在 langchain 中一起使用 Chain 和 Parser

Question

langchain 文档包含此示例，用于配置和调用PydanticOutputParser

# 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.
    @validator('setup')
    def question_ends_with_question_mark(cls, field):
        if field[-1] != '?':
            raise ValueError("Badly formed question!")
        return field

# And a query intented to prompt a language model to populate the data structure.
joke_query = "Tell me a joke."

# 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()}
)

然而，实际进行 API 调用的代码有点奇怪：

model_name = 'text-davinci-003'
temperature = 0.0
my_llm = OpenAI(model_name=model_name, temperature=temperature)

_input = prompt.format_prompt(query=joke_query)
output = my_llm(_input.to_string())

parser.parse(output)

这正是我们想要的：Joke(setup='Why did the chicken cross the road?', punchline='To get to the other side!')

然而，不用Chains于此似乎很奇怪。

我可以接近一下，如下：

chain = LLMChain(llm=my_llm, prompt=prompt)
chain.run(query=joke_query)

但这会返回原始的、未解析的文本：'\n{"setup": "Why did the chicken cross the road?", "punchline": "To get to the other side!"}'

是否有一种首选方法可以让Chain类充分利用 aParser并返回解析后的对象？我可以子类化和扩展LLMChain，但如果此功能尚不存在，我会感到惊讶。

Answer 1

您可以TransformChain为此使用 a ！

from langchain.chat_models import ChatOpenAI
from langchain.chains import LLMChain, TransformChain
from langchain.chains import SequentialChain


llm = ChatOpenAI(temperature=0.5)
llm_chain = LLMChain(
    prompt=prompt,
    llm=llm,
    output_key="json_string",
)

def parse_output(inputs: dict) -> dict:
    text = inputs["json_string"]
    return {"result": parser.parse(text)}

transform_chain = TransformChain(
    input_variables=["json_string"],
    output_variables=["result"],
    transform=parse_output
)

chain = SequentialChain(
    input_variables=["joke_query"],
    output_variables=["result"],
    chains=[llm_chain, transform_chain],
)

chain.run(query="Tell me a joke.")