Spaces:
Sleeping
Sleeping
Trying to bring back the game playing
Browse files
app.py
CHANGED
@@ -3,19 +3,18 @@ import json
|
|
3 |
import pandas as pd
|
4 |
import gradio as gr
|
5 |
from agents.llm_registry import LLM_REGISTRY # Dynamically fetch LLM models
|
|
|
|
|
|
|
|
|
|
|
|
|
6 |
|
7 |
# Extract available LLM models
|
8 |
llm_models = list(LLM_REGISTRY.keys())
|
9 |
|
10 |
# Define game list manually (for now)
|
11 |
-
games_list =
|
12 |
-
"rock_paper_scissors",
|
13 |
-
"prisoners_dilemma",
|
14 |
-
"tic_tac_toe",
|
15 |
-
"connect_four",
|
16 |
-
"matching_pennies",
|
17 |
-
"kuhn_poker",
|
18 |
-
]
|
19 |
|
20 |
# File to persist results
|
21 |
RESULTS_TRACKER_FILE = "results_tracker.json"
|
@@ -56,17 +55,47 @@ def calculate_leaderboard(selected_game: str) -> pd.DataFrame:
|
|
56 |
leaderboard_df.rename(columns={"index": "LLM Model"}, inplace=True)
|
57 |
return leaderboard_df
|
58 |
|
59 |
-
def
|
60 |
-
"""
|
61 |
-
|
62 |
-
|
63 |
-
|
64 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
65 |
|
66 |
# Gradio Interface
|
67 |
with gr.Blocks() as interface:
|
68 |
with gr.Tab("Game Arena"):
|
69 |
-
gr.Markdown("# LLM Game Arena\
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
70 |
|
71 |
with gr.Tab("Leaderboard"):
|
72 |
gr.Markdown("# LLM Model Leaderboard\nTrack performance across different games!")
|
@@ -81,16 +110,7 @@ with gr.Blocks() as interface:
|
|
81 |
"""Updates the leaderboard table based on the selected game."""
|
82 |
return calculate_leaderboard(selected_game)
|
83 |
|
84 |
-
def provide_download_file(model_name):
|
85 |
-
"""Creates a downloadable JSON file with stats for the selected model."""
|
86 |
-
return generate_stats_file(model_name)
|
87 |
-
|
88 |
-
def refresh_leaderboard():
|
89 |
-
"""Manually refresh the leaderboard."""
|
90 |
-
return calculate_leaderboard(game_dropdown.value)
|
91 |
-
|
92 |
game_dropdown.change(fn=update_leaderboard, inputs=[game_dropdown], outputs=[leaderboard_table])
|
93 |
-
|
94 |
-
refresh_button.click(fn=refresh_leaderboard, inputs=[], outputs=[leaderboard_table])
|
95 |
|
96 |
interface.launch()
|
|
|
3 |
import pandas as pd
|
4 |
import gradio as gr
|
5 |
from agents.llm_registry import LLM_REGISTRY # Dynamically fetch LLM models
|
6 |
+
from simulators.tic_tac_toe_simulator import TicTacToeSimulator
|
7 |
+
from simulators.prisoners_dilemma_simulator import PrisonersDilemmaSimulator
|
8 |
+
from simulators.rock_paper_scissors_simulator import RockPaperScissorsSimulator
|
9 |
+
from games_registry import GAMES_REGISTRY
|
10 |
+
from simulators.base_simulator import PlayerType
|
11 |
+
from typing import Dict
|
12 |
|
13 |
# Extract available LLM models
|
14 |
llm_models = list(LLM_REGISTRY.keys())
|
15 |
|
16 |
# Define game list manually (for now)
|
17 |
+
games_list = list(GAMES_REGISTRY.keys())
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
18 |
|
19 |
# File to persist results
|
20 |
RESULTS_TRACKER_FILE = "results_tracker.json"
|
|
|
55 |
leaderboard_df.rename(columns={"index": "LLM Model"}, inplace=True)
|
56 |
return leaderboard_df
|
57 |
|
58 |
+
def play_game(game_name, player1_type, player2_type, player1_model, player2_model, rounds):
|
59 |
+
"""Play the selected game with specified players."""
|
60 |
+
llms = {}
|
61 |
+
if player1_type == "llm":
|
62 |
+
llms["Player 1"] = player1_model
|
63 |
+
if player2_type == "llm":
|
64 |
+
llms["Player 2"] = player2_model
|
65 |
+
|
66 |
+
simulator_class = GAMES_REGISTRY[game_name]
|
67 |
+
simulator = simulator_class(game_name, llms=llms)
|
68 |
+
game_states = []
|
69 |
+
|
70 |
+
def log_fn(state):
|
71 |
+
"""Log current state and legal moves."""
|
72 |
+
current_player = state.current_player()
|
73 |
+
legal_moves = state.legal_actions(current_player)
|
74 |
+
board = str(state)
|
75 |
+
game_states.append(f"Current Player: {current_player}\nBoard:\n{board}\nLegal Moves: {legal_moves}")
|
76 |
+
|
77 |
+
results = simulator.simulate(rounds=int(rounds), log_fn=log_fn)
|
78 |
+
return "\n".join(game_states) + f"\nGame Result: {results}"
|
79 |
|
80 |
# Gradio Interface
|
81 |
with gr.Blocks() as interface:
|
82 |
with gr.Tab("Game Arena"):
|
83 |
+
gr.Markdown("# LLM Game Arena\nSelect a game and players to play against LLMs.")
|
84 |
+
|
85 |
+
game_dropdown = gr.Dropdown(choices=games_list, label="Select a Game", value=games_list[0])
|
86 |
+
player1_dropdown = gr.Dropdown(choices=["human", "random_bot", "llm"], label="Player 1 Type", value="llm")
|
87 |
+
player2_dropdown = gr.Dropdown(choices=["human", "random_bot", "llm"], label="Player 2 Type", value="random_bot")
|
88 |
+
player1_model_dropdown = gr.Dropdown(choices=llm_models, label="Player 1 Model", visible=False)
|
89 |
+
player2_model_dropdown = gr.Dropdown(choices=llm_models, label="Player 2 Model", visible=False)
|
90 |
+
rounds_slider = gr.Slider(1, 10, step=1, label="Rounds")
|
91 |
+
result_output = gr.Textbox(label="Game Result")
|
92 |
+
|
93 |
+
play_button = gr.Button("Play Game")
|
94 |
+
play_button.click(
|
95 |
+
play_game,
|
96 |
+
inputs=[game_dropdown, player1_dropdown, player2_dropdown, player1_model_dropdown, player2_model_dropdown, rounds_slider],
|
97 |
+
outputs=result_output,
|
98 |
+
)
|
99 |
|
100 |
with gr.Tab("Leaderboard"):
|
101 |
gr.Markdown("# LLM Model Leaderboard\nTrack performance across different games!")
|
|
|
110 |
"""Updates the leaderboard table based on the selected game."""
|
111 |
return calculate_leaderboard(selected_game)
|
112 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
113 |
game_dropdown.change(fn=update_leaderboard, inputs=[game_dropdown], outputs=[leaderboard_table])
|
114 |
+
refresh_button.click(fn=update_leaderboard, inputs=[game_dropdown], outputs=[leaderboard_table])
|
|
|
115 |
|
116 |
interface.launch()
|