Snow Globe multi-agent system for open-ended wargames with large language models
Project description
Snow Globe
Open-Ended Wargames with Large Language Models
Snow Globe, an ongoing applied research project and resulting software package, uses large language models (LLMs) for automated play of "open-ended" text-based wargames, such as seminar games and political wargames. LLMs enable a light, flexible architecture in which player actions are not restricted to predefined options. The system allows humans to play against or alongside AI agents with specific personas. Every stage of the wargame from scenario preparation to post-game analysis can be optionally carried out by AI, humans, or a combination thereof.
Read more here.
Installation
Build the Docker image and run a container.
./docker_setup.sh
Or, install Snow Globe from PyPI. For CPU only:
pip install llm-snowglobe
For GPU support:
CMAKE_ARGS="-DGGML_CUDA=on" pip install llm-snowglobe
AI Simulations
After installation, you can simulate a tabletop exercise about an AI incident response.
examples/haiwire.py
Or, simulate a political wargame about a geopolitical crisis.
examples/ac_sim.py
Human+AI Wargames
To play a game between a human and an AI player, launch the server and start a game:
snowglobe_server &
examples/ac_game.py
Then, open a browser window and navigate to:
http://localhost:8000
The terminal output will begin with the ID number for the human player. Type that number into the ID box in the browser window and click Log In.
Make sure to run snowglobe_server from the same file system location where you run the game. Game-related files will be stored in that location.
Custom Games
By default, Snow Globe uses a light LLM that runs locally. For better results, try the OpenAI API:
- Create an environment variable called
OPENAI_API_KEYwith your OpenAI API key in it. In a bash shell, for example:export OPENAI_API_KEY=your0key0here - In the example game you want to use with OpenAI, find the line that says
super().__init__()and change it to this:super().__init__(source='openai', model='gpt-4o')or similarly for whatever model you want to try.
Beyond that, you can use the examples as starting points for developing your own games, or you can create entirely new game formats using the Python classes provided by Snow Globe.
License
This repo is released under the Apache License Version 2.0, except for jQuery which is released under the MIT License.
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
Filter files by name, interpreter, ABI, and platform.
If you're not sure about the file name format, learn more about wheel file names.
Copy a direct link to the current filters
File details
Details for the file llm_snowglobe-1.0.0.tar.gz.
File metadata
- Download URL: llm_snowglobe-1.0.0.tar.gz
- Upload date:
- Size: 179.6 kB
- Tags: Source
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/6.2.0 CPython/3.12.3
File hashes
| Algorithm | Hash digest | |
|---|---|---|
| SHA256 |
5869216ba097515a585dd66a866638d476e63361ed3e11b476e29d7b885c148a
|
|
| MD5 |
07633cef8b3af006e94e40d9bad2ae68
|
|
| BLAKE2b-256 |
efce357a6b3916f25e3fffe3dc9eea79c927bdc0776f997ad6406e9f8a1569e8
|
File details
Details for the file llm_snowglobe-1.0.0-py3-none-any.whl.
File metadata
- Download URL: llm_snowglobe-1.0.0-py3-none-any.whl
- Upload date:
- Size: 178.4 kB
- Tags: Python 3
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/6.2.0 CPython/3.12.3
File hashes
| Algorithm | Hash digest | |
|---|---|---|
| SHA256 |
e126cda2fafcb0c4d1eda1f0af29b6c3eaa4d1435751c735d5791af206c98794
|
|
| MD5 |
ed064a89c37236720adcd9ad4655b37e
|
|
| BLAKE2b-256 |
260ac7c153107a12f0a570475f38ce80522befa3da1e842536ba75de539e0634
|