| import asyncio |
| import json |
| import os |
| import sys |
| import re |
| from pathlib import Path |
| from typing import List, Optional |
| import httpx |
| from dotenv import load_dotenv |
| from openai import OpenAI |
|
|
| ROOT = Path(__file__).resolve().parent |
| REPO_ROOT = ROOT.parent |
|
|
| if str(REPO_ROOT) not in sys.path: |
| sys.path.insert(0, str(REPO_ROOT)) |
|
|
| load_dotenv() |
|
|
| TASK_DIR = ROOT / "tasks" |
| SYSTEM_PROMPT = (ROOT / "system_prompt.txt").read_text().strip() |
|
|
| HF_TOKEN = os.getenv("HF_TOKEN") or os.getenv("API_KEY") |
| MODEL_NAME = os.getenv("MODEL_NAME", "gpt-4o") |
|
|
| BASE_URL = "https://vishvam10-cloudenv.hf.space" |
|
|
| TASK_ORDER = ["easy-task", "medium-task", "hard-task"] |
|
|
| MAX_PARSE_RETRIES = 3 |
| MAX_STEPS = 5 |
|
|
| client = OpenAI(base_url="https://api.openai.com/v1", api_key=HF_TOKEN) |
|
|
|
|
| def log_start(task: str, env: str, model: str) -> None: |
| print(f"[START] task={task} env={env} model={model}", flush=True) |
|
|
|
|
| def log_step( |
| step: int, action: str, reward: float, done: bool, error: Optional[str] |
| ) -> None: |
| error_val = error if error else "null" |
| done_val = str(done).lower() |
| action_safe = action.replace("\n", " ").replace("\r", "")[:120] |
| print( |
| f"[STEP] step={step} action={action_safe} reward={reward:.2f} " |
| f"done={done_val} error={error_val}", |
| flush=True, |
| ) |
|
|
|
|
| def log_end( |
| success: bool, steps: int, score: float, rewards: List[float] |
| ) -> None: |
| rewards_str = ",".join(f"{r:.2f}" for r in rewards) |
| print( |
| f"[END] success={str(success).lower()} steps={steps} " |
| f"score={score:.3f} rewards={rewards_str}", |
| flush=True, |
| ) |
|
|
|
|
| def load_task(name: str): |
| return json.loads((TASK_DIR / f"{name}.json").read_text()) |
|
|
|
|
| async def register_pipeline(task: dict): |
| async with httpx.AsyncClient(timeout=30) as http: |
| r = await http.post( |
| f"{BASE_URL}/api/pipelines", |
| json={ |
| "pipeline_id": task["pipeline_id"], |
| "broken_pipeline": task["broken_pipeline"], |
| "ideal_pipeline": task["ideal_pipeline"], |
| }, |
| ) |
|
|
| if r.status_code not in (200, 201, 400): |
| print( |
| f"[PIPELINE REGISTER ERROR] {r.status_code} {r.text}", |
| flush=True, |
| ) |
|
|
|
|
| async def env_reset(task_id: str, episode_id: str): |
| async with httpx.AsyncClient(timeout=30) as http: |
| r = await http.post( |
| f"{BASE_URL}/reset", |
| json={"task_id": task_id, "episode_id": episode_id}, |
| ) |
| r.raise_for_status() |
| return r.json() |
|
|
|
|
| async def env_step(action: dict): |
| async with httpx.AsyncClient(timeout=30) as http: |
| r = await http.post( |
| f"{BASE_URL}/step", |
| json={"action": action}, |
| ) |
| r.raise_for_status() |
| return r.json() |
|
|
|
|
| def clean_json(raw: str) -> str: |
| raw = raw.strip() |
| raw = re.sub(r"^```(?:json)?\s*", "", raw) |
| raw = re.sub(r"\s*```$", "", raw) |
|
|
| match = re.search(r"\{.*\}", raw, re.DOTALL) |
| if match: |
| return match.group(0) |
|
|
| return raw |
|
|
|
|
| def build_action(raw: str, task_id: str, episode_id: str): |
| try: |
| raw = clean_json(raw) |
| data = json.loads(raw) |
|
|
| if isinstance(data, dict) and "action" in data: |
| data = data["action"] |
|
|
| if not isinstance(data, dict): |
| return None |
|
|
| service = data.get("service") |
| operation = data.get("operation") |
|
|
| if not service or not operation: |
| return None |
|
|
| instance_id = data.get("instance_id") |
|
|
| if isinstance(service, str) and ":" in service: |
| service, parsed_instance = service.split(":", 1) |
| instance_id = instance_id or parsed_instance |
|
|
| return { |
| "task_id": task_id, |
| "episode_id": episode_id, |
| "service": service.lower(), |
| "instance_id": instance_id, |
| "operation": operation.strip(), |
| "payload": data.get("payload", {}) |
| if isinstance(data.get("payload"), dict) |
| else {}, |
| "principal_arn": data.get("principal_arn"), |
| "intent": data.get("intent"), |
| } |
|
|
| except Exception: |
| return None |
|
|
|
|
| def call_model(step, state, broken, ideal, history): |
| prompt = f""" |
| Return ONLY valid JSON. |
| |
| STEP {step} |
| STATE {json.dumps(state)[:1000]} |
| BROKEN {json.dumps(broken)[:1000]} |
| IDEAL {json.dumps(ideal)[:1000]} |
| HISTORY {history[-5:]} |
| """ |
|
|
| res = client.chat.completions.create( |
| model=MODEL_NAME, |
| messages=[ |
| {"role": "system", "content": SYSTEM_PROMPT}, |
| {"role": "user", "content": prompt}, |
| ], |
| temperature=0.3, |
| max_tokens=500, |
| ) |
|
|
| return (res.choices[0].message.content or "").strip() |
|
|
|
|
| async def run_episode(task): |
| task_id = task["pipeline_id"] |
| episode_id = f"{task_id}-episode-1" |
|
|
| log_start(task=task_id, env="http", model=MODEL_NAME) |
|
|
| result = await env_reset(task_id, episode_id) |
|
|
| await env_step( |
| { |
| "task_id": task_id, |
| "episode_id": episode_id, |
| "service": "internal", |
| "instance_id": None, |
| "operation": "InitializeEnvironment", |
| "payload": {}, |
| } |
| ) |
|
|
| await register_pipeline(task) |
|
|
| obs = {} |
| history = [] |
| rewards = [] |
| step_count = 0 |
|
|
| for step in range(MAX_STEPS): |
| if obs.get("done"): |
| break |
|
|
| error = None |
|
|
| try: |
| raw = call_model( |
| step, |
| obs, |
| task["broken_pipeline"], |
| task["ideal_pipeline"], |
| history, |
| ) |
|
|
| action = build_action(raw, task_id, episode_id) |
|
|
| if action is None: |
| error = "invalid_action" |
| log_step(step, "INVALID_ACTION", 0.0, False, error) |
| continue |
|
|
| result = await env_step(action) |
| obs = result["observation"] |
|
|
| reward = float(obs.get("reward", 0.0)) |
| done = bool(obs.get("done", False)) |
|
|
| rewards.append(reward) |
| history.append(raw) |
| step_count += 1 |
|
|
| log_step( |
| step=step_count, |
| action=str(action), |
| reward=reward, |
| done=done, |
| error=None, |
| ) |
|
|
| except Exception as e: |
| error = str(e) |
|
|
| log_step(step, "ERROR", 0.0, False, error) |
| break |
|
|
| if obs.get("done"): |
| break |
|
|
| success = obs.get("done", False) |
| score = sum(rewards) / (len(rewards) or 1) |
|
|
| log_end(success=success, steps=step_count, score=score, rewards=rewards) |
|
|
|
|
| async def main(): |
| for name in TASK_ORDER: |
| task = load_task(name) |
| await run_episode(task) |
|
|
|
|
| if __name__ == "__main__": |
| asyncio.run(main()) |
|
|
