[Bug] 通过火山引擎（字节跳动）api接入的DeepseekR1没有思考过程，Deepseek官方的api展示的markdown格式有问题 #6212
Comments
Title: [Bug] DeepseekR1 connected through Volcano Engine (ByteDance) API has no thinking process, and there is a problem with the markdown format displayed by Deepseek official API 📦 Deployment methodOfficial installation package 📌 Software versionv2.15.8 💻 System environmentWindows 📌 System version10 🌐 BrowserEdge 📌 Browser version/ 🐛 Question descriptionDeepseekR1, which is connected through the Volcano Engine (ByteDance) API, has no thinking process. I have a rough look at the source code and it should not have written the think part; there is a problem with the markdown format displayed by Deepseek's official API, whether it is V3 or R1 (chat and reasoner) No markdow can be displayed accurately, and it will swallow /n. 📷 Reproduction stepsNo response 🚦 Expected resultsNo response 📝 Supplementary informationNo response |
|
|
这不是同一个问题，楼主说的使用字节跳动、腾讯云的deepseek api接入时，没有显示思考过程。 |
This is not the same problem. The poster said that when using ByteDance and Tencent Cloud's deepseek API access, there is no reflection process. |
|
|
我调用的就是阿里云的deepseek ，nestgpt用官网最新的代码， CUSTOM_MODELS=-all,+deepseek-r1@DeepSeek ，就可以了 |
What I call is Alibaba Cloud's deepseek. nestgpt uses the latest code on the official website, CUSTOM_MODELS=-all,+deepseek-r1@DeepSeek, and that's fine. |
|
#6204 这个pr已经可以了
|
|
还有一个是Deepseek官方的API返回的内容输出为Markdown时/n会被吞掉，但查看响应时发现是有返回换行符的；用其他的大模型就没有Markdown格式的问题。 |
Another thing is that when the content returned by Deepseek's official API is Markdown, /n will be swallowed, but when viewing the response, it is found that there is a newline return; there is no problem with Markdown format when using other big models. |
已在 #6204 中修复（还没合并到main） |
Fixed in #6204 (not merged to main yet) |
请问一下火山方舟这个api怎么配置的，我的配置如下：
|
|
我是在环境变量里这样配置的 |
I configured it in the environment variable |
老哥，我是用的app。设置的模型服务商是：ByteDance，接口地址是：https://ark.cn-beijing.volces.com/api/v3，自定义模型是：+字节跳动-DeepSeek-r1@ByteDance=ep-202xxx-b5xsl，可是还是报错 |
|
app和docker版本都发布的很早吧（2024年11月份的） |
Both the app and docker versions were released very early (November 2024) |
想请问一下，我也是使用阿里的deepseek，这个程序的页面配置是要怎么选择的，还有配置文件.env.local里只要修改CUSTOM_MODELS就OK了么 |
I would like to ask, I also use Alibaba's deepseek, how to choose the page configuration of this program, and the configuration file. Is it OK if I just modify CUSTOM_MODELS in env.local? |
|
你的接口地址写错了，那里直接默认api/alibaba就行。模型名称那里不知道有没有问题我用的是字节的可以参考下项目配置
|
|
just run like below docker run -d --name chatgpt -p 127.0.0.1:3000:3000 \
-e CODE="xxx" \
-e BYTEDANCE_API_KEY="xxx"
-e CUSTOM_MODELS="-all,+deepseek-r1@ByteDance=bot-xxxx" \
-e DEFAULT_MODEL="deepseek-r1@ByteDance=bot-xxxx" yidadaa/chatgpt-next-web |
📦 部署方式
官方安装包
📌 软件版本
v2.15.8
💻 系统环境
Windows
📌 系统版本
10
🌐 浏览器
Edge
📌 浏览器版本
/
🐛 问题描述
通过火山引擎（字节跳动）api接入的DeepseekR1没有思考过程，粗略看了下源码应该是没写think部分；Deepseek官方的api展示的markdown格式有问题，无论是V3还是R1（chat和reasoner）都无法准确显示markdow，会吞掉/n。
📷 复现步骤
No response
🚦 期望结果
No response
📝 补充信息
No response
The text was updated successfully, but these errors were encountered: