Spaces:
Sleeping
Sleeping
File size: 11,278 Bytes
3c7cc35 56257d2 3c7cc35 56257d2 3c7cc35 56257d2 3c7cc35 56257d2 3c7cc35 56257d2 b0406e4 3c7cc35 b0406e4 56257d2 66fc5d8 b8214ff 56257d2 66fc5d8 b8214ff 3c7cc35 66fc5d8 b8214ff 3c7cc35 56257d2 3c7cc35 56257d2 3c7cc35 56257d2 3c7cc35 56257d2 3c7cc35 56257d2 3c7cc35 56257d2 3c7cc35 56257d2 3c7cc35 56257d2 3c7cc35 56257d2 3c7cc35 56257d2 3c7cc35 56257d2 66fc5d8 56257d2 66fc5d8 56257d2 dd054aa 56257d2 66fc5d8 dd054aa 56257d2 66fc5d8 56257d2 dd054aa 56257d2 66fc5d8 56257d2 3c7cc35 56257d2 3c7cc35 56257d2 3c7cc35 56257d2 3c7cc35 66fc5d8 56257d2 3c7cc35 56257d2 3c7cc35 56257d2 3c7cc35 56257d2 3c7cc35 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 | #!/usr/bin/env python3
"""
Baseline Inference Script for GDPR Auditor Environment.
Uses OpenAI Client + HTTP calls to the server to run a model against the environment.
Usage:
python inference.py
Environment Variables:
API_BASE_URL: The API endpoint (default: https://router.huggingface.co/v1)
MODEL_NAME: The model identifier (default: Qwen/Qwen2.5-72B-Instruct)
HF_TOKEN: Your Hugging Face / API key (required)
SERVER_URL: The environment server URL (default: http://localhost:7860)
Expected format for STDOUT:
[START] task=<task_name> env=<benchmark> model=<model_name>
[STEP] step=<n> action=<action_str> reward=<0.00> done=<true|false> error=<msg|null>
[END] success=<true|false> steps=<n> score=<score> rewards=<r1,r2,...,rn>
"""
import json
import os
import re
import textwrap
import requests
from typing import List, Optional
from dotenv import load_dotenv
load_dotenv()
from openai import OpenAI
# Force reload from .env file to override system env vars
import os
_env_file = os.path.join(os.path.dirname(os.path.abspath(__file__)), '.env')
if os.path.exists(_env_file):
with open(_env_file) as f:
for line in f:
line = line.strip()
if line and '=' in line and not line.startswith('#'):
key, val = line.split('=', 1)
os.environ[key] = val
API_BASE_URL = os.environ.get("API_BASE_URL", "https://router.huggingface.co/v1")
MODEL_NAME = os.environ.get("MODEL_NAME", "Qwen/Qwen2.5-72B-Instruct")
HF_TOKEN = os.environ.get("HF_TOKEN")
SERVER_URL = os.environ.get("SERVER_URL", "http://localhost:7860")
MAX_STEPS = 8
MAX_TOKENS = 512
TEMPERATURE = 0.7
SUCCESS_SCORE_THRESHOLD = 0.5
BENCHMARK = "gdpr_auditor_env"
SYSTEM_PROMPT = textwrap.dedent(
"""
You are an expert GDPR Compliance Auditor. Your job is to review privacy policies
and data practices to identify compliance violations.
IMPORTANT: This is a multi-step auditing task. You must systematically review
each section of the policy and data practice to find ALL issues.
Task Types:
- Easy (Clause Existence): Check if mandatory GDPR clauses are present
- Medium (Purpose Mapping): Match data practices to stated purposes
- Hard (Dark Patterns): Find contradictions between policy statements and actual practices
Output Format:
Your finding should clearly identify:
- The issue type (missing clause, contradiction, etc.)
- The specific violation
- The severity (if mentioned)
Examples of findings:
- "Missing Right to be Forgotten clause"
- "Health data shared with advertisers - contradicts policy"
- "Section 1 says never shared but Section 3 lists third-party partners"
Output ONLY your finding, nothing else."""
).strip()
def log_start(task, env, model):
print(f"[START] task={task} env={env} model={model}", flush=True)
def log_step(step, action, reward, done, error):
done_str = str(done).lower()
error_str = "null" if error is None else str(error).replace('\n', ' ').replace('\r', '')
print(f"[STEP] step={step} action={action} reward={reward:.4f} done={done_str} error={error_str}", flush=True)
def log_end(success, steps, score, rewards):
success_str = str(success).lower()
rewards_str = ",".join(f"{r:.4f}" for r in rewards)
print(f"[END] success={success_str} steps={steps} score={score:.4f} rewards={rewards_str}", flush=True)
def reset_env(task: str) -> dict:
"""Reset the environment via HTTP."""
resp = requests.get(f"{SERVER_URL}/reset", params={"task": task})
resp.raise_for_status()
return resp.json()
def step_env(message: str) -> dict:
"""Send action to environment via HTTP."""
resp = requests.post(f"{SERVER_URL}/step", json={"message": message})
resp.raise_for_status()
return resp.json()
def build_user_prompt(obs_data: dict, step_num: int) -> str:
documents = obs_data.get("documents", [])
data_practices = obs_data.get("data_practices", [])
requirements = obs_data.get("compliance_requirements", [])
flagged = obs_data.get("flagged_issues", [])
policy_text = documents[0].get("content", "") if documents else "No policy found"
prompt = f"""Privacy Policy Audit (Step {step_num}):
--- POLICY CONTENT ---
{policy_text[:1500]}
--- DATA PRACTICES ---
"""
for dp in data_practices:
prompt += f"- {dp.get('category')}: {dp.get('purpose')} (shared: {dp.get('shared_with_third_parties')})\n"
prompt += f"""
--- COMPLIANCE REQUIREMENTS ---
"""
for req in requirements:
prompt += f"- {req}\n"
if flagged:
prompt += f"\n--- ALREADY FLAGGED ---\n"
for f in flagged:
prompt += f"- {f}\n"
prompt += f"""
Task: {obs_data.get('task_name')} ({obs_data.get('difficulty')})
Provide your next compliance finding:"""
return prompt
def call_model(client: OpenAI, user_prompt: str, history: List[dict]) -> str:
messages = [{"role": "system", "content": SYSTEM_PROMPT}]
messages.extend(history)
messages.append({"role": "user", "content": user_prompt})
try:
completion = client.chat.completions.create(
model=MODEL_NAME,
messages=messages,
temperature=TEMPERATURE,
max_tokens=MAX_TOKENS,
stream=False,
)
text = (completion.choices[0].message.content or "").strip()
lines = text.split('\n')
for line in lines:
line = line.strip()
if line and len(line) > 10:
return line
return text if text else "No finding submitted"
except Exception as exc:
print(f"[DEBUG] Model request failed: {exc}", flush=True)
return "No finding submitted"
TASKS = {
"easy": {"task_id": "easy_clause_existence", "name": "Clause Existence Check", "difficulty": "easy"},
"medium": {"task_id": "medium_purpose_mapping", "name": "Purpose Mapping", "difficulty": "medium"},
"hard": {"task_id": "hard_dark_patterns", "name": "Dark Pattern Detection", "difficulty": "hard"},
"elite": {"task_id": "elite_multi_doc_reasoning", "name": "Multi-Document Reasoning", "difficulty": "elite"},
}
def run_task(client: OpenAI, task_key: str, verbose: bool = False) -> dict:
"""Run inference on a single task via HTTP."""
task = TASKS[task_key]
task_name = task["name"]
history: List[dict] = []
rewards: List[float] = []
steps_taken = 0
score = 0.0
success = False
error_msg = None
log_start(task=task_name, env=BENCHMARK, model=MODEL_NAME)
try:
result = reset_env(task_key)
obs_data = result.get("observation", {})
done = result.get("done", False)
for step in range(1, MAX_STEPS + 1):
if done:
break
user_prompt = build_user_prompt(obs_data, step)
response_text = call_model(client, user_prompt, history)
history.append({"role": "assistant", "content": response_text})
action_str = response_text[:60] + "..." if len(response_text) > 60 else response_text
action_log_str = action_str.replace('\n', ' ').replace('\r', '')
try:
result = step_env(response_text)
reward_data = result.get("reward", {})
if not reward_data:
print(f"[DEBUG] No reward in response: {result}", flush=True)
if isinstance(reward_data, dict):
reward = reward_data.get("value", 0.001) # default 0.001, never 0.0
else:
reward = float(reward_data)
done = result.get("done", False)
error_msg = None
obs_data = result.get("observation", {})
except Exception as exc:
error_msg = str(exc).replace('\n', ' ').replace('\r', '')
reward = 0.001 # 1e-6 formats as "0.0000" — use 0.001 so :.4f gives "0.0010"
done = True
obs_data = {}
rewards.append(reward)
steps_taken = step
log_step(step=step, action=action_log_str, reward=reward, done=done, error=error_msg)
if done:
break
# 1e-6 formats as "0.0000" with :.4f — validator parses that as 0.0 and fails
# Use 0.001 min so it formats as "0.0010", and 0.999 max so it formats as "0.9990"
_MIN_SCORE = 0.001
_MAX_SCORE = 0.999
score = max(rewards) if rewards else _MIN_SCORE
score = max(_MIN_SCORE, min(_MAX_SCORE, score))
success = score >= SUCCESS_SCORE_THRESHOLD
except Exception as exc:
error_msg = str(exc).replace('\n', ' ').replace('\r', '')
print(f"[DEBUG] Task execution error: {error_msg}", flush=True)
finally:
log_end(success=success, steps=steps_taken, score=score, rewards=rewards)
return {
"task_id": task["task_id"],
"task_name": task_name,
"score": score,
"success": success,
"steps": steps_taken,
"rewards": rewards,
}
def main():
print("=" * 60)
print("GDPR Auditor - Baseline Inference")
print("=" * 60)
print(f"API URL : {API_BASE_URL}")
print(f"Model : {MODEL_NAME}")
print(f"Server : {SERVER_URL}")
print()
if not HF_TOKEN:
print("ERROR: HF_TOKEN not set")
return
if not API_BASE_URL:
print("ERROR: API_BASE_URL not set")
return
if not MODEL_NAME:
print("ERROR: MODEL_NAME not set")
return
try:
client = OpenAI(
base_url=API_BASE_URL,
api_key=HF_TOKEN,
)
except Exception as e:
print(f"ERROR: Failed to create client: {e}")
return
try:
resp = requests.get(f"{SERVER_URL}/health", timeout=5)
if resp.status_code != 200:
print(f"ERROR: Server returned {resp.status_code}")
return
print("Server connection: OK")
except Exception as e:
print(f"ERROR: Cannot connect to server: {e}")
return
results = []
for task_key in ["easy", "medium", "hard", "elite"]:
task_name = TASKS[task_key]["name"]
print(f"\nRunning task: {task_name} ({task_key})...")
result = run_task(client, task_key)
results.append(result)
print(f" score={result['score']:.4f} steps={result['steps']}")
print("\n" + "=" * 60)
print("SUMMARY")
print("=" * 60)
for r in results:
print(f"[{r['task_id'][:1].upper()}] {r['task_id']:<30} score={r['score']:.4f}")
print("-" * 60)
avg = sum(r["score"] for r in results) / len(results)
print(f"Average score: {avg:.4f}")
output = {
"model": MODEL_NAME,
"api_url": API_BASE_URL,
"timestamp": str(__import__("datetime").datetime.now()),
"average_score": avg,
"results": results,
}
with open("inference_results.json", "w") as f:
json.dump(output, f, indent=2)
print(f"\nResults saved to: inference_results.json")
if __name__ == "__main__":
main()
|