Spaces:
Sleeping
Sleeping
Updated after name change
Browse files- ui/README.md +42 -0
- ui/__init__.py +7 -0
- ui/gradio_config_generator.py +445 -0
ui/README.md
CHANGED
@@ -0,0 +1,42 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# Gradio Interface Components
|
2 |
+
|
3 |
+
This directory contains the Gradio interface components for the Board Game Arena.
|
4 |
+
|
5 |
+
## Files
|
6 |
+
|
7 |
+
- `gradio_config_generator.py` - Configuration generator that bridges Gradio UI with the game infrastructure
|
8 |
+
- `__init__.py` - Package initialization
|
9 |
+
|
10 |
+
## Main App
|
11 |
+
|
12 |
+
The main Gradio app (`app.py`) is located in the root directory for HuggingFace Spaces compatibility.
|
13 |
+
|
14 |
+
## Running the App
|
15 |
+
|
16 |
+
From the project root directory:
|
17 |
+
|
18 |
+
```bash
|
19 |
+
python app.py
|
20 |
+
```
|
21 |
+
|
22 |
+
## Architecture
|
23 |
+
|
24 |
+
```
|
25 |
+
app.py (Gradio UI - in root directory for HF Spaces)
|
26 |
+
โ
|
27 |
+
ui/gradio_config_generator.py (Game configuration bridge)
|
28 |
+
โ
|
29 |
+
src/game_reasoning_arena/ (Core game library)
|
30 |
+
```
|
31 |
+
|
32 |
+
The Gradio app provides:
|
33 |
+
- Interactive game interface
|
34 |
+
- Performance leaderboards
|
35 |
+
- Metrics dashboards
|
36 |
+
- LLM reasoning analysis
|
37 |
+
|
38 |
+
|
39 |
+
## Uploading Results
|
40 |
+
|
41 |
+
- Go to **Leaderboard** tab โ **Upload .db**
|
42 |
+
- Files are stored in `scripts/results/` inside the Space
|
ui/__init__.py
ADDED
@@ -0,0 +1,7 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
"""
|
2 |
+
Gradio interface components for Board Game Arena.
|
3 |
+
"""
|
4 |
+
|
5 |
+
from .gradio_config_generator import run_game_with_existing_infrastructure
|
6 |
+
|
7 |
+
__all__ = ['run_game_with_existing_infrastructure']
|
ui/gradio_config_generator.py
ADDED
@@ -0,0 +1,445 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
#!/usr/bin/env python3
|
2 |
+
"""
|
3 |
+
Gradio Configuration Generator
|
4 |
+
|
5 |
+
This module creates configurations compatible with the existing runner.py and
|
6 |
+
simulate.py infrastructure, eliminating code duplication in the Gradio app.
|
7 |
+
"""
|
8 |
+
|
9 |
+
import tempfile
|
10 |
+
import yaml
|
11 |
+
from typing import Dict, Any, Tuple
|
12 |
+
import logging
|
13 |
+
|
14 |
+
logger = logging.getLogger(__name__)
|
15 |
+
|
16 |
+
|
17 |
+
def create_config_for_gradio_game(
|
18 |
+
game_name: str,
|
19 |
+
player1_type: str,
|
20 |
+
player2_type: str,
|
21 |
+
player1_model: str = None,
|
22 |
+
player2_model: str = None,
|
23 |
+
rounds: int = 1,
|
24 |
+
seed: int = 42,
|
25 |
+
use_ray: bool = False
|
26 |
+
) -> Dict[str, Any]:
|
27 |
+
"""
|
28 |
+
Create a configuration dictionary compatible with the existing
|
29 |
+
runner.py and simulate.py infrastructure.
|
30 |
+
|
31 |
+
Args:
|
32 |
+
game_name: Name of the game to play
|
33 |
+
player1_type: Type of player 1 (human, random, llm)
|
34 |
+
player2_type: Type of player 2 (human, random, llm)
|
35 |
+
player1_model: LLM model for player 1 (if applicable)
|
36 |
+
player2_model: LLM model for player 2 (if applicable)
|
37 |
+
rounds: Number of episodes to play
|
38 |
+
seed: Random seed for reproducibility
|
39 |
+
use_ray: Whether to use Ray for parallel processing
|
40 |
+
|
41 |
+
Returns:
|
42 |
+
Configuration dictionary compatible with runner.py
|
43 |
+
"""
|
44 |
+
|
45 |
+
# Base configuration structure (matches default_simulation_config)
|
46 |
+
config = {
|
47 |
+
"env_config": {
|
48 |
+
"game_name": game_name,
|
49 |
+
"max_game_rounds": None,
|
50 |
+
},
|
51 |
+
"num_episodes": rounds,
|
52 |
+
"seed": seed,
|
53 |
+
"use_ray": use_ray,
|
54 |
+
"mode": f"{player1_type}_vs_{player2_type}",
|
55 |
+
"agents": {},
|
56 |
+
"llm_backend": {
|
57 |
+
"max_tokens": 250,
|
58 |
+
"temperature": 0.1,
|
59 |
+
"default_model": "litellm_groq/gemma-7b-it",
|
60 |
+
},
|
61 |
+
"log_level": "INFO",
|
62 |
+
}
|
63 |
+
|
64 |
+
# Configure player agents
|
65 |
+
config["agents"]["player_0"] = _create_agent_config(
|
66 |
+
player1_type, player1_model)
|
67 |
+
config["agents"]["player_1"] = _create_agent_config(
|
68 |
+
player2_type, player2_model)
|
69 |
+
|
70 |
+
# Debug: Print the agent configurations
|
71 |
+
print("๐ CONFIG DEBUG: Agent configurations created:")
|
72 |
+
print(f" Player 0 config: {config['agents']['player_0']}")
|
73 |
+
print(f" Player 1 config: {config['agents']['player_1']}")
|
74 |
+
|
75 |
+
# Update backend default model if LLM is used
|
76 |
+
# Check player 1 first
|
77 |
+
if (player1_type == "llm" and player1_model) or player1_type.startswith("llm_"):
|
78 |
+
if player1_model:
|
79 |
+
config["llm_backend"]["default_model"] = player1_model
|
80 |
+
elif player1_type.startswith("llm_"):
|
81 |
+
# Extract model from player type (e.g., "llm_gpt2" -> "gpt2")
|
82 |
+
config["llm_backend"]["default_model"] = player1_type[4:]
|
83 |
+
# Check player 2 if player 1 doesn't have LLM
|
84 |
+
elif (player2_type == "llm" and player2_model) or player2_type.startswith("llm_"):
|
85 |
+
if player2_model:
|
86 |
+
config["llm_backend"]["default_model"] = player2_model
|
87 |
+
elif player2_type.startswith("llm_"):
|
88 |
+
# Extract model from player type (e.g., "llm_gpt2" -> "gpt2")
|
89 |
+
config["llm_backend"]["default_model"] = player2_type[4:]
|
90 |
+
|
91 |
+
return config
|
92 |
+
|
93 |
+
|
94 |
+
def _create_agent_config(player_type: str,
|
95 |
+
model: str = None) -> Dict[str, Any]:
|
96 |
+
"""
|
97 |
+
Create agent configuration based on player type and model.
|
98 |
+
|
99 |
+
Handles both Gradio-specific formats (e.g., "hf_gpt2", "random_bot")
|
100 |
+
and standard formats (e.g., "llm", "random").
|
101 |
+
|
102 |
+
Args:
|
103 |
+
player_type: Type of player (human, random, random_bot, hf_*, etc.)
|
104 |
+
model: Model name for LLM agents
|
105 |
+
|
106 |
+
Returns:
|
107 |
+
Agent configuration dictionary
|
108 |
+
"""
|
109 |
+
print("๐ง AGENT CONFIG DEBUG: Creating agent config for:")
|
110 |
+
print(f" player_type: {player_type}")
|
111 |
+
print(f" model: {model}")
|
112 |
+
|
113 |
+
# Handle Gradio-specific formats
|
114 |
+
if player_type == "random_bot":
|
115 |
+
config = {"type": "random"}
|
116 |
+
elif player_type.startswith("hf_"):
|
117 |
+
# Extract model from player type (e.g., "hf_gpt2" -> "gpt2")
|
118 |
+
model_from_type = player_type[3:] # Remove "hf_" prefix
|
119 |
+
|
120 |
+
# Use the hf_prefixed model name for LLM registry lookup
|
121 |
+
model_name = f"hf_{model_from_type}"
|
122 |
+
|
123 |
+
config = {
|
124 |
+
"type": "llm", # Use standard LLM agent type
|
125 |
+
"model": model_name # This will be looked up in LLM_REGISTRY
|
126 |
+
}
|
127 |
+
elif player_type.startswith("llm_"):
|
128 |
+
# For backwards compatibility with LiteLLM models
|
129 |
+
model_from_type = player_type[4:] # Remove "llm_" prefix
|
130 |
+
|
131 |
+
# Map display model names to actual model names with prefixes
|
132 |
+
model_name = model or model_from_type
|
133 |
+
if not model_name.startswith(("litellm_", "vllm_")):
|
134 |
+
# Add litellm_ prefix for LiteLLM models
|
135 |
+
model_name = f"litellm_{model_name}"
|
136 |
+
|
137 |
+
config = {
|
138 |
+
"type": "llm",
|
139 |
+
"model": model_name
|
140 |
+
}
|
141 |
+
elif player_type == "llm":
|
142 |
+
model_name = model or "litellm_groq/gemma-7b-it"
|
143 |
+
if not model_name.startswith(("litellm_", "vllm_")):
|
144 |
+
model_name = f"litellm_{model_name}"
|
145 |
+
config = {
|
146 |
+
"type": "llm",
|
147 |
+
"model": model_name
|
148 |
+
}
|
149 |
+
elif player_type == "random":
|
150 |
+
config = {"type": "random"}
|
151 |
+
elif player_type == "human":
|
152 |
+
config = {"type": "human"} # This might need additional handling
|
153 |
+
else:
|
154 |
+
# Default to random for unknown types
|
155 |
+
config = {"type": "random"}
|
156 |
+
|
157 |
+
print(f" โ Created config: {config}")
|
158 |
+
return config
|
159 |
+
|
160 |
+
|
161 |
+
def create_temporary_config_file(config: Dict[str, Any]) -> str:
|
162 |
+
"""
|
163 |
+
Create a temporary YAML config file that can be used with runner.py.
|
164 |
+
|
165 |
+
Args:
|
166 |
+
config: Configuration dictionary
|
167 |
+
|
168 |
+
Returns:
|
169 |
+
Path to the temporary config file
|
170 |
+
"""
|
171 |
+
# Create temporary file
|
172 |
+
temp_file = tempfile.NamedTemporaryFile(
|
173 |
+
mode='w',
|
174 |
+
suffix='.yaml',
|
175 |
+
delete=False
|
176 |
+
)
|
177 |
+
|
178 |
+
try:
|
179 |
+
yaml.dump(config, temp_file, default_flow_style=False)
|
180 |
+
temp_file.flush()
|
181 |
+
return temp_file.name
|
182 |
+
finally:
|
183 |
+
temp_file.close()
|
184 |
+
|
185 |
+
|
186 |
+
def run_game_with_existing_infrastructure(
|
187 |
+
game_name: str,
|
188 |
+
player1_type: str,
|
189 |
+
player2_type: str,
|
190 |
+
player1_model: str = None,
|
191 |
+
player2_model: str = None,
|
192 |
+
rounds: int = 1,
|
193 |
+
seed: int = 42
|
194 |
+
) -> str:
|
195 |
+
"""
|
196 |
+
Run a game using the existing runner.py and simulate.py infrastructure,
|
197 |
+
but capture detailed game logs for Gradio display.
|
198 |
+
|
199 |
+
This function reuses the existing simulation infrastructure while providing
|
200 |
+
detailed game output for the Gradio interface.
|
201 |
+
|
202 |
+
Args:
|
203 |
+
game_name: Name of the game to play
|
204 |
+
player1_type: Type of player 1
|
205 |
+
player2_type: Type of player 2
|
206 |
+
player1_model: LLM model for player 1 (if applicable)
|
207 |
+
player2_model: LLM model for player 2 (if applicable)
|
208 |
+
rounds: Number of episodes to play
|
209 |
+
seed: Random seed
|
210 |
+
|
211 |
+
Returns:
|
212 |
+
Detailed game simulation results as a string
|
213 |
+
"""
|
214 |
+
try:
|
215 |
+
# Import the existing infrastructure
|
216 |
+
from src.game_reasoning_arena.arena.utils.seeding import set_seed
|
217 |
+
from src.game_reasoning_arena.backends import initialize_llm_registry
|
218 |
+
from src.game_reasoning_arena.arena.games.registry import registry
|
219 |
+
from src.game_reasoning_arena.arena.agents.policy_manager import (
|
220 |
+
initialize_policies, policy_mapping_fn
|
221 |
+
)
|
222 |
+
|
223 |
+
# Create configuration
|
224 |
+
config = create_config_for_gradio_game(
|
225 |
+
game_name=game_name,
|
226 |
+
player1_type=player1_type,
|
227 |
+
player2_type=player2_type,
|
228 |
+
player1_model=player1_model,
|
229 |
+
player2_model=player2_model,
|
230 |
+
rounds=rounds,
|
231 |
+
seed=seed
|
232 |
+
)
|
233 |
+
|
234 |
+
# Set seed
|
235 |
+
set_seed(seed)
|
236 |
+
|
237 |
+
# Initialize LLM registry (required for simulate_game)
|
238 |
+
initialize_llm_registry(config)
|
239 |
+
|
240 |
+
# Use existing infrastructure but capture detailed logs
|
241 |
+
return _run_game_with_detailed_logging(game_name, config, seed)
|
242 |
+
|
243 |
+
except ImportError as e:
|
244 |
+
logger.error(f"Failed to import simulation infrastructure: {e}")
|
245 |
+
return f"Error: Simulation infrastructure not available. {e}"
|
246 |
+
except Exception as e:
|
247 |
+
logger.error(f"Game simulation failed: {e}")
|
248 |
+
return f"Error during game simulation: {e}"
|
249 |
+
|
250 |
+
|
251 |
+
def _run_game_with_detailed_logging(
|
252 |
+
game_name: str,
|
253 |
+
config: Dict[str, Any],
|
254 |
+
seed: int
|
255 |
+
) -> str:
|
256 |
+
"""
|
257 |
+
Run game simulation with detailed logging for Gradio display.
|
258 |
+
|
259 |
+
This reuses the existing infrastructure components but captures
|
260 |
+
detailed game state information for user display.
|
261 |
+
"""
|
262 |
+
from src.game_reasoning_arena.arena.games.registry import registry
|
263 |
+
from src.game_reasoning_arena.arena.agents.policy_manager import (
|
264 |
+
initialize_policies, policy_mapping_fn
|
265 |
+
)
|
266 |
+
|
267 |
+
# Initialize using existing infrastructure
|
268 |
+
policies_dict = initialize_policies(config, game_name, seed)
|
269 |
+
env = registry.make_env(game_name, config)
|
270 |
+
|
271 |
+
# Create player mapping (reusing existing logic)
|
272 |
+
player_to_agent = {}
|
273 |
+
for i, policy_name in enumerate(policies_dict.keys()):
|
274 |
+
player_to_agent[i] = policies_dict[policy_name]
|
275 |
+
|
276 |
+
game_log = []
|
277 |
+
|
278 |
+
# Add header
|
279 |
+
game_log.append("๐ฎ GAME SIMULATION RESULTS")
|
280 |
+
game_log.append("=" * 50)
|
281 |
+
game_log.append(f"Game: {game_name.replace('_', ' ').title()}")
|
282 |
+
game_log.append(f"Episodes: {config['num_episodes']}")
|
283 |
+
game_log.append("")
|
284 |
+
|
285 |
+
# Player information
|
286 |
+
game_log.append("๐ฅ PLAYERS:")
|
287 |
+
player1 = config["agents"]["player_0"]
|
288 |
+
player2 = config["agents"]["player_1"]
|
289 |
+
game_log.append(f" Player 0: {_format_player_info(player1)}")
|
290 |
+
game_log.append(f" Player 1: {_format_player_info(player2)}")
|
291 |
+
game_log.append("")
|
292 |
+
|
293 |
+
# Run episodes (reusing compute_actions logic from simulate.py)
|
294 |
+
for episode in range(config["num_episodes"]):
|
295 |
+
episode_seed = seed + episode
|
296 |
+
game_log.append(f"๐ฏ Episode {episode + 1}")
|
297 |
+
game_log.append("-" * 30)
|
298 |
+
|
299 |
+
observation_dict, _ = env.reset(seed=episode_seed)
|
300 |
+
terminated = truncated = False
|
301 |
+
step_count = 0
|
302 |
+
episode_rewards = {0: 0, 1: 0}
|
303 |
+
|
304 |
+
while not (terminated or truncated):
|
305 |
+
step_count += 1
|
306 |
+
game_log.append(f"\n๐ Step {step_count}")
|
307 |
+
|
308 |
+
# Show board state
|
309 |
+
try:
|
310 |
+
board = env.render_board(0)
|
311 |
+
game_log.append("Current board:")
|
312 |
+
game_log.append(board)
|
313 |
+
except:
|
314 |
+
game_log.append("Board state not available")
|
315 |
+
|
316 |
+
# Use the existing compute_actions logic from simulate.py
|
317 |
+
try:
|
318 |
+
action_dict = _compute_actions_for_gradio(
|
319 |
+
env, player_to_agent, observation_dict, game_log
|
320 |
+
)
|
321 |
+
except Exception as e:
|
322 |
+
game_log.append(f"โ Error computing actions: {e}")
|
323 |
+
truncated = True
|
324 |
+
break
|
325 |
+
|
326 |
+
# Step forward (reusing existing environment logic)
|
327 |
+
if not truncated:
|
328 |
+
observation_dict, rewards, terminated, truncated, _ = env.step(action_dict)
|
329 |
+
for player_id, reward in rewards.items():
|
330 |
+
episode_rewards[player_id] += reward
|
331 |
+
|
332 |
+
# Episode results
|
333 |
+
game_log.append(f"\n๐ Episode {episode + 1} Complete!")
|
334 |
+
try:
|
335 |
+
game_log.append("Final board:")
|
336 |
+
game_log.append(env.render_board(0))
|
337 |
+
except:
|
338 |
+
game_log.append("Final board state not available")
|
339 |
+
|
340 |
+
if episode_rewards[0] > episode_rewards[1]:
|
341 |
+
winner = "Player 0"
|
342 |
+
elif episode_rewards[1] > episode_rewards[0]:
|
343 |
+
winner = "Player 1"
|
344 |
+
else:
|
345 |
+
winner = "Draw"
|
346 |
+
|
347 |
+
game_log.append(f"๐ Winner: {winner}")
|
348 |
+
game_log.append(f"๐ Scores: Player 0={episode_rewards[0]}, Player 1={episode_rewards[1]}")
|
349 |
+
game_log.append("")
|
350 |
+
|
351 |
+
game_log.append("โ
Simulation completed successfully!")
|
352 |
+
game_log.append("Check the database logs for detailed move analysis.")
|
353 |
+
|
354 |
+
return "\n".join(game_log)
|
355 |
+
|
356 |
+
|
357 |
+
def _compute_actions_for_gradio(env, player_to_agent, observations, game_log):
|
358 |
+
"""
|
359 |
+
Compute actions and log details for Gradio display.
|
360 |
+
This reuses the compute_actions logic from simulate.py.
|
361 |
+
"""
|
362 |
+
if env.state.is_simultaneous_node():
|
363 |
+
# Simultaneous-move game
|
364 |
+
actions = {}
|
365 |
+
for player in player_to_agent:
|
366 |
+
agent_response = player_to_agent[player](observations[player])
|
367 |
+
action, reasoning = _extract_action_and_reasoning(agent_response)
|
368 |
+
actions[player] = action
|
369 |
+
|
370 |
+
game_log.append(f" Player {player} chooses action {action}")
|
371 |
+
if reasoning and reasoning != "None":
|
372 |
+
reasoning_preview = reasoning[:100] + ("..." if len(reasoning) > 100 else "")
|
373 |
+
game_log.append(f" Reasoning: {reasoning_preview}")
|
374 |
+
return actions
|
375 |
+
else:
|
376 |
+
# Turn-based game
|
377 |
+
current_player = env.state.current_player()
|
378 |
+
game_log.append(f"Player {current_player}'s turn")
|
379 |
+
|
380 |
+
agent_response = player_to_agent[current_player](observations[current_player])
|
381 |
+
action, reasoning = _extract_action_and_reasoning(agent_response)
|
382 |
+
|
383 |
+
game_log.append(f" Player {current_player} chooses action {action}")
|
384 |
+
if reasoning and reasoning != "None":
|
385 |
+
reasoning_preview = reasoning[:100] + ("..." if len(reasoning) > 100 else "")
|
386 |
+
game_log.append(f" Reasoning: {reasoning_preview}")
|
387 |
+
|
388 |
+
return {current_player: action}
|
389 |
+
|
390 |
+
|
391 |
+
def _extract_action_and_reasoning(agent_response):
|
392 |
+
"""Extract action and reasoning from agent response."""
|
393 |
+
if isinstance(agent_response, dict) and "action" in agent_response:
|
394 |
+
action = agent_response.get("action", -1)
|
395 |
+
reasoning = agent_response.get("reasoning", "None")
|
396 |
+
return action, reasoning
|
397 |
+
else:
|
398 |
+
return agent_response, "None"
|
399 |
+
|
400 |
+
|
401 |
+
def _format_player_info(player_config: Dict[str, Any]) -> str:
|
402 |
+
"""Format player information for display."""
|
403 |
+
player_type = player_config["type"]
|
404 |
+
if player_type == "llm":
|
405 |
+
model = player_config.get("model", "unknown")
|
406 |
+
return f"LLM ({model})"
|
407 |
+
else:
|
408 |
+
return player_type.replace("_", " ").title()
|
409 |
+
|
410 |
+
|
411 |
+
# For backward compatibility and easy integration
|
412 |
+
def create_gradio_compatible_config(
|
413 |
+
game_name: str,
|
414 |
+
player1_type: str,
|
415 |
+
player2_type: str,
|
416 |
+
player1_model: str = None,
|
417 |
+
player2_model: str = None,
|
418 |
+
rounds: int = 1
|
419 |
+
) -> Tuple[Dict[str, Any], str]:
|
420 |
+
"""
|
421 |
+
Create both a config dict and a temp file for maximum compatibility.
|
422 |
+
|
423 |
+
Returns:
|
424 |
+
Tuple of (config_dict, temp_file_path)
|
425 |
+
"""
|
426 |
+
config = create_config_for_gradio_game(
|
427 |
+
game_name, player1_type, player2_type,
|
428 |
+
player1_model, player2_model, rounds
|
429 |
+
)
|
430 |
+
temp_file = create_temporary_config_file(config)
|
431 |
+
return config, temp_file
|
432 |
+
|
433 |
+
|
434 |
+
if __name__ == "__main__":
|
435 |
+
# Example usage
|
436 |
+
config = create_config_for_gradio_game(
|
437 |
+
game_name="tic_tac_toe",
|
438 |
+
player1_type="llm",
|
439 |
+
player2_type="random",
|
440 |
+
player1_model="litellm_groq/llama-3.1-8b-instant",
|
441 |
+
rounds=3
|
442 |
+
)
|
443 |
+
|
444 |
+
print("Generated configuration:")
|
445 |
+
print(yaml.dump(config, default_flow_style=False))
|