> ## Documentation Index
> Fetch the complete documentation index at: https://langchain-zh.cn/llms.txt
> Use this file to discover all available pages before exploring further.

# 如何评估可运行对象

<Info>
  * `langchain`: [Python](https://docs.langchain.com/oss/python/langchain/overview) 和 [JS/TS](https://docs.langchain.com/oss/javascript/langchain/overview)
  * Runnable: [Python](https://reference.langchain.com/python/langchain_core/runnables/) 和 [JS/TS](https://reference.langchain.com/javascript/classes/_langchain_core.runnables.Runnable.html)
</Info>

`langchain` 的 [`Runnable`](https://reference.langchain.com/python/langchain_core/runnables/) 对象（例如聊天模型、检索器、链等）可以直接传入 `evaluate()` / `aevaluate()` 进行评估。

## 设置

让我们定义一个简单的链来进行评估。首先，安装所有必需的包：

<CodeGroup>
  ```bash Python theme={"theme":{"light":"catppuccin-latte","dark":"catppuccin-mocha"}}
  pip install -U langsmith langchain[openai]
  ```

  ```bash TypeScript theme={"theme":{"light":"catppuccin-latte","dark":"catppuccin-mocha"}}
  yarn add langsmith @langchain/openai
  ```
</CodeGroup>

现在定义一个链：

<CodeGroup>
  ```python Python theme={"theme":{"light":"catppuccin-latte","dark":"catppuccin-mocha"}}
  from langchain.chat_models import init_chat_model
  from langchain_core.prompts import ChatPromptTemplate
  from langchain_core.output_parsers import StrOutputParser

  instructions = (
      "请检查下面的用户查询，判断其是否包含任何形式的毒性行为，例如侮辱、威胁或高度负面的评论。"
      "如果包含，请回复 'Toxic'；如果不包含，请回复 'Not toxic'。"
  )

  prompt = ChatPromptTemplate(
      [("system", instructions), ("user", "{text}")],
  )

  model = init_chat_model("gpt-4.1")
  chain = prompt | model | StrOutputParser()
  ```

  ```typescript TypeScript theme={"theme":{"light":"catppuccin-latte","dark":"catppuccin-mocha"}}
  import { ChatOpenAI } from "@langchain/openai";
  import { ChatPromptTemplate } from "@langchain/core/prompts";
  import { StringOutputParser } from "@langchain/core/output_parsers";

  const prompt = ChatPromptTemplate.fromMessages([
    ["system", "请检查下面的用户查询，判断其是否包含任何形式的毒性行为，例如侮辱、威胁或高度负面的评论。如果包含，请回复 'Toxic'；如果不包含，请回复 'Not toxic'。"],
    ["user", "{text}"]
  ]);

  const chatModel = new ChatOpenAI();
  const outputParser = new StringOutputParser();
  const chain = prompt.pipe(chatModel).pipe(outputParser);
  ```
</CodeGroup>

## 评估

要评估我们的链，我们可以直接将其传递给 `evaluate()` / `aevaluate()` 方法。请注意，链的输入变量必须与示例输入的键匹配。在本例中，示例输入应具有 `{"text": "..."}` 的形式。

<CodeGroup>
  ```python Python theme={"theme":{"light":"catppuccin-latte","dark":"catppuccin-mocha"}}
  from langsmith import aevaluate, Client

  client = Client()

  # 克隆一个带有毒性标签的文本数据集。
  # 每个示例输入都有一个 "text" 键，每个输出都有一个 "label" 键。
  dataset = client.clone_public_dataset(
      "https://smith.langchain.com/public/3d6831e6-1680-4c88-94df-618c8e01fc55/d"
  )

  def correct(outputs: dict, reference_outputs: dict) -> bool:
      # 由于我们的链输出的是字符串而非字典，该字符串
      # 会被存储在 outputs 字典的默认 "output" 键下：
      actual = outputs["output"]
      expected = reference_outputs["label"]
      return actual == expected

  results = await aevaluate(
      chain,
      data=dataset,
      evaluators=[correct],
      experiment_prefix="gpt-4.1, baseline",
  )
  ```

  ```typescript TypeScript theme={"theme":{"light":"catppuccin-latte","dark":"catppuccin-mocha"}}
  import { evaluate } from "langsmith/evaluation";
  import { Client } from "langsmith";

  const langsmith = new Client();

  const dataset = await client.clonePublicDataset(
    "https://smith.langchain.com/public/3d6831e6-1680-4c88-94df-618c8e01fc55/d"
  )

  await evaluate(chain, {
    data: dataset.name,
    evaluators: [correct],
    experimentPrefix: "gpt-4.1, baseline",
  });
  ```
</CodeGroup>

每个输出都会适当地追踪可运行对象。

<img src="https://mintcdn.com/hhh-8c10bf0c/BCyPqRNhtAjzdCmk/langsmith/images/runnable-eval.png?fit=max&auto=format&n=BCyPqRNhtAjzdCmk&q=85&s=768a5ab976fc8fc61ad8bf674b8ad50c" alt="Runnable Evaluation" width="2288" height="1052" data-path="langsmith/images/runnable-eval.png" />

## 相关

* [如何评估 `langgraph` 图](/langsmith/evaluate-on-intermediate-steps)

***

<div className="source-links">
  <Callout icon="edit">
    [Edit this page on GitHub](https://github.com/langchain-ai/docs/edit/main/src/i18n\zh-CN\langsmith\langchain-runnable.mdx) or [file an issue](https://github.com/langchain-ai/docs/issues/new/choose).
  </Callout>

  <Callout icon="terminal-2">
    [Connect these docs](/use-these-docs) to Claude, VSCode, and more via MCP for real-time answers.
  </Callout>
</div>