本文将演示使用大语言模型从文本中提炼结构化信息。  
我们将用 `llama3.1` 和 `deepseek` 做一个简单的对比。
> 由于 `langchain` 可能对不同大模型支持程度不同，不同大模型的特点也不同，所以这个对比并不能说明哪个模型更好。

准备

在正式开始撸代码之前，需要准备一下编程环境。

1. 计算机  
本文涉及的所有代码可以在没有显存的环境中执行。 我使用的机器配置为：  
    - CPU: Intel i5-8400 2.80GHz  
    - 内存: 16GB  

2. Visual Studio Code 和 venv
这是很受欢迎的开发工具，相关文章的代码可以在 `Visual Studio Code` 中开发和调试。 我们用 `python` 的 `venv` 创建虚拟环境, 详见：  

3. Ollama
在 `Ollama` 平台上部署本地大模型非常方便，基于此平台，我们可以让 `langchain` 使用 `llama3.1`、`qwen2.5` 等各种本地大模型。详见：  

定义数据格式

 
我们先用 `Pydantic` 来定义要提取的数据格式。
class Person(BaseModel):
    """Information about a person."""

    # ^ Doc-string for the entity Person.
    # This doc-string is sent to the LLM as the description of the schema Person,
    # and it can help to improve extraction results.

    # Note that:
    # 1. Each field is an `optional` -- this allows the model to decline to extract it!
    # 2. Each field has a `description` -- this description is used by the LLM.
    # Having a good description can help improve extraction results.
    name: Optional[str] = Field(default=None, description="The name of the person")
    hair_color: Optional[str] = Field(
        default=None, description="The color of the person's hair if known"
    )
    height_in_meters: Optional[float] = Field(
        default=None, description="Height measured in meters"
    )

　 

设置提示词

prompt_template = ChatPromptTemplate.from_messages(
    [
        (
            "system",
            "You are an expert extraction algorithm. "
            "Only extract relevant information from the text. "
            "If you do not know the value of an attribute asked to extract, "
            "return null for the attribute's value.",
        ),
        # Please see the how-to about improving performance with
        # reference examples.
        # MessagesPlaceholder('examples'),
        ("human", "{text}"),
    ]
)

 

提取Person

from langchain_ollama import ChatOllama

def single_entry(model_name,text):

    structured_llm = ChatOllama(model=model_name,temperature=0.5,verbose=True).with_structured_output(schema=Person)

    prompt = prompt_template.invoke({"text": text})
    response = structured_llm.invoke(prompt)
    return response

 

我们用下面的方法做一下测试。
text = "Alan Smith is 1.83 meters tall and has blond hair."
response = single_entry("llama3.1",text)
print(f'\n llama3.1 response:\n{response}')

response = single_entry("deepseek-r1",text)
print(f'\n deepseek-r1 response:\n{response}')

text = "Alan Smith is 6 feet tall and has blond hair."
response = single_entry("llama3.1",text)
print(f'\n llama3.1 response:\n{response}')

response = single_entry("deepseek-r1",text)
print(f'\n deepseek-r1 response:\n{response}')

　　

测试结果：
   llama3.1 deepseek-r1-tool-calling
text = "Alan Smith is 1.83 meters tall and has blond hair." name='Alan Smith' hair_color='blond' height_in_meters=1.83 name='Alan Smith' hair_color='blond' height_in_meters=1.83
text = "Alan Smith is 6 feet tall and has blond hair." name='Alan Smith' hair_color='blond' height_in_meters=None name=None hair_color='blond' height_in_meters=1.8288
 
第一个问题比较简单，两个模型表现都不错；第二个问题难了一些，大模型需要能把英尺转换成米，我们可以看出 `llama3.1` 没有做转换，没有解析出来 height_in_meters ,`deepseek-r1` 则成功的将英尺转换成米，但是却没有提取出 name 。

提取多个 Person

def single_entry(model_name,text):

    structured_llm = ChatOllama(model=model_name,temperature=0.5,verbose=True).with_structured_output(schema=Person)

    prompt = prompt_template.invoke({"text": text})
    response = structured_llm.invoke(prompt)
    return response

from typing import List

class Data(BaseModel):
    """Extracted data about people."""

    # Creates a model so that we can extract multiple entities.
    people: List[Person]

def multiple_entry(model_name,text):
    structured_llm = ChatOllama(model=model_name,temperature=0.5,verbose=True).with_structured_output(schema=Data)  
    response = structured_llm.invoke([text])
    prompt = prompt_template.invoke({"text": text})
    response = structured_llm.invoke(prompt)
    return response

　　

我们对比测试一下：
text = "Alan Smith is 1.83 meters tall and has blond hair. John Doe is 1.72 meters tall and has brown hair."
response = multiple_entry("llama3.1",text)
print(f'\n llama3.1 response:\n{response}')

response = multiple_entry("MFDoom/deepseek-r1-tool-calling:7b",text)
print(f'\n deepseek response:\n{response}')

text = "Alan Smith is 1.88 meters tall and has blond hair. John Doe is 7 feet tall and has brown hair."
response = multiple_entry("llama3.1",text)
print(f'\n llama3.1 response:\n{response}')

response = multiple_entry("MFDoom/deepseek-r1-tool-calling:7b",text)
print(f'\n deepseek response:\n{response}')

　

  llama3.1  deepseek-r1-tool-calling
Alan Smith is 1.83 meters tall and has blond hair. John Doe is 1.72 meters tall and has brown hair. people=[Person(name='Alan Smith', hair_color=None, height_in_meters=None), Person(name='John Doe', hair_color=None, height_in_meters=None)] 失败
Alan Smith is 1.88 meters tall and has blond hair. John Doe is 7 feet tall and has brown hair. people=[Person(name=None, hair_color=None, height_in_meters=None), Person(name=None, hair_color=None, height_in_meters=None)] people=[Person(name=None, hair_color='blond', height_in_meters=None)]
 
这次表现都不好。当然，估计用参数更多的大模型能够正确解析出内容。

总结

 
通过以上的演练，我们可以看出 `llama3.1` 和 `deepseek` 在提取单个对象的时候表现还可以，但是提取列表都不行。`deepseek` 似乎考虑问题更复杂，推理时间也长一些；另外，`deepseek` 似乎稳定性差一点：有时候推理可以成功，有时候会失败。

代码

本文涉及的所有代码以及相关资源都已经共享，参见：
- [github]
- [gitee]

参考:

🪐祝好运🪐