AgentGym

AgentGym provides environments to train LLM (Lange Language Model) based Agents.

Environments

Pokemon-Red	Werewolf (social game)

Werewolf Game

AI v.s. AI environment playing Werewolf Game.

Werewolf is a game where each player deceives the others while trying to hunt down the werewolf before the whole village becomes food for the beast.

It is also a reproduction of papers Language Agents with Reinforcement Learning for Strategic Play in the Werewolf Game and Exploring Large Language Models for Communication Games: An Empirical Study on Werewolf

Pokemon Red

Language-Vision Model (LVM like GPT-vision) plays Pokemon-Red (GameBoy Game)

We give the model some screenshots, and let it predicts which button to press next.

Build on its non-LVM traditional RL environment

Preparation

Install the environment: conda env create -f environment.yml

conda activate camelgym

Set the configuration

Create ~/config/config.yaml

Copy and paste the following in it:

llm:
  api_type: "openai"  # or azure / ollama / open_llm etc. Check LLMType for more options
  model: "gpt-4-turbo-preview"  # or gpt-3.5-turbo-1106 / gpt-4-1106-preview
  base_url: "https://api.openai.com/v1"  # or forward url / other llm url
  api_key: "YOUR_API_KEY"

Run

Werewolf Game

Start the game:

python werewolf_game/start_game.py

Pokemon

1. Copy your legally obtained Pokemon Red ROM into the base directory. You can find this using google, it should be 1MB. Rename it to PokemonRed.gb if it is not already. The sha1 sum should be ea9bcae617fdf159b045185467ae58b2e4a48b9a, which you can verify by running shasum PokemonRed.gb.
2. Move into the baselines/ directory:
   cd baselines
3. Export your OpenAI API:

export OPENAI_API_KEY=<insert your OpenAI API key>
OPENAI_API_BASE_URL=<inert your OpenAI API BASE URL>  #(Should you utilize an OpenAI proxy service, kindly specify this)

4. Run:
   python ./pokemon/baselines/run_baseline_parallel_fast.py

TODO Env

1. robotic simulator
2. cell
3. minecraft
4. amongus
5. trade
6. town