Chain-of-Thought (CoT) Agent

Chain-of-Thought (CoT) Agent#

The Chain-of-Thought (CoT) Agent in Synthora is not a separate class but a configuration of the VanillaAgent with a specialized prompt. By using the ZeroShotCoTPrompt, the agent is guided to perform step-by-step reasoning, enabling it to solve problems that require logical breakdowns or intermediate steps.

Overview#

Purpose#

  • The CoT Agent leverages Chain-of-Thought prompting to explicitly generate intermediate steps in its reasoning process.

  • It builds on the VanillaAgent while providing additional capabilities for structured reasoning.

Key Features#

  1. Step-by-Step Reasoning: Breaks problems into smaller, manageable reasoning steps.

  2. Predefined Prompt: Utilizes a specialized ZeroShotCoTPrompt for CoT reasoning.

Implementation#

Here is how the CoT Agent is created and used:

Code Example#

from synthora.agents import VanillaAgent
from synthora.callbacks import RichOutputHandler
from synthora.prompts.buildin import ZeroShotCoTPrompt

# Create a CoT Agent with step-by-step reasoning
agent = VanillaAgent.default(
    ZeroShotCoTPrompt,
    handlers=[RichOutputHandler()]  # Adds rich output formatting
)

# Enable streaming for real-time output
agent.model.set_stream(True)

# Execute a reasoning task
response = agent.run("How many letters 'r' in the word 'strawberry'?")
print(response.unwrap())  # Outputs: "The word 'strawberry' contains 3 letters 'r'."

Example Walkthrough#

Input#

How many letters 'r' in the word 'strawberry'?

Reasoning Steps (CoT Output)#

  1. The word is “strawberry”.

  2. Count the occurrences of the letter ‘r’:

    • First ‘r’ in position 4.

    • Second ‘r’ in position 8.

    • Third ‘r’ in position 9.

  3. There are 3 ‘r’s in the word “strawberry”.

Final Output#

The word "strawberry" contains 3 letters 'r'.

Key Advantages of the CoT Agent#

  1. Enhanced Problem-Solving:

    • CoT reasoning improves accuracy on tasks that benefit from intermediate steps.

  2. Transparency:

    • Provides insights into the reasoning process, making it easier to debug and validate results.

Summary#

The Chain-of-Thought (CoT) Agent is a powerful extension of the Synthora framework, tailored for tasks requiring logical breakdowns and reasoning transparency. By integrating step-by-step reasoning, predefined prompts, and rich output formatting, it provides a robust tool for solving complex problems in an intuitive and explainable manner.

Contents