Table of Contents

Enum PromptTemplateType

Namespace
AiDotNet.Enums
Assembly
AiDotNet.dll

Represents different types of prompt templates for language model interactions.

public enum PromptTemplateType

Fields

ChainOfThought = 5

Template for chain-of-thought reasoning with step-by-step problem solving.

For Beginners: Chain-of-thought templates encourage the model to think step-by-step.

Think of it like showing your work in math class:

  • Instead of jumping to the answer
  • The model explains its reasoning process
  • Each step builds on the previous one
  • The final answer is more accurate and explainable

Example without chain-of-thought: Q: "If a store has 24 apples and sells 3/4 of them, how many are left?" A: "6 apples"

Example with chain-of-thought: Q: "If a store has 24 apples and sells 3/4 of them, how many are left?" A: "Let me solve this step by step: 1. The store has 24 apples 2. They sell 3/4 of them 3. 3/4 of 24 = 24 × 3/4 = 18 apples sold 4. Apples left = 24 - 18 = 6 apples Therefore, 6 apples are left."

Benefits:

  • Better accuracy on complex problems
  • Explainable reasoning
  • Easier to debug errors
  • More reliable for multi-step tasks

Use this when:

  • Tasks require multi-step reasoning
  • Accuracy is critical
  • You need to understand how the model reached its answer
Chat = 2

Template for structured message-based conversations with roles (system, user, assistant).

For Beginners: Chat templates structure conversations with different roles.

Modern language models understand different roles:

  • System: Instructions for how the assistant should behave
  • User: The person asking questions
  • Assistant: The AI's responses

Example: System: "You are a helpful math tutor for elementary school students." User: "What is 25 + 17?" Assistant: "Great question! Let's solve this step by step..."

This structure helps the model:

  • Understand its role and constraints (system message)
  • Distinguish between user questions and its own responses
  • Maintain consistent behavior across a conversation

Use this when:

  • Building chatbots or conversational AI
  • You need to set specific behavior guidelines
  • Working with multi-turn conversations
FewShot = 1

Template with few-shot learning examples to guide the model's output.

For Beginners: Few-shot templates include examples to show the model what you want.

Think of it like teaching by example:

  • Instead of just telling the model what to do, you show it examples
  • The model learns the pattern from your examples
  • Then applies that pattern to new inputs

Example: Template with examples: "Classify sentiment as positive or negative.

Example: 'This movie was amazing!' → Positive Example: 'I hated the food.' → Negative Example: 'Best day ever!' → Positive

Now classify: '{text}'"

The examples help the model understand exactly what format and type of response you want.

Use this when:

  • The task is complex or ambiguous
  • You want to control the output format
  • A few examples significantly improve results
Optimized = 3

Template optimized through automated prompt engineering techniques.

For Beginners: Optimized templates are automatically refined to get better results.

Think of it like A/B testing:

  • You provide a basic template
  • The system tries multiple variations
  • It measures which variation gets the best results
  • The winning version becomes your optimized template

Optimization can include:

  • Testing different wordings or phrasings
  • Trying different instruction orderings
  • Adding or removing examples
  • Adjusting tone and specificity

Example process: Original: "Summarize this text: {text}" Variation 1: "Provide a concise summary of the following text in 2-3 sentences: {text}" Variation 2: "What are the key points in this text? {text}" → System tests all variations and picks the one with best performance

Use this when:

  • You want the best possible results
  • You have evaluation metrics to measure success
  • You're willing to invest time in optimization
ReAct = 6

Template for ReAct (Reasoning + Acting) pattern combining thought and action.

For Beginners: ReAct templates combine reasoning with actions in an iterative loop.

ReAct stands for Reasoning + Acting:

  • The model alternates between thinking and doing
  • Each thought informs the next action
  • Each action result informs the next thought
  • This continues until the task is complete

The pattern:

  1. Thought: "I need to find X to answer this question"
  2. Action: Search for X
  3. Observation: Here's what I found about X
  4. Thought: "Now I need to verify Y"
  5. Action: Calculate Y
  6. Observation: Y equals Z
  7. Thought: "I have enough information to answer"
  8. Answer: Final response

Example: User: "Who is the current leader of the country where the Eiffel Tower is located?"

Thought: "I need to find out which country the Eiffel Tower is in." Action: search("Eiffel Tower location") Observation: "The Eiffel Tower is in Paris, France."

Thought: "Now I need to find the current leader of France." Action: search("current president of France") Observation: "Emmanuel Macron is the current President of France."

Thought: "I have all the information needed to answer." Answer: "Emmanuel Macron is the current leader of France, where the Eiffel Tower is located."

Use this when:

  • Tasks require both reasoning and external actions
  • You need to combine multiple tools/data sources
  • The solution path isn't straightforward
  • You want transparent, step-by-step problem solving
Simple = 0

Simple template with variable substitution using placeholders.

For Beginners: The simplest type of template where you replace placeholders with values.

Example: Template: "Translate the following {source_language} text to {target_language}: {text}" Variables: source_language="English", target_language="Spanish", text="Hello world" Result: "Translate the following English text to Spanish: Hello world"

Use this when:

  • You have a simple prompt structure
  • You need to insert specific values into predetermined positions
  • You don't need examples or complex formatting
Tool = 4

Template that includes function/tool calling capabilities.

For Beginners: Tool templates allow the model to use external functions and APIs.

Think of tools like giving the model superpowers:

  • Instead of just answering from its training, the model can:
    • Search the web
    • Query databases
    • Do calculations
    • Access current information
    • Execute code

How it works:

  1. You define available tools (e.g., "get_weather", "search_database")
  2. The model decides when to use these tools
  3. The model generates structured tool calls
  4. Your code executes the tools
  5. The model uses the results to answer the user

Example: User: "What's the weather in Paris?" Model: Calls tool: get_weather(city="Paris") Tool returns: {"temperature": 18, "condition": "Sunny"} Model: "The weather in Paris is sunny with a temperature of 18°C."

Use this when:

  • You need real-time or external data
  • The task requires computation or database access
  • You want to extend the model's capabilities

Remarks

For Beginners: Prompt templates are pre-structured formats for communicating with language models.

Think of templates like email templates:

  • Instead of writing each email from scratch, you use a template with placeholders
  • You fill in the specific details (name, date, etc.) for each email
  • The overall structure and tone remain consistent

Prompt templates work the same way:

  • You create a template with placeholders for variable content
  • You fill in specific values when you need to use the template
  • The language model receives a well-structured, consistent prompt

Different template types serve different purposes, from simple variable substitution to complex multi-turn conversations with examples and tool usage.

Edit this page