Spaces:
No application file
No application file
File size: 4,214 Bytes
b618214 | 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 | import asyncio
import os
import json
from openai import OpenAI
from my_env_v4 import MyEnvV4Env, MyEnvV4Action
API_KEY = os.getenv("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")
API_BASE_URL = os.getenv("API_BASE_URL", "https://router.huggingface.co/v1")
MODEL_NAME = os.getenv("MODEL_NAME", "Qwen/Qwen2.5-7B-Instruct")
TASK_NAME = "triage"
BENCHMARK = "healthcare_env"
client = OpenAI(api_key=API_KEY, base_url=API_BASE_URL)
# ================= RULE-BASED OVERRIDE =================
HIGH_RISK_KEYWORDS = [
"chest pain", "breathing", "unconscious",
"bleeding", "fainting", "severe", "pressure"
]
def rule_override(text):
t = text.lower()
for k in HIGH_RISK_KEYWORDS:
if k in t:
return MyEnvV4Action(
action_type="finalize",
urgency="high",
department="emergency",
escalate=True
)
return None
# ================= FEW-SHOT PROMPT =================
def build_prompt(obs):
return f"""
You are a medical triage assistant.
Determine:
- urgency: low / medium / high
- department: general / dermatology / cardiology / emergency
- escalate: true / false
Examples:
Patient: mild headache since morning
Answer:
{{"action_type":"finalize","urgency":"low","department":"general","escalate":false}}
Patient: small rash and itching
Answer:
{{"action_type":"finalize","urgency":"low","department":"dermatology","escalate":false}}
Patient: chest pain with sweating
Answer:
{{"action_type":"finalize","urgency":"high","department":"emergency","escalate":true}}
Patient: occasional chest discomfort
Answer:
{{"action_type":"finalize","urgency":"medium","department":"cardiology","escalate":false}}
Patient: severe breathing problem
Answer:
{{"action_type":"finalize","urgency":"high","department":"emergency","escalate":true}}
Rules:
- High = life-threatening
- Medium = persistent symptoms
- Low = mild symptoms
Now classify:
Patient: {obs.current_query.message}
Return ONLY JSON:
"""
# ================= SAFE PARSER =================
def safe_parse(text):
try:
text = text.strip()
if text.startswith("```"):
text = text.split("```")[1]
data = json.loads(text)
return MyEnvV4Action(
action_type="finalize",
urgency=data.get("urgency", "medium"),
department=data.get("department", "general"),
escalate=data.get("escalate", False)
)
except:
return MyEnvV4Action(
action_type="finalize",
urgency="medium",
department="general",
escalate=False
)
# ================= MAIN =================
async def main():
env = MyEnvV4Env()
obs = env.reset()
rewards = []
step = 0
done = False
last_error = None
print(f"[START] task={TASK_NAME} env={BENCHMARK} model={MODEL_NAME}")
while not done and step < 20:
step += 1
try:
# 🔥 RULE-BASED SHORTCUT
action = rule_override(obs.current_query.message)
# 🤖 If no rule → call API
if not action:
res = client.chat.completions.create(
model=MODEL_NAME,
messages=[{"role": "user", "content": build_prompt(obs)}],
temperature=0.0,
max_tokens=120
)
action = safe_parse(res.choices[0].message.content)
obs, reward, done, info = env.step(action)
last_error = info.get("error", None)
except Exception as e:
action = "error"
reward = 0.0
done = True
last_error = str(e)
rewards.append(reward)
print(f"[STEP] step={step} action={action} reward={reward:.2f} done={str(done).lower()} error={last_error if last_error else 'null'}")
env.close()
score = env.normalized_score()
success = score >= 0.5
reward_str = ",".join([f"{r:.2f}" for r in rewards])
print(f"[END] success={str(success).lower()} steps={step} score={score:.2f} rewards={reward_str}")
# ================= ENTRY =================
if __name__ == "__main__":
asyncio.run(main())
|