Spaces:
Sleeping
Sleeping
File size: 19,178 Bytes
d119afc | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 | from __future__ import annotations
import re
from dataclasses import dataclass, field
from enum import StrEnum
from typing import Any, Callable, TextIO
from persistentpoker_bench.adapters.litellm_adapter import LiteLLMConfig
from persistentpoker_bench.cards import cards_to_notation
from persistentpoker_bench.hand_runner import (
DecisionAgent,
DecisionEnvelope,
HandObserver,
HandRunResult,
HandRunnerConfig,
run_seeded_hand,
)
from persistentpoker_bench.pool import PersistentPool
from persistentpoker_bench.retries import RetryPolicy
from persistentpoker_bench.runtime_agents import LiteLLMRuntimeAgent
from persistentpoker_bench.schemas import LLMDecision, WinnerPoolDecision
class PlaySeatKind(StrEnum):
HUMAN = "human"
PASSIVE_BOT = "passive_bot"
LITELLM = "litellm"
@dataclass(frozen=True, slots=True)
class HumanCommand:
action: str
amount: int | None
winner_pool_decision: WinnerPoolDecision
@dataclass(frozen=True, slots=True)
class PlaySeatSpec:
name: str
kind: PlaySeatKind
provider: str | None = None
model_id: str | None = None
litellm_model: str | None = None
temperature: float = 0.0
max_tokens: int = 400
timeout: float = 60.0
prefer_json_mode: bool = True
max_attempts: int = 3
initial_delay_seconds: float = 0.25
backoff_multiplier: float = 2.0
extra_kwargs: dict[str, Any] = field(default_factory=dict)
@dataclass(frozen=True, slots=True)
class PlaySessionConfig:
seats: tuple[PlaySeatSpec, ...]
hand_count: int
hand_runner_config: HandRunnerConfig
replay_out: str | None = None
@property
def player_names(self) -> tuple[str, ...]:
return tuple(seat.name for seat in self.seats)
@dataclass(slots=True)
class HumanDecisionAgent(DecisionAgent):
seat_name: str
input_fn: Callable[[str], str] = input
output: TextIO | None = None
def decide(
self,
*,
prompt_bundle: Any,
game_snapshot: dict[str, Any],
legal_actions_snapshot: dict[str, Any],
player_index: int,
hand_state: Any,
persistent_pool: PersistentPool,
) -> DecisionEnvelope:
output = self.output
while True:
if output is not None:
output.write(
f"\nTour de {self.seat_name}. Commande attendue: "
"fold | check | call | all_in | bet <montant> | raise <montant> "
"[; reset|continue]\n"
)
output.write(f"Actions légales: {format_legal_actions(legal_actions_snapshot)}\n")
raw_text = self.input_fn("> ").strip()
try:
command = parse_human_command(raw_text)
except ValueError as exc:
if output is not None:
output.write(f"Commande invalide: {exc}\n")
continue
decision = LLMDecision(
action=command.action,
amount=command.amount,
believed_pool=persistent_pool.notation_snapshot(),
winner_pool_decision=command.winner_pool_decision,
reasoning="human-terminal",
)
return DecisionEnvelope(
decision=decision,
raw_text=raw_text,
parse_mode="human",
attempts=1,
provider="human",
model_id=f"human:{self.seat_name}",
latency_seconds=0.0,
usage=_zero_usage(),
)
@dataclass(slots=True)
class PassiveBotAgent(DecisionAgent):
seat_name: str
def decide(
self,
*,
prompt_bundle: Any,
game_snapshot: dict[str, Any],
legal_actions_snapshot: dict[str, Any],
player_index: int,
hand_state: Any,
persistent_pool: PersistentPool,
) -> DecisionEnvelope:
if legal_actions_snapshot["can_check"]:
action = "check"
amount = None
elif legal_actions_snapshot["can_call"]:
action = "call"
amount = None
else:
action = "fold"
amount = None
return DecisionEnvelope(
decision=LLMDecision(
action=action,
amount=amount,
believed_pool=persistent_pool.notation_snapshot(),
winner_pool_decision=WinnerPoolDecision.CONTINUE,
reasoning="passive-bot",
),
raw_text=action,
parse_mode="passive-bot",
attempts=1,
provider="bot",
model_id="passive-bot",
latency_seconds=0.0,
usage=_zero_usage(),
)
@dataclass(slots=True)
class TerminalHandObserver(HandObserver):
output: TextIO
visible_hole_seats: set[int]
def on_hand_started(
self,
*,
hand_id: str,
seed: int,
hand_state: Any,
persistent_pool: PersistentPool,
) -> None:
self.output.write(f"\n=== {hand_id} | seed={seed} ===\n")
self.output.write(render_table_state(hand_state, persistent_pool, visible_hole_seats=self.visible_hole_seats))
def on_action(
self,
*,
hand_state: Any,
persistent_pool: PersistentPool,
event: dict[str, Any],
) -> None:
if event.get("event_type") == "tiebreak":
self.output.write(
f"\nDépartage au dé [{event['context']}] -> seat {int(event['winner']) + 1}\n"
)
return
executed = event["executed_action"]
amount_suffix = f" {executed['amount']}" if executed.get("amount") is not None else ""
self.output.write(
f"\n{event['player_name']} -> {executed['action']}{amount_suffix} "
f"(memoire={event['memory']['multiset_accuracy']:.2f})\n"
)
self.output.write(render_table_state(hand_state, persistent_pool, visible_hole_seats=self.visible_hole_seats))
def on_street_advanced(
self,
*,
hand_state: Any,
persistent_pool: PersistentPool,
) -> None:
self.output.write(f"\n--- Street: {hand_state.street.value} ---\n")
self.output.write(render_table_state(hand_state, persistent_pool, visible_hole_seats=self.visible_hole_seats))
def on_hand_completed(
self,
*,
result: HandRunResult,
) -> None:
self.output.write("\n=== Résultat de la main ===\n")
self.output.write(render_hand_result(result))
def play_terminal_match(
*,
player_names: tuple[str, ...],
human_seats: tuple[int, ...],
hand_count: int,
config: HandRunnerConfig,
output: TextIO,
input_fn: Callable[[str], str] = input,
) -> tuple[HandRunResult, ...]:
seats = tuple(
PlaySeatSpec(
name=seat_name,
kind=PlaySeatKind.HUMAN if seat_index in human_seats else PlaySeatKind.PASSIVE_BOT,
)
for seat_index, seat_name in enumerate(player_names)
)
return run_play_session(
PlaySessionConfig(seats=seats, hand_count=hand_count, hand_runner_config=config),
input_fn=input_fn,
output=output,
observer_factory=lambda visible_hole_seats: TerminalHandObserver(
output=output,
visible_hole_seats=visible_hole_seats,
),
)
def run_play_session(
session_config: PlaySessionConfig,
*,
input_fn: Callable[[str], str] = input,
output: TextIO | None = None,
observer_factory: Callable[[set[int]], HandObserver] | None = None,
) -> tuple[HandRunResult, ...]:
pool = PersistentPool()
results: list[HandRunResult] = []
current_stacks = [session_config.hand_runner_config.starting_stack for _ in session_config.seats]
current_button_index = 0
visible_hole_seats = {
seat_index
for seat_index, seat in enumerate(session_config.seats)
if seat.kind is PlaySeatKind.HUMAN
}
for hand_number in range(1, session_config.hand_count + 1):
if _count_live_stacks(current_stacks) <= 1:
break
observer = (
observer_factory(visible_hole_seats)
if observer_factory is not None
else None
)
current_button_index = _resolve_button_for_next_hand(current_button_index, current_stacks)
decision_agents = build_play_agents(
session_config.seats,
input_fn=input_fn,
output=output,
)
result = run_seeded_hand(
player_names=session_config.player_names,
decision_agents=decision_agents,
persistent_pool=pool,
config=session_config.hand_runner_config,
starting_stacks=current_stacks,
button_index=current_button_index,
hand_number=hand_number,
observer=observer,
)
results.append(result)
current_stacks = list(result.ending_stacks_snapshot)
if _count_live_stacks(current_stacks) <= 1:
break
current_button_index = _next_live_seat(current_button_index, current_stacks)
return tuple(results)
def _count_live_stacks(stacks: list[int] | tuple[int, ...]) -> int:
return sum(1 for stack in stacks if stack > 0)
def _resolve_button_for_next_hand(button_index: int, stacks: list[int] | tuple[int, ...]) -> int:
if stacks[button_index] > 0:
return button_index
return _next_live_seat(button_index, stacks)
def _next_live_seat(start_index: int, stacks: list[int] | tuple[int, ...]) -> int:
player_count = len(stacks)
for offset in range(1, player_count + 1):
candidate = (start_index + offset) % player_count
if stacks[candidate] > 0:
return candidate
return start_index
def parse_play_session_config(payload: dict[str, Any]) -> PlaySessionConfig:
seats_payload = payload.get("players")
if not isinstance(seats_payload, list) or not seats_payload:
raise ValueError("play config must include a non-empty 'players' list.")
seats = tuple(_parse_play_seat_spec(seat_payload) for seat_payload in seats_payload)
return PlaySessionConfig(
seats=seats,
hand_count=int(payload.get("hand_count", 1)),
hand_runner_config=HandRunnerConfig(
seed=int(payload.get("seed", 20260428)),
hand_id_prefix=str(payload.get("hand_id_prefix", "hand")),
starting_stack=int(payload.get("starting_stack", 2000)),
small_blind=int(payload.get("small_blind", 10)),
big_blind=int(payload.get("big_blind", 20)),
),
replay_out=str(payload["replay_out"]) if "replay_out" in payload else None,
)
def build_play_agents(
seats: tuple[PlaySeatSpec, ...],
*,
input_fn: Callable[[str], str] = input,
output: TextIO | None = None,
) -> dict[int, DecisionAgent]:
return {
seat_index: _build_seat_agent(
seat,
input_fn=input_fn,
output=output,
)
for seat_index, seat in enumerate(seats)
}
def parse_human_command(raw_text: str) -> HumanCommand:
text = raw_text.strip().lower()
if not text:
raise ValueError("commande vide")
pool_decision = WinnerPoolDecision.CONTINUE
if ";" in text:
action_text, pool_text = [part.strip() for part in text.split(";", maxsplit=1)]
text = action_text
if pool_text:
pool_decision = _parse_pool_decision(pool_text)
match = re.fullmatch(r"(fold|check|call|all_in)", text)
if match:
return HumanCommand(
action=match.group(1),
amount=None,
winner_pool_decision=pool_decision,
)
amount_match = re.fullmatch(r"(bet|raise)\s+(\d+)", text)
if amount_match:
return HumanCommand(
action=amount_match.group(1),
amount=int(amount_match.group(2)),
winner_pool_decision=pool_decision,
)
raise ValueError("format non reconnu")
def format_legal_actions(legal_actions_snapshot: dict[str, Any]) -> str:
options: list[str] = []
if legal_actions_snapshot["can_fold"]:
options.append("fold")
if legal_actions_snapshot["can_check"]:
options.append("check")
if legal_actions_snapshot["can_call"]:
options.append(f"call({legal_actions_snapshot['call_amount']})")
if legal_actions_snapshot["can_bet"]:
options.append(
f"bet[{legal_actions_snapshot['min_bet_to']}-{legal_actions_snapshot['max_to']}]"
)
if legal_actions_snapshot["can_raise"]:
options.append(
f"raise[{legal_actions_snapshot['min_raise_to']}-{legal_actions_snapshot['max_to']}]"
)
if legal_actions_snapshot["can_all_in"]:
options.append(f"all_in({legal_actions_snapshot['max_to']})")
return ", ".join(options)
def render_table_state(
hand_state: Any,
persistent_pool: PersistentPool,
*,
visible_hole_seats: set[int] | None = None,
) -> str:
visible_hole_seats = visible_hole_seats or set()
lines = [
f"Street: {hand_state.street.value}",
f"Board: {' '.join(cards_to_notation(hand_state.community_cards)) or '-'}",
f"Pool public: {' '.join(persistent_pool.notation_snapshot()) or '-'}",
f"Pot: {hand_state.pot_total} | Mise courante: {hand_state.current_bet}",
]
for player in hand_state.players:
hole = "?? ??"
if player.seat in visible_hole_seats and len(player.hole_cards) == 2:
hole = " ".join(cards_to_notation(player.hole_cards))
actor_marker = " <- to act" if hand_state.actor_index == player.seat and hand_state.pending_actor_indices else ""
lines.append(
f"P{player.seat + 1} {player.name}: stack={player.stack} "
f"committed={player.committed_total} folded={player.folded} all_in={player.all_in} "
f"cards={hole}{actor_marker}"
)
return "\n".join(lines) + "\n"
def render_hand_result(result: HandRunResult) -> str:
lines = [
f"Winner pool decision: {result.winner_pool_decision}",
f"Board final: {' '.join(cards_to_notation(result.hand_state.community_cards))}",
f"Pool après main: {' '.join(result.persistent_pool_after) or '-'}",
]
for player in result.hand_state.players:
cards = " ".join(cards_to_notation(player.hole_cards)) if len(player.hole_cards) == 2 else "- -"
payout = result.showdown_result.payouts[player.seat] if result.showdown_result is not None else 0
lines.append(f"P{player.seat + 1} {player.name}: {cards} | payout={payout}")
if result.tiebreak_events:
lines.append("Départages au dé:")
for event in result.tiebreak_events:
lines.append(f"- {event['context']} -> winner seat {int(event['winner']) + 1}")
return "\n".join(lines) + "\n"
def _parse_play_seat_spec(payload: dict[str, Any]) -> PlaySeatSpec:
if not isinstance(payload, dict):
raise ValueError("Each player spec must be an object.")
try:
kind = PlaySeatKind(str(payload["kind"]).strip().lower())
except KeyError as exc:
raise ValueError("Each player spec must include a 'kind'.") from exc
except ValueError as exc:
raise ValueError(f"Unsupported player kind: {payload.get('kind')!r}") from exc
name = str(payload.get("name", "")).strip()
if not name:
raise ValueError("Each player spec must include a non-empty 'name'.")
return PlaySeatSpec(
name=name,
kind=kind,
provider=str(payload["provider"]).strip() if "provider" in payload else None,
model_id=str(payload["model_id"]).strip() if "model_id" in payload else None,
litellm_model=str(payload["litellm_model"]).strip() if "litellm_model" in payload else None,
temperature=float(payload.get("temperature", 0.0)),
max_tokens=int(payload.get("max_tokens", 400)),
timeout=float(payload.get("timeout", 60.0)),
prefer_json_mode=bool(payload.get("prefer_json_mode", True)),
max_attempts=int(payload.get("max_attempts", 3)),
initial_delay_seconds=float(payload.get("initial_delay_seconds", 0.25)),
backoff_multiplier=float(payload.get("backoff_multiplier", 2.0)),
extra_kwargs=dict(payload.get("extra_kwargs", {})),
)
def _build_seat_agent(
seat: PlaySeatSpec,
*,
input_fn: Callable[[str], str],
output: TextIO | None,
) -> DecisionAgent:
if seat.kind is PlaySeatKind.HUMAN:
return HumanDecisionAgent(seat_name=seat.name, input_fn=input_fn, output=output)
if seat.kind is PlaySeatKind.PASSIVE_BOT:
return PassiveBotAgent(seat_name=seat.name)
if seat.kind is PlaySeatKind.LITELLM:
if not seat.provider or not seat.model_id:
raise ValueError("LiteLLM seats require both 'provider' and 'model_id'.")
litellm_model = seat.litellm_model or _default_litellm_model_for_seat(seat)
return LiteLLMRuntimeAgent(
provider=seat.provider,
config=LiteLLMConfig(
model=litellm_model,
temperature=seat.temperature,
max_tokens=seat.max_tokens,
timeout=seat.timeout,
prefer_json_mode=seat.prefer_json_mode,
extra_kwargs=seat.extra_kwargs,
),
retry_policy=RetryPolicy(
max_attempts=seat.max_attempts,
initial_delay_seconds=seat.initial_delay_seconds,
backoff_multiplier=seat.backoff_multiplier,
),
)
raise ValueError(f"Unsupported seat kind: {seat.kind!r}")
def _parse_pool_decision(text: str) -> WinnerPoolDecision:
normalized = text.strip().lower()
if normalized in {WinnerPoolDecision.RESET.value, WinnerPoolDecision.CONTINUE.value}:
return WinnerPoolDecision(normalized)
raise ValueError("la partie après ';' doit être reset ou continue")
def _zero_usage() -> dict[str, Any]:
return {
"prompt_tokens": 0,
"completion_tokens": 0,
"total_tokens": 0,
"cached_tokens": 0,
"cache_creation_input_tokens": 0,
"cache_read_input_tokens": 0,
"estimated_cost": 0.0,
}
def _default_litellm_model_for_seat(seat: PlaySeatSpec) -> str:
assert seat.provider is not None
assert seat.model_id is not None
provider = seat.provider.strip().lower()
if provider == "openai":
return f"openai/{seat.model_id}"
if provider == "xai":
return f"xai/{seat.model_id}"
if provider == "deepseek":
return f"deepseek/{seat.model_id}"
if provider in {"gemini", "google"}:
raise ValueError(
"Gemini seats require an explicit 'litellm_model'. "
"Example: 'gemini/gemini-3.1-pro-preview' or another provider-specific route you have validated."
)
if provider == "qwen":
raise ValueError(
"Qwen seats require an explicit 'litellm_model' and typically provider-specific API base configuration."
)
return seat.model_id
|