Few shot Prompt Templates

Few-shot prompting is a powerful technique that teaches a language model a specific pattern by providing a handful of examples at runtime. This approach avoids hardcoding lengthy instructions and lets the LLM infer the desired transformation from context.

Table of Contents

1. Prerequisites
2. Defining Examples
3. Creating an Example Prompt
4. Building the Few-Shot Prompt
5. Assembling the Chat Prompt Template
6. Inspecting the Prompt Structure
7. Formatting for Invocation
8. Invoking the Model
9. References

1. Prerequisites

Install the required packages and set your OpenAI API key:

pip install langchain langchain_openai

export OPENAI_API_KEY="your_api_key_here"

from langchain.prompts import (
    ChatPromptTemplate,
    FewShotChatMessagePromptTemplate,
)
from langchain_openai import ChatOpenAI

2. Defining Examples

Prepare a small list of example pairs. Each item maps an input (country name) to its output (the reversed string):

examples = [
    {"input": "India",     "output": "aidnI"},
    {"input": "Canada",    "output": "adanaC"},
    {"input": "Australia", "output": "ailartsuA"},
]

3. Creating an Example Prompt

Use ChatPromptTemplate to describe how each example should appear in the conversation:

example_prompt = ChatPromptTemplate.from_messages(
    [
        ("human", "{input}"),
        ("ai",    "{output}"),
    ]
)

4. Building the Few-Shot Prompt

Combine your individual example template with the list of examples via FewShotChatMessagePromptTemplate:

few_shot_prompt = FewShotChatMessagePromptTemplate(
    example_prompt=example_prompt,
    examples=examples,
)

5. Assembling the Chat Prompt Template

Wrap the system instruction, the few-shot examples, and the final human query into a single ChatPromptTemplate:

prompt_template = ChatPromptTemplate.from_messages(
    [
        ("system", "You are a linguistic specialist."),
        few_shot_prompt,
        ("human",  "{input}"),
    ]
)

6. Inspecting the Prompt Structure

Print out the internal representation to verify the sequence of messages:

print(prompt_template)

This helps ensure your prompt layout and placeholders are correct.

7. Formatting for Invocation

Populate the template with a new input—e.g., "Brazil"—to generate the messages you’ll send to the LLM:

messages = prompt_template.format_messages(input="Brazil")
print(messages)

Example output:

[
  SystemMessage(content='You are a linguistic specialist.'),
  HumanMessage(content='India'),
  AIMessage(content='aidnI'),
  HumanMessage(content='Canada'),
  AIMessage(content='adanaC'),
  HumanMessage(content='Australia'),
  AIMessage(content='ailartsuA'),
  HumanMessage(content='Brazil'),
]

8. Invoking the Model

Pass the formatted messages to the ChatOpenAI model and print the response:

model = ChatOpenAI()
response = model.invoke(messages)
print(response.content)

Expected result:

'lizarB'

Notice the model inferred the reverse‐text pattern purely from examples—no explicit instruction was needed.

9. References

• LangChain Prompt Templates
• OpenAI Python SDK
• Few-Shot Learning on Wikipedia