Fix/8267 mimo reasoning content

[NayukiChiba]

Modifications / 改动点

• 消息过滤时增加reasoning_content判断，保留仅含思考内容的assistant消息
• 自动为MiMo推理模型的assistant历史消息注入空reasoning_content，满足API要求
• 通过模型名称集合和xiaomimimo.com端点双重判断是否为MiMo推理模型
• 添加单元测试覆盖不同模型识别、字段注入、端点检测和已有内容保留等场景
  close [Bug] mimo-v2.5适配问题 #8267

在openai_source.py中参考deepseek的reasoning_content回传，增加了mimo的几个模型回传reasoning_content功能

• This is NOT a breaking change. / 这不是一个破坏性变更。

Screenshots or Test Results / 运行截图或测试结果

test_mimo_reasoning_model_adds_empty_reasoning_content：5 个推理模型名触发补 reasoning_content: ""
test_mimo_non_reasoning_model_does_not_add_reasoning_content：TTS类模型不触发
test_mimo_reasoning_preserves_existing_reasoning_content：已有的think ->reasoning_content不会被覆盖

---

Checklist / 检查清单

• 😊 If there are new features added in the PR, I have discussed it with the authors through issues/emails, etc.
  / 如果 PR 中有新加入的功能，已经通过 Issue / 邮件等方式和作者讨论过。

• 👀 My changes have been well-tested, and "Verification Steps" and "Screenshots" have been provided above.
  / 我的更改经过了良好的测试，并已在上方提供了“验证步骤”和“运行截图”。

• 🤓 I have ensured that no new dependencies are introduced, OR if new dependencies are introduced, they have been added to the appropriate locations in requirements.txt and pyproject.toml.
  / 我确保没有引入新依赖库，或者引入了新依赖库的同时将其添加到 requirements.txt 和 pyproject.toml 文件相应位置。

• 😮 My changes do not introduce malicious code.
  / 我的更改没有引入恶意代码。

Summary by Sourcery

Handle MiMo reasoning models’ assistant messages to comply with API reasoning_content requirements and prevent unintended filtering.

Bug Fixes:

• Prevent assistant messages that only contain reasoning_content from being incorrectly filtered as empty.
• Ensure MiMo reasoning models do not fail with 400 errors by always including reasoning_content in assistant history messages when required.

Enhancements:

• Detect MiMo reasoning models via a dedicated model-name set and inject empty reasoning_content for assistant messages when missing.
• Add tests covering MiMo reasoning vs non-reasoning models, reasoning_content injection behavior, and preservation of existing reasoning content.

[sourcery-ai]

Hey - I've found 1 issue, and left some high level feedback:

• The mimo_reasoning_models set is duplicated between the implementation and tests; consider defining a single shared constant (or using a pattern-based check) to avoid future drift when models are added or renamed.
• The PR description mentions using both model-name and xiaomimimo.com endpoint checks to detect MiMo reasoning models, but the current code only checks model names; either add the host-based check (similar to is_deepseek_reasoning) or update the description for consistency.

Prompt for AI Agents

Please address the comments from this code review:

## Overall Comments
- The `mimo_reasoning_models` set is duplicated between the implementation and tests; consider defining a single shared constant (or using a pattern-based check) to avoid future drift when models are added or renamed.
- The PR description mentions using both model-name and `xiaomimimo.com` endpoint checks to detect MiMo reasoning models, but the current code only checks model names; either add the host-based check (similar to `is_deepseek_reasoning`) or update the description for consistency.

## Individual Comments

### Comment 1
<location path="tests/test_openai_source.py" line_range="1808-1810" />
<code_context>
+]
+
+
+@pytest.mark.asyncio
+@pytest.mark.parametrize("model", MIMO_REASONING_MODELS)
+async def test_mimo_reasoning_model_adds_empty_reasoning_content(model: str):
+    """MiMo 推理模型：assistant 消息缺少 reasoning_content 时自动补空字符串"""
+    provider = _make_provider()
</code_context>
<issue_to_address>
**suggestion (testing):** Extend tests to cover interaction with the assistant-message filtering logic when only reasoning_content is present

MiMo tests currently cover injection/preservation of `reasoning_content`, but not its interaction with the "empty assistant message" filter. Please add a test where an assistant message has `reasoning_content` set (possibly injected) and empty/omitted `content`, run it through the filtering flow, and assert that the message is not dropped. This will guard against regressions where MiMo-related messages are filtered out.

Suggested implementation:

```python
        payloads = {
            "model": model,
            "messages": [
                {"role": "user", "content": "hello"},
                {
                    "role": "assistant",
                    "content": "I will help with that.",
                },
                # assistant 消息只包含 reasoning_content，content 为空，用于验证过滤逻辑不会将其丢弃
                {
                    "role": "assistant",
                    "content": "",
                    "reasoning_content": "Thinking about how to help with that.",
                },
            ],

```

To fully implement the requested coverage (asserting the message is not dropped by the "empty assistant message" filter), you should:

1. Locate where this test currently asserts against the serialized / filtered messages (for example, inspecting the outgoing request body captured by an HTTP mock, or the result of an internal normalization/filtering helper).
2. Extend those assertions to verify that the assistant message with `content == ""` and non-empty `reasoning_content` is still present after filtering, e.g.:
   - There is an assistant message in the filtered list whose `reasoning_content` matches `"Thinking about how to help with that."`, and
   - That message has been preserved (i.e., not removed by the empty-assistant-message filter), even though `content` is empty or omitted.
3. If other tests (e.g. something like `test_empty_assistant_messages_are_filtered_out`) already exercise the filter via a specific helper or by inspecting a particular mocked client call, mirror that pattern here so that this MiMo-specific test runs the messages through the exact same filtering flow before asserting on them.
</issue_to_address>

---

Sourcery is free for open source - if you like our reviews please consider sharing them ✨

• X
• Mastodon
• LinkedIn
• Facebook

_{Help me be more useful! Please click 👍 or 👎 on each comment and I'll use the feedback to improve your reviews.}

[sourcery-ai]

suggestion (testing): Extend tests to cover interaction with the assistant-message filtering logic when only reasoning_content is present

MiMo tests currently cover injection/preservation of reasoning_content, but not its interaction with the "empty assistant message" filter. Please add a test where an assistant message has reasoning_content set (possibly injected) and empty/omitted content, run it through the filtering flow, and assert that the message is not dropped. This will guard against regressions where MiMo-related messages are filtered out.

Suggested implementation:

        payloads = {
            "model": model,
            "messages": [
                {"role": "user", "content": "hello"},
                {
                    "role": "assistant",
                    "content": "I will help with that.",
                },
                # assistant 消息只包含 reasoning_content，content 为空，用于验证过滤逻辑不会将其丢弃
                {
                    "role": "assistant",
                    "content": "",
                    "reasoning_content": "Thinking about how to help with that.",
                },
            ],

To fully implement the requested coverage (asserting the message is not dropped by the "empty assistant message" filter), you should:

1. Locate where this test currently asserts against the serialized / filtered messages (for example, inspecting the outgoing request body captured by an HTTP mock, or the result of an internal normalization/filtering helper).
2. Extend those assertions to verify that the assistant message with content == "" and non-empty reasoning_content is still present after filtering, e.g.:
   • There is an assistant message in the filtered list whose reasoning_content matches "Thinking about how to help with that.", and
   • That message has been preserved (i.e., not removed by the empty-assistant-message filter), even though content is empty or omitted.
3. If other tests (e.g. something like test_empty_assistant_messages_are_filtered_out) already exercise the filter via a specific helper or by inspecting a particular mocked client call, mirror that pattern here so that this MiMo-specific test runs the messages through the exact same filtering flow before asserting on them.

[gemini-code-assist]

Code Review

This pull request introduces support for MiMo reasoning models by ensuring that assistant messages include a reasoning_content field, which prevents 400 errors from the API. The changes include updated message filtering, payload conversion logic for specific MiMo models, and new unit tests. Feedback suggests refining the identification of MiMo models by also checking the endpoint host and optimizing performance by moving the model name set to a constant.

[gemini-code-assist]

根据 PR 描述，应该通过模型名称集合和 xiaomimimo.com 端点双重判断是否为 MiMo 推理模型。目前代码中缺少了对端点的判断。此外，根据项目规则，新功能的实现应配套相应的单元测试。

        is_mimo_reasoning = (
            model in mimo_reasoning_models
            or "xiaomimimo.com" in self.client.base_url.host
        )

References

1. New functionality, such as handling attachments, should be accompanied by corresponding unit tests.

[gemini-code-assist]

为了提高性能，建议将 mimo_reasoning_models 定义为类常量或模块级常量，以避免在每次 _finally_convert_payload 调用时都重新创建该集合。