feat: add rerank provider for Text Embeddings Inference (TEI)

[Takeoff0518]

Text Embeddings Inference(TEI) 是由 HuggingFace 开源的一个专门用于部署和提供 embedding 和 reranking 模型的工具。

Closes #9266

TEI 的 OpenAPI

Modifications / 改动点

添加了 TEIRerankProvider 以及相关默认配置，使得框架可以使用 TEI rerank provider 提供 rerank 模型。

由于一个 TEI 容器实例在启动时就绑定了一个固定的模型，所以不需要再单独指定模型。

• 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

---

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

Add HuggingFace Text Embeddings Inference (TEI) as a rerank provider and wire it into the provider system and dashboard configuration.

New Features:

• Introduce TEIRerankProvider to integrate HuggingFace TEI reranking into the provider framework.
• Add default configuration and UI metadata for the TEI rerank provider, including truncation, score format, and return-text options.

Enhancements:

• Register the TEI rerank provider in the dynamic provider manager and expose a HuggingFace icon in the dashboard for visual identification.

[gemini-code-assist]

Warning

You have reached your daily quota limit. Please wait up to 24 hours and I will start processing your requests again!

[sourcery-ai]

Hey - I've found 3 issues, and left some high level feedback:

• The TEIRerankProvider creates an aiohttp.ClientSession in __init__; consider lazy-initializing the session in an async method (or ensuring this matches the existing pattern) to avoid potential event loop issues when constructing providers synchronously.
• After terminate, self.client is set to None, but rerank only checks self.client/self.client.closed and does not recreate the session; you may want to add logic to reinitialize the client or clearly prevent further use after termination.
• Error handling currently raises generic Exception instances with user-facing messages; using more specific exception types (or a provider-specific error class) would make upstream error handling and debugging more precise.

Prompt for AI Agents

Please address the comments from this code review:

## Overall Comments
- The `TEIRerankProvider` creates an `aiohttp.ClientSession` in `__init__`; consider lazy-initializing the session in an async method (or ensuring this matches the existing pattern) to avoid potential event loop issues when constructing providers synchronously.
- After `terminate`, `self.client` is set to `None`, but `rerank` only checks `self.client`/`self.client.closed` and does not recreate the session; you may want to add logic to reinitialize the client or clearly prevent further use after termination.
- Error handling currently raises generic `Exception` instances with user-facing messages; using more specific exception types (or a provider-specific error class) would make upstream error handling and debugging more precise.

## Individual Comments

### Comment 1
<location path="astrbot/core/provider/sources/tei_rerank_source.py" line_range="26-27" />
<code_context>
+        self.base_url = provider_config.get("rerank_api_base", "http://127.0.0.1:8080").rstrip("/")
+        self.timeout = provider_config.get("timeout", 20)
+        self.truncate = provider_config.get("tei_rerank_truncate", False)
+        self.truncation_direction = provider_config.get(
+            "tei_rerank_truncation_direction", "Right"
+        )
+        self.raw_scores = provider_config.get("tei_rerank_raw_scores", False)
</code_context>
<issue_to_address>
**issue (bug_risk):** Normalize or validate truncation_direction to match TEI API expectations.

TEI’s `/rerank` API expects lowercase values like `"left"` / `"right"`, but this defaults to `"Right"` and forwards it unchanged. To avoid request failures or ignored params, normalize the value (e.g., `self.truncation_direction.lower()`) and/or validate it against the allowed options before sending the request.
</issue_to_address>

### Comment 2
<location path="astrbot/core/provider/sources/tei_rerank_source.py" line_range="46-48" />
<code_context>
+        documents: list[str],
+        top_n: int | None = None,
+    ) -> list[RerankResult]:
+        if not self.client or self.client.closed:
+            logger.error("[TEI Rerank] Client session is not initialized or closed")
+            return []
+        if not documents:
+            logger.warning("[TEI Rerank] Document list is empty, returning empty results")
</code_context>
<issue_to_address>
**issue (bug_risk):** Returning an empty list when the client session is closed can silently hide real errors.

When `self.client` is `None` or closed, we log an error but still return an empty list. This is indistinguishable from a legitimate "no results" response and can obscure session/connection failures. It would be better to either recreate the client here or raise an exception so callers can detect and handle the failure explicitly.
</issue_to_address>

### Comment 3
<location path="astrbot/core/provider/sources/tei_rerank_source.py" line_range="55-57" />
<code_context>
+        if not query.strip():
+            logger.warning("[TEI Rerank] Query is empty, returning empty results")
+            return []
+        payload: dict = {
+            "query": query,
+            "texts": documents,
+        }
+
</code_context>
<issue_to_address>
**suggestion (performance):** Consider passing top_n to TEI so the server can limit results for performance.

Currently TEI returns all reranked documents and you truncate with `results = results[:top_n]`. If `/rerank` supports a `top_n` parameter, consider adding it to `payload` when provided so the service can do the truncation, reducing both work and response size.

Suggested implementation:

```python
        if not query.strip():
            logger.warning("[TEI Rerank] Query is empty, returning empty results")
            return []

        payload: dict = {
            "query": query,
            "texts": documents,
        }

        if top_n is not None:
            payload["top_n"] = top_n

```

1. Ensure the method signature includes a `top_n: int | None = None` (or equivalent) parameter if it does not already.
2. Verify that wherever this method is called, `top_n` is passed appropriately.
3. Confirm that the HTTP request to the TEI `/rerank` endpoint uses the `payload` dictionary defined here as its body.
</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.}

[Takeoff0518]

All things done.

[Takeoff0518]

新的提交 b539b91 可以正常检索与召回。