JohnHeibel/Automatic_Multi-instance_GPT-4

Implementation of the GPT-4 and GPT-3.5-Turbo apis that automatically creates multiple instances with different assigned apis. These are then combined for the final response.

Automatic Multi-Instance GPT (AMIG)

Overview

Automatic Multi-Instance GPT is designed to provide an interface for utilizing multiple GPT-based AI models for iteratively answering a given question. It calls a main GPT instance that creates a list of related questions. This is then processed by another GPT instance to convert it to JSON. This file is parsed to delegate a variable number of child instances. Each child instance can be assigned a tool to use. Currently, only internal knowledge and search are implemented. Search is a simple implementation that gets the top 3 search results and uses GPT to summarize the page. These summaries are passed to the child instance. After each child has processed, the results are passed into a single final summarizer GPT instance that tries to answer the main question.

Usage

1. Create file Keys.py
2. Add key values of the form

GOOGLE_API_KEY = "[GOOGLE_API_KEY]"
GOOGLE_CSE_KEY = "[GOOGLE_CSE_KEY]"
OPEN_AI_KEY = "[OPEN_AI_KEY]"

3. Run GPT-Run.py

Example Usage

import GPT_Pipeline

def main():
  question = input("What is your question?\n")
  # Create an instance of SemiRecursivePipeline
  pipeline = SemiRecursivePipeline(main_question=question)

  # Get the answer to the question using the process method
  answer = pipeline.process()

  print(answer)

if __name__ == '__main__':
  main()

Dependencies

• OpenAI API
• GoogleAPIClient
• BeautifulSoup
• Requests
• Logging
• Tiktoken
• Multiprocessing