Spaces:
Sleeping
Sleeping
File size: 16,725 Bytes
6ceefb7 bec85f3 6ceefb7 bec85f3 6ceefb7 bec85f3 6ceefb7 bec85f3 6ceefb7 bec85f3 6ceefb7 bec85f3 6ceefb7 bec85f3 6ceefb7 bec85f3 6ceefb7 bec85f3 6ceefb7 bec85f3 6ceefb7 bec85f3 6ceefb7 bec85f3 6ceefb7 bec85f3 6ceefb7 bec85f3 6ceefb7 bec85f3 6ceefb7 bec85f3 6ceefb7 bec85f3 6ceefb7 bec85f3 6ceefb7 bec85f3 6ceefb7 bec85f3 6ceefb7 bec85f3 6ceefb7 bec85f3 6ceefb7 bec85f3 6ceefb7 bec85f3 6ceefb7 bec85f3 6ceefb7 bec85f3 6ceefb7 bec85f3 6ceefb7 bec85f3 6ceefb7 bec85f3 6ceefb7 bec85f3 6ceefb7 bec85f3 6ceefb7 bec85f3 6ceefb7 bec85f3 6ceefb7 bec85f3 6ceefb7 bec85f3 6ceefb7 bec85f3 6ceefb7 bec85f3 6ceefb7 bec85f3 6ceefb7 bec85f3 6ceefb7 bec85f3 6ceefb7 bec85f3 6ceefb7 bec85f3 6ceefb7 bec85f3 6ceefb7 bec85f3 6ceefb7 bec85f3 6ceefb7 bec85f3 6ceefb7 bec85f3 6ceefb7 bec85f3 6ceefb7 bec85f3 6ceefb7 bec85f3 6ceefb7 bec85f3 6ceefb7 bec85f3 6ceefb7 bec85f3 6ceefb7 bec85f3 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 |
import json
import gradio as gr
from typing import Dict, Any
import os
from dataclasses import dataclass
import re
from llama_cpp import Llama
from huggingface_hub import hf_hub_download
@dataclass
class LocalModelConfig:
"""تنظیمات مدل محلی GGUF - Qwen2.5-32B"""
repo_id: str = "Qwen/Qwen2.5-32B-Instruct-GGUF"
filename: str = "qwen2.5-32b-instruct-q4_k_m.gguf"
max_tokens: int = 8000
temperature: float = 0.3
top_p: float = 0.8
n_ctx: int = 4096
n_threads: int = 4 # کمتر برای Spaces
n_gpu_layers: int = 50
class LocalCerebrasAnonymizer:
"""سیستم ناشناسسازی متون مالی فارسی با مدل محلی"""
def __init__(self):
self.config = LocalModelConfig()
self.llm = None
self.model_loaded = False
def load_model(self) -> str:
"""بارگذاری مدل از HuggingFace"""
try:
print(f"🤖 درحال دانلود مدل از HuggingFace...")
print(f"📦 Repo: {self.config.repo_id}")
print(f"📄 Filename: {self.config.filename}")
# دانلود مدل
model_path = hf_hub_download(
repo_id=self.config.repo_id,
filename=self.config.filename,
local_dir="./models",
local_dir_use_symlinks=False
)
print(f"✅ مدل دانلود شد: {model_path}")
print(f"🤖 درحال بارگذاری مدل...")
self.llm = Llama(
model_path=model_path,
n_ctx=self.config.n_ctx,
n_threads=self.config.n_threads,
n_gpu_layers=self.config.n_gpu_layers,
verbose=False
)
self.model_loaded = True
print("✅ مدل با موفقیت بارگذاری شد\n")
return "✅ مدل آماده است"
except Exception as e:
error_msg = f"❌ خطا: {str(e)}"
print(error_msg)
return error_msg
def _create_system_prompt(self) -> str:
"""دستورالعمل سیستمی"""
return """شما یک سیستم ناشناسسازی متون مالی فارسی هستید.
⚠️ CRITICAL: در پاسخ نهایی خود، فقط و فقط متن ناشناسسازی شده را برگردانید، بدون هیچ توضیح، تحلیل، یا تگ اضافی.
## قوانین اندیسگذاری:
1. **ترتیب پیوسته**: company-01, company-02, ... | person-01, person-02, ... | amount-01, amount-02, ... | percent-01, percent-02, ...
2. **ثبات**: اگر "همراه اول" → company-01 شد، در تمام متن همان باشد
3. **نام مستعار**: "فاما" = "فولاد مبارکه" → هر دو company-01
## انواع موجودیت:
- **company-XX**: شرکتها، بانکها، سازمانها
- **person-XX**: نام و نام خانوادگی اشخاص
- **amount-XX**: مبالغ - واحد را حفظ کن
- **percent-XX**: درصدها
## مثال:
ورودی: ایران خودرو در اسفند 1402 حدود 23 هزار میلیارد درآمد کسب کرد که 4.58 درصد افزایش داشت.
خروجی: company-01 در اسفند 1402 حدود amount-01 درآمد کسب کرد که percent-01 افزایش داشت.
⚠️ یادآوری: فقط متن ناشناسشده."""
def anonymize_text(self, text: str) -> Dict[str, Any]:
"""ناشناسسازی متن"""
if not self.model_loaded:
return {"success": False, "error": "مدل بارگذاری نشده است"}
if not text.strip():
return {"success": False, "error": "متن ورودی خالی است"}
try:
messages = [
{"role": "system", "content": self._create_system_prompt()},
{"role": "user", "content": text}
]
prompt = self._format_prompt(messages)
print(f"⏳ پردازش متن... (طول: {len(text)} کاراکتر)")
response = self.llm(
prompt,
max_tokens=self.config.max_tokens,
temperature=self.config.temperature,
top_p=self.config.top_p,
stop=["</s>", "[/INST]", "### User:"]
)
content = response["choices"][0]["text"].strip()
# پاکسازی
content = self._remove_thinking_tags(content)
content = self._clean_markdown(content)
content = self._clean_explanations(content)
content = content.strip()
analysis = self._analyze_anonymized_text(content)
return {
"success": True,
"anonymized_text": content,
"entities": analysis["entities"],
"statistics": analysis["statistics"],
"detailed_analysis": analysis["detailed_analysis"],
"quality_check": self._validate_anonymized_text(content)
}
except Exception as e:
return {"success": False, "error": f"خطا: {str(e)}"}
def _format_prompt(self, messages: list) -> str:
"""فرمت prompt برای Qwen2.5"""
formatted = ""
for message in messages:
role = message["role"]
content = message["content"]
if role == "system":
formatted += f"{content}\n\n"
elif role == "user":
formatted += f"[INST] {content} [/INST]\n"
elif role == "assistant":
formatted += f"{content}\n\n"
return formatted
def _remove_thinking_tags(self, content: str) -> str:
content = re.sub(r'<think>.*?</think>', '', content, flags=re.DOTALL)
content = re.sub(r'</?think>', '', content)
return content.strip()
def _clean_markdown(self, content: str) -> str:
if "```" in content:
lines = content.split('\n')
clean_lines = []
skip = False
for line in lines:
if line.strip().startswith('```'):
skip = not skip
continue
if not skip:
clean_lines.append(line)
content = '\n'.join(clean_lines)
return content
def _clean_explanations(self, content: str) -> str:
lines = content.split('\n')
clean_lines = []
for line in lines:
if any(word in line.lower() for word in
['okay', 'let me', 'here is', 'خروجی', 'نتیجه', 'پاسخ:', 'assistant', '[inst]']):
continue
clean_lines.append(line)
return '\n'.join(clean_lines).strip()
def _analyze_anonymized_text(self, text: str) -> Dict[str, Any]:
companies = re.findall(r'company-(\d+)', text)
persons = re.findall(r'person-(\d+)', text)
amounts = re.findall(r'amount-(\d+)', text)
percents = re.findall(r'percent-(\d+)', text)
statistics = {
"company": len(set(companies)),
"person": len(set(persons)),
"amount": len(set(amounts)),
"percent": len(set(percents)),
"total": len(companies) + len(persons) + len(amounts) + len(percents)
}
entities = {
"companies": sorted(list(set(companies)), key=lambda x: int(x)),
"persons": sorted(list(set(persons)), key=lambda x: int(x)),
"amounts": sorted(list(set(amounts)), key=lambda x: int(x)),
"percents": sorted(list(set(percents)), key=lambda x: int(x))
}
detailed_analysis = {
"preserved_dates": len(re.findall(r'\d{4}/\d{1,2}/\d{1,2}|\d{1,2}\s+\w+\s+\d{4}', text)),
"financial_indicators": len(re.findall(r'\b(EPS|P/E|ARPU|NPL|ROE|ROA)\b', text)),
"units_preserved": len(re.findall(r'(میلیارد|میلیون|هزار|تومان|ریال|درهم|دلار)', text))
}
return {
"statistics": statistics,
"entities": entities,
"detailed_analysis": detailed_analysis
}
def _validate_anonymized_text(self, text: str) -> Dict[str, Any]:
companies = re.findall(r'company-(\d+)', text)
persons = re.findall(r'person-(\d+)', text)
amounts = re.findall(r'amount-(\d+)', text)
percents = re.findall(r'percent-(\d+)', text)
validation_issues = []
for entity_type, indices in [("company", companies), ("person", persons),
("amount", amounts), ("percent", percents)]:
if indices:
unique_indices = sorted(list(set([int(x) for x in indices])))
if unique_indices[0] != 1:
validation_issues.append(f"⚠️ {entity_type} از 01 شروع نشده")
expected = list(range(1, len(unique_indices) + 1))
if unique_indices != expected:
validation_issues.append(f"⚠️ {entity_type} پیوسته نیست")
return {
"is_valid": len(validation_issues) == 0,
"issues": validation_issues,
"entity_counts": {
"company": len(set(companies)),
"person": len(set(persons)),
"amount": len(set(amounts)),
"percent": len(set(percents))
}
}
# ========== رابط کاربری ==========
anonymizer = LocalCerebrasAnonymizer()
def create_interface():
custom_css = """
.gradio-container {
font-family: 'Tahoma', 'Arial', sans-serif !important;
direction: rtl;
max-width: 1400px;
margin: 0 auto;
}
.info-box {
background-color: #e3f2fd;
border: 2px solid #2196F3;
border-radius: 12px;
padding: 15px;
color: #0d47a1;
margin: 10px 0;
}
.local-box {
background-color: #e8f5e9;
border: 2px solid #4caf50;
border-radius: 12px;
padding: 15px;
color: #1b5e20;
margin: 10px 0;
}
.result-box {
background-color: #f8f9fa;
border: 2px solid #e9ecef;
border-radius: 12px;
padding: 20px;
}
"""
with gr.Blocks(css=custom_css, title="ناشناسساز Qwen2.5", theme=gr.themes.Soft()) as interface:
gr.Markdown("""
# 🔒 سیستم ناشناسسازی متون مالی فارسی
### 🚀 Qwen 2.5-32B (HuggingFace Spaces)
""")
gr.Markdown("""
<div class="info-box">
📊 <strong>مدل:</strong> Qwen2.5-32B-Instruct-Q4_K_M<br>
🌐 <strong>منبع:</strong> HuggingFace Hub<br>
💾 <strong>حجم:</strong> ~20 GB (Q4 quantization)<br>
⚡ <strong>سرعت:</strong> بستگی به GPU Spaces دارد
</div>
""")
status_box = gr.Textbox(label="📋 وضعیت", interactive=False, value="⏳ درحال بارگذاری مدل...")
load_btn = gr.Button("🤖 بارگذاری مدل", variant="primary", size="lg")
with gr.Row(visible=False) as input_section:
with gr.Column(scale=1):
input_text = gr.Textbox(
label="📝 متن ورودی",
placeholder="متن خود را اینجا وارد کنید...",
lines=12,
max_lines=25
)
with gr.Row():
anonymize_btn = gr.Button("🔒 ناشناسسازی", variant="primary", size="lg")
clear_btn = gr.Button("🗑️ پاک کردن", variant="secondary")
with gr.Column(scale=1):
output_text = gr.Textbox(
label="🎯 متن ناشناسسازی شده",
lines=12,
max_lines=25,
elem_classes=["result-box"]
)
with gr.Row(visible=False) as output_section:
with gr.Column():
statistics_output = gr.Markdown(label="📊 آمار")
with gr.Column():
quality_output = gr.Markdown(label="✅ کیفیت")
with gr.Row(visible=False) as output_section2:
entities_output = gr.Markdown(label="🏷️ موجودیتها")
detailed_output = gr.Markdown(label="🔍 تحلیل")
def load_model_action():
"""بارگذاری مدل"""
msg = anonymizer.load_model()
return (
gr.Textbox(value=msg),
gr.Row(visible=True),
gr.Row(visible=True),
gr.Row(visible=True)
)
def process_text(text):
"""پردازش متن"""
if not text.strip():
return ("", "❌ متن خالی است", "", "", "", "")
result = anonymizer.anonymize_text(text)
if not result["success"]:
return ("", f"❌ {result['error']}", "", "", "", "")
stats = result.get("statistics", {})
stats_md = f"""📊 **آمار:**
🏢 شرکت: {stats.get('company', 0)}
👤 اشخاص: {stats.get('person', 0)}
💰 مبالغ: {stats.get('amount', 0)}
📊 درصدها: {stats.get('percent', 0)}
🔢 کل: {stats.get('total', 0)}"""
quality = result.get("quality_check", {})
quality_md = f"""✅ **کنترل کیفیت:**
{'✅ موفق' if quality.get('is_valid') else '❌ مشکل'}
"""
if quality.get("issues"):
quality_md += "\n**مشکلات:**\n"
for issue in quality["issues"]:
quality_md += f"• {issue}\n"
entities = result.get("entities", {})
entities_md = "🏷️ **موجودیتها:**\n"
if entities.get("companies"):
entities_md += f"\n🏢 company-{', company-'.join(entities['companies'])}"
if entities.get("persons"):
entities_md += f"\n👤 person-{', person-'.join(entities['persons'])}"
if entities.get("amounts"):
entities_md += f"\n💰 amount-{', amount-'.join(entities['amounts'])}"
if entities.get("percents"):
entities_md += f"\n📊 percent-{', percent-'.join(entities['percents'])}"
detailed = result.get("detailed_analysis", {})
detailed_md = f"""🔍 **تحلیل:**
📅 تاریخ: {detailed.get('preserved_dates', 0)}
📈 شاخص: {detailed.get('financial_indicators', 0)}
📏 واحد: {detailed.get('units_preserved', 0)}"""
return (
result["anonymized_text"],
stats_md,
quality_md,
entities_md,
detailed_md,
"✅ موفق"
)
def clear_all():
return "", "", "", "", "", ""
load_btn.click(
fn=load_model_action,
outputs=[status_box, input_section, output_section, output_section2]
)
anonymize_btn.click(
fn=process_text,
inputs=[input_text],
outputs=[output_text, statistics_output, quality_output, entities_output, detailed_output, status_box]
)
clear_btn.click(
fn=clear_all,
outputs=[input_text, output_text, statistics_output, quality_output, entities_output, detailed_output]
)
gr.Examples(
examples=[
["ایران خودرو در اسفندماه حدود 23 هزار میلیارد تومان درآمد کسب کرد که 4.58 درصد افزایش داشت."],
["مجمع پتروشیمی برگزار شد. وانیا نیک تدبیر را بازرس انتخاب کردند."],
],
inputs=input_text,
label="📚 مثالها"
)
return interface
if __name__ == "__main__":
interface = create_interface()
interface.launch() |