如何使用输出解析器将LLM响应解析为结构化格式

语言模型输出文本。但有时您希望获得比纯文本更结构化的信息。虽然一些模型提供商支持内置方法来返回结构化输出，但并非所有都支持。

输出解析器是帮助组织语言模型响应的类。输出解析器必须实现两个主要方法：

"Get format instructions"：一个方法，返回一个字符串，其中包含关于语言模型输出应如何格式化的说明。
"Parse"：一个方法，接收一个字符串（假定为语言模型的响应）并将其解析为某种结构。

还有一个可选方法：

"Parse with prompt"：一个方法，接收一个字符串（假定为语言模型的响应）和一个提示（假定为生成该响应的提示），并将其解析为某种结构。提供提示主要是为了在输出解析器希望以某种方式重试或修复输出时，需要从提示中获取信息。

开始

下面我们将介绍主要的输出解析器类型，即 PydanticOutputParser。

from langchain_core.output_parsers import PydanticOutputParser
from langchain_core.prompts import PromptTemplate
from langchain_openai import OpenAI
from pydantic import BaseModel, Field, model_validator

model = OpenAI(model_name="gpt-3.5-turbo-instruct", temperature=0.0)


# Define your desired data structure.
class Joke(BaseModel):
    setup: str = Field(description="question to set up a joke")
    punchline: str = Field(description="answer to resolve the joke")

    # You can add custom validation logic easily with Pydantic.
    @model_validator(mode="before")
    @classmethod
    def question_ends_with_question_mark(cls, values: dict) -> dict:
        setup = values.get("setup")
        if setup and setup[-1] != "?":
            raise ValueError("Badly formed question!")
        return values


# Set up a parser + inject instructions into the prompt template.
parser = PydanticOutputParser(pydantic_object=Joke)

prompt = PromptTemplate(
    template="Answer the user query.\n{format_instructions}\n{query}\n",
    input_variables=["query"],
    partial_variables={"format_instructions": parser.get_format_instructions()},
)

# And a query intended to prompt a language model to populate the data structure.
prompt_and_model = prompt | model
output = prompt_and_model.invoke({"query": "Tell me a joke."})
parser.invoke(output)

API 参考：PydanticOutputParser | PromptTemplate | OpenAI

Joke(setup='Why did the tomato turn red?', punchline='Because it saw the salad dressing!')

LCEL

输出解析器实现了可运行接口，这是LangChain 表达式语言 (LCEL) 的基本构建块。这意味着它们支持 invoke、ainvoke、stream、astream、batch、abatch、astream_log 调用。

输出解析器接受字符串或 BaseMessage 作为输入，并可以返回任意类型。

parser.invoke(output)

Joke(setup='Why did the tomato turn red?', punchline='Because it saw the salad dressing!')

除了手动调用解析器之外，我们也可以将其添加到我们的 Runnable 序列中

chain = prompt | model | parser
chain.invoke({"query": "Tell me a joke."})

Joke(setup='Why did the tomato turn red?', punchline='Because it saw the salad dressing!')

虽然所有解析器都支持流式接口，但只有某些解析器可以流式传输部分解析的对象，因为这高度依赖于输出类型。无法构建部分对象的解析器将只生成完全解析的输出。

例如，SimpleJsonOutputParser 可以流式传输部分输出

from langchain.output_parsers.json import SimpleJsonOutputParser

json_prompt = PromptTemplate.from_template(
    "Return a JSON object with an `answer` key that answers the following question: {question}"
)
json_parser = SimpleJsonOutputParser()
json_chain = json_prompt | model | json_parser

API 参考：SimpleJsonOutputParser

list(json_chain.stream({"question": "Who invented the microscope?"}))

[{},
 {'answer': ''},
 {'answer': 'Ant'},
 {'answer': 'Anton'},
 {'answer': 'Antonie'},
 {'answer': 'Antonie van'},
 {'answer': 'Antonie van Lee'},
 {'answer': 'Antonie van Leeu'},
 {'answer': 'Antonie van Leeuwen'},
 {'answer': 'Antonie van Leeuwenho'},
 {'answer': 'Antonie van Leeuwenhoek'}]

类似地，对于 PydanticOutputParser

list(chain.stream({"query": "Tell me a joke."}))

[Joke(setup='Why did the tomato turn red?', punchline=''),
 Joke(setup='Why did the tomato turn red?', punchline='Because'),
 Joke(setup='Why did the tomato turn red?', punchline='Because it'),
 Joke(setup='Why did the tomato turn red?', punchline='Because it saw'),
 Joke(setup='Why did the tomato turn red?', punchline='Because it saw the'),
 Joke(setup='Why did the tomato turn red?', punchline='Because it saw the salad'),
 Joke(setup='Why did the tomato turn red?', punchline='Because it saw the salad dressing'),
 Joke(setup='Why did the tomato turn red?', punchline='Because it saw the salad dressing!')]

开始​

LCEL​

开始

LCEL