Spaces:
Sleeping
Sleeping
Trying to bring back the games tab and the files download
Browse files
app.py
CHANGED
@@ -14,11 +14,34 @@ from typing import Dict
|
|
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"
|
21 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
22 |
# Load or initialize the results tracker
|
23 |
if os.path.exists(RESULTS_TRACKER_FILE):
|
24 |
with open(RESULTS_TRACKER_FILE, "r") as f:
|
@@ -37,10 +60,10 @@ def save_results_tracker():
|
|
37 |
|
38 |
def calculate_leaderboard(selected_game: str) -> pd.DataFrame:
|
39 |
"""Generate a structured leaderboard table for the selected game."""
|
40 |
-
leaderboard_df = pd.DataFrame(index=llm_models,
|
41 |
-
columns=["# games", "moves/game",
|
42 |
"illegal-moves", "win-rate", "vs Random"])
|
43 |
-
|
44 |
for llm in llm_models:
|
45 |
game_stats = results_tracker[llm].get(selected_game, {})
|
46 |
leaderboard_df.loc[llm] = [
|
@@ -50,7 +73,7 @@ def calculate_leaderboard(selected_game: str) -> pd.DataFrame:
|
|
50 |
f"{game_stats.get('win-rate', 0):.1f}%",
|
51 |
f"{game_stats.get('vs Random', 0):.1f}%"
|
52 |
]
|
53 |
-
|
54 |
leaderboard_df = leaderboard_df.reset_index()
|
55 |
leaderboard_df.rename(columns={"index": "LLM Model"}, inplace=True)
|
56 |
return leaderboard_df
|
@@ -73,7 +96,7 @@ def play_game(game_name, player1_type, player2_type, player1_model, player2_mode
|
|
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 |
|
@@ -81,7 +104,7 @@ def play_game(game_name, player1_type, player2_type, player1_model, player2_mode
|
|
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")
|
@@ -99,17 +122,18 @@ with gr.Blocks() as interface:
|
|
99 |
|
100 |
with gr.Tab("Leaderboard"):
|
101 |
gr.Markdown("# LLM Model Leaderboard\nTrack performance across different games!")
|
102 |
-
|
103 |
game_dropdown = gr.Dropdown(choices=games_list, label="Select Game", value=games_list[0])
|
104 |
leaderboard_table = gr.Dataframe(value=calculate_leaderboard(games_list[0]), label="Leaderboard")
|
105 |
model_dropdown = gr.Dropdown(choices=llm_models, label="Select LLM Model")
|
106 |
download_button = gr.File(label="Download Statistics File")
|
107 |
refresh_button = gr.Button("Refresh Leaderboard")
|
108 |
-
|
109 |
def update_leaderboard(selected_game):
|
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 |
|
|
|
14 |
llm_models = list(LLM_REGISTRY.keys())
|
15 |
|
16 |
# Define game list manually (for now)
|
17 |
+
#games_list = list(GAMES_REGISTRY.keys())
|
18 |
+
games_list = [
|
19 |
+
"rock_paper_scissors",
|
20 |
+
"prisoners_dilemma",
|
21 |
+
"tic_tac_toe",
|
22 |
+
"connect_four",
|
23 |
+
"matching_pennies",
|
24 |
+
"kuhn_poker",
|
25 |
+
]
|
26 |
|
27 |
# File to persist results
|
28 |
RESULTS_TRACKER_FILE = "results_tracker.json"
|
29 |
|
30 |
+
def generate_stats_file(model_name: str):
|
31 |
+
"""Generate a JSON file with detailed statistics for the selected LLM model."""
|
32 |
+
file_path = f"{model_name}_stats.json"
|
33 |
+
with open(file_path, "w") as f:
|
34 |
+
json.dump(results_tracker.get(model_name, {}), f, indent=4)
|
35 |
+
return file_path
|
36 |
+
|
37 |
+
def provide_download_file(model_name):
|
38 |
+
"""Creates a downloadable JSON file with stats for the selected model."""
|
39 |
+
return generate_stats_file(model_name)
|
40 |
+
|
41 |
+
def refresh_leaderboard():
|
42 |
+
"""Manually refresh the leaderboard."""
|
43 |
+
return calculate_leaderboard(game_dropdown.value)
|
44 |
+
|
45 |
# Load or initialize the results tracker
|
46 |
if os.path.exists(RESULTS_TRACKER_FILE):
|
47 |
with open(RESULTS_TRACKER_FILE, "r") as f:
|
|
|
60 |
|
61 |
def calculate_leaderboard(selected_game: str) -> pd.DataFrame:
|
62 |
"""Generate a structured leaderboard table for the selected game."""
|
63 |
+
leaderboard_df = pd.DataFrame(index=llm_models,
|
64 |
+
columns=["# games", "moves/game",
|
65 |
"illegal-moves", "win-rate", "vs Random"])
|
66 |
+
|
67 |
for llm in llm_models:
|
68 |
game_stats = results_tracker[llm].get(selected_game, {})
|
69 |
leaderboard_df.loc[llm] = [
|
|
|
73 |
f"{game_stats.get('win-rate', 0):.1f}%",
|
74 |
f"{game_stats.get('vs Random', 0):.1f}%"
|
75 |
]
|
76 |
+
|
77 |
leaderboard_df = leaderboard_df.reset_index()
|
78 |
leaderboard_df.rename(columns={"index": "LLM Model"}, inplace=True)
|
79 |
return leaderboard_df
|
|
|
96 |
legal_moves = state.legal_actions(current_player)
|
97 |
board = str(state)
|
98 |
game_states.append(f"Current Player: {current_player}\nBoard:\n{board}\nLegal Moves: {legal_moves}")
|
99 |
+
|
100 |
results = simulator.simulate(rounds=int(rounds), log_fn=log_fn)
|
101 |
return "\n".join(game_states) + f"\nGame Result: {results}"
|
102 |
|
|
|
104 |
with gr.Blocks() as interface:
|
105 |
with gr.Tab("Game Arena"):
|
106 |
gr.Markdown("# LLM Game Arena\nSelect a game and players to play against LLMs.")
|
107 |
+
|
108 |
game_dropdown = gr.Dropdown(choices=games_list, label="Select a Game", value=games_list[0])
|
109 |
player1_dropdown = gr.Dropdown(choices=["human", "random_bot", "llm"], label="Player 1 Type", value="llm")
|
110 |
player2_dropdown = gr.Dropdown(choices=["human", "random_bot", "llm"], label="Player 2 Type", value="random_bot")
|
|
|
122 |
|
123 |
with gr.Tab("Leaderboard"):
|
124 |
gr.Markdown("# LLM Model Leaderboard\nTrack performance across different games!")
|
125 |
+
|
126 |
game_dropdown = gr.Dropdown(choices=games_list, label="Select Game", value=games_list[0])
|
127 |
leaderboard_table = gr.Dataframe(value=calculate_leaderboard(games_list[0]), label="Leaderboard")
|
128 |
model_dropdown = gr.Dropdown(choices=llm_models, label="Select LLM Model")
|
129 |
download_button = gr.File(label="Download Statistics File")
|
130 |
refresh_button = gr.Button("Refresh Leaderboard")
|
131 |
+
|
132 |
def update_leaderboard(selected_game):
|
133 |
"""Updates the leaderboard table based on the selected game."""
|
134 |
return calculate_leaderboard(selected_game)
|
135 |
+
|
136 |
+
model_dropdown.change(fn=provide_download_file, inputs=[model_dropdown], outputs=[download_button])
|
137 |
game_dropdown.change(fn=update_leaderboard, inputs=[game_dropdown], outputs=[leaderboard_table])
|
138 |
refresh_button.click(fn=update_leaderboard, inputs=[game_dropdown], outputs=[leaderboard_table])
|
139 |
|