Spam Detection with Large Language Models

This Python script utilizes a Large Language Model (LLM) to classify messages as either "SPAM" or "NOT_SPAM". It provides a simple interactive interface to test the model's capabilities.

Prerequisites

• OpenAI API Key: You'll need an OpenAI API key to run this script. Obtain one from https://platform.openai.com/.

• Environment Variables: Set the following environment variables:

  • CLOUD_MODEL_NAME: The name of the LLM to use (default: llama-3.1-8b-instruct).
  • CLOUD_BASE_URL: The base URL of the OpenAI API endpoint.
  • CLOUD_API_KEY: Your OpenAI API key.

  You can set these environment variables in your terminal before running the script, or in your .bashrc or equivalent configuration file.

• uv package manager

Usage

1. Run the script:

   uv run --env-file .env main.py

2. Interactive Input: The script will prompt you to enter a message. You can:

   • Type your own message to be analyzed.
   • Enter "1" to use the sample spam message: Buy V-Bucks for free at http://perdu.com !
   • Enter "2" to use the sample non-spam message: More trees in our streets !

3. Output: The script will print the classification result (either "SPAM" or "NOT_SPAM") based on the message you provided.

Code Description

• SpamDetection Class: This Pydantic model defines the structure of the output data from the LLM. It has one field, spam, which must be either "SPAM" or "NOT_SPAM".
• send_for_analysis(msg) Function:
  • Takes a message (msg) as input.
  • Uses the OpenAI API to send the message to the LLM.
  • Parses the LLM's response (which is expected to be in JSON format) and returns it as a Python dictionary.
• main() Function:
  • Prints an introductory message and a list of sample messages.
  • Takes user input to determine which message to send to the LLM.
  • Calls the send_for_analysis() function to get the classification result.
  • Prints the classification result to the console.

Configuration

• MODEL: The LLM to use. Defaults to llama-3.1-8b-instruct. You can override this by setting the CLOUD_MODEL_NAME environment variable.

Example

If you enter "1" (the spam message), the output will be: SPAM

$ uv run --env-file .env main.py

Send content for spam analysis by a LLM.
Sample messages :
(SPAM) 1. Buy V-Bucks for free at http://perdu.com !
(NOT_SPAM) 2. More trees in our streets !
Type your own message or select 1 or 2 to continue.
> 2
Spam analysis result: NOT_SPAM

Resources

• Scaleway Generative API
• Pydantic doc
• uv repository