Skip to content

A Collection of Competitive Text-Based Games for Language Model Evaluation and Reinforcement Learning

License

Notifications You must be signed in to change notification settings

LeonGuertler/TextArena

Repository files navigation

PyPI version Discord Website

TextArena  

TextArena is a flexible and extensible framework for training, evaluating, and benchmarking models in text-based games. It follows an OpenAI Gym-style interface, making it straightforward to integrate with a wide range of reinforcement learning and language model frameworks.


Example

Installation

Install TextArena directly from PyPI:

pip install textarena

Play Offline

Run the following command to set your OpenRouter API key:

export OPENROUTER_API_KEY="YOUR_OPENROUTER_API_KEY"

Then run the following code to play offline:

import textarena as ta

# Initialize agents
agents = {
    0: ta.agents.OpenRouterAgent(model_name="GPT-4o-mini"),
    1: ta.agents.OpenRouterAgent(model_name="anthropic/claude-3.5-haiku"),
}

# Initialize environment from subset and wrap it
env = ta.make(env_id="SpellingBee-v0")
env = ta.wrappers.LLMObservationWrapper(env=env)
env = ta.wrappers.SimpleRenderWrapper(
    env=env,
    player_names={0: "GPT-4o-mini", 1: "claude-3.5-haiku"},
)

env.reset(num_players=len(agents))
done = False
while not done:
    player_id, observation = env.get_observation()
    action = agents[player_id](observation)
    done, info = env.step(action=action)
rewards = env.close()

Play Online

If you want to evaluate your model against other submitted models and humans, you can simply change the .make to .make_online. Please make sure that the model_name is unique and that the email address provided is correct.

import textarena as ta
 
model_name = "Standard GPT-4o LLM"
model_description = "Standard OpenAI GPT-4o model."
email = "[email protected]"


# Initialize agent
agent = ta.agents.OpenRouterAgent(model_name="gpt-4o") 


env = ta.make_online(
    env_id=["SpellingBee-v0", "SimpleNegotiation-v0", "Poker-v0"], 
    model_name=model_name,
    model_description=model_description,
    email=email
)
env = ta.wrappers.LLMObservationWrapper(env=env)


env.reset(num_players=1)

done = False
while not done:
    player_id, observation = env.get_observation()
    action = agent(observation)
    done, info = env.step(action=action)


rewards = env.close()

Implementation Status

Game Players Offline Play Online Play Documentation
CarPuzzle 1
Crosswords 1
FifteenPuzzle 1
GuessTheNumber 1
GuessWho 1
Hangman 1
LogicPuzzle 1
Mastermind 1
MathProof 1
Minesweeper 1
Sudoku 1
TowerOfHanoi 1
TwentyQuestions 1
WordLadder 1
WordSearch 1
Wordle 1
AirLandAndSea † 2
BattleOfSexes ‡ 2
Battleship 2
Brass 2
Breakthrough ¶ 2
Checkers 2
Chess 2
ConnectFour 2
Debate 2
DontSayIt 2
DracoGame ‡ 2
DuopolisticCompetition ‡ 2
EscalationGame ‡ 2
Hive † 2
HotColdGame ‡ 2
IntegrativeDistributiveNegotiation § 2
IteratedPrisonersDilemma 2
Jaipur 2
KuhnPoker ¶ 2
LetterAuction 2
MemoryGame 2
MonopolyGame ‡ 2
Nim ¶ 2
Othello (Reversi) 2
PigDice ¶ 2
PrisonersDilemma ‡ 2
Santorini † 2
ScenarioPlanning 2
SeaBattle † 2
SimpleBlindAuction ¶ 2
SimpleNegotiation 2
SpellingBee 2
SpiteAndMalice 2
StagHunt ‡ 2
Stratego 2
Taboo 2
Tak 2
TicTacToe 2
TriGame ‡ 2
TruthAndDeception 2
UltimateTicTacToe 2
WaitGoGame ‡ 2
WordChains 2
ArcticScavengers † 3+
AreYouTheTraitor † 3+
BlindAuction 3–15
CharacterConclave 3–15
Codenames† 4
LiarsDice 2–15
Negotiation 3–15
Pit † 3+
Poker 2–15
Snake 2–15
Surround 2–15
TwoRoomsAndABoom † 6+
Diplomacy 3–7
7 Wonders 3+
Bohnanza 3+
Codenames 4+
Risk 3+
SettlersOfCatan 2–4
TerraformingMars 1–5
Werewolf 5+

† Games from LLM Arena: Studying the Impact of Domain Expertise and Problem Complexity in LLM Competitions

‡ Games from Language Model Negotiations: Theory-of-Mind vs. Complexity of the Game

§ Games from Negotiating with Humans by LLMs via Strategic Reasoning

¶ These games were added because they are part of Language Models Make Better Players than Solvers in Cooperative Games

About

A Collection of Competitive Text-Based Games for Language Model Evaluation and Reinforcement Learning

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages