Prompt

Prompt#

The Prompt functionality in synthora provides a way to create and manage customizable prompt templates with dynamic argument formatting. This system integrates with agents like VanillaAgent and allows for flexible prompt handling during runtime.

Example Usage#

Here is an example of how to use BasePrompt with a VanillaAgent:

from synthora.agents.vanilla_agent import VanillaAgent
from synthora.callbacks import RichOutputHandler
from synthora.prompts import BasePrompt

# Define a prompt template
prompt = BasePrompt("Your name is {username}.")

# Create a default VanillaAgent using the prompt and a callback handler
agent = VanillaAgent.default(prompt, handlers=[RichOutputHandler()])

# Run the agent with input and a template argument
agent.run("what is your name?", username="John Doe")

Output#

The agent will dynamically replace {username} in the prompt with "John Doe" and process the query accordingly.

BasePrompt Class#

The BasePrompt class extends Python’s built-in str class, adding functionality for working with template strings and argument validation.

Key Features#

  1. String-like Behavior

    • BasePrompt is a subclass of str, so it behaves like a standard string.

  2. Template Argument Extraction

    • Dynamically extracts placeholder arguments from a template.

  3. Safe Formatting

    • Validates arguments against the template and only substitutes recognized placeholders.

  4. Integration with Pydantic

    • Supports serialization through Pydantic for seamless configuration handling.

Properties and Methods#

args (Property)#

Extracts all unique placeholders (arguments) in the prompt template.

Example#
prompt = BasePrompt("Your name is {username} and your role is {role}.")
print(prompt.args)  # Output: {'username', 'role'}

format (Method)#

Formats the prompt template by substituting placeholders with the provided arguments. It ensures that only the placeholders present in the template are replaced.

Parameters#

  • **kwargs: Key-value pairs for substituting template placeholders.

Example#
prompt = BasePrompt("Your name is {username}.")
formatted_prompt = prompt.format(username="Alice")
print(formatted_prompt)  # Output: "Your name is Alice."

Built-in Prompts#

The following table summarizes the built-in prompts provided in synthora, their purposes, and typical use cases:

Prompt Name

Description

Use Case

ZeroShotCoTPrompt

A prompt designed for zero-shot chain-of-thought reasoning, encouraging step-by-step explanations.

Use when requiring the model to provide detailed reasoning without prior examples.

ZeroShotReactPrompt

A zero-shot prompt for reasoning and acting (ReAct) in tasks requiring a mix of analysis and action.

Suitable for scenarios requiring both reasoning and action, without additional examples.

FewShotReactPrompt

A few-shot ReAct prompt providing the model with examples to guide reasoning and action.

Use when examples improve performance in reasoning and acting tasks.

ZeroShotTOTProposePrompt

A zero-shot Tree-of-Thought (ToT) prompt for proposing multiple reasoning paths.

Ideal for brainstorming or exploring diverse problem-solving approaches.

ZeroShotTOTEvalPrompt

A zero-shot ToT evaluation prompt for assessing and ranking proposed solutions or ideas.

Use when evaluating multiple proposed solutions to select the best one.

VanillaPrompt

A simple, general-purpose prompt template for straightforward interactions.

Suitable for standard queries where minimal guidance is required.

These prompts are pre-configured to cover a wide range of interaction styles and can be customized further as needed for specific applications.

Summary#

The BasePrompt class provides a robust and extensible way to handle dynamic prompt templates in synthora. By leveraging its integration with agents, templates, and Pydantic, it simplifies the process of creating personalized and context-aware AI interactions.

Contents