Files changed (1) hide show
  1. inference.py +9 -7
inference.py CHANGED
@@ -18,7 +18,6 @@ SPACE_URL = "https://arrowman123-customer-supp-env.hf.space"
18
  BENCHMARK = "openenv"
19
  MAX_STEPS = 20
20
 
21
- # Task names must match openenv.yaml and cycle easy → medium → hard
22
  TASK_NAMES = ["easy", "medium", "hard"]
23
 
24
 
@@ -66,25 +65,25 @@ def log_step(step, action, reward, done, error=None):
66
  flush=True
67
  )
68
 
69
- def log_end(success, steps, rewards):
70
  success_str = "true" if success else "false"
71
  rewards_str = ",".join(f"{r:.2f}" for r in rewards)
72
- print(f"[END] success={success_str} steps={steps} rewards={rewards_str}", flush=True)
73
 
74
 
75
  # MAIN
76
  async def main():
77
- # HF_TOKEN is used as the API key per guidelines
78
  client = OpenAI(base_url=API_BASE_URL, api_key=API_KEY)
79
 
80
  for task_name in TASK_NAMES:
81
  step_rewards = []
 
82
  success = False
83
 
84
  log_start(task=task_name, env=BENCHMARK, model=MODEL_NAME)
85
 
86
  try:
87
- res = safe_post(f"{SPACE_URL}/reset")
88
  result = res.json()
89
 
90
  state = result["observation"]["echoed_message"]
@@ -107,13 +106,16 @@ async def main():
107
 
108
  log_step(step=step, action=action, reward=reward, done=done, error=None)
109
 
110
- success = True
 
 
111
 
112
  except Exception as e:
113
  print(f"[ERROR] task={task_name} {e}", flush=True)
 
114
 
115
  finally:
116
- log_end(success=success, steps=len(step_rewards), rewards=step_rewards)
117
 
118
 
119
  if __name__ == "__main__":
 
18
  BENCHMARK = "openenv"
19
  MAX_STEPS = 20
20
 
 
21
  TASK_NAMES = ["easy", "medium", "hard"]
22
 
23
 
 
65
  flush=True
66
  )
67
 
68
+ def log_end(success, steps, score, rewards):
69
  success_str = "true" if success else "false"
70
  rewards_str = ",".join(f"{r:.2f}" for r in rewards)
71
+ print(f"[END] success={success_str} steps={steps} score={score:.4f} rewards={rewards_str}", flush=True)
72
 
73
 
74
  # MAIN
75
  async def main():
 
76
  client = OpenAI(base_url=API_BASE_URL, api_key=API_KEY)
77
 
78
  for task_name in TASK_NAMES:
79
  step_rewards = []
80
+ score = 0.5
81
  success = False
82
 
83
  log_start(task=task_name, env=BENCHMARK, model=MODEL_NAME)
84
 
85
  try:
86
+ res = safe_post(f"{SPACE_URL}/reset", json={"task": task_name})
87
  result = res.json()
88
 
89
  state = result["observation"]["echoed_message"]
 
106
 
107
  log_step(step=step, action=action, reward=reward, done=done, error=None)
108
 
109
+ raw_score = float(result.get("info", {}).get("score", 0.5))
110
+ score = max(0.05, min(0.95, raw_score))
111
+ success = True
112
 
113
  except Exception as e:
114
  print(f"[ERROR] task={task_name} {e}", flush=True)
115
+ score = 0.1
116
 
117
  finally:
118
+ log_end(success=success, steps=len(step_rewards), score=score, rewards=step_rewards)
119
 
120
 
121
  if __name__ == "__main__":