Skip to main content

MAgIC: Investigation of Large Language Model Powered Multi-Agent in Cognition, Adaptability, Rationality and Collaboration

Project description

Paper link


MAgIC is the benchmark to assess LLM-powered multi-agents' capabilities including cognition, adaptability, rationality and collaboration. Based on it, you can quantitatively measure your own LLM's ability and compare it with the cutting-edge LLMs.


pip install MAgIC_LLM==0.8.0

Need to assign the OPENAI_API_KEY, if you are trying to use openai api

export OPENAI_API_KEY=""


To assess your own large language model, follow the below instructions

import MAgIC_LLM
import time
import openai
import google.generativeai as palm
import cohere
from anthropic import Anthropic, HUMAN_PROMPT, AI_PROMPT

# Here we use GPT-4-turbo as the example
def chatbox(messages,temperature,max_tokens):
    response = openai.ChatCompletion.create(model="gpt-4-1106-preview",
                                        temperature = temperature,
    response = response['choices'][0]['message']['content']
    return response

# configure the path you want to save the assessment results
path = 'result.json' 

# Here is the name of your own LLM
test_player_model_name = 'My_LLM',path,test_player_model_name,PGM=False)

If any interruption happens in the process, you can just simple re-excute your programme and it will continue to assess your LLM from the point where interruption happens.

Currently, PGM method can be used to enhance your own LLM, the usage is below:,path,test_player_model_name,PGM=True)


MIT License

Project details

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

MAgIC_LLM-0.8.0.tar.gz (116.4 kB view hashes)

Uploaded Source

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page