DarkyCodez/precision-ag-env

1import os2import requests3from openai import OpenAI4 5# Required Environment Variables6API_BASE_URL = os.getenv("API_BASE_URL", "https://router.huggingface.co/v1")7MODEL_NAME = os.getenv("MODEL_NAME", "Qwen/Qwen2.5-72B-Instruct")8API_KEY = os.getenv("HF_TOKEN", "dummy-token")9ENV_URL = "http://0.0.0.0:7860"10 11TASKS = ["task_thirsty_crop", "task_nutrient_balance", "task_heatwave_crisis"]12 13client = OpenAI(base_url=API_BASE_URL, api_key=API_KEY)14 15SYSTEM_PROMPT = """You are an agricultural AI managing a greenhouse.16Observations include soil moisture, nitrogen levels, and crop health.17Choose one action per turn:180: Do nothing191: Irrigate202: Fertilize213: Harvest22Reply ONLY with a single integer (0, 1, 2, or 3)."""23 24def main():25    for task in TASKS:26        print(f"[START] task={task} env=precision-resilience-env model={MODEL_NAME}", flush=True)27 28        try:29            res = requests.post(f"{ENV_URL}/reset", json={"task_id": task}).json()30            obs = res.get("observation", {})31            32            rewards = []33            step = 134            done = False35 36            while not done and step <= 15:37                user_prompt = f"Observation: {obs}\nAction (0-3):"38                try:39                    completion = client.chat.completions.create(40                        model=MODEL_NAME,41                        messages=[42                            {"role": "system", "content": SYSTEM_PROMPT},43                            {"role": "user", "content": user_prompt}44                        ],45                        temperature=0.1,46                        max_tokens=547                    )48                    action_str = completion.choices[0].message.content.strip()49                    action_int = int(action_str)50                except Exception:51                    action_int = 052 53                step_res = requests.post(f"{ENV_URL}/step", json={"action": action_int}).json()54                obs = step_res.get("observation", {})55                reward = float(step_res.get("reward", 0.0))56                done = step_res.get("done", False)57 58                rewards.append(reward)59 60                print(f"[STEP] step={step} action={action_int} reward={reward:.2f} done={str(done).lower()} error=null", flush=True)61                step += 162 63            total_score = sum(rewards) / len(rewards) if rewards else 0.064            total_score = min(max(total_score, 0.0), 1.0)65            success = total_score > 0.566            rewards_str = ",".join(f"{r:.2f}" for r in rewards)67 68            print(f"[END] success={str(success).lower()} steps={step-1} score={total_score:.3f} rewards={rewards_str}", flush=True)69 70        except Exception as e:71            print(f"[END] success=false steps=0 score=0.00 rewards=0.00", flush=True)72 73if __name__ == "__main__":74    main()