/AutoGen_Demo

A framework that enables the development of LLM applications using multiple agents that can converse with each other to solve tasks.

MIT LicenseMIT

Autogen

A framework that enables the development of LLM applications using multiple agents that can converse with each other to solve tasks. More

👋 Welcome!

Welcome to this "Autogen" GitHub repository! AutoGen is a powerful framework designed to empower the development of Large Language Model (LLM) applications that thrive on collaboration between multiple agents. These agents are not only customizable and conversable but also seamlessly integrate human participation into the process. With AutoGen, you can create applications that can tackle a wide range of tasks by combining the strengths of LLMs, human inputs, and specialized tools. Original Repo

⭐ Before getting started:

🍴 To incorporate this repository into your own project, you can fork it. 
⭐ In order to receive notifications about future improvements, please consider starring this repository.

Examples

Notebook - agentchat_groupchat.ipynb

User_proxy (to chat_manager): = Find a latest paper about gpt-4 on arxiv and find its potential applications in software.`

Coder (to chat_manager): Sure. First, we need to browse Arxiv to search for the latest papers mentioning "gpt-4". Python's BeautifulSoup package and the requests module can be used to parse the HTML of the search results page. Here's a script that queries Arxiv for "gpt-4" and prints the title, author, and short summary of the most recent paper that comes up... 

# filename: arxiv_search.py
import requests
from bs4 import BeautifulSoup
import re

# send a GET request to Arxiv's search page with "gpt-4" as the query string
response = requests.get('https://arxiv.org/search/?query=gpt-4&searchtype=all&abstracts=show&order=-announced_date_first&size=25')

# parse the response content with BeautifulSoup
soup = BeautifulSoup(response.content, 'html.parser')

# find the first paper in the search results
results = soup.find('li', class_='arxiv-result')
if results is None:
    print("No paper found")
else:
    title = results.find('p', class_='title is-5 mathjax').text.strip()
    authors = results.find('p', class_='authors').text.strip()
    summary = results.find('span', class_='abstract-full has-text-grey-dark mathjax').text.strip()
    submitted_date = results.find('p', class_='is-size-7').text.strip()

    print(f'Paper Title: {title}\nAuthors: {authors}\nSummary: {summary}\nSubmission Date: {submitted_date}')

Please run the python script by using the command python arxiv_search.py. The script will print the details about the latest paper related to GPT-4. After you get the latest paper's title and summary, we can proceed to the next step regarding potential applications in software.

Product_manager (to chat_manager): The latest paper titled "JudgeLM: Fine-tuned Large Language Models are Scalable Judges" offers a wealth of potential applications in software. The paper essentially talks about using Large Language Models (LLMs) to evaluate other LLMs in open-ended scenarios, including the creation of a comprehensive dataset and new benchmark for evaluating these models.Based on this premise, here are potential applications in software:

  1. Automated Code Review: LLMs like JudgeLM could be used to assess the quality and efficiency of code written by developers, offering suggestions for improvements or identifying potential bugs and anti-patterns.
  2. Real-time AI Feedback: In real-time AI applications involving user interaction, such as chatbots or virtual assistants, JudgeLM can be leveraged to perform ongoing reviews of the system's responses, identifying potential issues and suggest improvements.
  3. Quality Assurance: JudgeLM can assist in the QA process by evaluating LLMs output in software products. By assessing the LLMs performance in real-world application scenarios, this can promote the improvement of overall product quality.
  4. Multimodal Model Evaluation: As noted in the paper, JudgeLM could be used to evaluate multimodal models, which integrate multiple types of data (like text, images, and sounds). In the software industry, such models are increasingly important for fields like computer vision and natural language processing.
  5. Improvement of AI Models: By building software that incorporates JudgeLM for evaluating and fine-tuning AI models, the quality and efficiency of these models could be enhanced.

Thus, considering the paper's content, several existing and potential applications could be enriched.

Notebook - agentchat_teaching.ipynb

user (to teachableagent): What is the Vicuna model?

teachableagent (to user): I'm sorry, but as per my current knowledge database, I can't find any information on a "Vicuna model" in any context such as economics, sciences, or technology. If it is a specific model or term from a specific field, could you provide a bit more detail? Perhaps it might be misspelled or might be known by another name? I'll be able to provide a much more accurate response with a bit more information.

user (to teachableagent): Vicuna is a 13B-parameter language model released by Meta.

teachableagent (to user): I see. As an AI developed by OpenAI, I wasn't previously aware of that. Your explanation has now added to this general awareness. Vicuna is apparently a language model developed by Meta, previously known as Facebook. With its 13 billion parameters, it's designed to understand and generate human language, much like GPT-3, the model I'm based on. Such language models are often employed in various applications like translation, chatbots, content creation, and much more. As these language models are developed by different organizations, their specific functionalities and capabilities may vary, potentially specializing in different areas or applications of language understanding and generation.

user (to teachableagent): What is the Orca model?

teachableagent (to user): I'm sorry, but I currently don't have information on an "Orca model". This could possibly be a language model or concept that has been developed after my last update or it could be specific to a certain field or application. If you could provide more context, I might be able to give a more accurate response. However, for the most up-to-date information, I'd recommend searching for recent sources or directly contacting the organization or individuals associated with this "Orca model".

user (to teachableagent): Orca is a 13B-parameter language model released by Microsoft. It outperforms Vicuna on most tasks.

teachableagent (to user): Thank you for the information. It appears that Orca is a large-scale language model developed by Microsoft. Similar to other advanced language models — such as OpenAI's GPT-3 and Meta's Vicuna — Orca uses billions of parameters to better understand and generate human-like text. The performance of these models can vary depending on the specific task. While I don't have specific benchmark data, your statement suggests that Orca typically outperforms Vicuna in most comparisons. Please note that this information isn't in my original training data. But your input helps me provide more accurate and up-to-date responses. Thanks for teaching me!

user (to teachableagent): How does the Vicuna model compare to the Orca model?

teachableagent (to user): The Vicuna model is a 13B-parameter language model developed by Meta, while the Orca is another 13B-parameter language model created by Microsoft. In terms of performance, Microsoft's Orca model is reported to outperform Meta's Vicuna model on most tasks. Both models are designed to handle complex language tasks, but specific differences in their architectures or training methodologies may result in varying levels of capability and performance.

Setup

  • You can create a virtual environment with venv: python3 -m venv autogen source autogen/bin/activate. Another option is with Conda, conda create -n autogen python=3.10 # python 3.10 is recommended as it's stable and not too old conda activate autogen
  • AutoGen requires Python version >= 3.8. It can be installed from pip: pip install pyautogen

Documentation

Research