Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
|
@@ -1,452 +1,620 @@
|
|
| 1 |
-
import gradio as gr
|
| 2 |
-
import os
|
| 3 |
-
from llama_index.core import Settings
|
| 4 |
-
from documents_prep import load_json_documents, load_table_documents, load_image_documents
|
| 5 |
-
from utils import get_llm_model, get_embedding_model, get_reranker_model, answer_question
|
| 6 |
-
from my_logging import log_message
|
| 7 |
-
from index_retriever import create_vector_index, create_query_engine
|
| 8 |
-
import sys
|
| 9 |
-
from config import (
|
| 10 |
-
HF_REPO_ID, HF_TOKEN, DOWNLOAD_DIR, CHUNKS_FILENAME,
|
| 11 |
-
JSON_FILES_DIR, TABLE_DATA_DIR, IMAGE_DATA_DIR, DEFAULT_MODEL, AVAILABLE_MODELS
|
| 12 |
-
)
|
| 13 |
-
|
| 14 |
-
|
| 15 |
-
def merge_table_chunks(chunk_info):
|
| 16 |
-
merged = {}
|
| 17 |
-
|
| 18 |
-
for chunk in chunk_info:
|
| 19 |
-
doc_type = chunk.get('type', 'text')
|
| 20 |
-
doc_id = chunk.get('document_id', 'unknown')
|
| 21 |
-
|
| 22 |
-
if doc_type == 'table' or doc_type == 'table_row':
|
| 23 |
-
table_num = chunk.get('table_number', '')
|
| 24 |
-
key = f"{doc_id}_{table_num}"
|
| 25 |
-
|
| 26 |
-
if key not in merged:
|
| 27 |
-
merged[key] = {
|
| 28 |
-
'document_id': doc_id,
|
| 29 |
-
'type': 'table',
|
| 30 |
-
'table_number': table_num,
|
| 31 |
-
'section_id': chunk.get('section_id', 'unknown'),
|
| 32 |
-
'chunk_text': chunk.get('chunk_text', '')
|
| 33 |
-
}
|
| 34 |
-
else:
|
| 35 |
-
merged[key]['chunk_text'] += '\n' + chunk.get('chunk_text', '')
|
| 36 |
-
else:
|
| 37 |
-
unique_key = f"{doc_id}_{chunk.get('section_id', '')}_{chunk.get('chunk_id', 0)}"
|
| 38 |
-
merged[unique_key] = chunk
|
| 39 |
-
|
| 40 |
-
return list(merged.values())
|
| 41 |
-
|
| 42 |
-
|
| 43 |
-
def create_chunks_display_html(chunk_info):
|
| 44 |
-
if not chunk_info:
|
| 45 |
-
return "<div style='padding: 20px; text-align: center; color: black;'>Нет данных о чанках</div>"
|
| 46 |
-
|
| 47 |
-
merged_chunks = merge_table_chunks(chunk_info)
|
| 48 |
-
|
| 49 |
-
html = "<div style='max-height: 500px; overflow-y: auto; padding: 10px; color: black;'>"
|
| 50 |
-
html += f"<h4 style='color: black;'>Найдено релевантных чанков: {len(merged_chunks)}</h4>"
|
| 51 |
-
|
| 52 |
-
for i, chunk in enumerate(merged_chunks):
|
| 53 |
-
bg_color = "#f8f9fa" if i % 2 == 0 else "#e9ecef"
|
| 54 |
-
section_display = get_section_display(chunk)
|
| 55 |
-
formatted_content = get_formatted_content(chunk)
|
| 56 |
-
|
| 57 |
-
html += f"""
|
| 58 |
-
<div style='background-color: {bg_color}; padding: 10px; margin: 5px 0; border-radius: 5px; border-left: 4px solid #007bff; color: black;'>
|
| 59 |
-
<strong style='color: black;'>Документ:</strong> <span style='color: black;'>{chunk['document_id']}</span><br>
|
| 60 |
-
<strong style='color: black;'>Раздел:</strong> <span style='color: black;'>{section_display}</span><br>
|
| 61 |
-
<strong style='color: black;'>Содержание:</strong><br>
|
| 62 |
-
<div style='background-color: white; padding: 8px; margin-top: 5px; border-radius: 3px; font-family: monospace; font-size: 12px; color: black; max-height: 200px; overflow-y: auto;'>
|
| 63 |
-
{formatted_content}
|
| 64 |
-
</div>
|
| 65 |
-
</div>
|
| 66 |
-
"""
|
| 67 |
-
|
| 68 |
-
html += "</div>"
|
| 69 |
-
return html
|
| 70 |
-
|
| 71 |
-
def get_section_display(chunk):
|
| 72 |
-
section_path = chunk.get('section_path', '')
|
| 73 |
-
section_id = chunk.get('section_id', 'unknown')
|
| 74 |
-
doc_type = chunk.get('type', 'text')
|
| 75 |
-
|
| 76 |
-
if doc_type == 'table' and chunk.get('table_number'):
|
| 77 |
-
table_num = chunk.get('table_number')
|
| 78 |
-
if not str(table_num).startswith('№'):
|
| 79 |
-
table_num = f"№{table_num}"
|
| 80 |
-
return f"таблица {table_num}"
|
| 81 |
-
|
| 82 |
-
if doc_type == 'image' and chunk.get('image_number'):
|
| 83 |
-
image_num = chunk.get('image_number')
|
| 84 |
-
if not str(image_num).startswith('№'):
|
| 85 |
-
image_num = f"№{image_num}"
|
| 86 |
-
return f"рисунок {image_num}"
|
| 87 |
-
|
| 88 |
-
if section_path:
|
| 89 |
-
return section_path
|
| 90 |
-
elif section_id and section_id != 'unknown':
|
| 91 |
-
return section_id
|
| 92 |
-
|
| 93 |
-
return section_id
|
| 94 |
-
|
| 95 |
-
def get_formatted_content(chunk):
|
| 96 |
-
document_id = chunk.get('document_id', 'unknown')
|
| 97 |
-
section_path = chunk.get('section_path', '')
|
| 98 |
-
section_id = chunk.get('section_id', 'unknown')
|
| 99 |
-
section_text = chunk.get('section_text', '')
|
| 100 |
-
parent_section = chunk.get('parent_section', '')
|
| 101 |
-
parent_title = chunk.get('parent_title', '')
|
| 102 |
-
level = chunk.get('level', '')
|
| 103 |
-
chunk_text = chunk.get('chunk_text', '')
|
| 104 |
-
doc_type = chunk.get('type', 'text')
|
| 105 |
-
|
| 106 |
-
# For text documents
|
| 107 |
-
if level in ['subsection', 'sub_subsection', 'sub_sub_subsection'] and parent_section:
|
| 108 |
-
current_section = section_path if section_path else section_id
|
| 109 |
-
parent_info = f"{parent_section} ({parent_title})" if parent_title else parent_section
|
| 110 |
-
return f"В разделе {parent_info} в документе {document_id}, пункт {current_section}: {chunk_text}"
|
| 111 |
-
else:
|
| 112 |
-
current_section = section_path if section_path else section_id
|
| 113 |
-
clean_text = chunk_text
|
| 114 |
-
if section_text and chunk_text.startswith(section_text):
|
| 115 |
-
section_title = section_text
|
| 116 |
-
elif chunk_text.startswith(f"{current_section} "):
|
| 117 |
-
clean_text = chunk_text[len(f"{current_section} "):].strip()
|
| 118 |
-
section_title = section_text if section_text else f"{current_section} {clean_text.split('.')[0] if '.' in clean_text else clean_text[:50]}"
|
| 119 |
-
else:
|
| 120 |
-
section_title = section_text if section_text else current_section
|
| 121 |
-
|
| 122 |
-
return f"В разделе {current_section} в документе {document_id}, пункт {section_title}: {clean_text}"
|
| 123 |
-
|
| 124 |
-
def initialize_system(repo_id, hf_token, download_dir, chunks_filename=None,
|
| 125 |
-
json_files_dir=None, table_data_dir=None, image_data_dir=None,
|
| 126 |
-
use_json_instead_csv=False):
|
| 127 |
-
try:
|
| 128 |
-
log_message("Инициализация системы")
|
| 129 |
-
os.makedirs(download_dir, exist_ok=True)
|
| 130 |
-
from config import CHUNK_SIZE, CHUNK_OVERLAP
|
| 131 |
-
from llama_index.core.text_splitter import TokenTextSplitter
|
| 132 |
-
|
| 133 |
-
embed_model = get_embedding_model()
|
| 134 |
-
llm = get_llm_model(DEFAULT_MODEL)
|
| 135 |
-
reranker = get_reranker_model()
|
| 136 |
-
|
| 137 |
-
Settings.embed_model = embed_model
|
| 138 |
-
Settings.llm = llm
|
| 139 |
-
Settings.text_splitter = TokenTextSplitter(
|
| 140 |
-
chunk_size=CHUNK_SIZE,
|
| 141 |
-
chunk_overlap=CHUNK_OVERLAP,
|
| 142 |
-
separator=" ",
|
| 143 |
-
backup_separators=["\n", ".", "!", "?"]
|
| 144 |
-
)
|
| 145 |
-
|
| 146 |
-
log_message(f"Configured chunk size: {CHUNK_SIZE} tokens")
|
| 147 |
-
log_message(f"Configured chunk overlap: {CHUNK_OVERLAP} tokens")
|
| 148 |
-
|
| 149 |
-
all_documents = []
|
| 150 |
-
chunks_df = None
|
| 151 |
-
|
| 152 |
-
# CHANGED: Use load_all_documents instead of loading separately
|
| 153 |
-
if use_json_instead_csv and json_files_dir:
|
| 154 |
-
log_message("Используем JSON файлы вместо CSV")
|
| 155 |
-
from documents_prep import load_all_documents
|
| 156 |
-
|
| 157 |
-
# This will handle text, tables, and images all together with proper logging
|
| 158 |
-
all_documents = load_all_documents(
|
| 159 |
-
repo_id=repo_id,
|
| 160 |
-
hf_token=hf_token,
|
| 161 |
-
json_dir=json_files_dir,
|
| 162 |
-
table_dir=table_data_dir if table_data_dir else "",
|
| 163 |
-
image_dir=image_data_dir if image_data_dir else ""
|
| 164 |
-
)
|
| 165 |
-
else:
|
| 166 |
-
# OLD PATH: Loading separately (fallback)
|
| 167 |
-
if chunks_filename:
|
| 168 |
-
log_message("Загружаем данные из CSV")
|
| 169 |
-
|
| 170 |
-
if table_data_dir:
|
| 171 |
-
log_message("Добавляю табличные данные")
|
| 172 |
-
from documents_prep import load_table_documents
|
| 173 |
-
|
| 174 |
-
table_chunks = load_table_documents(repo_id, hf_token, table_data_dir)
|
| 175 |
-
log_message(f"Загружено {len(table_chunks)} табличных чанков")
|
| 176 |
-
all_documents.extend(table_chunks)
|
| 177 |
-
|
| 178 |
-
if image_data_dir:
|
| 179 |
-
log_message("Добавляю данные изображений")
|
| 180 |
-
from documents_prep import load_image_documents
|
| 181 |
-
|
| 182 |
-
image_documents = load_image_documents(repo_id, hf_token, image_data_dir)
|
| 183 |
-
log_message(f"Загружено {len(image_documents)} документов изображений")
|
| 184 |
-
all_documents.extend(image_documents)
|
| 185 |
-
|
| 186 |
-
log_message(f"Всего документов после всей обработки: {len(all_documents)}")
|
| 187 |
-
|
| 188 |
-
vector_index = create_vector_index(all_documents)
|
| 189 |
-
query_engine = create_query_engine(vector_index)
|
| 190 |
-
|
| 191 |
-
# Create chunk_info for display (extract from documents metadata)
|
| 192 |
-
chunk_info = []
|
| 193 |
-
for doc in all_documents:
|
| 194 |
-
chunk_info.append({
|
| 195 |
-
'document_id': doc.metadata.get('document_id', 'unknown'),
|
| 196 |
-
'section_id': doc.metadata.get('section_id', 'unknown'),
|
| 197 |
-
'type': doc.metadata.get('type', 'text'),
|
| 198 |
-
'chunk_text': doc.text[:200] + '...' if len(doc.text) > 200 else doc.text,
|
| 199 |
-
'table_number': doc.metadata.get('table_number', ''),
|
| 200 |
-
'image_number': doc.metadata.get('image_number', ''),
|
| 201 |
-
'section': doc.metadata.get('section', ''),
|
| 202 |
-
'connection_type': doc.metadata.get('connection_type', '') # ADD THIS
|
| 203 |
-
})
|
| 204 |
-
|
| 205 |
-
log_message(f"Система успешно инициализирована")
|
| 206 |
-
return query_engine, chunks_df, reranker, vector_index, chunk_info
|
| 207 |
-
|
| 208 |
-
except Exception as e:
|
| 209 |
-
log_message(f"Ошибка инициализации: {str(e)}")
|
| 210 |
-
import traceback
|
| 211 |
-
log_message(traceback.format_exc())
|
| 212 |
-
return None, None, None, None, []
|
| 213 |
-
|
| 214 |
-
def switch_model(model_name, vector_index):
|
| 215 |
-
from llama_index.core import Settings
|
| 216 |
-
from index_retriever import create_query_engine
|
| 217 |
-
|
| 218 |
-
try:
|
| 219 |
-
log_message(f"Переключение на модель: {model_name}")
|
| 220 |
-
|
| 221 |
-
new_llm = get_llm_model(model_name)
|
| 222 |
-
Settings.llm = new_llm
|
| 223 |
-
|
| 224 |
-
if vector_index is not None:
|
| 225 |
-
new_query_engine = create_query_engine(vector_index)
|
| 226 |
-
log_message(f"Модель успешно переключена на: {model_name}")
|
| 227 |
-
return new_query_engine, f"✅ Модель переключена на: {model_name}"
|
| 228 |
-
else:
|
| 229 |
-
return None, "❌ Ошибка: система не инициализирована"
|
| 230 |
-
|
| 231 |
-
except Exception as e:
|
| 232 |
-
error_msg = f"Ошибка переключения модели: {str(e)}"
|
| 233 |
-
log_message(error_msg)
|
| 234 |
-
return None, f"❌ {error_msg}"
|
| 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 |
-
|
| 426 |
-
|
| 427 |
-
|
| 428 |
-
|
| 429 |
-
|
| 430 |
-
|
| 431 |
-
|
| 432 |
-
|
| 433 |
-
|
| 434 |
-
|
| 435 |
-
|
| 436 |
-
|
| 437 |
-
|
| 438 |
-
|
| 439 |
-
|
| 440 |
-
|
| 441 |
-
|
| 442 |
-
|
| 443 |
-
|
| 444 |
-
|
| 445 |
-
|
| 446 |
-
|
| 447 |
-
|
| 448 |
-
|
| 449 |
-
|
| 450 |
-
|
| 451 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 452 |
main()
|
|
|
|
| 1 |
+
import gradio as gr
|
| 2 |
+
import os
|
| 3 |
+
from llama_index.core import Settings
|
| 4 |
+
from documents_prep import load_json_documents, load_table_documents, load_image_documents
|
| 5 |
+
from utils import get_llm_model, get_embedding_model, get_reranker_model, answer_question
|
| 6 |
+
from my_logging import log_message
|
| 7 |
+
from index_retriever import create_vector_index, create_query_engine
|
| 8 |
+
import sys
|
| 9 |
+
from config import (
|
| 10 |
+
HF_REPO_ID, HF_TOKEN, DOWNLOAD_DIR, CHUNKS_FILENAME,
|
| 11 |
+
JSON_FILES_DIR, TABLE_DATA_DIR, IMAGE_DATA_DIR, DEFAULT_MODEL, AVAILABLE_MODELS
|
| 12 |
+
)
|
| 13 |
+
|
| 14 |
+
|
| 15 |
+
def merge_table_chunks(chunk_info):
|
| 16 |
+
merged = {}
|
| 17 |
+
|
| 18 |
+
for chunk in chunk_info:
|
| 19 |
+
doc_type = chunk.get('type', 'text')
|
| 20 |
+
doc_id = chunk.get('document_id', 'unknown')
|
| 21 |
+
|
| 22 |
+
if doc_type == 'table' or doc_type == 'table_row':
|
| 23 |
+
table_num = chunk.get('table_number', '')
|
| 24 |
+
key = f"{doc_id}_{table_num}"
|
| 25 |
+
|
| 26 |
+
if key not in merged:
|
| 27 |
+
merged[key] = {
|
| 28 |
+
'document_id': doc_id,
|
| 29 |
+
'type': 'table',
|
| 30 |
+
'table_number': table_num,
|
| 31 |
+
'section_id': chunk.get('section_id', 'unknown'),
|
| 32 |
+
'chunk_text': chunk.get('chunk_text', '')
|
| 33 |
+
}
|
| 34 |
+
else:
|
| 35 |
+
merged[key]['chunk_text'] += '\n' + chunk.get('chunk_text', '')
|
| 36 |
+
else:
|
| 37 |
+
unique_key = f"{doc_id}_{chunk.get('section_id', '')}_{chunk.get('chunk_id', 0)}"
|
| 38 |
+
merged[unique_key] = chunk
|
| 39 |
+
|
| 40 |
+
return list(merged.values())
|
| 41 |
+
|
| 42 |
+
|
| 43 |
+
def create_chunks_display_html(chunk_info):
|
| 44 |
+
if not chunk_info:
|
| 45 |
+
return "<div style='padding: 20px; text-align: center; color: black;'>Нет данных о чанках</div>"
|
| 46 |
+
|
| 47 |
+
merged_chunks = merge_table_chunks(chunk_info)
|
| 48 |
+
|
| 49 |
+
html = "<div style='max-height: 500px; overflow-y: auto; padding: 10px; color: black;'>"
|
| 50 |
+
html += f"<h4 style='color: black;'>Найдено релевантных чанков: {len(merged_chunks)}</h4>"
|
| 51 |
+
|
| 52 |
+
for i, chunk in enumerate(merged_chunks):
|
| 53 |
+
bg_color = "#f8f9fa" if i % 2 == 0 else "#e9ecef"
|
| 54 |
+
section_display = get_section_display(chunk)
|
| 55 |
+
formatted_content = get_formatted_content(chunk)
|
| 56 |
+
|
| 57 |
+
html += f"""
|
| 58 |
+
<div style='background-color: {bg_color}; padding: 10px; margin: 5px 0; border-radius: 5px; border-left: 4px solid #007bff; color: black;'>
|
| 59 |
+
<strong style='color: black;'>Документ:</strong> <span style='color: black;'>{chunk['document_id']}</span><br>
|
| 60 |
+
<strong style='color: black;'>Раздел:</strong> <span style='color: black;'>{section_display}</span><br>
|
| 61 |
+
<strong style='color: black;'>Содержание:</strong><br>
|
| 62 |
+
<div style='background-color: white; padding: 8px; margin-top: 5px; border-radius: 3px; font-family: monospace; font-size: 12px; color: black; max-height: 200px; overflow-y: auto;'>
|
| 63 |
+
{formatted_content}
|
| 64 |
+
</div>
|
| 65 |
+
</div>
|
| 66 |
+
"""
|
| 67 |
+
|
| 68 |
+
html += "</div>"
|
| 69 |
+
return html
|
| 70 |
+
|
| 71 |
+
def get_section_display(chunk):
|
| 72 |
+
section_path = chunk.get('section_path', '')
|
| 73 |
+
section_id = chunk.get('section_id', 'unknown')
|
| 74 |
+
doc_type = chunk.get('type', 'text')
|
| 75 |
+
|
| 76 |
+
if doc_type == 'table' and chunk.get('table_number'):
|
| 77 |
+
table_num = chunk.get('table_number')
|
| 78 |
+
if not str(table_num).startswith('№'):
|
| 79 |
+
table_num = f"№{table_num}"
|
| 80 |
+
return f"таблица {table_num}"
|
| 81 |
+
|
| 82 |
+
if doc_type == 'image' and chunk.get('image_number'):
|
| 83 |
+
image_num = chunk.get('image_number')
|
| 84 |
+
if not str(image_num).startswith('№'):
|
| 85 |
+
image_num = f"№{image_num}"
|
| 86 |
+
return f"рисунок {image_num}"
|
| 87 |
+
|
| 88 |
+
if section_path:
|
| 89 |
+
return section_path
|
| 90 |
+
elif section_id and section_id != 'unknown':
|
| 91 |
+
return section_id
|
| 92 |
+
|
| 93 |
+
return section_id
|
| 94 |
+
|
| 95 |
+
def get_formatted_content(chunk):
|
| 96 |
+
document_id = chunk.get('document_id', 'unknown')
|
| 97 |
+
section_path = chunk.get('section_path', '')
|
| 98 |
+
section_id = chunk.get('section_id', 'unknown')
|
| 99 |
+
section_text = chunk.get('section_text', '')
|
| 100 |
+
parent_section = chunk.get('parent_section', '')
|
| 101 |
+
parent_title = chunk.get('parent_title', '')
|
| 102 |
+
level = chunk.get('level', '')
|
| 103 |
+
chunk_text = chunk.get('chunk_text', '')
|
| 104 |
+
doc_type = chunk.get('type', 'text')
|
| 105 |
+
|
| 106 |
+
# For text documents
|
| 107 |
+
if level in ['subsection', 'sub_subsection', 'sub_sub_subsection'] and parent_section:
|
| 108 |
+
current_section = section_path if section_path else section_id
|
| 109 |
+
parent_info = f"{parent_section} ({parent_title})" if parent_title else parent_section
|
| 110 |
+
return f"В разделе {parent_info} в документе {document_id}, пункт {current_section}: {chunk_text}"
|
| 111 |
+
else:
|
| 112 |
+
current_section = section_path if section_path else section_id
|
| 113 |
+
clean_text = chunk_text
|
| 114 |
+
if section_text and chunk_text.startswith(section_text):
|
| 115 |
+
section_title = section_text
|
| 116 |
+
elif chunk_text.startswith(f"{current_section} "):
|
| 117 |
+
clean_text = chunk_text[len(f"{current_section} "):].strip()
|
| 118 |
+
section_title = section_text if section_text else f"{current_section} {clean_text.split('.')[0] if '.' in clean_text else clean_text[:50]}"
|
| 119 |
+
else:
|
| 120 |
+
section_title = section_text if section_text else current_section
|
| 121 |
+
|
| 122 |
+
return f"В разделе {current_section} в документе {document_id}, пункт {section_title}: {clean_text}"
|
| 123 |
+
|
| 124 |
+
def initialize_system(repo_id, hf_token, download_dir, chunks_filename=None,
|
| 125 |
+
json_files_dir=None, table_data_dir=None, image_data_dir=None,
|
| 126 |
+
use_json_instead_csv=False):
|
| 127 |
+
try:
|
| 128 |
+
log_message("Инициализация системы")
|
| 129 |
+
os.makedirs(download_dir, exist_ok=True)
|
| 130 |
+
from config import CHUNK_SIZE, CHUNK_OVERLAP
|
| 131 |
+
from llama_index.core.text_splitter import TokenTextSplitter
|
| 132 |
+
|
| 133 |
+
embed_model = get_embedding_model()
|
| 134 |
+
llm = get_llm_model(DEFAULT_MODEL)
|
| 135 |
+
reranker = get_reranker_model()
|
| 136 |
+
|
| 137 |
+
Settings.embed_model = embed_model
|
| 138 |
+
Settings.llm = llm
|
| 139 |
+
Settings.text_splitter = TokenTextSplitter(
|
| 140 |
+
chunk_size=CHUNK_SIZE,
|
| 141 |
+
chunk_overlap=CHUNK_OVERLAP,
|
| 142 |
+
separator=" ",
|
| 143 |
+
backup_separators=["\n", ".", "!", "?"]
|
| 144 |
+
)
|
| 145 |
+
|
| 146 |
+
log_message(f"Configured chunk size: {CHUNK_SIZE} tokens")
|
| 147 |
+
log_message(f"Configured chunk overlap: {CHUNK_OVERLAP} tokens")
|
| 148 |
+
|
| 149 |
+
all_documents = []
|
| 150 |
+
chunks_df = None
|
| 151 |
+
|
| 152 |
+
# CHANGED: Use load_all_documents instead of loading separately
|
| 153 |
+
if use_json_instead_csv and json_files_dir:
|
| 154 |
+
log_message("Используем JSON файлы вместо CSV")
|
| 155 |
+
from documents_prep import load_all_documents
|
| 156 |
+
|
| 157 |
+
# This will handle text, tables, and images all together with proper logging
|
| 158 |
+
all_documents = load_all_documents(
|
| 159 |
+
repo_id=repo_id,
|
| 160 |
+
hf_token=hf_token,
|
| 161 |
+
json_dir=json_files_dir,
|
| 162 |
+
table_dir=table_data_dir if table_data_dir else "",
|
| 163 |
+
image_dir=image_data_dir if image_data_dir else ""
|
| 164 |
+
)
|
| 165 |
+
else:
|
| 166 |
+
# OLD PATH: Loading separately (fallback)
|
| 167 |
+
if chunks_filename:
|
| 168 |
+
log_message("Загружаем данные из CSV")
|
| 169 |
+
|
| 170 |
+
if table_data_dir:
|
| 171 |
+
log_message("Добавляю табличные данные")
|
| 172 |
+
from documents_prep import load_table_documents
|
| 173 |
+
|
| 174 |
+
table_chunks = load_table_documents(repo_id, hf_token, table_data_dir)
|
| 175 |
+
log_message(f"Загружено {len(table_chunks)} табличных чанков")
|
| 176 |
+
all_documents.extend(table_chunks)
|
| 177 |
+
|
| 178 |
+
if image_data_dir:
|
| 179 |
+
log_message("Добавляю данные изображений")
|
| 180 |
+
from documents_prep import load_image_documents
|
| 181 |
+
|
| 182 |
+
image_documents = load_image_documents(repo_id, hf_token, image_data_dir)
|
| 183 |
+
log_message(f"Загружено {len(image_documents)} документов изображений")
|
| 184 |
+
all_documents.extend(image_documents)
|
| 185 |
+
|
| 186 |
+
log_message(f"Всего документов после всей обработки: {len(all_documents)}")
|
| 187 |
+
|
| 188 |
+
vector_index = create_vector_index(all_documents)
|
| 189 |
+
query_engine = create_query_engine(vector_index)
|
| 190 |
+
|
| 191 |
+
# Create chunk_info for display (extract from documents metadata)
|
| 192 |
+
chunk_info = []
|
| 193 |
+
for doc in all_documents:
|
| 194 |
+
chunk_info.append({
|
| 195 |
+
'document_id': doc.metadata.get('document_id', 'unknown'),
|
| 196 |
+
'section_id': doc.metadata.get('section_id', 'unknown'),
|
| 197 |
+
'type': doc.metadata.get('type', 'text'),
|
| 198 |
+
'chunk_text': doc.text[:200] + '...' if len(doc.text) > 200 else doc.text,
|
| 199 |
+
'table_number': doc.metadata.get('table_number', ''),
|
| 200 |
+
'image_number': doc.metadata.get('image_number', ''),
|
| 201 |
+
'section': doc.metadata.get('section', ''),
|
| 202 |
+
'connection_type': doc.metadata.get('connection_type', '') # ADD THIS
|
| 203 |
+
})
|
| 204 |
+
|
| 205 |
+
log_message(f"Система успешно инициализирована")
|
| 206 |
+
return query_engine, chunks_df, reranker, vector_index, chunk_info
|
| 207 |
+
|
| 208 |
+
except Exception as e:
|
| 209 |
+
log_message(f"Ошибка инициализации: {str(e)}")
|
| 210 |
+
import traceback
|
| 211 |
+
log_message(traceback.format_exc())
|
| 212 |
+
return None, None, None, None, []
|
| 213 |
+
|
| 214 |
+
def switch_model(model_name, vector_index):
|
| 215 |
+
from llama_index.core import Settings
|
| 216 |
+
from index_retriever import create_query_engine
|
| 217 |
+
|
| 218 |
+
try:
|
| 219 |
+
log_message(f"Переключение на модель: {model_name}")
|
| 220 |
+
|
| 221 |
+
new_llm = get_llm_model(model_name)
|
| 222 |
+
Settings.llm = new_llm
|
| 223 |
+
|
| 224 |
+
if vector_index is not None:
|
| 225 |
+
new_query_engine = create_query_engine(vector_index)
|
| 226 |
+
log_message(f"Модель успешно переключена на: {model_name}")
|
| 227 |
+
return new_query_engine, f"✅ Модель переключена на: {model_name}"
|
| 228 |
+
else:
|
| 229 |
+
return None, "❌ Ошибка: система не инициализирована"
|
| 230 |
+
|
| 231 |
+
except Exception as e:
|
| 232 |
+
error_msg = f"Ошибка переключения модели: {str(e)}"
|
| 233 |
+
log_message(error_msg)
|
| 234 |
+
return None, f"❌ {error_msg}"
|
| 235 |
+
|
| 236 |
+
# Add these global variables near the top with other globals
|
| 237 |
+
retrieval_params = {
|
| 238 |
+
'vector_top_k': 50,
|
| 239 |
+
'bm25_top_k': 50,
|
| 240 |
+
'similarity_cutoff': 0.55,
|
| 241 |
+
'hybrid_top_k': 100,
|
| 242 |
+
'rerank_top_k': 20
|
| 243 |
+
}
|
| 244 |
+
|
| 245 |
+
# MODIFIED: Update create_query_engine call signature
|
| 246 |
+
def create_query_engine(vector_index, vector_top_k=50, bm25_top_k=50,
|
| 247 |
+
similarity_cutoff=0.55, hybrid_top_k=100):
|
| 248 |
+
try:
|
| 249 |
+
from config import CUSTOM_PROMPT
|
| 250 |
+
from index_retriever import create_query_engine as create_index_query_engine
|
| 251 |
+
|
| 252 |
+
# Pass parameters to the index_retriever function
|
| 253 |
+
query_engine = create_index_query_engine(
|
| 254 |
+
vector_index=vector_index,
|
| 255 |
+
vector_top_k=vector_top_k,
|
| 256 |
+
bm25_top_k=bm25_top_k,
|
| 257 |
+
similarity_cutoff=similarity_cutoff,
|
| 258 |
+
hybrid_top_k=hybrid_top_k
|
| 259 |
+
)
|
| 260 |
+
|
| 261 |
+
log_message(f"Query engine created with params: vector_top_k={vector_top_k}, "
|
| 262 |
+
f"bm25_top_k={bm25_top_k}, cutoff={similarity_cutoff}, hybrid_top_k={hybrid_top_k}")
|
| 263 |
+
return query_engine
|
| 264 |
+
|
| 265 |
+
except Exception as e:
|
| 266 |
+
log_message(f"Ошибка создания query engine: {str(e)}")
|
| 267 |
+
raise
|
| 268 |
+
|
| 269 |
+
# MODIFIED: Update answer_question to use global retrieval_params
|
| 270 |
+
def main_answer_question(question):
|
| 271 |
+
global query_engine, reranker, current_model, chunks_df, retrieval_params
|
| 272 |
+
if not question.strip():
|
| 273 |
+
return ("<div style='color: black;'>Пожалуйста, введите вопрос</div>",
|
| 274 |
+
"<div style='color: black;'>Источники появятся после обработки запроса</div>",
|
| 275 |
+
"<div style='color: black;'>Чанки появятся после обработки запроса</div>")
|
| 276 |
+
|
| 277 |
+
try:
|
| 278 |
+
answer_html, sources_html, chunks_html = answer_question(
|
| 279 |
+
question, query_engine, reranker, current_model, chunks_df,
|
| 280 |
+
rerank_top_k=retrieval_params['rerank_top_k']
|
| 281 |
+
)
|
| 282 |
+
return answer_html, sources_html, chunks_html
|
| 283 |
+
|
| 284 |
+
except Exception as e:
|
| 285 |
+
log_message(f"Ошибка при ответе на вопрос: {str(e)}")
|
| 286 |
+
return (f"<div style='color: red;'>Ошибка: {str(e)}</div>",
|
| 287 |
+
"<div style='color: black;'>Источники недоступны из-за ошибки</div>",
|
| 288 |
+
"<div style='color: black;'>Чанки недоступны из-за ошибки</div>")
|
| 289 |
+
|
| 290 |
+
# NEW: Function to update retrieval parameters and recreate query engine
|
| 291 |
+
def update_retrieval_params(vector_top_k, bm25_top_k, similarity_cutoff, hybrid_top_k, rerank_top_k):
|
| 292 |
+
global query_engine, vector_index, retrieval_params
|
| 293 |
+
|
| 294 |
+
try:
|
| 295 |
+
retrieval_params['vector_top_k'] = vector_top_k
|
| 296 |
+
retrieval_params['bm25_top_k'] = bm25_top_k
|
| 297 |
+
retrieval_params['similarity_cutoff'] = similarity_cutoff
|
| 298 |
+
retrieval_params['hybrid_top_k'] = hybrid_top_k
|
| 299 |
+
retrieval_params['rerank_top_k'] = rerank_top_k
|
| 300 |
+
|
| 301 |
+
# Recreate query engine with new parameters
|
| 302 |
+
if vector_index is not None:
|
| 303 |
+
query_engine = create_query_engine(
|
| 304 |
+
vector_index=vector_index,
|
| 305 |
+
vector_top_k=vector_top_k,
|
| 306 |
+
bm25_top_k=bm25_top_k,
|
| 307 |
+
similarity_cutoff=similarity_cutoff,
|
| 308 |
+
hybrid_top_k=hybrid_top_k
|
| 309 |
+
)
|
| 310 |
+
log_message(f"Параметры поиска обновлены: vector_top_k={vector_top_k}, "
|
| 311 |
+
f"bm25_top_k={bm25_top_k}, cutoff={similarity_cutoff}, "
|
| 312 |
+
f"hybrid_top_k={hybrid_top_k}, rerank_top_k={rerank_top_k}")
|
| 313 |
+
return f"✅ Параметры обновлены"
|
| 314 |
+
else:
|
| 315 |
+
return "❌ Система не инициализирована"
|
| 316 |
+
except Exception as e:
|
| 317 |
+
error_msg = f"Ошибка обновления параметров: {str(e)}"
|
| 318 |
+
log_message(error_msg)
|
| 319 |
+
return f"❌ {error_msg}"
|
| 320 |
+
|
| 321 |
+
def retrieve_chunks(question: str, top_k: int = 20) -> list:
|
| 322 |
+
from index_retriever import rerank_nodes
|
| 323 |
+
global query_engine, reranker
|
| 324 |
+
|
| 325 |
+
if query_engine is None:
|
| 326 |
+
return []
|
| 327 |
+
|
| 328 |
+
try:
|
| 329 |
+
retrieved_nodes = query_engine.retriever.retrieve(question)
|
| 330 |
+
log_message(f"Получено {len(retrieved_nodes)} узлов")
|
| 331 |
+
|
| 332 |
+
reranked_nodes = rerank_nodes(
|
| 333 |
+
question,
|
| 334 |
+
retrieved_nodes,
|
| 335 |
+
reranker,
|
| 336 |
+
top_k=top_k,
|
| 337 |
+
min_score_threshold=0.5
|
| 338 |
+
)
|
| 339 |
+
|
| 340 |
+
chunks_data = []
|
| 341 |
+
for i, node in enumerate(reranked_nodes):
|
| 342 |
+
metadata = node.metadata if hasattr(node, 'metadata') else {}
|
| 343 |
+
chunk = {
|
| 344 |
+
'rank': i + 1,
|
| 345 |
+
'document_id': metadata.get('document_id', 'unknown'),
|
| 346 |
+
'section_id': metadata.get('section_id', ''),
|
| 347 |
+
'section_path': metadata.get('section_path', ''),
|
| 348 |
+
'section_text': metadata.get('section_text', ''),
|
| 349 |
+
'type': metadata.get('type', 'text'),
|
| 350 |
+
'table_number': metadata.get('table_number', ''),
|
| 351 |
+
'image_number': metadata.get('image_number', ''),
|
| 352 |
+
'text': node.text
|
| 353 |
+
}
|
| 354 |
+
chunks_data.append(chunk)
|
| 355 |
+
|
| 356 |
+
log_message(f"Возвращено {len(chunks_data)} чанков")
|
| 357 |
+
return chunks_data
|
| 358 |
+
|
| 359 |
+
except Exception as e:
|
| 360 |
+
log_message(f"Ошибка получения чанков: {str(e)}")
|
| 361 |
+
return []
|
| 362 |
+
|
| 363 |
+
|
| 364 |
+
def create_demo_interface(answer_question_func, switch_model_func, current_model, chunk_info=None):
|
| 365 |
+
with gr.Blocks(title="AIEXP - AI Expert для нормативной документации", theme=gr.themes.Soft()) as demo:
|
| 366 |
+
gr.api(retrieve_chunks, api_name="retrieve_chunks")
|
| 367 |
+
|
| 368 |
+
gr.Markdown("""
|
| 369 |
+
# AIEXP - Artificial Intelligence Expert
|
| 370 |
+
|
| 371 |
+
## Инструмент для работы с нормативной документацией
|
| 372 |
+
""")
|
| 373 |
+
|
| 374 |
+
with gr.Tab("Поиск по нормативным документам"):
|
| 375 |
+
gr.Markdown("### Задайте вопрос по нормативной документации")
|
| 376 |
+
|
| 377 |
+
with gr.Row():
|
| 378 |
+
with gr.Column(scale=2):
|
| 379 |
+
model_dropdown = gr.Dropdown(
|
| 380 |
+
choices=list(AVAILABLE_MODELS.keys()),
|
| 381 |
+
value=current_model,
|
| 382 |
+
label="Выберите языковую модель",
|
| 383 |
+
info="Выберите модель для генерации ответов"
|
| 384 |
+
)
|
| 385 |
+
with gr.Column(scale=1):
|
| 386 |
+
switch_btn = gr.Button("Переключить модель", variant="secondary")
|
| 387 |
+
model_status = gr.Textbox(
|
| 388 |
+
value=f"Текущая модель: {current_model}",
|
| 389 |
+
label="Статус модели",
|
| 390 |
+
interactive=False
|
| 391 |
+
)
|
| 392 |
+
|
| 393 |
+
with gr.Row():
|
| 394 |
+
with gr.Column(scale=3):
|
| 395 |
+
question_input = gr.Textbox(
|
| 396 |
+
label="Ваш вопрос к базе знаний",
|
| 397 |
+
placeholder="Введите вопрос по нормативным документам...",
|
| 398 |
+
lines=3
|
| 399 |
+
)
|
| 400 |
+
ask_btn = gr.Button("Найти ответ", variant="primary", size="lg")
|
| 401 |
+
|
| 402 |
+
gr.Examples(
|
| 403 |
+
examples=[
|
| 404 |
+
"О чем этот рисунок: ГОСТ Р 50.04.07-2022 Приложение Л. Л.1.5 Рисунок Л.2",
|
| 405 |
+
"Л.9 Формула в ГОСТ Р 50.04.07 - 2022 что и о чем там?",
|
| 406 |
+
"Какой стандарт устанавливает порядок признания протоколов испытаний продукции в об��асти использования атомной энергии?",
|
| 407 |
+
"Кто несет ответственность за организацию и проведение признания протоколов испытаний продукции?",
|
| 408 |
+
"В каких случаях могут быть признаны протоколы испытаний, проведенные лабораториями?",
|
| 409 |
+
"В какой таблице можно найти информацию о методы исследований при аттестационных испытаниях технологии термической обработки заготовок из легированных сталей? Какой документ и какой раздел?"
|
| 410 |
+
],
|
| 411 |
+
inputs=question_input
|
| 412 |
+
)
|
| 413 |
+
|
| 414 |
+
with gr.Row():
|
| 415 |
+
with gr.Column(scale=2):
|
| 416 |
+
answer_output = gr.HTML(
|
| 417 |
+
label="",
|
| 418 |
+
value=f"<div style='background-color: #2d3748; color: white; padding: 20px; border-radius: 10px; text-align: center;'>Здесь появится ответ на ваш вопрос...<br><small>Текущая модель: {current_model}</small></div>",
|
| 419 |
+
)
|
| 420 |
+
|
| 421 |
+
with gr.Column(scale=1):
|
| 422 |
+
sources_output = gr.HTML(
|
| 423 |
+
label="",
|
| 424 |
+
value="<div style='background-color: #2d3748; color: white; padding: 20px; border-radius: 10px; text-align: center;'>Здесь появятся релевантные чанки...</div>",
|
| 425 |
+
)
|
| 426 |
+
|
| 427 |
+
with gr.Column(scale=1):
|
| 428 |
+
chunks_output = gr.HTML(
|
| 429 |
+
label="Релевантные чанки",
|
| 430 |
+
value="<div style='background-color: #2d3748; color: white; padding: 20px; border-radius: 10px; text-align: center;'>Здесь появятся релевантные чанки...</div>",
|
| 431 |
+
)
|
| 432 |
+
|
| 433 |
+
# NEW TAB: Retrieval Parameters
|
| 434 |
+
with gr.Tab("⚙️ Параметры поиска"):
|
| 435 |
+
gr.Markdown("### Настройка параметров векторного поиска и переранжирования")
|
| 436 |
+
|
| 437 |
+
with gr.Row():
|
| 438 |
+
with gr.Column():
|
| 439 |
+
vector_top_k = gr.Slider(
|
| 440 |
+
minimum=10,
|
| 441 |
+
maximum=200,
|
| 442 |
+
value=50,
|
| 443 |
+
step=10,
|
| 444 |
+
label="Vector Top K",
|
| 445 |
+
info="Количество результатов из векторного поиска"
|
| 446 |
+
)
|
| 447 |
+
|
| 448 |
+
with gr.Column():
|
| 449 |
+
bm25_top_k = gr.Slider(
|
| 450 |
+
minimum=10,
|
| 451 |
+
maximum=200,
|
| 452 |
+
value=50,
|
| 453 |
+
step=10,
|
| 454 |
+
label="BM25 Top K",
|
| 455 |
+
info="Количество результатов из BM25 поиска"
|
| 456 |
+
)
|
| 457 |
+
|
| 458 |
+
with gr.Row():
|
| 459 |
+
with gr.Column():
|
| 460 |
+
similarity_cutoff = gr.Slider(
|
| 461 |
+
minimum=0.0,
|
| 462 |
+
maximum=1.0,
|
| 463 |
+
value=0.55,
|
| 464 |
+
step=0.05,
|
| 465 |
+
label="Similarity Cutoff",
|
| 466 |
+
info="Минимальный порог схожести для векторного поиска"
|
| 467 |
+
)
|
| 468 |
+
|
| 469 |
+
with gr.Column():
|
| 470 |
+
hybrid_top_k = gr.Slider(
|
| 471 |
+
minimum=10,
|
| 472 |
+
maximum=300,
|
| 473 |
+
value=100,
|
| 474 |
+
step=10,
|
| 475 |
+
label="Hybrid Top K",
|
| 476 |
+
info="Количество результатов из гибридного поиска"
|
| 477 |
+
)
|
| 478 |
+
|
| 479 |
+
with gr.Row():
|
| 480 |
+
with gr.Column():
|
| 481 |
+
rerank_top_k = gr.Slider(
|
| 482 |
+
minimum=5,
|
| 483 |
+
maximum=100,
|
| 484 |
+
value=20,
|
| 485 |
+
step=5,
|
| 486 |
+
label="Rerank Top K",
|
| 487 |
+
info="Количество результатов после переранжирования"
|
| 488 |
+
)
|
| 489 |
+
|
| 490 |
+
with gr.Column():
|
| 491 |
+
update_btn = gr.Button("Применить параметры", variant="primary")
|
| 492 |
+
update_status = gr.Textbox(
|
| 493 |
+
value="Параметры готовы к применению",
|
| 494 |
+
label="Статус",
|
| 495 |
+
interactive=False
|
| 496 |
+
)
|
| 497 |
+
|
| 498 |
+
gr.Markdown("""
|
| 499 |
+
### Рекомендации:
|
| 500 |
+
- **Vector Top K**: Увеличьте для более полного поиска по семантике (50-100)
|
| 501 |
+
- **BM25 Top K**: Увеличьте для лучшего поиска по ключевым словам (30-80)
|
| 502 |
+
- **Similarity Cutoff**: Снизьте для более мягких критериев (0.3-0.6), повысьте для строгих (0.7-0.9)
|
| 503 |
+
- **Hybrid Top K**: Объединённые результаты (100-150)
|
| 504 |
+
- **Rerank Top K**: Финальные результаты (10-30)
|
| 505 |
+
""")
|
| 506 |
+
|
| 507 |
+
update_btn.click(
|
| 508 |
+
fn=update_retrieval_params,
|
| 509 |
+
inputs=[vector_top_k, bm25_top_k, similarity_cutoff, hybrid_top_k, rerank_top_k],
|
| 510 |
+
outputs=[update_status]
|
| 511 |
+
)
|
| 512 |
+
|
| 513 |
+
# Display current parameters
|
| 514 |
+
gr.Markdown("### Текущие параметры:")
|
| 515 |
+
current_params_display = gr.Textbox(
|
| 516 |
+
value="Vector: 50 | BM25: 50 | Cutoff: 0.55 | Hybrid: 100 | Rerank: 20",
|
| 517 |
+
label="",
|
| 518 |
+
interactive=False,
|
| 519 |
+
lines=2
|
| 520 |
+
)
|
| 521 |
+
|
| 522 |
+
def display_current_params():
|
| 523 |
+
return f"""Vector Top K: {retrieval_params['vector_top_k']}
|
| 524 |
+
BM25 Top K: {retrieval_params['bm25_top_k']}
|
| 525 |
+
Similarity Cutoff: {retrieval_params['similarity_cutoff']}
|
| 526 |
+
Hybrid Top K: {retrieval_params['hybrid_top_k']}
|
| 527 |
+
Rerank Top K: {retrieval_params['rerank_top_k']}"""
|
| 528 |
+
|
| 529 |
+
# Refresh params display on tab change
|
| 530 |
+
demo.load(
|
| 531 |
+
fn=display_current_params,
|
| 532 |
+
outputs=[current_params_display]
|
| 533 |
+
)
|
| 534 |
+
|
| 535 |
+
update_btn.click(
|
| 536 |
+
fn=display_current_params,
|
| 537 |
+
outputs=[current_params_display]
|
| 538 |
+
)
|
| 539 |
+
|
| 540 |
+
# Original tab logic
|
| 541 |
+
switch_btn.click(
|
| 542 |
+
fn=switch_model_func,
|
| 543 |
+
inputs=[model_dropdown],
|
| 544 |
+
outputs=[model_status]
|
| 545 |
+
)
|
| 546 |
+
|
| 547 |
+
ask_btn.click(
|
| 548 |
+
fn=answer_question_func,
|
| 549 |
+
inputs=[question_input],
|
| 550 |
+
outputs=[answer_output, sources_output, chunks_output]
|
| 551 |
+
)
|
| 552 |
+
|
| 553 |
+
question_input.submit(
|
| 554 |
+
fn=answer_question_func,
|
| 555 |
+
inputs=[question_input],
|
| 556 |
+
outputs=[answer_output, sources_output, chunks_output]
|
| 557 |
+
)
|
| 558 |
+
return demo
|
| 559 |
+
|
| 560 |
+
|
| 561 |
+
query_engine = None
|
| 562 |
+
chunks_df = None
|
| 563 |
+
reranker = None
|
| 564 |
+
vector_index = None
|
| 565 |
+
current_model = DEFAULT_MODEL
|
| 566 |
+
|
| 567 |
+
def main_switch_model(model_name):
|
| 568 |
+
global query_engine, vector_index, current_model
|
| 569 |
+
|
| 570 |
+
new_query_engine, status_message = switch_model(model_name, vector_index)
|
| 571 |
+
if new_query_engine:
|
| 572 |
+
query_engine = new_query_engine
|
| 573 |
+
current_model = model_name
|
| 574 |
+
|
| 575 |
+
return status_message
|
| 576 |
+
|
| 577 |
+
|
| 578 |
+
|
| 579 |
+
|
| 580 |
+
def main():
|
| 581 |
+
global query_engine, chunks_df, reranker, vector_index, current_model
|
| 582 |
+
GOOGLE_API_KEY = os.getenv("GOOGLE_API_KEY", "")
|
| 583 |
+
if GOOGLE_API_KEY:
|
| 584 |
+
log_message("Использование Google API для модели генерации текста")
|
| 585 |
+
else:
|
| 586 |
+
log_message("Google API ключ не найден, использование локальной модели")
|
| 587 |
+
log_message("Запуск AIEXP - AI Expert для нормативной документации")
|
| 588 |
+
query_engine, chunks_df, reranker, vector_index, chunk_info = initialize_system(
|
| 589 |
+
repo_id=HF_REPO_ID,
|
| 590 |
+
hf_token=HF_TOKEN,
|
| 591 |
+
download_dir=DOWNLOAD_DIR,
|
| 592 |
+
json_files_dir=JSON_FILES_DIR,
|
| 593 |
+
table_data_dir=TABLE_DATA_DIR,
|
| 594 |
+
image_data_dir=IMAGE_DATA_DIR,
|
| 595 |
+
use_json_instead_csv=True,
|
| 596 |
+
)
|
| 597 |
+
|
| 598 |
+
if query_engine:
|
| 599 |
+
log_message("Запуск веб-интерфейса")
|
| 600 |
+
demo = create_demo_interface(
|
| 601 |
+
answer_question_func=main_answer_question,
|
| 602 |
+
switch_model_func=main_switch_model,
|
| 603 |
+
current_model=current_model,
|
| 604 |
+
chunk_info=chunk_info
|
| 605 |
+
)
|
| 606 |
+
demo.api = "retrieve_chunks"
|
| 607 |
+
demo.queue()
|
| 608 |
+
|
| 609 |
+
demo.launch(
|
| 610 |
+
server_name="0.0.0.0",
|
| 611 |
+
server_port=7860,
|
| 612 |
+
share=True,
|
| 613 |
+
debug=False
|
| 614 |
+
)
|
| 615 |
+
else:
|
| 616 |
+
log_message("Невозможно запустить приложение из-за ошибки инициализации")
|
| 617 |
+
sys.exit(1)
|
| 618 |
+
|
| 619 |
+
if __name__ == "__main__":
|
| 620 |
main()
|