the lower left corner to automatically enter the key. More convenient Can this customized version be used directly fork?
Bot detected the issue body's language is not English, translate it automatically.
Same problem
Bot detected the issue body's language is not English, translate it automatically.
The same is true for me. The paid pro version is fine, probably the restrictions have been added.
SILICONFLOW_URL https://api.siliconflow.cn/v1/ like this then run
Don't do that, it will return 404 not found.
V3和R1模型的普通版无法使用，但是需要使用付费余额的V3和R1模型Pro是可以使用的。但deepseek-ai/DeepSeek-V2.5也无法通过配置环境变量使用，希望能够修改适配。
Bot detected the issue body's language is not English, translate it automatically.
The normal version of the V3 and R1 model cannot be used, but the V3 and R1 model Pro that require a paid balance are available. However, deepseek-ai/DeepSeek-V2.5 cannot be used through configuration environment variables, and I hope to modify the adaptation.
Bot detected the issue body's language is not English, translate it automatically.
The vicious bug is still there