附件或者图片能否使用base64实现

[a8851625]

如题

少一个服务，部署起来也简单些

[Issues-translate-bot]

Bot detected the issue body's language is not English, translate it automatically.

---

Title: Can attachments or pictures be implemented using base64?

As title

One less service and easier to deploy

[Sh1n3zZ]

Hi,

chatnio-blob-service 项目附带支持 base64 的选项，不需要其他依赖、支持使用 vercel 进行一键部署
具体请查看 https://github.com/Deeptrain-Community/chatnio-blob-service/ 项目的 README.md

Regrads,
Junhai Deng

[AnnaStreeter]

提问题前建议看文档

[Issues-translate-bot]

Bot detected the issue body's language is not English, translate it automatically.

---

It is recommended to read the documentation before asking questions

[a8851625]

但是请求到到llm后端的时候是通过 chatnio-blob-service 配置的公网域名url；

我指的是直接到llm后端的时候 就是 base64。

[Issues-translate-bot]

Bot detected the issue body's language is not English, translate it automatically.

---

But when the request reaches the llm backend, it is the public domain name URL configured through chatnio-blob-service;

What I mean is that when going directly to the llm backend, it is base64.

[Sh1n3zZ]

Hi,

较为冗余，建议使用 vercel 一键快速部署。

Regards,
Junhai Deng

[Issues-translate-bot]

Bot detected the issue body's language is not English, translate it automatically.

---

Hi,

It is relatively redundant. It is recommended to use vercel for one-click rapid deployment.

Regards,
Junhai Deng

[a8851625]

好的，确实是支持，没看清楚；抱歉，不过内容貌似不全：

{"model":"qwen","messages":[{"role":"user","content":"hi"},{"role":"assistant","content":"Hi! How can I help you today? "},{"role":"user","content":"hi"},{"role":"assistant","content":"Hi! How can I assist you today?"},{"role":"user","content":"file\n[[20240618-104928.jpeg]]\ndata:image/jpeg;base64,/9j/4AAQSkZJRgABAQAAAQA ...\n\n\n图片的内容是啥？"}],"max_tokens":2000,"stream":true,"presence_penalty":0,"frequency_penalty":0,"temperature":0.8,"top_p":1}

[Issues-translate-bot]

Bot detected the issue body's language is not English, translate it automatically.

---

Okay, it is indeed supported, but I didn’t see it clearly; I’m sorry, but the content seems incomplete:

{"model":"qwen","messages":[{"role":"user","content":"hi"},{"role":"assistant","content":"Hi! How can I help you today? "},{"role":"user","content":"hi"},{"role":"assistant","content":"Hi! How can I assist you today?" },{"role":"user","content":"```file\n[[20240618-104928.jpeg]]\ndata:image/jpeg;base64,/9j/4AAQSkZJRgABAQAAAQA ...\n` ``\n\nWhat is the content of the picture? "}],"max_tokens":2000,"stream":true,"presence_penalty":0,"frequency_penalty":0,"temperature":0.8,"top_p": 1}

[Sh1n3zZ]

Hi,

你请求的模型 qwen 不支持图片识别，所以被自动省略了。
当模型不支持图片识别时，经过 base64 编码后的图片会被认为是文本，过长的文本（token）会导致过多的费用，所以采取判断省略用于防止出现该情况。

解决方案：使用支持图片识别的模型。如：gpt-4-vision gpt-4o 等

Regards,
Junhai Deng

[Issues-translate-bot]

Bot detected the issue body's language is not English, translate it automatically.

---

Hi,

The model qwen you requested does not support image recognition, so it is automatically omitted.
When the model does not support image recognition, the base64-encoded image will be considered text. Too long text (token) will result in excessive costs, so judgment omission is used to prevent this situation.

Solution: Use a model that supports image recognition. Such as: gpt-4-vision gpt-4o etc.

Regards,
Junhai Deng

[zmh-program]

hey there, your model qwen is not a vision-ability common model, u can use the models mapping or try other models like gpt-4-turbo, claude-3-opus and so on.