[Q&A] 使用 API 中转服务，无法使用插件

[MartialBE]

💻 系统环境 | Operating System

macOS

🌐 浏览器 | Browser

Chrome

🐛 问题描述 | Bug Description

我将国内的讯飞/ 百度 模型 转换成OpenAI API 格式，然后使用 lobe 调用插件时，lobe 无法正确识别函数调用。
好像是和匹配有关。 OpenAI 的 stream 会把 tool_calls 拆分了输出。 而国内的模型函数调用都是一次输出，所以我在转换的时候，函数调用也是一次性输出。

🚦 期望结果 | Expected Behavior

No response

📷 复现步骤 | Recurrence Steps

使用的工具：https://github.com/MartialBE/one-api

开启 天气插件，询问天气：

下面是 我调用API输出的结果：

data: {"id":"as-fthf7qvbyg","object":"chat.completion.chunk","created":1704277500,"model":"ERNIE-Bot-4","choices":[{"index":0,"delta":{"tool_calls":[{"id":"as-fthf7qvbyg","type":"function","function":{"name":"realtime-weather____fetchCurrentWeather","arguments":"{\"city\":\"北京\"}"},"index":0}]},"finish_reason":"tool_calls"}]}

data: [DONE]

下面是 lobe 处理的输出结果：

{"tool_calls":[ {"id": "as-fthf7qvbyg", "type": "function", "function": {"name": "realtime-weather____fetchCurrentWeather", "arguments": "

lobe处理到 arguments 就没有下文了。导致无法正常调用插件。

📝 补充信息 | Additional Information

No response

[lobehubbot]

👀 @MartialBE

Thank you for raising an issue. We will investigate into the matter and get back to you as soon as possible.
Please make sure you have given us as much context as possible.
非常感谢您提交 issue。我们会尽快调查此事，并尽快回复您。 请确保您已经提供了尽可能多的背景信息。

[arvinxx]

这个是不是应该提给 one-api ？我理解这个似乎是因为oneapi 没有按照openai标准方案输出fc 的调用吧？

[lobehubbot]

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

---

Should this be submitted to one-api? I understand this seems to be because oneapi does not output fc calls according to the openai standard scheme, right?

[MartialBE]

是我想错了。
我之前想的是 lobe 是否可以通过finish_reason=tool_calls来判断结束 tool_calls的拼装。所以应该可以支持一次性的tool_calls的读取。
这确实不是lobe的问题。我再去研究一下openai的stream输出。把oneapi那边输出的格式符合openai的规范。

[lobehubbot]

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

---

I was wrong.
What I was thinking before was whether lobe could use finish_reason=tool_calls to determine the end of the assembly of tool_calls. So it should be possible to support one-time reading of tool_calls.
This is really not a lobe problem. I will study the stream output of openai again. Make the format output by oneapi comply with openai's specifications.

[lobehubbot]

✅ @MartialBE

This issue is closed, If you have any questions, you can comment and reply.
此问题已经关闭。如果您有任何问题，可以留言并回复。