Spaces:
Sleeping
Sleeping
File size: 16,723 Bytes
f9cf02d cfe0896 dda0d73 f9cf02d cfe0896 54e0639 cfe0896 75ca235 339abf5 cfe0896 711b9b5 cfe0896 711b9b5 cfe0896 f9cf02d 54e0639 cfe0896 54e0639 cfe0896 dda0d73 75ca235 cfe0896 75ca235 cfe0896 75ca235 cfe0896 f9cf02d cfe0896 339abf5 cfe0896 339abf5 cfe0896 54e0639 cfe0896 54e0639 cfe0896 54e0639 cfe0896 54e0639 cfe0896 54e0639 cfe0896 54e0639 cfe0896 54e0639 75ca235 cfe0896 dda0d73 cfe0896 f9cf02d cfe0896 54e0639 cfe0896 54e0639 f9cf02d cfe0896 75ca235 cfe0896 339abf5 cfe0896 75ca235 cfe0896 dda0d73 cfe0896 f9cf02d 339abf5 dda0d73 cfe0896 75ca235 cfe0896 54e0639 75ca235 54e0639 75ca235 54e0639 75ca235 54e0639 75ca235 54e0639 cfe0896 | 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 | """
Submission inference entrypoint.
Mandatory environment variables:
- API_BASE_URL
- MODEL_NAME
- HF_TOKEN
Stdout contract:
- [START] task=<task_name> env=<benchmark> model=<model_name>
- [STEP] step=<n> action=<action_str> reward=<0.00> done=<true|false> error=<msg|null>
- [END] success=<true|false> steps=<n> score=<0.xxxx> rewards=<r1,r2,...,rn>
All deterministic heuristics live in ``server/heuristics.py`` and are
shared with the FastAPI server so that changes remain in one place.
"""
from __future__ import annotations
import json
import os
import re
import sys
from collections import defaultdict
from pathlib import Path
from openai import OpenAI
# Make package importable when run from repo root.
PROJECT_ROOT = Path(__file__).resolve().parent
if str(PROJECT_ROOT) not in sys.path:
sys.path.insert(0, str(PROJECT_ROOT))
from models import ActionModel
from server.environment import BugTriageEnv
from server.graders import BugTriageGrader
from server.policy import recommend_action
def _load_simple_env_file(dotenv_path: Path) -> None:
"""Load simple KEY=VALUE lines without failing on stray shell commands."""
if not dotenv_path.exists():
return
for raw_line in dotenv_path.read_text(encoding="utf-8").splitlines():
line = raw_line.strip()
if not line or line.startswith("#") or "=" not in line:
continue
key, value = line.split("=", 1)
key = key.strip()
value = value.strip().strip('"').strip("'")
if key:
os.environ.setdefault(key, value)
_load_simple_env_file(PROJECT_ROOT / ".env")
HF_TOKEN = os.getenv("HF_TOKEN")
OPENAI_API_KEY = os.getenv("OPENAI_API_KEY")
API_BASE_URL = os.getenv("API_BASE_URL", "https://router.huggingface.co/v1")
MODEL_NAME = os.getenv("MODEL_NAME", "Qwen/Qwen2.5-72B-Instruct")
API_KEY = HF_TOKEN or OPENAI_API_KEY
BENCHMARK = os.getenv("OPENENV_BENCHMARK", "bug-triage-openenv")
TASKS = [
t.strip()
for t in os.getenv(
"OPENENV_TASKS",
"bug_triage_easy,bug_triage_medium,bug_triage_hard",
).split(",")
if t.strip()
]
SEED = int(os.getenv("OPENENV_SEED", "42"))
MAX_STEPS_PER_TICKET = int(os.getenv("MAX_STEPS_PER_TICKET", "4"))
MAX_STEPS = int(os.getenv("MAX_STEPS", "200"))
TEMPERATURE = float(os.getenv("TEMPERATURE", "0"))
MAX_TOKENS = int(os.getenv("MAX_TOKENS", "220"))
# COMPONENT_TEAM_MAP, COMPONENT_KEYWORDS, SERVICE_COMPONENT_HINTS and all
# heuristic inference functions are imported from server.heuristics so they
# are shared with the FastAPI server without duplication.
class ActionParseError(ValueError):
"""Raised when a model response cannot be converted into a valid action."""
def _b(value: bool) -> str:
return "true" if value else "false"
def _as_bool(value: str | None) -> bool:
if value is None:
return False
return value.strip().lower() in {"1", "true", "yes", "on"}
def _sanitize(text: str) -> str:
return " ".join(str(text).replace("\n", " ").replace("\r", " ").split())
def _strict_unit_interval(value: float, eps: float = 1e-6) -> float:
"""Clamp numeric scores to the strict open interval (0, 1)."""
return max(eps, min(1.0 - eps, float(value)))
def _emit(line: str) -> None:
"""Emit one stdout log line immediately."""
print(line, flush=True)
def _action_to_log(action: ActionModel) -> str:
if _as_bool(os.getenv("OPENENV_LOG_ACTION_JSON")):
payload = action.model_dump(exclude_none=True)
return json.dumps(payload, separators=(",", ":"))
def _value(raw: object, max_len: int = 48) -> str:
text = _sanitize(str(raw)).replace(" ", "_")
if len(text) > max_len:
return f"{text[:max_len - 3]}..."
return text
if action.action_type == "classify" and action.classify:
return (
"classify("
f"sev={action.classify.severity},"
f"pri={action.classify.priority},"
f"comp={_value(action.classify.component)}"
")"
)
if action.action_type == "assign" and action.assign:
return f"assign(team={_value(action.assign.team)})"
if action.action_type == "mark_duplicate" and action.mark_duplicate:
return f"mark_duplicate(canonical={_value(action.mark_duplicate.canonical_ticket_id)})"
if action.action_type == "request_info" and action.request_info:
return f"request_info(type={_value(action.request_info.info_type)})"
if action.action_type == "defer" and action.defer:
return f"defer(reason={_value(action.defer.reason)})"
if action.action_type == "close" and action.close:
return f"close(reason={_value(action.close.reason)})"
if action.action_type == "escalate_incident" and action.escalate_incident:
return f"escalate_incident(why={_value(action.escalate_incident.justification)})"
if action.action_type == "next_ticket":
return "next_ticket"
payload = action.model_dump(exclude_none=True)
return f"raw:{json.dumps(payload, separators=(',', ':'))}"
def _fallback_action(observation, plans: dict[str, dict]) -> ActionModel:
"""Deterministic fallback policy used when the LLM is unavailable."""
ticket = observation.current_ticket
if ticket is None:
return ActionModel(action_type="next_ticket", next_ticket={})
history = plans.setdefault(ticket.ticket_id, {}).setdefault("history", [])
action, _ = recommend_action(observation=observation, action_history=list(history))
return action
def _guard_action(
action: ActionModel,
observation,
action_history_by_ticket: dict[str, list[str]],
steps_by_ticket: dict[str, int],
) -> ActionModel:
ticket = observation.current_ticket
if ticket is None:
return action
ticket_id = ticket.ticket_id
history = action_history_by_ticket[ticket_id]
ticket_steps = steps_by_ticket[ticket_id]
if ticket_steps >= MAX_STEPS_PER_TICKET and action.action_type != "next_ticket":
return ActionModel(action_type="next_ticket", next_ticket={})
if action.action_type == "request_info" and "request_info" in history:
return ActionModel(action_type="next_ticket", next_ticket={})
if action.action_type == "classify" and "classify" in history:
return ActionModel(action_type="next_ticket", next_ticket={})
if action.action_type == "assign" and "assign" in history:
return ActionModel(action_type="next_ticket", next_ticket={})
if action.action_type == "mark_duplicate" and "mark_duplicate" in history:
return ActionModel(action_type="next_ticket", next_ticket={})
if action.action_type == "escalate_incident" and "escalate_incident" in history:
return ActionModel(action_type="next_ticket", next_ticket={})
if len(history) >= 2 and history[-1] == history[-2] == action.action_type and action.action_type != "next_ticket":
return ActionModel(action_type="next_ticket", next_ticket={})
return action
def _build_prompt(observation) -> str:
ticket = observation.current_ticket
if ticket is None:
return '{"action_type":"next_ticket","next_ticket":{}}'
return f"""Return ONLY JSON for the next bug-triage action.
Ticket ID: {ticket.ticket_id}
Title: {ticket.title}
Description: {ticket.description}
Reporter: {ticket.reporter_type}
Service: {ticket.service}
Tier: {ticket.customer_tier}
Repro Steps Present: {ticket.repro_steps_present}
Logs Present: {ticket.logs_present}
Suspected Duplicates: {ticket.suspected_duplicate_ids}
Last Result: {observation.last_action_result}
Available Teams: {observation.available_teams}
Available Components: {observation.available_components}
Allowed action_type values:
classify, assign, mark_duplicate, request_info, defer, close, escalate_incident, next_ticket
Rules:
- Do not repeat request_info on the same ticket.
- Avoid loops. If uncertain, use classify or next_ticket.
- Do not include markdown fences, analysis, or <think> tags.
- Output exactly one valid JSON object only.
"""
def _message_to_text(content: object) -> str:
if isinstance(content, str):
return content
if isinstance(content, list):
chunks: list[str] = []
for item in content:
if isinstance(item, dict) and item.get("type") == "text":
chunks.append(str(item.get("text", "")))
continue
text_value = getattr(item, "text", None)
if text_value:
chunks.append(str(text_value))
return "".join(chunks)
return "" if content is None else str(content)
def _extract_json_objects(text: str) -> list[str]:
objects: list[str] = []
start: int | None = None
depth = 0
in_string = False
escaped = False
for index, char in enumerate(text):
if start is None:
if char == "{":
start = index
depth = 1
in_string = False
escaped = False
continue
if in_string:
if escaped:
escaped = False
elif char == "\\":
escaped = True
elif char == '"':
in_string = False
continue
if char == '"':
in_string = True
elif char == "{":
depth += 1
elif char == "}":
depth -= 1
if depth == 0:
objects.append(text[start:index + 1])
start = None
return objects
def _parse_action(raw: str) -> ActionModel:
text = re.sub(r"<think>.*?</think>", " ", raw, flags=re.IGNORECASE | re.DOTALL).strip()
candidates: list[str] = [text]
if "```json" in text:
start = text.find("```json") + 7
end = text.find("```", start)
if end != -1:
candidates.append(text[start:end].strip())
elif "```" in text:
start = text.find("```") + 3
end = text.find("```", start)
if end != -1:
candidates.append(text[start:end].strip())
candidates.extend(_extract_json_objects(text))
seen: set[str] = set()
for candidate in candidates:
candidate = candidate.strip()
if not candidate or candidate in seen:
continue
seen.add(candidate)
try:
data = json.loads(candidate)
return ActionModel(**data)
except (json.JSONDecodeError, TypeError, ValueError):
continue
raise ActionParseError(f"Could not parse model action from response: {_sanitize(text[:200])}")
def _request_model_action(client: OpenAI, observation) -> ActionModel:
response = client.chat.completions.create(
model=MODEL_NAME,
messages=[
{
"role": "system",
"content": "You are an expert bug triage assistant. Return one JSON object only.",
},
{"role": "user", "content": _build_prompt(observation)},
],
temperature=TEMPERATURE,
max_tokens=MAX_TOKENS,
)
raw = _message_to_text(response.choices[0].message.content)
return _parse_action(raw)
def _run_task(task_id: str, env: BugTriageEnv, client: OpenAI | None) -> None:
_emit(f"[START] task={task_id} env={BENCHMARK} model={MODEL_NAME}")
step_no = 0
rewards: list[str] = []
success = False
score = 0.5
api_disabled = False
plans: dict[str, dict] = {}
action_history_by_ticket: dict[str, list[str]] = defaultdict(list)
steps_by_ticket: dict[str, int] = defaultdict(int)
done = False
episode_actions: list[dict] = []
info: dict = {"metrics": {}}
try:
obs = env.reset(task_id=task_id, seed=SEED)
while not done and step_no < MAX_STEPS:
step_no += 1
current_ticket_id = obs.current_ticket.ticket_id if obs.current_ticket else None
if client is not None and not api_disabled:
try:
action = _request_model_action(client, obs)
except ActionParseError:
action = _fallback_action(obs, plans)
except Exception as exc: # noqa: BLE001
print(
f"[WARN] Model API error; switching to offline fallback: {_sanitize(str(exc))}",
file=sys.stderr,
flush=True,
)
api_disabled = True
action = _fallback_action(obs, plans)
else:
action = _fallback_action(obs, plans)
action = _guard_action(action, obs, action_history_by_ticket, steps_by_ticket)
err_value = "null"
try:
obs, reward, done, info = env.step(action)
reward_value = f"{reward.step_reward:.2f}"
rewards.append(reward_value)
last_action_error = info.get("last_action_error") if isinstance(info, dict) else None
validation_error = info.get("validation_error") if isinstance(info, dict) else None
error_raw = last_action_error if last_action_error else validation_error
if error_raw:
err_value = _sanitize(error_raw)
_emit(
f"[STEP] step={step_no:03d} action={_action_to_log(action)} "
f"reward={reward_value} done={_b(bool(done))} error={err_value}"
)
episode_actions.append(action.model_dump(exclude_none=True))
if current_ticket_id:
action_history_by_ticket[current_ticket_id].append(action.action_type)
steps_by_ticket[current_ticket_id] += 1
plans.setdefault(current_ticket_id, {}).setdefault("history", []).append(action.action_type)
except Exception as exc:
err_value = _sanitize(str(exc))
_emit(
f"[STEP] step={step_no:03d} action={_action_to_log(action)} "
f"reward=0.00 done=true error={err_value}"
)
rewards.append("0.00")
done = True
try:
grader = BugTriageGrader(task_id=task_id)
ground_truths = [
gt.model_dump() for gt in env.current_task.ground_truths
] if env.current_task else []
grader_result = grader.grade_episode(
episode_actions=[{"action": a} for a in episode_actions],
ground_truths=ground_truths,
metrics=info.get("metrics", {}) if isinstance(info, dict) else {},
)
score = _strict_unit_interval(grader_result.score)
success = bool(grader_result.passed)
except Exception as exc: # noqa: BLE001
print(
f"[WARN] Grader failed for task '{task_id}'; falling back to reward average: "
f"{_sanitize(str(exc))}",
file=sys.stderr,
flush=True,
)
success = False
parsed_rewards: list[float] = []
for reward_text in rewards:
try:
parsed_rewards.append(float(reward_text))
except ValueError:
continue
avg_reward = (sum(parsed_rewards) / len(parsed_rewards)) if parsed_rewards else 0.5
score = _strict_unit_interval(avg_reward)
finally:
if hasattr(env, "close"):
try:
env.close()
except Exception:
pass
rewards_csv = ",".join(rewards)
_emit(
f"[END] success={_b(success)} steps={step_no:03d} "
f"score={score:.6f} rewards={rewards_csv}"
)
def main() -> int:
offline_mode = _as_bool(os.getenv("OPENENV_OFFLINE"))
client: OpenAI | None = None
if not offline_mode and not API_KEY:
print(
"No HF_TOKEN or OPENAI_API_KEY found. Falling back to offline mode.",
file=sys.stderr,
flush=True,
)
offline_mode = True
if not offline_mode:
try:
client = OpenAI(api_key=API_KEY, base_url=API_BASE_URL, max_retries=0, timeout=30)
except Exception as exc:
print(
f"Failed to initialize OpenAI client ({_sanitize(exc)}). "
"Falling back to offline mode.",
file=sys.stderr,
flush=True,
)
client = None
env = BugTriageEnv()
for task_id in TASKS:
_run_task(task_id=task_id, env=env, client=client)
return 0
if __name__ == "__main__":
raise SystemExit(main())
|