LangChain Few-Shot Prompt Templates(two)

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

This demo shows how the example selector picks the most relevant example to help the LLM answer a user's question effectively.

Step 1: Understand What We're Building

We'll create a tool that:

  • Uses "semantic similarity" to find the most relevant example for a user's question (instead of using all examples).
  • Teaches the LLM to answer questions by first asking follow-ups (just like the examples).
  • Produces a clear, step-by-step answer to a real user question.

Complete Code Demo

python 复制代码
# 1. Import tools we need (from the original text's Part 2)
from langchain.prompts.few_shot import FewShotPromptTemplate
from langchain.prompts.prompt import PromptTemplate
from langchain.prompts.example_selector import SemanticSimilarityExampleSelector
from langchain.vectorstores import Chroma
from langchain.embeddings import OpenAIEmbeddings
from langchain.chat_models import ChatOpenAI
import os

# 2. Set your OpenAI API key (get one from https://platform.openai.com/)
os.environ["OPENAI_API_KEY"] = "your-openai-api-key-here"

# 3. Define our few-shot examples (same as original text)
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
"""
  }
]

# 4. Create the "example formatter" (how each example is displayed)
example_prompt = PromptTemplate(
    input_variables=["question", "answer"],
    template="Question: {question}\n{answer}"  # Format: "Question: X\nAnswer: Y"
)

# 5. Create the Example Selector (the key part from Part 2!)
# This finds the example most similar to the user's question
example_selector = SemanticSimilarityExampleSelector.from_examples(
    examples,  # Our list of examples
    OpenAIEmbeddings(),  # Converts text to "meaning numbers" (to check similarity)
    Chroma,  # Stores these numbers to quickly find matches
    k=1  # Pick only the 1 most similar example
)

# 6. Create the few-shot prompt template (using the selector)
prompt = FewShotPromptTemplate(
    example_selector=example_selector,  # Use the selector instead of all examples
    example_prompt=example_prompt,  # How to format the selected example
    suffix="Question: {input}",  # The user's question at the end
    input_variables=["input"]  # The user's question is called "input"
)

# 7. Define the user's question (let's ask something similar to our examples)
user_question = "Who was the paternal grandfather of George Washington?"

# 8. Generate the full prompt (with only the most relevant example)
full_prompt = prompt.format(input=user_question)
print("=== Generated Prompt (with 1 relevant example) ===")
print(full_prompt)
print("\n=== LLM's Answer ===")

# 9. Use the LLM to generate a response
llm = ChatOpenAI(model_name="gpt-3.5-turbo", temperature=0)  # "temperature=0" = consistent answers
response = llm.predict(full_prompt)
print(response)

What You'll See When You Run It

复制代码
=== Generated Prompt (with 1 relevant example) ===
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 paternal grandfather of George Washington?

=== LLM's Answer ===
Are follow up questions needed here: Yes.
Follow up: Who was the father of George Washington?
Intermediate answer: The father of George Washington was Augustine Washington.
Follow up: Who was the father of Augustine Washington?
Intermediate answer: The father of Augustine Washington was Lawrence Washington.
So the final answer is: Lawrence Washington

Simple Explanation

  1. Example Selector: Instead of shoving all 4 examples into the prompt, we use a "smart selector." It checks which example is most like the user's question (using "meaning numbers" from the embedding model) and only includes that one. This keeps the prompt short and focused.

  2. How It Works for Our Question : The user asked about George Washington's paternal grandfather . The selector noticed our 3rd example was about his maternal grandfather (similar topic!) and picked that one.

  3. LLM Follows the Pattern: The LLM sees the selected example and copies its style: first asking follow-ups ("Who was George Washington's father?"), then finding intermediate answers, and finally giving a clear final answer.

Why This Matters

  • Faster & More Accurate: The LLM doesn't waste time reading irrelevant examples.
  • Scalable: If you have 100 examples, the selector will still pick the 1 best one---no need to clutter the prompt.

Just replace "your-openai-api-key-here" with your actual API key, and you're good to go!

相关推荐
YIAN6 小时前
从 LangChain 到 DeepAgents:手把手拆解 createAgent、Middleware 与复杂 Agent
javascript·langchain·deepseek
syso_稻草人1 天前
qwen-image-2.1 姿态控制拆解:SDPose 提骨架 + Fun-ControlNet-Union 锁姿势,换人换景不换姿势
人工智能·stable diffusion·prompt·提示词优化·qwen-image
K71 天前
别再搜「LangChain 教程」了——你学的 Chain 已经被判了死刑
langchain·ai编程
xyphf_和派孔明2 天前
第六章: AI Agent长期记忆的存储引擎介绍
langchain·llm·agent
xyphf_和派孔明2 天前
第四章: AI Agent短期记忆的信息管理机制设计
langchain·llm·agent
叼烟扛炮2 天前
[LangChain] 学习笔记(13):并发不是越大越好——batch / gather / 信号量 / 配额治理
笔记·学习·langchain
叼烟扛炮2 天前
[LangChain] 学习笔记(09):工具(Tool)定义与调用闭环 · 理论+实战完整版
笔记·学习·langchain
forest_092 天前
《AI 编程 Harness 实战手册》04|审批这道门:点了一百次“允许”,还剩多少作用?
人工智能·算法·prompt·agi
仍然.3 天前
LangChain核心组件---消息
langchain
xyphf_和派孔明3 天前
第十章: Agent Skill的调用机制解读
langchain·llm·agent