LangChain 提示模板之少样本示例(二)

https://python.langchain.com.cn/docs/modules/model_io/prompts/prompt_templates/few_shot_examples

一、先明确原文核心目标（Use Case）

原文开篇就定了任务：为"自问自答与搜索"配置少样本示例 ------简单说，就是让LLM遇到问题时，先判断"是否需要追问中间问题"，再一步步推导最终答案（比如问"谁活得更长"，先追问"两人各自活了多少岁"，再对比）。

所有代码和步骤，都是为了实现这个任务。

二、第一部分：直接使用示例集（原文第一大板块）

原文第一步讲"手动准备好所有示例，直接传给模板"，分3个具体步骤，每个步骤都对应原文代码。

步骤1：创建少样本示例集（原文核心代码）

原文说"每个示例是包含'输入变量'的字典"，这里的输入变量是question（问题）和answer（带追问步骤的答案），代码完全照搬原文：

# 原文代码：少样本示例列表（4个自问自答案例）
from langchain.prompts.few_shot import FewShotPromptTemplate
from langchain.prompts.prompt import PromptTemplate

examples = [
  {
    "question": "Who lived longer, Muhammad Ali or Alan Turing?",
    "answer": 
"""
Are follow up questions needed here: Yes.
Follow up: How old was Muhammad Ali when he died?
Intermediate answer: Muhammad Ali was 74 years old when he died.
Follow up: How old was Alan Turing when he died?
Intermediate answer: Alan Turing was 41 years old when he died.
So the final answer is: Muhammad Ali
"""
  },
  {
    "question": "When was the founder of craigslist born?",
    "answer": 
"""
Are follow up questions needed here: Yes.
Follow up: Who was the founder of craigslist?
Intermediate answer: Craigslist was founded by Craig Newmark.
Follow up: When was Craig Newmark born?
Intermediate answer: Craig Newmark was born on December 6, 1952.
So the final answer is: December 6, 1952
"""
  },
  {
    "question": "Who was the maternal grandfather of George Washington?",
    "answer":
"""
Are follow up questions needed here: Yes.
Follow up: Who was the mother of George Washington?
Intermediate answer: The mother of George Washington was Mary Ball Washington.
Follow up: Who was the father of Mary Ball Washington?
Intermediate answer: The father of Mary Ball Washington was Joseph Ball.
So the final answer is: Joseph Ball
"""
  },
  {
    "question": "Are both the directors of Jaws and Casino Royale from the same country?",
    "answer":
"""
Are follow up questions needed here: Yes.
Follow up: Who is the director of Jaws?
Intermediate Answer: The director of Jaws is Steven Spielberg.
Follow up: Where is Steven Spielberg from?
Intermediate Answer: The United States.
Follow up: Who is the director of Casino Royale?
Intermediate Answer: The director of Casino Royale is Martin Campbell.
Follow up: Where is Martin Campbell from?
Intermediate Answer: New Zealand.
So the final answer is: No
"""
  }
]

• 原文关键说明：每个字典里的question和answer是"固定输入变量名"，后面的模板会用这两个名字取数据，不能随便改。

步骤2：创建"单个示例的格式化程序"（原文叫 example_prompt）

原文说这一步是"定义单个示例要怎么显示"------把每个examples里的字典（question+answer）转成固定格式的文本，代码照搬原文：

# 原文代码：单个示例的格式化模板
example_prompt = PromptTemplate(
    input_variables=["question", "answer"],  # 必须和示例字典的key对应
    template="Question: {question}\n{answer}"  # 单个示例的显示格式：先问题，再答案
)

# 原文测试：打印第一个示例的格式化结果
print(example_prompt.format(**examples[0]))

• 原文运行结果（你会看到的内容）：

  Question: Who lived longer, Muhammad Ali or Alan Turing?

    Are follow up questions needed here: Yes.
    Follow up: How old was Muhammad Ali when he died?
    Intermediate answer: Muhammad Ali was 74 years old when he died.
    Follow up: How old was Alan Turing when he died?
    Intermediate answer: Alan Turing was 41 years old when he died.
    So the final answer is: Muhammad Ali

• 难点解释：**examples[0]是"解包字典"------把examples[0]里的question和answer自动传给example_prompt的input_variables，不用手动写question=examples[0]["question"], answer=examples[0]["answer"]。

步骤3：创建少样本提示模板（FewShotPromptTemplate）

原文说这是"把示例集和格式化程序组合起来"，再加上"最终要问的问题"，代码照搬原文：

# 原文代码：创建完整的少样本提示模板
prompt = FewShotPromptTemplate(
    examples=examples,  # 步骤1准备的所有示例
    example_prompt=example_prompt,  # 步骤2定义的单个示例格式化模板
    suffix="Question: {input}",  # 示例之后的"最终问题"（{input}是用户要传的问题）
    input_variables=["input"]  # 告诉模板：用户需要传的参数是"input"（即最终问题）
)

# 原文测试：传入最终问题，生成完整提示
print(prompt.format(input="Who was the father of Mary Ball Washington?"))

• 原文运行结果（你会看到的内容）：
  模板会自动把所有示例按example_prompt格式拼好，最后加上你传的问题：

  Question: Who lived longer, Muhammad Ali or Alan Turing?

    Are follow up questions needed here: Yes.
    Follow up: How old was Muhammad Ali when he died?
    Intermediate answer: Muhammad Ali was 74 years old when he died.
    Follow up: How old was Alan Turing when he died?
    Intermediate answer: Alan Turing was 41 years old when he died.
    So the final answer is: Muhammad Ali
    
    
    Question: When was the founder of craigslist born?
    
    Are follow up questions needed here: Yes.
    Follow up: Who was the founder of craigslist?
    Intermediate answer: Craigslist was founded by Craig Newmark.
    Follow up: When was Craig Newmark born?
    Intermediate answer: Craig Newmark was born on December 6, 1952.
    So the final answer is: December 6, 1952
    
    
    Question: Who was the maternal grandfather of George Washington?
    
    Are follow up questions needed here: Yes.
    Follow up: Who was the mother of George Washington?
    Intermediate answer: The mother of George Washington was Mary Ball Washington.
    Follow up: Who was the father of Mary Ball Washington?
    Intermediate answer: The father of Mary Ball Washington was Joseph Ball.
    So the final answer is: Joseph Ball
    
    
    Question: Are both the directors of Jaws and Casino Royale from the same country?
    
    Are follow up questions needed here: Yes.
    Follow up: Who is the director of Jaws?
    Intermediate Answer: The director of Jaws is Steven Spielberg.
    Follow up: Where is Steven Spielberg from?
    Intermediate Answer: The United States.
    Follow up: Who is the director of Casino Royale?
    Intermediate Answer: The director of Casino Royale is Martin Campbell.
    Follow up: Where is Martin Campbell from?
    Intermediate Answer: New Zealand.
    So the final answer is: No
    
    
    Question: Who was the father of Mary Ball Washington?

• 原文关键说明：suffix是"示例的后缀"，也就是所有示例显示完后，最后要问的问题；input_variables=["input"]表示你调用format时，必须传input参数（即你的问题）。

三、第二部分：使用示例选择器（原文第二大板块）

原文说这是"进阶用法"------不把所有示例都放进提示，而是"根据用户问题，选最相似的1个示例"（减少提示长度，让LLM更聚焦），分3个步骤，代码完全对应原文。

步骤1：创建示例选择器（SemanticSimilarityExampleSelector）

原文用"语义相似度"选示例（比如用户问"Mary Ball Washington的父亲是谁"，选和"George Washington外祖父"最像的示例），代码照搬原文：

# 原文代码：导入需要的工具（向量存储、嵌入模型、示例选择器）
from langchain.prompts.example_selector import SemanticSimilarityExampleSelector
from langchain.vectorstores import Chroma
from langchain.embeddings import OpenAIEmbeddings

# 原文代码：创建示例选择器
example_selector = SemanticSimilarityExampleSelector.from_examples(
    examples,  # 还是步骤1的示例集（所有可选的示例）
    OpenAIEmbeddings(),  # 用OpenAI的嵌入模型（把文字转成"能算相似度的数字"）
    Chroma,  # 用Chroma向量存储（存嵌入后的数字，方便找相似）
    k=1  # 只选"最相似的1个示例"
)

• 难点解释：
  • "嵌入模型（OpenAIEmbeddings）"：把"问题文字"转成一串数字（比如"苹果"转成[0.1, 0.2,...]），数字越像，文字语义越近；
  • "Chroma"：专门存这些数字的工具，能快速找出和"用户问题的数字"最像的"示例数字"；
  • "k=1"：只选1个最像的示例，原文用k=1，你也可以改k=2选2个。

步骤2：测试示例选择器（选最相似的示例）

原文用"Who was the father of Mary Ball Washington?"这个问题测试，看选哪个示例，代码照搬原文：

# 原文代码：定义要测试的问题
question = "Who was the father of Mary Ball Washington?"

# 原文代码：根据问题选最相似的示例
selected_examples = example_selector.select_examples({"question": question})

# 原文代码：打印选中的示例
print(f"Examples most similar to the input: {question}")
for example in selected_examples:
    print("\n")
    for k, v in example.items():
        print(f"{k}: {v}")

• 原文运行结果（你会看到的内容）：
  会选中"George Washington外祖父"的示例，因为两个问题都和"Mary Ball Washington的亲属"相关：

  Running Chroma using direct local API.
  Using DuckDB in-memory for database. Data will be transient.
  Examples most similar to the input: Who was the father of Mary Ball Washington?

  question: Who was the maternal grandfather of George Washington?
  answer:
  Are follow up questions needed here: Yes.
  Follow up: Who was the mother of George Washington?
  Intermediate answer: The mother of George Washington was Mary Ball Washington.
  Follow up: Who was the father of Mary Ball Washington?
  Intermediate answer: The father of Mary Ball Washington was Joseph Ball.
  So the final answer is: Joseph Ball

步骤3：用示例选择器创建少样本模板

原文说"把之前的examples参数换成example_selector"，其他不变，代码照搬原文：

# 原文代码：创建少样本模板（用示例选择器代替所有示例）
prompt = FewShotPromptTemplate(
    example_selector=example_selector,  # 步骤1创建的示例选择器
    example_prompt=example_prompt,  # 还是步骤2的单个示例格式化模板
    suffix="Question: {input}",  # 还是最终问题的格式
    input_variables=["input"]  # 还是用户传"input"参数
)

# 原文测试：传入问题，生成完整提示
print(prompt.format(input="Who was the father of Mary Ball Washington?"))

• 原文运行结果（你会看到的内容）：
  这次只显示"最相似的1个示例"，而不是所有4个，最后加问题：

  Question: Who was the maternal grandfather of George Washington?

    Are follow up questions needed here: Yes.
    Follow up: Who was the mother of George Washington?
    Intermediate answer: The mother of George Washington was Mary Ball Washington.
    Follow up: Who was the father of Mary Ball Washington?
    Intermediate answer: The father of Mary Ball Washington was Joseph Ball.
    So the final answer is: Joseph Ball
    
    
    Question: Who was the father of Mary Ball Washington?

• 原文关键说明：这种方式的好处是"提示更短"，LLM不用看无关示例，回答更准、更快。

四、原文核心总结（只提炼原文内容）

1. 少样本提示模板的两种用法 ：
   • 直接用examples：把所有示例都放进提示，适合示例少的场景；
   • 用example_selector：按语义相似度选最相关的示例，适合示例多的场景。
2. 必须有的3个核心组件 ：
   • 示例集（examples）：字典列表，每个字典含input_variables对应的key；
   • 单个示例格式化模板（example_prompt）：定义单个示例的显示格式；
   • 少样本模板（FewShotPromptTemplate）：组合示例、格式化模板和最终问题。
3. 示例选择器的核心工具 ：
   • SemanticSimilarityExampleSelector：按语义选示例；
   • OpenAIEmbeddings：把文字转成相似度数字；
   • Chroma：存储数字并找相似。