跳至主要内容

如何解析 JSON 输出

虽然一些模型提供商支持返回结构化输出的内置方式，但并非所有模型都支持。我们可以使用输出解析器来帮助用户通过提示指定任意 JSON 架构，查询模型以获取符合该架构的输出，最后将该架构解析为 JSON。

注意

请记住，大型语言模型是泄漏的抽象！你需要使用具有足够容量的 LLM 才能生成格式良好的 JSON。

先决条件

本指南假设你熟悉以下概念

The JsonOutputParser is one built-in option for prompting for and then parsing JSON output.

选择你的聊天模型

安装依赖项

提示

查看 此部分了解有关安装集成包的一般说明.

yarn add @langchain/openai

添加环境变量

OPENAI_API_KEY=your-api-key

实例化模型

import { ChatOpenAI } from "@langchain/openai";

const model = new ChatOpenAI({
  model: "gpt-4o-mini",
  temperature: 0
});
import { ChatOpenAI } from "@langchain/openai";
const model = new ChatOpenAI({
  model: "gpt-4o",
  temperature: 0,
});

import { JsonOutputParser } from "@langchain/core/output_parsers";
import { ChatPromptTemplate } from "@langchain/core/prompts";

// Define your desired data structure. Only used for typing the parser output.
interface Joke {
  setup: string;
  punchline: string;
}

// A query and format instructions used to prompt a language model.
const jokeQuery = "Tell me a joke.";
const formatInstructions =
  "Respond with a valid JSON object, containing two fields: 'setup' and 'punchline'.";

// Set up a parser + inject instructions into the prompt template.
const parser = new JsonOutputParser<Joke>();

const prompt = ChatPromptTemplate.fromTemplate(
  "Answer the user query.\n{format_instructions}\n{query}\n"
);

const partialedPrompt = await prompt.partial({
  format_instructions: formatInstructions,
});

const chain = partialedPrompt.pipe(model).pipe(parser);

await chain.invoke({ query: jokeQuery });
{
  setup: "Why don't scientists trust atoms?",
  punchline: "Because they make up everything!"
}

流式传输​

The JsonOutputParser also supports streaming partial chunks. This is useful when the model returns partial JSON output in multiple chunks. The parser will keep track of the partial chunks and return the final JSON output when the model finishes generating the output.

for await (const s of await chain.stream({ query: jokeQuery })) {
  console.log(s);
}
{}
{ setup: "" }
{ setup: "Why" }
{ setup: "Why don't" }
{ setup: "Why don't scientists" }
{ setup: "Why don't scientists trust" }
{ setup: "Why don't scientists trust atoms" }
{ setup: "Why don't scientists trust atoms?", punchline: "" }
{ setup: "Why don't scientists trust atoms?", punchline: "Because" }
{
  setup: "Why don't scientists trust atoms?",
  punchline: "Because they"
}
{
  setup: "Why don't scientists trust atoms?",
  punchline: "Because they make"
}
{
  setup: "Why don't scientists trust atoms?",
  punchline: "Because they make up"
}
{
  setup: "Why don't scientists trust atoms?",
  punchline: "Because they make up everything"
}
{
  setup: "Why don't scientists trust atoms?",
  punchline: "Because they make up everything!"
}

下一步​

你已经了解了一种提示模型返回结构化 JSON 的方法。接下来，查看获取结构化输出的更广泛指南，了解其他技术。

此页面是否有帮助？


你也可以在 GitHub 上留下详细的反馈 GitHub.