Prompt Composition with LangChain’s PipelinePromptTemplate

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

Learning Guide: Prompt Composition with LangChain's PipelinePromptTemplate

This guide simplifies how to combine multiple prompts for reuse (using LangChain's PipelinePromptTemplate), while keeping all original code, examples, and key points exactly as they appear in the link.

1. What is PipelinePromptTemplate?

It's a LangChain tool to reuse parts of prompts. It has two key parts:

  • Final Prompt : The last, complete prompt you get after combining all parts. It uses placeholders (like {introduction}, {example}) to "hold space" for other small prompts.
  • Pipeline Prompts: A list of "name + small prompt" pairs. Each small prompt is formatted first, then put into the final prompt using its name (to match the placeholder).

2. Step-by-Step Code (Exact as Original)

We'll follow the original code step by step. Each code block is unchanged, and we'll explain what it does simply.

Step 1: Import Needed Tools

First, we get the two tools we need from LangChain:

python 复制代码
from langchain.prompts.pipeline import PipelinePromptTemplate
from langchain.prompts.prompt import PromptTemplate
  • PipelinePromptTemplate: Helps combine multiple prompts.
  • PromptTemplate: Makes single, reusable prompt templates.

Step 2: Make the Final Prompt Template

This is the "big" prompt that will hold all the small parts. It uses 3 placeholders:

python 复制代码
full_template = """{introduction}
{example}
{start}"""
full_prompt = PromptTemplate.from_template(full_template)
  • full_template: The structure of the final prompt (with placeholders).
  • PromptTemplate.from_template(): Turns the text structure into a LangChain "prompt object" (so we can use it later).

Step 3: Make Small Reusable Prompts

We create 3 small prompts (each is a reusable part). Each has its own variables:

1. Introduction Prompt (sets who to impersonate)
python 复制代码
introduction_template = """You are impersonating {person}."""
introduction_prompt = PromptTemplate.from_template(introduction_template)
  • Uses {person}: We'll fill this in later (e.g., "Elon Musk").
2. Example Prompt (gives a sample interaction)
python 复制代码
example_template = """Here's an example of an interaction:
Q: {example_q}
A: {example_a}"""
example_prompt = PromptTemplate.from_template(example_template)
  • Uses {example_q} (sample question) and {example_a} (sample answer).
3. Start Prompt (asks for a real response)
python 复制代码
start_template = """Now, do this for real!
Q: {input}
A:"""
start_prompt = PromptTemplate.from_template(start_template)
  • Uses {input}: The real question we want to ask later.

We make a list to connect each small prompt to its placeholder in the final prompt:

python 复制代码
input_prompts = [
    ("introduction", introduction_prompt),  # "introduction" → matches {introduction}
    ("example", example_prompt),            # "example" → matches {example}
    ("start", start_prompt)                 # "start" → matches {start}
]

Step 5: Create the PipelinePromptTemplate

We put the final prompt and the small prompt list together:

python 复制代码
pipeline_prompt = PipelinePromptTemplate(final_prompt=full_prompt, pipeline_prompts=input_prompts)

Step 6: Check Required Variables

To use the pipeline, we need to know all variables we must fill in. The original code shows these variables:

python 复制代码
pipeline_prompt.input_variables
# Output: ['example_a', 'person', 'example_q', 'input']
  • These come from the small prompts: person (from introduction), example_q/example_a (from example), input (from start).

3. Generate the Final Prompt

We fill in all required variables and print the result. The code and output are exactly as in the original:

Code to Format the Prompt

python 复制代码
print(pipeline_prompt.format(
    person="Elon Musk",
    example_q="What's your favorite car?",
    example_a="Telsa",
    input="What's your favorite social media site?"
))

Final Output

复制代码
You are impersonating Elon Musk.
    Here's an example of an interaction: 
    
    Q: What's your favorite car?
    A: Telsa
    Now, do this for real!
    
    Q: What's your favorite social media site?
    A:

Key Takeaway (No Extra Info)

PipelinePromptTemplate helps you reuse prompt parts (like the "impersonate" or "example" sections) so you don't rewrite code. All parts combine to make one final prompt, and you only need to fill in the required variables.

相关推荐
Resean02233 小时前
Spring 源码分析(二)配置类的解析与扫描
java·后端·spring
小毛球.3 小时前
第 3 篇:IDEA 最新版下载与完整安装教程(Mac)
java·macos·intellij-idea·idea
Misnearch5 小时前
ScheduledExecutorService
java·线程池
ashiho6 小时前
【LeetCode hot 100】3. 无重复字符的最长子串 (三种解法)
java·数据结构·后端·算法·leetcode·滑动窗口
T_Wizard6 小时前
AI 回复为什么要转圈 3 秒才「啪」一下全出来?把流式输出拆开看,顺便踩了 6 个坑
java·ai编程
CoderYanger6 小时前
Java EE 进阶:9.MyBatis 操作数据库(进阶)(第二弹)
java·数据库·面试·职场和发展·java-ee·tomcat·mybatis
仍然.8 小时前
LangChain核心组件---消息
langchain
小满、8 小时前
Java 值传递
java·java 值传递
IT枫斗者枫哥8 小时前
Java 异步导出只测最终行数,为什么下载时仍会少一行?
java·后端
张紫娃8 小时前
idea 如何看一个类的 .class 文件
java·ide·intellij-idea