inference.py

import json
import os
import sys
import time
import textwrap
from typing import List, Optional
from openai import OpenAI
import httpx

# ---------------------------------------------------------------------------
# Config (MANDATORY per Checklist)
# ---------------------------------------------------------------------------
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 = os.getenv("HF_TOKEN") or os.getenv("API_KEY")

# Environment Server URL
ENV_URL = os.getenv("ENV_URL", "http://localhost:7860")

if not API_KEY:
    # We print and exit to avoid unhandled exceptions later
    print("ERROR: HF_TOKEN or API_KEY environment variable is required", flush=True)
    sys.exit(1)

# OpenAI Client configured via environment variables
client = OpenAI(api_key=API_KEY, base_url=API_BASE_URL)

# ---------------------------------------------------------------------------
# Logging Utilities
# ---------------------------------------------------------------------------
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()
    print(f"[STEP] step={step} action={action} reward={reward:.2f} 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} score={score:.3f} rewards={rewards_str}", flush=True)

# ---------------------------------------------------------------------------
# Agent Logic
# ---------------------------------------------------------------------------
SYSTEM_PROMPT = """You are an Eco-Resilient Logistics Agent. 
Your goal is to fulfill orders while minimizing CO2.
Available Actions: {"action_type": "order_parts | produce | offset | skip", "part_type": "chips | sensors | batteries | casing", "quantity": count, "mode": "sea | air | rail | road", "product": "EcoPhone | GreenTab"}
Respond ONLY with a valid JSON object."""

def get_action(obs) -> dict:
    prompt = f"Current Observation: {json.dumps(obs)}\nChoose next action:"
    try:
        response = client.chat.completions.create(
            model=MODEL_NAME,
            messages=[
                {"role": "system", "content": SYSTEM_PROMPT},
                {"role": "user", "content": prompt}
            ],
            response_format={"type": "json_object"},
            timeout=15.0
        )
        return json.loads(response.choices[0].message.content)
    except Exception as e:
        # Emergency fallback to prevent script crash
        print(f"[DEBUG] Model error: {e}", file=sys.stderr)
        return {"action_type": "skip"}

# ---------------------------------------------------------------------------
# Runner
# ---------------------------------------------------------------------------
def run_task(task_name: str):
    success = False
    score = 0.01  # Initialize to valid strictly-positive value
    steps_taken = 0
    rewards = []
    
    log_start(task=task_name, env="netzero-nav", model=MODEL_NAME)
    
    try:
        with httpx.Client(base_url=ENV_URL, timeout=30.0) as app:
            # Reset environment
            resp = app.post("/reset", json={"task": task_name})
            obs = resp.json()
            
            done = False
            while not done and steps_taken < 50:
                steps_taken += 1
                action_json = get_action(obs)
                
                # Take step
                resp = app.post("/step", json=action_json).json()
                
                obs = resp["observation"]
                reward = float(resp["reward"] or 0.0)
                done = resp["done"]
                info = resp.get("info", {})
                error = info.get("error")
                
                rewards.append(reward)
                
                # Format action for logs
                act_type = action_json.get("action_type", "skip")
                act_part = action_json.get("part_type", "")
                act_str = f"{act_type}-{act_part}" if act_part else act_type
                
                log_step(step=steps_taken, action=act_str, reward=reward, done=done, error=error)
                
                if done:
                    score = float(info.get("final_score", 0.01))
                    # Clamp strictly between 0 and 1
                    score = max(0.01, min(0.99, score))
                    success = score >= 0.95
    
    except Exception as e:
        print(f"[DEBUG] Runtime error during task {task_name}: {e}", file=sys.stderr)
    
    finally:
        # Final safety clamp before mandatory STDOUT
        score = max(0.01, min(0.99, score))
        log_end(success=success, steps=steps_taken, score=score, rewards=rewards)

if __name__ == "__main__":
    for task in ["easy", "medium", "hard"]:
        run_task(task)