/consistent

Visualise the consistency of an LLM's answers

Primary LanguageGoMIT LicenseMIT

consistent

Want to visualise the consistency of an LLM's answers?

consistent is a CLI tool that allows users to ask an LLM the same yes/no question multiple times and visualise its consistency as a bar chart.

This tool currently supports models available via the Anthropic API.

Demo

Table of Contents

Installation

To install consistent, follow these steps:

  1. Ensure you have Go installed on your system (version 1.16 or later).
  2. Clone this repository:
    git clone https://github.com/rossheat/consistent.git
    
  3. Navigate to the project directory:
    cd consistent
    
  4. Build the binary:
    go build -o consistent
    
  5. (Optional) Move the binary to a directory in your PATH for easy access:
    sudo mv consistent /usr/local/bin/
    

Usage

To use consistent, you'll need an Anthropic API key. Run the tool with the following command:

consistent -key YOUR_ANTHROPIC_API_KEY

After running the command, you'll be prompted to enter a yes/no question. The tool will then send this question to multiple instances of the specified Anthropic model and visualise the consistency of the responses.

CLI Arguments

consistent supports the following command-line arguments:

  • -key (required): Your Anthropic API key.
  • -debug (optional): Start the program in debug mode. Default: false.
  • -model (optional): The name of the Anthropic model you'd like to question. Default: "claude-3-5-sonnet-20240620".
  • -instances (optional): The number of times your question is sent to the model API. Default: 50.
  • -delay (optional): Milliseconds of delay between calling the Anthropic API. Default: 500.

Example with all arguments:

consistent -key YOUR_API_KEY -debug -model claude-3-5-sonnet-20240620 -instances 25 -delay 1000

Contributing

Contributions to consistent are welcome! Please feel free to submit a Pull Request.

License

This project is made available under the terms of the MIT License