[Bug] NVIDIA Provider 测试连通性报错 "OpenAI completion has no choices" ——max_tokens 未正确写入请求 payload

[knight-de-ficus]

What happened / 发生了什么

配置页面测试 nvidia/minimaxai/minimax-m3 时，测试连通性报错 "OpenAI completion has no choices. response_id="。
实际对话时，调用该模型返回为空，报错见日志。

临时解决措施

此措施可以临时解决此bug：模型提供商 - nvidia/minimaxai/minimax-m3 - 自定义请求体参数 - 新键名max_tokens，值类型number - 指定为8192 - 保存

根因

NVIDIA NIM API 的Minimax模型要求请求体中必须包含 max_tokens（其他特定参数如 temperature未测试），否则返回空 choices：

{
  "id": "",
  "choices": [],
  "created": 0,
  "model": "",
  "object": "chat.completion",
  "usage": null
}

而 AstrBot 的 test() 方法调用 text_chat 时，组装出的 payload 只包含 messages 和 model，没有 max_tokens。

Reproduce / 如何复现？

1. 在 AstrBot Desktop 中添加 NVIDIA 提供商，模型选择 minimaxai/minimax-m3
2. 点击「测试连通性」
3. 报错 "OpenAI completion has no choices. response_id="

AstrBot version, deployment method (e.g., Windows Docker Desktop deployment), provider used, and messaging platform used. / AstrBot 版本、部署方式（如 Windows Docker Desktop 部署）、使用的提供商、使用的消息平台适配器

AstrBot 版本： 4.26.4
部署方式： Astrbot Desktop，Windows版本
提供商： NVIDIA（type=openai_chat_completion），模型 nvidia/minimaxai/minimax-m3
消息平台： Astrbot Desktop

OS

Windows

Logs / 报错日志

报错现象

[2026-07-11 13:19:21.348] [Core]
[INFO]
[core.event_bus:74]: [default] [webchat(webchat)] astrbot/astrbot: 你好 
[2026-07-11 13:19:21.367] [Core]
[INFO]
[respond.stage:206]: Prepare to send - astrbot/astrbot: 
[2026-07-11 13:19:21.367] [Core]
[INFO]
[respond.stage:222]: 应用流式输出(webchat)
[2026-07-11 13:19:22.343] [Core]
[WARN]
[v4.26.4] [runners.tool_loop_agent_runner:556]: Chat Model nvidia/minimaxai/minimax-m3 request error: Internal server error
Traceback (most recent call last):
  File "\\?\E:\Astrbot\backend\app\astrbot\core\agent\runners\tool_loop_agent_runner.py", line 511, in _iter_llm_responses_with_fallback
    async for attempt in retrying:
  File "E:\Astrbot\backend\python\Lib\site-packages\tenacity\asyncio\__init__.py", line 170, in __anext__
    do = await self.iter(retry_state=self._retry_state)
         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "E:\Astrbot\backend\python\Lib\site-packages\tenacity\asyncio\__init__.py", line 157, in iter
    result = await action(retry_state)
             ^^^^^^^^^^^^^^^^^^^^^^^^^
  File "E:\Astrbot\backend\python\Lib\site-packages\tenacity\_utils.py", line 111, in inner
    return call(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^
  File "E:\Astrbot\backend\python\Lib\site-packages\tenacity\__init__.py", line 393, in <lambda>
    self._add_action_func(lambda rs: rs.outcome.result())
                                     ^^^^^^^^^^^^^^^^^^^
  File "E:\Astrbot\backend\python\Lib\concurrent\futures\_base.py", line 449, in result
    return self.__get_result()
           ^^^^^^^^^^^^^^^^^^^
  File "E:\Astrbot\backend\python\Lib\concurrent\futures\_base.py", line 401, in __get_result
    raise self._exception
  File "\\?\E:\Astrbot\backend\app\astrbot\core\agent\runners\tool_loop_agent_runner.py", line 515, in _iter_llm_responses_with_fallback
    async for resp in self._iter_llm_responses(
  File "\\?\E:\Astrbot\backend\app\astrbot\core\agent\runners\tool_loop_agent_runner.py", line 475, in _iter_llm_responses
    async for resp in stream:  # type: ignore
  File "\\?\E:\Astrbot\backend\app\astrbot\core\provider\sources\openai_source.py", line 1294, in text_chat_stream
    ) = await self._handle_api_error(
        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "\\?\E:\Astrbot\backend\app\astrbot\core\provider\sources\openai_source.py", line 1158, in _handle_api_error
    raise e
  File "\\?\E:\Astrbot\backend\app\astrbot\core\provider\sources\openai_source.py", line 1277, in text_chat_stream
    async for response in self._query_stream(
  File "\\?\E:\Astrbot\backend\app\astrbot\core\provider\sources\openai_source.py", line 625, in _query_stream
    async for chunk in stream:
  File "E:\Astrbot\backend\python\Lib\site-packages\openai\_streaming.py", line 158, in __aiter__
    async for item in self._iterator:
  File "E:\Astrbot\backend\python\Lib\site-packages\openai\_streaming.py", line 205, in __stream__
    raise APIError(
openai.APIError: Internal server error
[2026-07-11 13:19:22.351] [Core]
[ERRO]
[v4.26.4] [core.astr_main_agent:1049]: Failed to generate webchat title for session ea9adb41-e79b-4338-95e7-d476c5353fd3: OpenAI completion has no choices. response_id=
Traceback (most recent call last):
  File "\\?\E:\Astrbot\backend\app\astrbot\core\astr_main_agent.py", line 1037, in _handle_webchat
    llm_resp = await prov.text_chat(
               ^^^^^^^^^^^^^^^^^^^^^
  File "\\?\E:\Astrbot\backend\app\astrbot\core\provider\sources\openai_source.py", line 1217, in text_chat
    ) = await self._handle_api_error(
        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "\\?\E:\Astrbot\backend\app\astrbot\core\provider\sources\openai_source.py", line 1158, in _handle_api_error
    raise e
  File "\\?\E:\Astrbot\backend\app\astrbot\core\provider\sources\openai_source.py", line 1201, in text_chat
    llm_response = await self._query(
                   ^^^^^^^^^^^^^^^^^^
  File "\\?\E:\Astrbot\backend\app\astrbot\core\provider\sources\openai_source.py", line 569, in _query
    llm_response = await self._parse_openai_completion(completion, tools)
                   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "\\?\E:\Astrbot\backend\app\astrbot\core\provider\sources\openai_source.py", line 824, in _parse_openai_completion
    raise EmptyModelOutputError(
astrbot.core.exceptions.EmptyModelOutputError: OpenAI completion has no choices. response_id=

底层报错

raise EmptyModelOutputError(
    f"OpenAI completion has no choices. response_id={completion.id}"
)

底层 API 返回的裸响应：

{"id":"","choices":[],"created":0,"model":"","object":"chat.completion","usage":null}

Are you willing to submit a PR? / 你愿意提交 PR 吗？

• Yes!

Code of Conduct

• I have read and agree to abide by the project's Code of Conduct。

[dosubot]

Hi @knight-de-ficus! I'm Dosu and I'm helping the AstrBotTeam team.

确认了你的分析是正确的。

根因确认： test() 方法继承自父类 Provider，调用 text_chat 时仅传递了 prompt 参数 [1]：

async def test(self, timeout: float = 45.0) -> None:
    await asyncio.wait_for(
        self.text_chat(prompt="REPLY `PONG` ONLY"),
        timeout=timeout,
    )

而 _prepare_chat_payload() 构建的 payload 也仅包含 messages 和 model [2]：

payloads = {"messages": context_query, "model": model}

没有任何逻辑为 max_tokens 设置默认值，所以对于 NVIDIA NIM API 的 Minimax 模型（要求必须包含 max_tokens），请求体中缺少该字段导致返回空 choices。

你的临时方案（自定义请求体参数中添加 max_tokens）会通过 custom_extra_body 注入到请求中 [3]，这是目前有效的解决方法。

可能的修复方向：

• 在 test() 方法或 payload 构建逻辑中，当用户配置了 max_tokens（或自定义请求体参数中包含它）时，确保测试连通性时也带上该参数
• 或者为特定需要 max_tokens 的提供商设置一个合理默认值

To reply, just mention @dosu.

---

Share context across your team and agents. Try Dosu.