Spaces:
Sleeping
Sleeping
Commit
·
147e01b
1
Parent(s):
9160af0
new version of rag
Browse files
app.py
CHANGED
|
@@ -1,354 +1,392 @@
|
|
| 1 |
-
import gradio as gr
|
| 2 |
import os
|
| 3 |
-
|
| 4 |
-
|
| 5 |
-
from
|
| 6 |
-
|
| 7 |
-
from
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 8 |
import sys
|
| 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 |
-
return html
|
| 41 |
|
| 42 |
-
def
|
| 43 |
-
|
| 44 |
-
section_id = chunk.get('section_id', 'unknown')
|
| 45 |
-
doc_type = chunk.get('type', 'text')
|
| 46 |
|
| 47 |
-
|
| 48 |
-
|
| 49 |
-
if not str(table_num).startswith('№'):
|
| 50 |
-
table_num = f"№{table_num}"
|
| 51 |
-
return f"таблица {table_num}"
|
| 52 |
|
| 53 |
-
|
| 54 |
-
|
| 55 |
-
|
| 56 |
-
|
| 57 |
-
return f"рисунок {image_num}"
|
| 58 |
|
| 59 |
-
|
| 60 |
-
|
| 61 |
-
elif section_id and section_id != 'unknown':
|
| 62 |
-
return section_id
|
| 63 |
|
| 64 |
-
|
| 65 |
-
|
| 66 |
-
def get_formatted_content(chunk):
|
| 67 |
-
document_id = chunk.get('document_id', 'unknown')
|
| 68 |
-
section_path = chunk.get('section_path', '')
|
| 69 |
-
section_id = chunk.get('section_id', 'unknown')
|
| 70 |
-
section_text = chunk.get('section_text', '')
|
| 71 |
-
parent_section = chunk.get('parent_section', '')
|
| 72 |
-
parent_title = chunk.get('parent_title', '')
|
| 73 |
-
level = chunk.get('level', '')
|
| 74 |
-
chunk_text = chunk.get('chunk_text', '')
|
| 75 |
-
doc_type = chunk.get('type', 'text')
|
| 76 |
|
| 77 |
-
|
| 78 |
-
|
| 79 |
-
|
| 80 |
-
|
| 81 |
-
|
| 82 |
-
else:
|
| 83 |
-
current_section = section_path if section_path else section_id
|
| 84 |
-
clean_text = chunk_text
|
| 85 |
-
if section_text and chunk_text.startswith(section_text):
|
| 86 |
-
section_title = section_text
|
| 87 |
-
elif chunk_text.startswith(f"{current_section} "):
|
| 88 |
-
clean_text = chunk_text[len(f"{current_section} "):].strip()
|
| 89 |
-
section_title = section_text if section_text else f"{current_section} {clean_text.split('.')[0] if '.' in clean_text else clean_text[:50]}"
|
| 90 |
-
else:
|
| 91 |
-
section_title = section_text if section_text else current_section
|
| 92 |
-
|
| 93 |
-
return f"В разделе {current_section} в документе {document_id}, пункт {section_title}: {clean_text}"
|
| 94 |
-
|
| 95 |
-
def initialize_system(repo_id, hf_token, download_dir, chunks_filename=None,
|
| 96 |
-
json_files_dir=None, table_data_dir=None, image_data_dir=None,
|
| 97 |
-
use_json_instead_csv=False):
|
| 98 |
-
try:
|
| 99 |
-
from documents_prep import process_documents_with_chunking
|
| 100 |
-
log_message("Инициализация системы")
|
| 101 |
-
os.makedirs(download_dir, exist_ok=True)
|
| 102 |
-
from config import CHUNK_SIZE, CHUNK_OVERLAP
|
| 103 |
-
from llama_index.core.text_splitter import TokenTextSplitter
|
| 104 |
-
|
| 105 |
-
embed_model = get_embedding_model()
|
| 106 |
-
llm = get_llm_model(DEFAULT_MODEL)
|
| 107 |
-
reranker = get_reranker_model()
|
| 108 |
-
|
| 109 |
-
Settings.embed_model = embed_model
|
| 110 |
-
Settings.llm = llm
|
| 111 |
-
Settings.text_splitter = TokenTextSplitter(
|
| 112 |
-
chunk_size=CHUNK_SIZE,
|
| 113 |
-
chunk_overlap=CHUNK_OVERLAP,
|
| 114 |
-
separator=" ",
|
| 115 |
-
backup_separators=["\n", ".", "!", "?"]
|
| 116 |
-
)
|
| 117 |
-
|
| 118 |
-
log_message(f"Configured chunk size: {CHUNK_SIZE} tokens")
|
| 119 |
-
log_message(f"Configured chunk overlap: {CHUNK_OVERLAP} tokens")
|
| 120 |
|
| 121 |
-
|
| 122 |
-
|
| 123 |
-
|
| 124 |
-
|
| 125 |
-
|
| 126 |
-
|
| 127 |
-
json_documents, json_chunk_info = load_json_documents(repo_id, hf_token, json_files_dir, download_dir)
|
| 128 |
-
all_documents.extend(json_documents)
|
| 129 |
-
chunk_info.extend(json_chunk_info)
|
| 130 |
-
else:
|
| 131 |
-
if chunks_filename:
|
| 132 |
-
log_message("Загружаем данные из CSV")
|
| 133 |
-
csv_documents, chunks_df = load_csv_chunks(repo_id, hf_token, chunks_filename, download_dir)
|
| 134 |
-
all_documents.extend(csv_documents)
|
| 135 |
-
|
| 136 |
-
if table_data_dir:
|
| 137 |
-
log_message("Добавляю табличные данные")
|
| 138 |
-
table_documents = load_table_data(repo_id, hf_token, table_data_dir)
|
| 139 |
-
log_message(f"Загружено {len(table_documents)} табличных документов")
|
| 140 |
-
|
| 141 |
-
# Process table documents through chunking
|
| 142 |
-
chunked_table_docs, table_chunk_info = process_documents_with_chunking(table_documents)
|
| 143 |
-
all_documents.extend(chunked_table_docs)
|
| 144 |
-
chunk_info.extend(table_chunk_info)
|
| 145 |
-
|
| 146 |
-
if image_data_dir:
|
| 147 |
-
log_message("Добавляю данные изображений")
|
| 148 |
-
image_documents = load_image_data(repo_id, hf_token, image_data_dir)
|
| 149 |
-
log_message(f"Загружено {len(image_documents)} документов изображений")
|
| 150 |
-
|
| 151 |
-
# Process image documents through chunking
|
| 152 |
-
chunked_image_docs, image_chunk_info = process_documents_with_chunking(image_documents)
|
| 153 |
-
all_documents.extend(chunked_image_docs)
|
| 154 |
-
chunk_info.extend(image_chunk_info)
|
| 155 |
|
| 156 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 157 |
|
| 158 |
-
|
| 159 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 160 |
|
| 161 |
-
|
| 162 |
-
return query_engine, chunks_df, reranker, vector_index, chunk_info
|
| 163 |
|
| 164 |
except Exception as e:
|
| 165 |
-
|
| 166 |
-
return
|
| 167 |
|
| 168 |
-
def switch_model(model_name, vector_index):
|
| 169 |
-
from llama_index.core import Settings
|
| 170 |
-
from index_retriever import create_query_engine
|
| 171 |
-
|
| 172 |
try:
|
| 173 |
-
|
| 174 |
-
|
| 175 |
new_llm = get_llm_model(model_name)
|
| 176 |
Settings.llm = new_llm
|
| 177 |
|
| 178 |
-
|
| 179 |
-
|
| 180 |
-
log_message(f"Модель успешно переключена на: {model_name}")
|
| 181 |
-
return new_query_engine, f"✅ Модель переключена на: {model_name}"
|
| 182 |
-
else:
|
| 183 |
-
return None, "❌ Ошибка: система не инициализирована"
|
| 184 |
-
|
| 185 |
-
except Exception as e:
|
| 186 |
-
error_msg = f"Ошибка переключения модели: {str(e)}"
|
| 187 |
-
log_message(error_msg)
|
| 188 |
-
return None, f"❌ {error_msg}"
|
| 189 |
-
|
| 190 |
-
def main_answer_question(question):
|
| 191 |
-
global query_engine, reranker, current_model, chunks_df
|
| 192 |
-
if not question.strip():
|
| 193 |
-
return ("<div style='color: black;'>Пожалуйста, введите вопрос</div>",
|
| 194 |
-
"<div style='color: black;'>Источники появятся после обработки запроса</div>",
|
| 195 |
-
"<div style='color: black;'>Чанки появятся после обработки запроса</div>")
|
| 196 |
-
|
| 197 |
-
try:
|
| 198 |
-
# Call the answer_question function which returns 3 values
|
| 199 |
-
answer_html, sources_html, chunks_html = answer_question(question, query_engine, reranker, current_model, chunks_df)
|
| 200 |
-
return answer_html, sources_html, chunks_html
|
| 201 |
-
|
| 202 |
except Exception as e:
|
| 203 |
-
|
| 204 |
-
return (f"<div style='color: red;'>Ошибка: {str(e)}</div>",
|
| 205 |
-
"<div style='color: black;'>Источники недоступны из-за ошибки</div>",
|
| 206 |
-
"<div style='color: black;'>Чанки недоступны из-за ошибки</div>")
|
| 207 |
-
|
| 208 |
-
|
| 209 |
-
|
| 210 |
-
def create_demo_interface(answer_question_func, switch_model_func, current_model, chunk_info=None):
|
| 211 |
-
with gr.Blocks(title="AIEXP - AI Expert для нормативной документации", theme=gr.themes.Soft()) as demo:
|
| 212 |
-
|
| 213 |
-
gr.Markdown("""
|
| 214 |
-
# AIEXP - Artificial Intelligence Expert
|
| 215 |
-
|
| 216 |
-
## Инструмент для работы с нормативной документацией
|
| 217 |
-
""")
|
| 218 |
-
|
| 219 |
-
with gr.Tab("Поиск по нормативным документам"):
|
| 220 |
-
gr.Markdown("### Задайте вопрос по нормативной документации")
|
| 221 |
-
|
| 222 |
-
with gr.Row():
|
| 223 |
-
with gr.Column(scale=2):
|
| 224 |
-
model_dropdown = gr.Dropdown(
|
| 225 |
-
choices=list(AVAILABLE_MODELS.keys()),
|
| 226 |
-
value=current_model,
|
| 227 |
-
label="Выберите языковую модель",
|
| 228 |
-
info="Выберите модель для генерации ответов"
|
| 229 |
-
)
|
| 230 |
-
with gr.Column(scale=1):
|
| 231 |
-
switch_btn = gr.Button("Переключить модель", variant="secondary")
|
| 232 |
-
model_status = gr.Textbox(
|
| 233 |
-
value=f"Текущая модель: {current_model}",
|
| 234 |
-
label="Статус модели",
|
| 235 |
-
interactive=False
|
| 236 |
-
)
|
| 237 |
-
|
| 238 |
-
with gr.Row():
|
| 239 |
-
with gr.Column(scale=3):
|
| 240 |
-
question_input = gr.Textbox(
|
| 241 |
-
label="Ваш вопрос к базе знаний",
|
| 242 |
-
placeholder="Введите вопрос по нормативным документам...",
|
| 243 |
-
lines=3
|
| 244 |
-
)
|
| 245 |
-
ask_btn = gr.Button("Найти ответ", variant="primary", size="lg")
|
| 246 |
-
|
| 247 |
-
gr.Examples(
|
| 248 |
-
examples=[
|
| 249 |
-
"О чем этот рисунок: ГОСТ Р 50.04.07-2022 Приложение Л. Л.1.5 Рисунок Л.2",
|
| 250 |
-
"Л.9 Формула в ГОСТ Р 50.04.07 - 2022 что и о чем там?",
|
| 251 |
-
"Какой стандарт устанавливает порядок признания протоколов испытаний продукции в области использования атомной энергии?",
|
| 252 |
-
"Кто несет ответственность за организацию и проведение признания протоколов испытаний продукции?",
|
| 253 |
-
"В каких случаях могут быть признаны протоколы испытаний, проведенные лабораториями?",
|
| 254 |
-
"В какой таблице можно найти информацию о методы исследований при аттестационных испытаниях технологии термической обработки заготовок из легированных сталей? Какой документ и какой раздел?"
|
| 255 |
-
],
|
| 256 |
-
inputs=question_input
|
| 257 |
-
)
|
| 258 |
-
|
| 259 |
-
with gr.Row():
|
| 260 |
-
with gr.Column(scale=2):
|
| 261 |
-
answer_output = gr.HTML(
|
| 262 |
-
label="",
|
| 263 |
-
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>",
|
| 264 |
-
)
|
| 265 |
-
|
| 266 |
-
with gr.Column(scale=1):
|
| 267 |
-
sources_output = gr.HTML(
|
| 268 |
-
label="",
|
| 269 |
-
value="<div style='background-color: #2d3748; color: white; padding: 20px; border-radius: 10px; text-align: center;'>Здесь появятся релевантные чанки...</div>",
|
| 270 |
-
)
|
| 271 |
-
|
| 272 |
-
with gr.Column(scale=1):
|
| 273 |
-
chunks_output = gr.HTML(
|
| 274 |
-
label="Релевантные чанки",
|
| 275 |
-
value="<div style='background-color: #2d3748; color: white; padding: 20px; border-radius: 10px; text-align: center;'>Здесь появятся релевантные чанки...</div>",
|
| 276 |
-
)
|
| 277 |
-
|
| 278 |
-
switch_btn.click(
|
| 279 |
-
fn=switch_model_func,
|
| 280 |
-
inputs=[model_dropdown],
|
| 281 |
-
outputs=[model_status]
|
| 282 |
-
)
|
| 283 |
-
|
| 284 |
-
ask_btn.click(
|
| 285 |
-
fn=answer_question_func,
|
| 286 |
-
inputs=[question_input],
|
| 287 |
-
outputs=[answer_output, sources_output, chunks_output]
|
| 288 |
-
)
|
| 289 |
-
|
| 290 |
-
question_input.submit(
|
| 291 |
-
fn=answer_question_func,
|
| 292 |
-
inputs=[question_input],
|
| 293 |
-
outputs=[answer_output, sources_output, chunks_output]
|
| 294 |
-
)
|
| 295 |
-
return demo
|
| 296 |
-
|
| 297 |
|
| 298 |
query_engine = None
|
| 299 |
-
chunks_df = None
|
| 300 |
-
reranker = None
|
| 301 |
vector_index = None
|
|
|
|
|
|
|
| 302 |
current_model = DEFAULT_MODEL
|
| 303 |
|
| 304 |
def main_answer_question(question):
|
| 305 |
-
global query_engine
|
| 306 |
-
|
| 307 |
-
question, query_engine, reranker, current_model, chunks_df
|
| 308 |
-
)
|
| 309 |
-
return answer_html, sources_html, chunks_html
|
| 310 |
|
| 311 |
def main_switch_model(model_name):
|
| 312 |
-
global query_engine, vector_index, current_model
|
| 313 |
-
|
| 314 |
-
new_query_engine, status_message = switch_model(model_name, vector_index)
|
| 315 |
if new_query_engine:
|
| 316 |
query_engine = new_query_engine
|
| 317 |
current_model = model_name
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 318 |
|
| 319 |
-
return
|
| 320 |
|
| 321 |
def main():
|
| 322 |
-
global query_engine,
|
| 323 |
|
| 324 |
-
|
| 325 |
|
| 326 |
-
query_engine,
|
| 327 |
-
repo_id=HF_REPO_ID,
|
| 328 |
-
hf_token=HF_TOKEN,
|
| 329 |
-
download_dir=DOWNLOAD_DIR,
|
| 330 |
-
json_files_dir=JSON_FILES_DIR,
|
| 331 |
-
table_data_dir=TABLE_DATA_DIR,
|
| 332 |
-
image_data_dir=IMAGE_DATA_DIR,
|
| 333 |
-
use_json_instead_csv=True,
|
| 334 |
-
)
|
| 335 |
|
| 336 |
if query_engine:
|
| 337 |
-
|
| 338 |
-
demo =
|
| 339 |
-
answer_question_func=main_answer_question,
|
| 340 |
-
switch_model_func=main_switch_model,
|
| 341 |
-
current_model=current_model,
|
| 342 |
-
chunk_info=chunk_info
|
| 343 |
-
)
|
| 344 |
demo.launch(
|
| 345 |
server_name="0.0.0.0",
|
| 346 |
server_port=7860,
|
| 347 |
-
share=True
|
| 348 |
-
debug=False
|
| 349 |
)
|
| 350 |
else:
|
| 351 |
-
|
| 352 |
sys.exit(1)
|
| 353 |
|
| 354 |
if __name__ == "__main__":
|
|
|
|
|
|
|
| 1 |
import os
|
| 2 |
+
import json
|
| 3 |
+
import zipfile
|
| 4 |
+
from typing import List, Dict, Any
|
| 5 |
+
import pandas as pd
|
| 6 |
+
from huggingface_hub import hf_hub_download, list_repo_files
|
| 7 |
+
from llama_index.core import Document, VectorStoreIndex, KeywordTableIndex, Settings
|
| 8 |
+
from llama_index.core.retrievers import VectorIndexRetriever, QueryFusionRetriever
|
| 9 |
+
from llama_index.retrievers.bm25 import BM25Retriever
|
| 10 |
+
from llama_index.core.query_engine import RetrieverQueryEngine
|
| 11 |
+
from llama_index.core.response_synthesizers import get_response_synthesizer, ResponseMode
|
| 12 |
+
from llama_index.core.text_splitter import SentenceSplitter
|
| 13 |
+
from sentence_transformers import SentenceTransformer
|
| 14 |
+
import gradio as gr
|
| 15 |
import sys
|
| 16 |
+
|
| 17 |
+
GOOGLE_API_KEY = os.getenv('GOOGLE_API_KEY')
|
| 18 |
+
OPENAI_API_KEY = os.getenv('OPENAI_API_KEY')
|
| 19 |
+
HF_REPO_ID = "MrSimple01/AIEXP_RAG_FILES"
|
| 20 |
+
HF_TOKEN = os.getenv('HF_TOKEN')
|
| 21 |
+
|
| 22 |
+
AVAILABLE_MODELS = {
|
| 23 |
+
"Gemini 2.5 Flash": {
|
| 24 |
+
"provider": "google",
|
| 25 |
+
"model_name": "gemini-2.5-flash",
|
| 26 |
+
"api_key": GOOGLE_API_KEY
|
| 27 |
+
},
|
| 28 |
+
"Gemini 2.5 Pro": {
|
| 29 |
+
"provider": "google",
|
| 30 |
+
"model_name": "gemini-2.5-pro",
|
| 31 |
+
"api_key": GOOGLE_API_KEY
|
| 32 |
+
},
|
| 33 |
+
"GPT-4o": {
|
| 34 |
+
"provider": "openai",
|
| 35 |
+
"model_name": "gpt-4o",
|
| 36 |
+
"api_key": OPENAI_API_KEY
|
| 37 |
+
},
|
| 38 |
+
"GPT-4o Mini": {
|
| 39 |
+
"provider": "openai",
|
| 40 |
+
"model_name": "gpt-4o-mini",
|
| 41 |
+
"api_key": OPENAI_API_KEY
|
| 42 |
+
},
|
| 43 |
+
"GPT-5": {
|
| 44 |
+
"provider": "openai",
|
| 45 |
+
"model_name": "gpt-5",
|
| 46 |
+
"api_key": OPENAI_API_KEY
|
| 47 |
+
}
|
| 48 |
+
}
|
| 49 |
+
|
| 50 |
+
DEFAULT_MODEL = "Gemini 2.5 Flash"
|
| 51 |
+
DOWNLOAD_DIR = "rag_files"
|
| 52 |
+
JSON_FILES_DIR = "JSON"
|
| 53 |
+
TABLE_DATA_DIR = "Табличные данные_JSON"
|
| 54 |
+
IMAGE_DATA_DIR = "Изображения"
|
| 55 |
+
CHUNK_SIZE = 512
|
| 56 |
+
CHUNK_OVERLAP = 50
|
| 57 |
+
TABLE_MAX_ROWS_PER_CHUNK = 30
|
| 58 |
+
|
| 59 |
+
os.makedirs(DOWNLOAD_DIR, exist_ok=True)
|
| 60 |
+
|
| 61 |
+
def get_llm_model(model_name):
|
| 62 |
+
config = AVAILABLE_MODELS[model_name]
|
| 63 |
+
if config["provider"] == "google":
|
| 64 |
+
from llama_index.llms.gemini import Gemini
|
| 65 |
+
return Gemini(model=config["model_name"], api_key=config["api_key"])
|
| 66 |
+
else:
|
| 67 |
+
from llama_index.llms.openai import OpenAI
|
| 68 |
+
return OpenAI(model=config["model_name"], api_key=config["api_key"])
|
| 69 |
+
|
| 70 |
+
def get_embedding_model():
|
| 71 |
+
from llama_index.embeddings.huggingface import HuggingFaceEmbedding
|
| 72 |
+
return HuggingFaceEmbedding(model_name="sentence-transformers/all-MiniLM-L6-v2")
|
| 73 |
+
|
| 74 |
+
def list_zip_files_in_repo(repo_id: str) -> List[str]:
|
| 75 |
+
files = list_repo_files(repo_id, token=HF_TOKEN)
|
| 76 |
+
return [f for f in files if f.startswith(JSON_FILES_DIR) and f.endswith('.zip')]
|
| 77 |
+
|
| 78 |
+
def download_file_from_hf(repo_id: str, path_in_repo: str, dest_dir: str) -> str:
|
| 79 |
+
local_path = hf_hub_download(repo_id=repo_id, filename=path_in_repo, repo_type="dataset", token=HF_TOKEN)
|
| 80 |
+
base = os.path.basename(local_path)
|
| 81 |
+
dst = os.path.join(dest_dir, base)
|
| 82 |
+
if local_path != dst:
|
| 83 |
+
try:
|
| 84 |
+
with open(local_path, 'rb') as r, open(dst, 'wb') as w:
|
| 85 |
+
w.write(r.read())
|
| 86 |
+
except Exception:
|
| 87 |
+
pass
|
| 88 |
+
return dst
|
| 89 |
+
|
| 90 |
+
def read_jsons_from_zip(zip_path: str) -> List[Dict[str, Any]]:
|
| 91 |
+
docs = []
|
| 92 |
+
with zipfile.ZipFile(zip_path, 'r') as z:
|
| 93 |
+
for name in z.namelist():
|
| 94 |
+
if name.lower().endswith('.json'):
|
| 95 |
+
with z.open(name) as f:
|
| 96 |
+
try:
|
| 97 |
+
text = f.read().decode('utf-8')
|
| 98 |
+
data = json.loads(text)
|
| 99 |
+
docs.append(data)
|
| 100 |
+
except Exception as e:
|
| 101 |
+
print(f"Failed to load {name} in {zip_path}: {e}")
|
| 102 |
+
return docs
|
| 103 |
+
|
| 104 |
+
def chunk_text_field(text: str, doc_meta: Dict[str, Any], splitter: SentenceSplitter) -> List[Document]:
|
| 105 |
+
nodes = splitter.split_text(text)
|
| 106 |
+
chunks = []
|
| 107 |
+
for i, node_text in enumerate(nodes):
|
| 108 |
+
md = dict(doc_meta)
|
| 109 |
+
md.update({
|
| 110 |
+
'chunk_id': f"{md.get('document_id','unknown')}_text_{i}",
|
| 111 |
+
'chunk_type': 'text'
|
| 112 |
+
})
|
| 113 |
+
chunks.append(Document(text=node_text, metadata=md))
|
| 114 |
+
return chunks
|
| 115 |
+
|
| 116 |
+
def chunk_table(table: Dict[str, Any], table_meta: Dict[str, Any], max_rows: int = TABLE_MAX_ROWS_PER_CHUNK) -> List[Document]:
|
| 117 |
+
headers = table.get('headers') or []
|
| 118 |
+
rows = table.get('data') or []
|
| 119 |
+
if not rows:
|
| 120 |
+
text = table.get('table_description') or table.get('table_title') or ''
|
| 121 |
+
md = {**table_meta, 'chunk_type': 'table', 'chunk_id': f"{table_meta.get('document_id')}_table_single"}
|
| 122 |
+
return [Document(text=text, metadata=md)]
|
| 123 |
+
|
| 124 |
+
chunks = []
|
| 125 |
+
for i in range(0, len(rows), max_rows):
|
| 126 |
+
block = rows[i:i+max_rows]
|
| 127 |
+
lines = []
|
| 128 |
+
lines.append(f"Table {table_meta.get('table_number','?')} - {table_meta.get('table_title','')}")
|
| 129 |
+
lines.append(f"Headers: {headers}")
|
| 130 |
+
for r in block:
|
| 131 |
+
row_items = [f"{k}: {v}" for k, v in r.items()]
|
| 132 |
+
lines.append(" | ".join(row_items))
|
| 133 |
+
chunk_text = "\n".join(lines)
|
| 134 |
+
md = dict(table_meta)
|
| 135 |
+
md.update({'chunk_type': 'table', 'chunk_id': f"{table_meta.get('document_id')}_table_{i // max_rows}"})
|
| 136 |
+
chunks.append(Document(text=chunk_text, metadata=md))
|
| 137 |
+
return chunks
|
| 138 |
+
|
| 139 |
+
def chunk_image(image_entry: Dict[str, Any], image_meta: Dict[str, Any]) -> Document:
|
| 140 |
+
txt = f"Image: {image_entry.get('Название изображения') or image_entry.get('title','')}. "
|
| 141 |
+
txt += f"Описание: {image_entry.get('Описание изображение') or image_entry.get('description','')}. "
|
| 142 |
+
txt += f"Файл: {image_entry.get('Файл изображения') or image_entry.get('file','')}."
|
| 143 |
+
md = dict(image_meta)
|
| 144 |
+
md.update({'chunk_type': 'image', 'chunk_id': f"{image_meta.get('document_id')}_image_{image_entry.get('№ Изображения','0')}"})
|
| 145 |
+
return Document(text=txt, metadata=md)
|
| 146 |
+
|
| 147 |
+
def build_chunks_from_repo(repo_id: str) -> List[Document]:
|
| 148 |
+
zip_paths = list_zip_files_in_repo(repo_id)
|
| 149 |
+
print(f"Found {len(zip_paths)} zip files under {JSON_FILES_DIR} in repo {repo_id}")
|
| 150 |
+
|
| 151 |
+
splitter = SentenceSplitter(chunk_size=CHUNK_SIZE, chunk_overlap=CHUNK_OVERLAP)
|
| 152 |
+
all_chunks = []
|
| 153 |
+
|
| 154 |
+
for remote_path in zip_paths:
|
| 155 |
+
print(f"Downloading {remote_path}...")
|
| 156 |
+
local_zip = download_file_from_hf(repo_id, remote_path, DOWNLOAD_DIR)
|
| 157 |
+
print(f"Parsing {local_zip}...")
|
| 158 |
+
json_docs = read_jsons_from_zip(local_zip)
|
| 159 |
+
for doc in json_docs:
|
| 160 |
+
doc_meta = doc.get('document_metadata', {})
|
| 161 |
+
doc_id = doc_meta.get('document_id') or doc_meta.get('document_name') or 'unknown_doc'
|
| 162 |
+
base_meta = {'document_id': doc_id, 'document_name': doc_meta.get('document_name','')}
|
| 163 |
+
|
| 164 |
+
for sec in doc.get('sections', []):
|
| 165 |
+
sec_meta = dict(base_meta)
|
| 166 |
+
sec_meta.update({'section_id': sec.get('section_id'), 'section_title': None})
|
| 167 |
+
text = sec.get('section_text') or sec.get('text') or ''
|
| 168 |
+
if text and text.strip():
|
| 169 |
+
chunks = chunk_text_field(text, sec_meta, splitter)
|
| 170 |
+
all_chunks.extend(chunks)
|
| 171 |
+
|
| 172 |
+
for sheet in doc.get('sheets', []) + doc.get('tables', []) if (doc.get('sheets') or doc.get('tables')) else []:
|
| 173 |
+
table_meta = dict(base_meta)
|
| 174 |
+
table_meta.update({
|
| 175 |
+
'sheet_name': sheet.get('sheet_name') or sheet.get('table_title'),
|
| 176 |
+
'section': sheet.get('section'),
|
| 177 |
+
'table_number': sheet.get('table_number'),
|
| 178 |
+
'table_title': sheet.get('table_title')
|
| 179 |
+
})
|
| 180 |
+
table_chunks = chunk_table(sheet, table_meta, max_rows=TABLE_MAX_ROWS_PER_CHUNK)
|
| 181 |
+
all_chunks.extend(table_chunks)
|
| 182 |
+
|
| 183 |
+
for img in doc.get('images', []) or doc.get('image_data', []) or doc.get('image_entries', []):
|
| 184 |
+
img_meta = dict(base_meta)
|
| 185 |
+
chunk = chunk_image(img, img_meta)
|
| 186 |
+
all_chunks.append(chunk)
|
| 187 |
+
|
| 188 |
+
print(f"Built total {len(all_chunks)} chunks")
|
| 189 |
+
return all_chunks
|
| 190 |
+
|
| 191 |
+
def create_hybrid_index(documents):
|
| 192 |
+
print("Creating vector index...")
|
| 193 |
+
vector_index = VectorStoreIndex.from_documents(documents)
|
| 194 |
|
| 195 |
+
print("Creating keyword index...")
|
| 196 |
+
keyword_index = KeywordTableIndex.from_documents(documents)
|
| 197 |
|
| 198 |
+
return vector_index, keyword_index
|
| 199 |
+
|
| 200 |
+
def create_fusion_retriever(vector_index, keyword_index, documents):
|
| 201 |
+
vector_retriever = VectorIndexRetriever(index=vector_index, similarity_top_k=5)
|
| 202 |
+
|
| 203 |
+
bm25_retriever = BM25Retriever.from_defaults(
|
| 204 |
+
docstore=vector_index.docstore,
|
| 205 |
+
similarity_top_k=5
|
| 206 |
+
)
|
| 207 |
+
|
| 208 |
+
fusion_retriever = QueryFusionRetriever(
|
| 209 |
+
[vector_retriever, bm25_retriever],
|
| 210 |
+
similarity_top_k=5,
|
| 211 |
+
num_queries=1,
|
| 212 |
+
mode="reciprocal_rerank",
|
| 213 |
+
use_async=False
|
| 214 |
+
)
|
| 215 |
+
|
| 216 |
+
return fusion_retriever
|
| 217 |
+
|
| 218 |
+
def create_query_engine(vector_index, keyword_index, documents):
|
| 219 |
+
fusion_retriever = create_fusion_retriever(vector_index, keyword_index, documents)
|
| 220 |
+
|
| 221 |
+
response_synthesizer = get_response_synthesizer(
|
| 222 |
+
response_mode=ResponseMode.COMPACT,
|
| 223 |
+
use_async=False
|
| 224 |
+
)
|
| 225 |
+
|
| 226 |
+
query_engine = RetrieverQueryEngine(
|
| 227 |
+
retriever=fusion_retriever,
|
| 228 |
+
response_synthesizer=response_synthesizer
|
| 229 |
+
)
|
| 230 |
|
| 231 |
+
return query_engine
|
|
|
|
| 232 |
|
| 233 |
+
def initialize_system():
|
| 234 |
+
print("Initializing system...")
|
|
|
|
|
|
|
| 235 |
|
| 236 |
+
embed_model = get_embedding_model()
|
| 237 |
+
llm = get_llm_model(DEFAULT_MODEL)
|
|
|
|
|
|
|
|
|
|
| 238 |
|
| 239 |
+
Settings.embed_model = embed_model
|
| 240 |
+
Settings.llm = llm
|
| 241 |
+
Settings.chunk_size = CHUNK_SIZE
|
| 242 |
+
Settings.chunk_overlap = CHUNK_OVERLAP
|
|
|
|
| 243 |
|
| 244 |
+
print("Loading documents...")
|
| 245 |
+
documents = build_chunks_from_repo(HF_REPO_ID)
|
|
|
|
|
|
|
| 246 |
|
| 247 |
+
print("Creating indices...")
|
| 248 |
+
vector_index, keyword_index = create_hybrid_index(documents)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 249 |
|
| 250 |
+
print("Creating query engine...")
|
| 251 |
+
query_engine = create_query_engine(vector_index, keyword_index, documents)
|
| 252 |
+
|
| 253 |
+
print("System initialized successfully!")
|
| 254 |
+
return query_engine, vector_index, keyword_index, documents
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 255 |
|
| 256 |
+
def answer_question(question, query_engine):
|
| 257 |
+
if not question.strip():
|
| 258 |
+
return "<div style='color: black;'>Please enter a question</div>"
|
| 259 |
+
|
| 260 |
+
try:
|
| 261 |
+
response = query_engine.query(question)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 262 |
|
| 263 |
+
answer_html = f"""
|
| 264 |
+
<div style='background-color: #f8f9fa; padding: 20px; border-radius: 10px; color: black;'>
|
| 265 |
+
<h3 style='color: #007bff;'>Answer:</h3>
|
| 266 |
+
<p>{response.response}</p>
|
| 267 |
+
</div>
|
| 268 |
+
"""
|
| 269 |
|
| 270 |
+
sources_html = "<div style='background-color: #e9ecef; padding: 15px; border-radius: 8px; color: black;'>"
|
| 271 |
+
sources_html += "<h4>Sources:</h4>"
|
| 272 |
+
for i, node in enumerate(response.source_nodes):
|
| 273 |
+
sources_html += f"""
|
| 274 |
+
<div style='margin: 10px 0; padding: 10px; background-color: white; border-left: 3px solid #007bff;'>
|
| 275 |
+
<strong>Document {i+1}:</strong> {node.metadata.get('document_id', 'unknown')}<br>
|
| 276 |
+
<strong>Score:</strong> {node.score:.3f}<br>
|
| 277 |
+
<strong>Text:</strong> {node.text[:200]}...
|
| 278 |
+
</div>
|
| 279 |
+
"""
|
| 280 |
+
sources_html += "</div>"
|
| 281 |
|
| 282 |
+
return answer_html, sources_html
|
|
|
|
| 283 |
|
| 284 |
except Exception as e:
|
| 285 |
+
error_html = f"<div style='color: red;'>Error: {str(e)}</div>"
|
| 286 |
+
return error_html, error_html
|
| 287 |
|
| 288 |
+
def switch_model(model_name, vector_index, keyword_index, documents):
|
|
|
|
|
|
|
|
|
|
| 289 |
try:
|
| 290 |
+
print(f"Switching to model: {model_name}")
|
|
|
|
| 291 |
new_llm = get_llm_model(model_name)
|
| 292 |
Settings.llm = new_llm
|
| 293 |
|
| 294 |
+
new_query_engine = create_query_engine(vector_index, keyword_index, documents)
|
| 295 |
+
return new_query_engine, f"✅ Model switched to: {model_name}"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 296 |
except Exception as e:
|
| 297 |
+
return None, f"❌ Error: {str(e)}"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 298 |
|
| 299 |
query_engine = None
|
|
|
|
|
|
|
| 300 |
vector_index = None
|
| 301 |
+
keyword_index = None
|
| 302 |
+
documents = None
|
| 303 |
current_model = DEFAULT_MODEL
|
| 304 |
|
| 305 |
def main_answer_question(question):
|
| 306 |
+
global query_engine
|
| 307 |
+
return answer_question(question, query_engine)
|
|
|
|
|
|
|
|
|
|
| 308 |
|
| 309 |
def main_switch_model(model_name):
|
| 310 |
+
global query_engine, vector_index, keyword_index, documents, current_model
|
| 311 |
+
new_query_engine, status = switch_model(model_name, vector_index, keyword_index, documents)
|
|
|
|
| 312 |
if new_query_engine:
|
| 313 |
query_engine = new_query_engine
|
| 314 |
current_model = model_name
|
| 315 |
+
return status
|
| 316 |
+
|
| 317 |
+
def create_interface():
|
| 318 |
+
with gr.Blocks(title="AIEXP - RAG System", theme=gr.themes.Soft()) as demo:
|
| 319 |
+
gr.Markdown("# AIEXP - AI Expert for Regulatory Documentation")
|
| 320 |
+
|
| 321 |
+
with gr.Row():
|
| 322 |
+
model_dropdown = gr.Dropdown(
|
| 323 |
+
choices=list(AVAILABLE_MODELS.keys()),
|
| 324 |
+
value=current_model,
|
| 325 |
+
label="Select Language Model"
|
| 326 |
+
)
|
| 327 |
+
switch_btn = gr.Button("Switch Model")
|
| 328 |
+
model_status = gr.Textbox(
|
| 329 |
+
value=f"Current model: {current_model}",
|
| 330 |
+
label="Model Status",
|
| 331 |
+
interactive=False
|
| 332 |
+
)
|
| 333 |
+
|
| 334 |
+
with gr.Row():
|
| 335 |
+
question_input = gr.Textbox(
|
| 336 |
+
label="Your Question",
|
| 337 |
+
placeholder="Ask a question about the documents...",
|
| 338 |
+
lines=3
|
| 339 |
+
)
|
| 340 |
+
|
| 341 |
+
ask_btn = gr.Button("Get Answer", variant="primary")
|
| 342 |
+
|
| 343 |
+
with gr.Row():
|
| 344 |
+
answer_output = gr.HTML(
|
| 345 |
+
label="Answer",
|
| 346 |
+
value="<div style='padding: 20px; text-align: center;'>Answer will appear here...</div>"
|
| 347 |
+
)
|
| 348 |
+
sources_output = gr.HTML(
|
| 349 |
+
label="Sources",
|
| 350 |
+
value="<div style='padding: 20px; text-align: center;'>Sources will appear here...</div>"
|
| 351 |
+
)
|
| 352 |
+
|
| 353 |
+
switch_btn.click(
|
| 354 |
+
fn=main_switch_model,
|
| 355 |
+
inputs=[model_dropdown],
|
| 356 |
+
outputs=[model_status]
|
| 357 |
+
)
|
| 358 |
+
|
| 359 |
+
ask_btn.click(
|
| 360 |
+
fn=main_answer_question,
|
| 361 |
+
inputs=[question_input],
|
| 362 |
+
outputs=[answer_output, sources_output]
|
| 363 |
+
)
|
| 364 |
+
|
| 365 |
+
question_input.submit(
|
| 366 |
+
fn=main_answer_question,
|
| 367 |
+
inputs=[question_input],
|
| 368 |
+
outputs=[answer_output, sources_output]
|
| 369 |
+
)
|
| 370 |
|
| 371 |
+
return demo
|
| 372 |
|
| 373 |
def main():
|
| 374 |
+
global query_engine, vector_index, keyword_index, documents
|
| 375 |
|
| 376 |
+
print("Starting AIEXP - AI Expert for Regulatory Documentation")
|
| 377 |
|
| 378 |
+
query_engine, vector_index, keyword_index, documents = initialize_system()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 379 |
|
| 380 |
if query_engine:
|
| 381 |
+
print("Launching web interface...")
|
| 382 |
+
demo = create_interface()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 383 |
demo.launch(
|
| 384 |
server_name="0.0.0.0",
|
| 385 |
server_port=7860,
|
| 386 |
+
share=True
|
|
|
|
| 387 |
)
|
| 388 |
else:
|
| 389 |
+
print("Failed to initialize system")
|
| 390 |
sys.exit(1)
|
| 391 |
|
| 392 |
if __name__ == "__main__":
|
app_1.py
CHANGED
|
@@ -1,501 +1,213 @@
|
|
| 1 |
import gradio as gr
|
| 2 |
-
from huggingface_hub import hf_hub_download
|
| 3 |
-
import faiss
|
| 4 |
-
import pandas as pd
|
| 5 |
import os
|
| 6 |
-
import
|
| 7 |
-
from
|
| 8 |
-
from
|
| 9 |
-
from
|
| 10 |
-
from
|
| 11 |
-
from llama_index.core.retrievers import VectorIndexRetriever
|
| 12 |
-
from llama_index.core.response_synthesizers import get_response_synthesizer, ResponseMode
|
| 13 |
-
from llama_index.core.prompts import PromptTemplate
|
| 14 |
-
import time
|
| 15 |
import sys
|
| 16 |
-
from config import
|
| 17 |
-
|
| 18 |
-
|
| 19 |
-
|
| 20 |
-
|
| 21 |
-
|
| 22 |
-
|
| 23 |
-
|
| 24 |
-
GOOGLE_API_KEY = os.getenv('GOOGLE_API_KEY')
|
| 25 |
-
|
| 26 |
-
CUSTOM_PROMPT_NEW = """
|
| 27 |
-
Вы являетесь высокоспециализированным Ассистентом для анализа документов (AIEXP). Ваша цель - предоставлять точные, корректные и контекстно релевантные ответы на основе анализа нормативной документации (НД). Все ваши ответы должны основываться исключительно на предоставленном контексте без использования внешних знаний или предположений.
|
| 28 |
-
|
| 29 |
-
КРИТИЧЕСКИ ВАЖНО: ВСЕ ОТВЕТЫ ДОЛЖНЫ БЫТЬ ТОЛЬКО НА РУССКОМ ЯЗЫКЕ! НИКОГДА НЕ ОТВЕЧАЙТЕ НА АНГЛИЙСКОМ!
|
| 30 |
-
|
| 31 |
-
История чата:
|
| 32 |
-
{chat_history}
|
| 33 |
-
|
| 34 |
-
|
| 35 |
-
ИНСТРУКЦИИ ПО ОБРАБОТКЕ КОНТЕКСТА:
|
| 36 |
-
|
| 37 |
-
1. АНАЛИЗ ТАБЛИЧНЫХ ДАННЫХ:
|
| 38 |
-
- Если в контексте есть информация начинающаяся с "Таблица", внимательно изучите её содержимое
|
| 39 |
-
- Извлекайте данные из строк с заголовками и данными таблицы
|
| 40 |
-
- Указывайте номер и название таблицы при ответе
|
| 41 |
-
- Структурируйте ответ на основе табличных данных
|
| 42 |
-
|
| 43 |
-
2. ОПРЕДЕЛЕНИЕ ТИПА ЗАДАЧИ:
|
| 44 |
-
Проанализируйте запрос пользователя и определите тип задачи:
|
| 45 |
-
|
| 46 |
-
1. КРАТКОЕ САММАРИ (ключевые слова: "кратко", "суммировать", "резюме", "основные моменты", "в двух словах"):
|
| 47 |
-
- Предоставьте структурированное резюме запрашиваемого раздела/пункта
|
| 48 |
-
- Выделите ключевые требования, процедуры или положения
|
| 49 |
-
- Используйте нумерованный список для лучшей читаемости
|
| 50 |
-
- Сохраняйте терминологию НД
|
| 51 |
-
|
| 52 |
-
2. ПОИСК ДОКУМЕНТА И ПУНКТА (ключевые слова: "найти", "где", "какой документ", "в каком разделе", "ссылка"):
|
| 53 |
-
- Укажите конкретный документ и его структурное расположение
|
| 54 |
-
- Предоставьте точные номера разделов/подразделов/пунктов
|
| 55 |
-
- Процитируйте релевантные фрагменты
|
| 56 |
-
- Если найдено несколько документов, перечислите все с указанием специфики каждого
|
| 57 |
-
|
| 58 |
-
3. ПРОВЕРКА КОРРЕКТНОСТИ (ключевые слова: "правильно ли", "соответствует ли", "проверить", "корректно", "нарушение"):
|
| 59 |
-
- Сопоставьте предоставленную информацию с требованиями НД
|
| 60 |
-
- Четко укажите: "СООТВЕТСТВУЕТ" или "НЕ СООТВЕТСТВУЕТ"
|
| 61 |
-
- Перечислите конкретные требования НД
|
| 62 |
-
- Укажите выявленные расхождения или подтвердите соответствие
|
| 63 |
-
- Процитируйте релевантные пункты НД
|
| 64 |
-
|
| 65 |
-
4. ПЛАН ДЕЙСТВИЙ (ключевые слова: "план", "алгоритм", "последовательность", "как действовать", "пошагово"):
|
| 66 |
-
- Создайте пронумерованный пошаговый план
|
| 67 |
-
- Каждый шаг должен содержать ссылку на соответствующий пункт НД
|
| 68 |
-
- Укажите необходимые документы или формы
|
| 69 |
-
- Добавьте временные рамки, если они указаны в НД
|
| 70 |
-
- Выделите критические требования или ограничения
|
| 71 |
-
|
| 72 |
-
5. УТОЧНЯЮЩИЕ ВОПРОСЫ (ключевые слова: "что это значит", "что означает", "объясните", "расскажите подробнее"):
|
| 73 |
-
- Используйте историю чата для понимания контекста
|
| 74 |
-
- Если вопрос относится к предыдущему обсуждению, опирайтесь на него
|
| 75 |
-
- Предоставьте подробное объяснение на основе НД
|
| 76 |
-
- Если контекст неясен, попросите уточнения
|
| 77 |
-
|
| 78 |
-
ПРАВИЛА ФОРМИРОВАНИЯ ОТВЕТОВ:
|
| 79 |
-
|
| 80 |
-
1. ОБЯЗАТЕЛЬНОЕ УКАЗАНИЕ ИСТОЧНИКОВ:
|
| 81 |
-
- Для каждого ответа указывайте: "Согласно [Название документа], раздел [X], пункт [X.X]: [Ваш ответ]"
|
| 82 |
-
- В конце ответа добавляйте: "Подробнее об этом можно узнать в документе [Название документа], раздел [X]."
|
| 83 |
-
- При отсутствии точного раздела: "Согласно документу [Название]: [Ваш ответ]"
|
| 84 |
-
|
| 85 |
-
2. СТРОГОЕ СЛЕДОВАНИЕ КОНТЕКСТУ:
|
| 86 |
-
- Если информация не найдена: "Информация по вашему запросу не была найдена в нормативной документации."
|
| 87 |
-
- НЕ используйте английский язык ни при каких обстоятельствах
|
| 88 |
-
- Используйте историю чата для понимания контекста вопросов
|
| 89 |
-
|
| 90 |
-
3. ИСПОЛЬЗОВАНИЕ ТЕРМИНОЛОГИИ НД:
|
| 91 |
-
- Применяйте официальную терминологию из документов
|
| 92 |
-
- Сохраняйте оригинальные формулировки ключевых требований
|
| 93 |
-
- При необходимости разъясняйте специальные термины на основе НД
|
| 94 |
-
|
| 95 |
-
4. СТРУКТУРИРОВАНИЕ ОТВЕТОВ:
|
| 96 |
-
- Основной ответ на русском языке
|
| 97 |
-
- Указание источника
|
| 98 |
-
- Дополнительная информация о документе
|
| 99 |
-
|
| 100 |
-
Контекст: {context_str}
|
| 101 |
-
|
| 102 |
-
Вопрос: {query_str}
|
| 103 |
-
|
| 104 |
-
Ответ (ТОЛЬКО НА РУССКОМ ЯЗЫКЕ):
|
| 105 |
-
"""
|
| 106 |
-
|
| 107 |
-
|
| 108 |
-
query_engine = None
|
| 109 |
-
chunks_df = None
|
| 110 |
-
chat_history = []
|
| 111 |
-
|
| 112 |
-
def log_message(message):
|
| 113 |
-
print(message, flush=True)
|
| 114 |
-
sys.stdout.flush()
|
| 115 |
-
|
| 116 |
-
def table_to_document(table_json):
|
| 117 |
-
document_id = table_json.get("document_id") or table_json.get("document", "unknown")
|
| 118 |
-
|
| 119 |
-
metadata = {
|
| 120 |
-
"document_id": document_id,
|
| 121 |
-
"section": table_json.get("section", ""),
|
| 122 |
-
"table_number": table_json.get("table_number", ""),
|
| 123 |
-
"table_title": table_json.get("table_title", ""),
|
| 124 |
-
}
|
| 125 |
|
| 126 |
-
|
| 127 |
-
|
| 128 |
|
| 129 |
-
|
| 130 |
-
|
| 131 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 132 |
|
| 133 |
-
|
| 134 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 135 |
|
| 136 |
-
if
|
| 137 |
-
|
|
|
|
|
|
|
|
|
|
| 138 |
|
| 139 |
-
|
| 140 |
-
|
| 141 |
-
|
| 142 |
-
|
| 143 |
-
|
| 144 |
-
row_str = " | ".join([f"{k}: {v}" for k,v in row.items()])
|
| 145 |
-
table_text += f"Строка {i+1}: {row_str}\n"
|
| 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 |
-
local_dir=download_dir,
|
| 172 |
-
repo_type="dataset",
|
| 173 |
-
token=HF_TOKEN
|
| 174 |
-
)
|
| 175 |
-
|
| 176 |
-
with open(local_path, 'r', encoding='utf-8') as f:
|
| 177 |
-
table_data = json.load(f)
|
| 178 |
-
|
| 179 |
-
log_message(f"📋 Структура JSON: {list(table_data.keys()) if isinstance(table_data, dict) else 'Список'}")
|
| 180 |
-
|
| 181 |
-
if isinstance(table_data, dict):
|
| 182 |
-
if 'sheets' in table_data:
|
| 183 |
-
for sheet in table_data['sheets']:
|
| 184 |
-
doc = table_to_document(sheet)
|
| 185 |
-
table_documents.append(doc)
|
| 186 |
-
log_message(f"✅ Создан документ из таблицы: {sheet.get('table_number', 'unknown')}")
|
| 187 |
-
else:
|
| 188 |
-
doc = table_to_document(table_data)
|
| 189 |
-
table_documents.append(doc)
|
| 190 |
-
log_message(f"✅ Создан документ из JSON объекта")
|
| 191 |
-
elif isinstance(table_data, list):
|
| 192 |
-
for table_json in table_data:
|
| 193 |
-
doc = table_to_document(table_json)
|
| 194 |
-
table_documents.append(doc)
|
| 195 |
-
log_message(f"✅ Создан документ из элемента списка")
|
| 196 |
-
|
| 197 |
-
except Exception as e:
|
| 198 |
-
log_message(f"❌ Ошибка обработки файла {file_path}: {str(e)}")
|
| 199 |
-
continue
|
| 200 |
-
|
| 201 |
-
log_message(f"✅ Создано {len(table_documents)} документов из таблиц")
|
| 202 |
-
return table_documents
|
| 203 |
-
|
| 204 |
-
except Exception as e:
|
| 205 |
-
log_message(f"❌ Ошибка загрузки табличных данных: {str(e)}")
|
| 206 |
-
return []
|
| 207 |
-
|
| 208 |
|
| 209 |
-
def
|
|
|
|
|
|
|
| 210 |
try:
|
| 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 |
-
log_message(f"✨ Улучшенный запрос: {improved_query}")
|
| 241 |
|
| 242 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
| 243 |
|
| 244 |
except Exception as e:
|
| 245 |
-
log_message(f"
|
| 246 |
-
return
|
| 247 |
-
|
| 248 |
|
| 249 |
-
def
|
| 250 |
-
|
| 251 |
-
|
| 252 |
-
|
| 253 |
-
history_text = ""
|
| 254 |
-
for i, (user_msg, bot_msg) in enumerate(chat_history[-5:], 1):
|
| 255 |
-
history_text += f"Сообщение {i}:\nПользователь: {user_msg}\nАссистент: {bot_msg}\n\n"
|
| 256 |
-
|
| 257 |
-
return history_text
|
| 258 |
-
|
| 259 |
-
|
| 260 |
-
def answer_question(question, history):
|
| 261 |
-
global query_engine, chunks_df, chat_history
|
| 262 |
-
|
| 263 |
-
if query_engine is None:
|
| 264 |
-
return history + [["", "❌ Система не инициализирована"]], ""
|
| 265 |
|
| 266 |
try:
|
| 267 |
-
|
| 268 |
-
|
| 269 |
-
log_message(f"🔍 Получен вопрос: {question}")
|
| 270 |
-
log_message(f"📜 История чата: {len(chat_history)} сообщений")
|
| 271 |
-
|
| 272 |
-
# Улучшаем запрос с учетом истории
|
| 273 |
-
improved_question = improve_query_with_history(question, chat_history)
|
| 274 |
-
log_message(f"🎯 Обработка улучшенного запроса: {improved_question}")
|
| 275 |
-
|
| 276 |
-
# Форматируем историю чата для промпта
|
| 277 |
-
chat_history_text = format_chat_history()
|
| 278 |
-
log_message(f"📝 Сформированная история для промпта: {len(chat_history_text)} символов")
|
| 279 |
-
|
| 280 |
-
log_message("🔎 Поиск релевантных чанков...")
|
| 281 |
-
retrieved_nodes = query_engine.retriever.retrieve(improved_question)
|
| 282 |
-
log_message(f"📊 Найдено {len(retrieved_nodes)} релевантных чанков")
|
| 283 |
-
|
| 284 |
-
# Логируем найденные чанки
|
| 285 |
-
for i, node in enumerate(retrieved_nodes[:3]):
|
| 286 |
-
log_message(f"📄 Чанк {i+1}: {node.text[:100]}...")
|
| 287 |
-
log_message(f"🏷️ Метаданные: {node.metadata}")
|
| 288 |
|
| 289 |
-
|
|
|
|
| 290 |
|
| 291 |
-
|
| 292 |
-
|
| 293 |
-
|
| 294 |
-
{
|
| 295 |
-
|
| 296 |
-
|
| 297 |
-
"""
|
| 298 |
-
|
| 299 |
-
response = query_engine.query(query_with_context)
|
| 300 |
-
|
| 301 |
-
end_time = time.time()
|
| 302 |
-
processing_time = end_time - start_time
|
| 303 |
-
|
| 304 |
-
bot_response = response.response
|
| 305 |
-
log_message(f"✅ Получен ответ: {bot_response[:100]}...")
|
| 306 |
-
|
| 307 |
-
# Проверяем, что ответ на русском языке
|
| 308 |
-
if any(english_word in bot_response.lower() for english_word in ['i am sorry', 'i cannot', 'the query', 'this request']):
|
| 309 |
-
log_message("⚠️ Обнаружен ответ на английском языке, форсируем русский ответ")
|
| 310 |
-
|
| 311 |
-
# Принудительно запрашиваем ответ на русском
|
| 312 |
-
russian_prompt = f"""
|
| 313 |
-
ВАЖНО: Отвечай ТОЛЬКО на русском языке!
|
| 314 |
-
|
| 315 |
-
Вопрос: {question}
|
| 316 |
-
История: {chat_history_text}
|
| 317 |
-
Контекст: {retrieved_nodes[0].text if retrieved_nodes else 'Нет контекста'}
|
| 318 |
-
|
| 319 |
-
Если информации недостаточно для ответа, скажи: "Недостаточно информации для ответа на ваш вопрос в предоставленной документации."
|
| 320 |
-
|
| 321 |
-
Ответ на русском языке:
|
| 322 |
-
"""
|
| 323 |
|
| 324 |
-
from llama_index.llms.google_genai import GoogleGenAI
|
| 325 |
-
llm = GoogleGenAI(model="gemini-2.0-flash", api_key=GOOGLE_API_KEY)
|
| 326 |
-
bot_response = llm.complete(russian_prompt).text.strip()
|
| 327 |
-
log_message(f"🔄 Исправленный ответ на русском: {bot_response[:100]}...")
|
| 328 |
-
|
| 329 |
-
# Обновляем историю чата
|
| 330 |
-
chat_history.append((question, bot_response))
|
| 331 |
-
|
| 332 |
-
if len(chat_history) > 10:
|
| 333 |
-
chat_history = chat_history[-10:]
|
| 334 |
-
|
| 335 |
-
log_message(f"💾 История чата обновлена. Всего сообщений: {len(chat_history)}")
|
| 336 |
-
|
| 337 |
-
sources_html = generate_sources_html(retrieved_nodes)
|
| 338 |
-
|
| 339 |
-
response_with_time = f"{bot_response}\n\n⏱️ Время обработки: {processing_time:.2f} сек"
|
| 340 |
-
|
| 341 |
-
history.append([question, response_with_time])
|
| 342 |
-
|
| 343 |
-
return history, sources_html
|
| 344 |
-
|
| 345 |
except Exception as e:
|
| 346 |
-
error_msg = f"
|
| 347 |
-
log_message(
|
| 348 |
-
|
| 349 |
-
return history, ""
|
| 350 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 351 |
|
| 352 |
-
def initialize_models():
|
| 353 |
-
global query_engine, chunks_df
|
| 354 |
-
|
| 355 |
try:
|
| 356 |
-
|
| 357 |
-
|
| 358 |
-
|
| 359 |
-
log_message("📥 Загрузка основных файлов...")
|
| 360 |
-
faiss_index_path = hf_hub_download(
|
| 361 |
-
repo_id=REPO_ID,
|
| 362 |
-
filename=faiss_index_filename,
|
| 363 |
-
local_dir=download_dir,
|
| 364 |
-
repo_type="dataset",
|
| 365 |
-
token=HF_TOKEN
|
| 366 |
-
)
|
| 367 |
-
|
| 368 |
-
chunks_csv_path = hf_hub_download(
|
| 369 |
-
repo_id=REPO_ID,
|
| 370 |
-
filename=chunks_filename,
|
| 371 |
-
local_dir=download_dir,
|
| 372 |
-
repo_type="dataset",
|
| 373 |
-
token=HF_TOKEN
|
| 374 |
-
)
|
| 375 |
-
|
| 376 |
-
log_message("📚 Загрузка индекса и данных...")
|
| 377 |
-
index_faiss = faiss.read_index(faiss_index_path)
|
| 378 |
-
chunks_df = pd.read_csv(chunks_csv_path)
|
| 379 |
-
log_message(f"📄 Загружено {len(chunks_df)} основных чанков")
|
| 380 |
-
log_message(f"📋 Колонки в chunks_df: {list(chunks_df.columns)}")
|
| 381 |
-
|
| 382 |
-
table_documents = download_table_data()
|
| 383 |
-
|
| 384 |
-
log_message("🤖 Настройка моделей...")
|
| 385 |
-
embed_model = HuggingFaceEmbedding(model_name="sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2")
|
| 386 |
-
llm = GoogleGenAI(model="gemini-2.0-flash", api_key=GOOGLE_API_KEY)
|
| 387 |
-
|
| 388 |
-
Settings.embed_model = embed_model
|
| 389 |
-
Settings.llm = llm
|
| 390 |
-
|
| 391 |
-
text_column = None
|
| 392 |
-
for col in chunks_df.columns:
|
| 393 |
-
if 'text' in col.lower() or 'content' in col.lower() or 'chunk' in col.lower():
|
| 394 |
-
text_column = col
|
| 395 |
-
break
|
| 396 |
-
|
| 397 |
-
if text_column is None:
|
| 398 |
-
text_column = chunks_df.columns[0]
|
| 399 |
-
|
| 400 |
-
log_message(f"📝 Используется колонка для текста: {text_column}")
|
| 401 |
-
|
| 402 |
-
documents = []
|
| 403 |
-
for i, (_, row) in enumerate(chunks_df.iterrows()):
|
| 404 |
-
doc = Document(
|
| 405 |
-
text=str(row[text_column]),
|
| 406 |
-
metadata={
|
| 407 |
-
"chunk_id": row.get('chunk_id', i),
|
| 408 |
-
"document_id": row.get('document_id', 'unknown')
|
| 409 |
-
}
|
| 410 |
-
)
|
| 411 |
-
documents.append(doc)
|
| 412 |
-
|
| 413 |
-
documents.extend(table_documents)
|
| 414 |
-
log_message(f"📋 Всего создано {len(documents)} документов ({len(chunks_df)} чанков + {len(table_documents)} таблиц)")
|
| 415 |
-
|
| 416 |
-
log_message("🔍 Построение векторного индекса...")
|
| 417 |
-
vector_index = VectorStoreIndex.from_documents(documents)
|
| 418 |
-
|
| 419 |
-
retriever = VectorIndexRetriever(
|
| 420 |
-
index=vector_index,
|
| 421 |
-
similarity_top_k=20,
|
| 422 |
-
similarity_cutoff=0.7
|
| 423 |
-
)
|
| 424 |
-
|
| 425 |
-
custom_prompt_template = PromptTemplate(CUSTOM_PROMPT_NEW)
|
| 426 |
-
response_synthesizer = get_response_synthesizer(
|
| 427 |
-
response_mode=ResponseMode.TREE_SUMMARIZE,
|
| 428 |
-
text_qa_template=custom_prompt_template
|
| 429 |
-
)
|
| 430 |
-
|
| 431 |
-
query_engine = RetrieverQueryEngine(
|
| 432 |
-
retriever=retriever,
|
| 433 |
-
response_synthesizer=response_synthesizer
|
| 434 |
-
)
|
| 435 |
-
|
| 436 |
-
log_message("✅ Система успешно инициализирована!")
|
| 437 |
-
return True
|
| 438 |
-
|
| 439 |
-
except Exception as e:
|
| 440 |
-
log_message(f"❌ Ошибка инициализации: {str(e)}")
|
| 441 |
-
return False
|
| 442 |
|
| 443 |
-
|
| 444 |
-
|
| 445 |
-
|
| 446 |
-
|
| 447 |
-
|
| 448 |
-
for node in nodes:
|
| 449 |
-
metadata = node.metadata if hasattr(node, 'metadata') else {}
|
| 450 |
-
doc_id = metadata.get('document_id', 'unknown')
|
| 451 |
-
if doc_id not in unique_docs:
|
| 452 |
-
unique_docs[doc_id] = []
|
| 453 |
-
unique_docs[doc_id].append(node)
|
| 454 |
-
|
| 455 |
-
for doc_id, doc_nodes in unique_docs.items():
|
| 456 |
-
if doc_id == 'unknown' or doc_id == 'Раздел документа':
|
| 457 |
-
continue
|
| 458 |
-
|
| 459 |
-
file_link = None
|
| 460 |
-
if chunks_df is not None and 'file_link' in chunks_df.columns:
|
| 461 |
-
doc_rows = chunks_df[chunks_df['document_id'] == doc_id]
|
| 462 |
-
if not doc_rows.empty:
|
| 463 |
-
file_link = doc_rows.iloc[0]['file_link']
|
| 464 |
-
|
| 465 |
-
html += f"<div style='margin-bottom: 15px; padding: 15px; border: 1px solid #4a5568; border-radius: 8px; background-color: #1a202c;'>"
|
| 466 |
-
html += f"<h4 style='margin: 0 0 10px 0; color: #63b3ed;'>📄 {doc_id}</h4>"
|
| 467 |
-
|
| 468 |
-
if file_link:
|
| 469 |
-
html += f"<a href='{file_link}' target='_blank' style='color: #68d391; text-decoration: none; font-size: 14px; display: inline-block; margin-bottom: 10px;'>🔗 Ссылка на документ</a><br>"
|
| 470 |
-
|
| 471 |
-
table_nodes = [node for node in doc_nodes if 'table_number' in node.metadata]
|
| 472 |
-
if table_nodes:
|
| 473 |
-
for node in table_nodes[:3]:
|
| 474 |
-
metadata = node.metadata
|
| 475 |
-
table_num = metadata.get('table_number', '')
|
| 476 |
-
table_title = metadata.get('table_title', 'Без названия')
|
| 477 |
-
if table_num and table_title != 'Без названия':
|
| 478 |
-
html += f"<p style='font-size: 12px; color: #a0aec0; margin: 5px 0;'>📊 {table_num}: {table_title}</p>"
|
| 479 |
-
|
| 480 |
-
html += "</div>"
|
| 481 |
-
|
| 482 |
-
html += "</div>"
|
| 483 |
-
return html
|
| 484 |
|
| 485 |
-
def clear_chat():
|
| 486 |
-
global chat_history
|
| 487 |
-
chat_history = []
|
| 488 |
-
log_message("🗑️ История чата очищена")
|
| 489 |
-
return [], ""
|
| 490 |
|
| 491 |
-
def handle_submit(message, history):
|
| 492 |
-
if not message.strip():
|
| 493 |
-
return history, ""
|
| 494 |
-
|
| 495 |
-
updated_history, sources = answer_question(message, history)
|
| 496 |
-
return updated_history, sources
|
| 497 |
|
| 498 |
-
def create_demo_interface():
|
| 499 |
with gr.Blocks(title="AIEXP - AI Expert для нормативной документации", theme=gr.themes.Soft()) as demo:
|
| 500 |
|
| 501 |
gr.Markdown("""
|
|
@@ -504,65 +216,131 @@ def create_demo_interface():
|
|
| 504 |
## Инструмент для работы с нормативной документацией
|
| 505 |
""")
|
| 506 |
|
| 507 |
-
with gr.Tab("
|
| 508 |
gr.Markdown("### Задайте вопрос по нормативной документации")
|
| 509 |
|
| 510 |
with gr.Row():
|
| 511 |
with gr.Column(scale=2):
|
| 512 |
-
|
| 513 |
-
|
| 514 |
-
|
| 515 |
-
|
|
|
|
| 516 |
)
|
| 517 |
-
|
| 518 |
-
|
| 519 |
-
|
| 520 |
-
|
| 521 |
-
|
| 522 |
-
|
| 523 |
-
|
| 524 |
-
|
| 525 |
-
|
| 526 |
-
|
| 527 |
-
|
| 528 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
| 529 |
|
| 530 |
gr.Examples(
|
| 531 |
examples=[
|
|
|
|
|
|
|
| 532 |
"Какой стандарт устанавливает порядок признания протоколов испытаний продукции в области использования атомной энергии?",
|
| 533 |
"Кто несет ответственность за организацию и проведение признания протоколов испытаний продукции?",
|
| 534 |
-
"В каких случаях могут быть признаны протоколы испытаний, проведенные
|
|
|
|
| 535 |
],
|
| 536 |
-
inputs=
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 537 |
)
|
| 538 |
|
| 539 |
with gr.Column(scale=1):
|
| 540 |
sources_output = gr.HTML(
|
| 541 |
-
label="
|
| 542 |
-
value="<div style='background-color: #2d3748; color: white; padding: 20px; border-radius: 10px; text-align: center;'>Здесь появятся
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 543 |
)
|
| 544 |
|
| 545 |
-
|
| 546 |
-
|
| 547 |
-
|
| 548 |
-
|
| 549 |
-
lambda: "", None, msg
|
| 550 |
)
|
| 551 |
|
| 552 |
-
|
| 553 |
-
|
|
|
|
|
|
|
| 554 |
)
|
| 555 |
|
| 556 |
-
|
| 557 |
-
|
|
|
|
|
|
|
|
|
|
| 558 |
return demo
|
| 559 |
|
| 560 |
-
|
| 561 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 562 |
|
| 563 |
-
|
| 564 |
-
|
| 565 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 566 |
demo.launch(
|
| 567 |
server_name="0.0.0.0",
|
| 568 |
server_port=7860,
|
|
@@ -570,5 +348,8 @@ if __name__ == "__main__":
|
|
| 570 |
debug=False
|
| 571 |
)
|
| 572 |
else:
|
| 573 |
-
log_message("
|
| 574 |
-
sys.exit(1)
|
|
|
|
|
|
|
|
|
|
|
|
| 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_data, load_image_data, load_csv_chunks
|
| 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 |
+
def create_chunks_display_html(chunk_info):
|
| 15 |
+
if not chunk_info:
|
| 16 |
+
return "<div style='padding: 20px; text-align: center; color: black;'>Нет данных о чанках</div>"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 17 |
|
| 18 |
+
html = "<div style='max-height: 500px; overflow-y: auto; padding: 10px; color: black;'>"
|
| 19 |
+
html += f"<h4 style='color: black;'>Найдено релевантных чанков: {len(chunk_info)}</h4>"
|
| 20 |
|
| 21 |
+
for i, chunk in enumerate(chunk_info):
|
| 22 |
+
bg_color = "#f8f9fa" if i % 2 == 0 else "#e9ecef"
|
| 23 |
+
|
| 24 |
+
# Get section display info
|
| 25 |
+
section_display = get_section_display(chunk)
|
| 26 |
+
formatted_content = get_formatted_content(chunk)
|
| 27 |
+
|
| 28 |
+
html += f"""
|
| 29 |
+
<div style='background-color: {bg_color}; padding: 10px; margin: 5px 0; border-radius: 5px; border-left: 4px solid #007bff; color: black;'>
|
| 30 |
+
<strong style='color: black;'>Документ:</strong> <span style='color: black;'>{chunk['document_id']}</span><br>
|
| 31 |
+
<strong style='color: black;'>Раздел:</strong> <span style='color: black;'>{section_display}</span><br>
|
| 32 |
+
<strong style='color: black;'>Содержание:</strong><br>
|
| 33 |
+
<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;'>
|
| 34 |
+
{formatted_content}
|
| 35 |
+
</div>
|
| 36 |
+
</div>
|
| 37 |
+
"""
|
| 38 |
|
| 39 |
+
html += "</div>"
|
| 40 |
+
return html
|
| 41 |
+
|
| 42 |
+
def get_section_display(chunk):
|
| 43 |
+
section_path = chunk.get('section_path', '')
|
| 44 |
+
section_id = chunk.get('section_id', 'unknown')
|
| 45 |
+
doc_type = chunk.get('type', 'text')
|
| 46 |
|
| 47 |
+
if doc_type == 'table' and chunk.get('table_number'):
|
| 48 |
+
table_num = chunk.get('table_number')
|
| 49 |
+
if not str(table_num).startswith('№'):
|
| 50 |
+
table_num = f"№{table_num}"
|
| 51 |
+
return f"таблица {table_num}"
|
| 52 |
|
| 53 |
+
if doc_type == 'image' and chunk.get('image_number'):
|
| 54 |
+
image_num = chunk.get('image_number')
|
| 55 |
+
if not str(image_num).startswith('№'):
|
| 56 |
+
image_num = f"№{image_num}"
|
| 57 |
+
return f"рисунок {image_num}"
|
|
|
|
|
|
|
| 58 |
|
| 59 |
+
if section_path:
|
| 60 |
+
return section_path
|
| 61 |
+
elif section_id and section_id != 'unknown':
|
| 62 |
+
return section_id
|
| 63 |
|
| 64 |
+
return section_id
|
| 65 |
+
|
| 66 |
+
def get_formatted_content(chunk):
|
| 67 |
+
document_id = chunk.get('document_id', 'unknown')
|
| 68 |
+
section_path = chunk.get('section_path', '')
|
| 69 |
+
section_id = chunk.get('section_id', 'unknown')
|
| 70 |
+
section_text = chunk.get('section_text', '')
|
| 71 |
+
parent_section = chunk.get('parent_section', '')
|
| 72 |
+
parent_title = chunk.get('parent_title', '')
|
| 73 |
+
level = chunk.get('level', '')
|
| 74 |
+
chunk_text = chunk.get('chunk_text', '')
|
| 75 |
+
doc_type = chunk.get('type', 'text')
|
| 76 |
|
| 77 |
+
# For text documents
|
| 78 |
+
if level in ['subsection', 'sub_subsection', 'sub_sub_subsection'] and parent_section:
|
| 79 |
+
current_section = section_path if section_path else section_id
|
| 80 |
+
parent_info = f"{parent_section} ({parent_title})" if parent_title else parent_section
|
| 81 |
+
return f"В разделе {parent_info} в документе {document_id}, пункт {current_section}: {chunk_text}"
|
| 82 |
+
else:
|
| 83 |
+
current_section = section_path if section_path else section_id
|
| 84 |
+
clean_text = chunk_text
|
| 85 |
+
if section_text and chunk_text.startswith(section_text):
|
| 86 |
+
section_title = section_text
|
| 87 |
+
elif chunk_text.startswith(f"{current_section} "):
|
| 88 |
+
clean_text = chunk_text[len(f"{current_section} "):].strip()
|
| 89 |
+
section_title = section_text if section_text else f"{current_section} {clean_text.split('.')[0] if '.' in clean_text else clean_text[:50]}"
|
| 90 |
+
else:
|
| 91 |
+
section_title = section_text if section_text else current_section
|
| 92 |
+
|
| 93 |
+
return f"В разделе {current_section} в документе {document_id}, пункт {section_title}: {clean_text}"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 94 |
|
| 95 |
+
def initialize_system(repo_id, hf_token, download_dir, chunks_filename=None,
|
| 96 |
+
json_files_dir=None, table_data_dir=None, image_data_dir=None,
|
| 97 |
+
use_json_instead_csv=False):
|
| 98 |
try:
|
| 99 |
+
from documents_prep import process_documents_with_chunking
|
| 100 |
+
log_message("Инициализация системы")
|
| 101 |
+
os.makedirs(download_dir, exist_ok=True)
|
| 102 |
+
from config import CHUNK_SIZE, CHUNK_OVERLAP
|
| 103 |
+
from llama_index.core.text_splitter import TokenTextSplitter
|
| 104 |
|
| 105 |
+
embed_model = get_embedding_model()
|
| 106 |
+
llm = get_llm_model(DEFAULT_MODEL)
|
| 107 |
+
reranker = get_reranker_model()
|
| 108 |
|
| 109 |
+
Settings.embed_model = embed_model
|
| 110 |
+
Settings.llm = llm
|
| 111 |
+
Settings.text_splitter = TokenTextSplitter(
|
| 112 |
+
chunk_size=CHUNK_SIZE,
|
| 113 |
+
chunk_overlap=CHUNK_OVERLAP,
|
| 114 |
+
separator=" ",
|
| 115 |
+
backup_separators=["\n", ".", "!", "?"]
|
| 116 |
+
)
|
| 117 |
|
| 118 |
+
log_message(f"Configured chunk size: {CHUNK_SIZE} tokens")
|
| 119 |
+
log_message(f"Configured chunk overlap: {CHUNK_OVERLAP} tokens")
|
| 120 |
+
|
| 121 |
+
all_documents = []
|
| 122 |
+
chunks_df = None
|
| 123 |
+
chunk_info = []
|
| 124 |
+
|
| 125 |
+
if use_json_instead_csv and json_files_dir:
|
| 126 |
+
log_message("Используем JSON файлы вместо CSV")
|
| 127 |
+
json_documents, json_chunk_info = load_json_documents(repo_id, hf_token, json_files_dir, download_dir)
|
| 128 |
+
all_documents.extend(json_documents)
|
| 129 |
+
chunk_info.extend(json_chunk_info)
|
| 130 |
+
else:
|
| 131 |
+
if chunks_filename:
|
| 132 |
+
log_message("Загружаем данные из CSV")
|
| 133 |
+
csv_documents, chunks_df = load_csv_chunks(repo_id, hf_token, chunks_filename, download_dir)
|
| 134 |
+
all_documents.extend(csv_documents)
|
| 135 |
+
|
| 136 |
+
if table_data_dir:
|
| 137 |
+
log_message("Добавляю табличные данные")
|
| 138 |
+
table_documents = load_table_data(repo_id, hf_token, table_data_dir)
|
| 139 |
+
log_message(f"Загружено {len(table_documents)} табличных документов")
|
| 140 |
+
|
| 141 |
+
# Process table documents through chunking
|
| 142 |
+
chunked_table_docs, table_chunk_info = process_documents_with_chunking(table_documents)
|
| 143 |
+
all_documents.extend(chunked_table_docs)
|
| 144 |
+
chunk_info.extend(table_chunk_info)
|
| 145 |
+
|
| 146 |
+
if image_data_dir:
|
| 147 |
+
log_message("Добавляю данные изображений")
|
| 148 |
+
image_documents = load_image_data(repo_id, hf_token, image_data_dir)
|
| 149 |
+
log_message(f"Загружено {len(image_documents)} документов изображений")
|
| 150 |
+
|
| 151 |
+
# Process image documents through chunking
|
| 152 |
+
chunked_image_docs, image_chunk_info = process_documents_with_chunking(image_documents)
|
| 153 |
+
all_documents.extend(chunked_image_docs)
|
| 154 |
+
chunk_info.extend(image_chunk_info)
|
| 155 |
|
| 156 |
+
log_message(f"Всего документов после всей обработки: {len(all_documents)}")
|
|
|
|
| 157 |
|
| 158 |
+
vector_index = create_vector_index(all_documents)
|
| 159 |
+
query_engine = create_query_engine(vector_index)
|
| 160 |
+
|
| 161 |
+
log_message(f"Система успешно инициализирована")
|
| 162 |
+
return query_engine, chunks_df, reranker, vector_index, chunk_info
|
| 163 |
|
| 164 |
except Exception as e:
|
| 165 |
+
log_message(f"Ошибка инициализации: {str(e)}")
|
| 166 |
+
return None, None, None, None, []
|
|
|
|
| 167 |
|
| 168 |
+
def switch_model(model_name, vector_index):
|
| 169 |
+
from llama_index.core import Settings
|
| 170 |
+
from index_retriever import create_query_engine
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 171 |
|
| 172 |
try:
|
| 173 |
+
log_message(f"Переключение на модель: {model_name}")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 174 |
|
| 175 |
+
new_llm = get_llm_model(model_name)
|
| 176 |
+
Settings.llm = new_llm
|
| 177 |
|
| 178 |
+
if vector_index is not None:
|
| 179 |
+
new_query_engine = create_query_engine(vector_index)
|
| 180 |
+
log_message(f"Модель успешно переключена на: {model_name}")
|
| 181 |
+
return new_query_engine, f"✅ Модель переключена на: {model_name}"
|
| 182 |
+
else:
|
| 183 |
+
return None, "❌ Ошибка: система не инициализирована"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 184 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 185 |
except Exception as e:
|
| 186 |
+
error_msg = f"Ошибка переключения модели: {str(e)}"
|
| 187 |
+
log_message(error_msg)
|
| 188 |
+
return None, f"❌ {error_msg}"
|
|
|
|
| 189 |
|
| 190 |
+
def main_answer_question(question):
|
| 191 |
+
global query_engine, reranker, current_model, chunks_df
|
| 192 |
+
if not question.strip():
|
| 193 |
+
return ("<div style='color: black;'>Пожалуйста, введите вопрос</div>",
|
| 194 |
+
"<div style='color: black;'>Источники появятся после обработки запроса</div>",
|
| 195 |
+
"<div style='color: black;'>Чанки появятся после обработки запроса</div>")
|
| 196 |
|
|
|
|
|
|
|
|
|
|
| 197 |
try:
|
| 198 |
+
# Call the answer_question function which returns 3 values
|
| 199 |
+
answer_html, sources_html, chunks_html = answer_question(question, query_engine, reranker, current_model, chunks_df)
|
| 200 |
+
return answer_html, sources_html, chunks_html
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 201 |
|
| 202 |
+
except Exception as e:
|
| 203 |
+
log_message(f"Ошибка при ответе на вопрос: {str(e)}")
|
| 204 |
+
return (f"<div style='color: red;'>Ошибка: {str(e)}</div>",
|
| 205 |
+
"<div style='color: black;'>Источники недоступны из-за ошибки</div>",
|
| 206 |
+
"<div style='color: black;'>Чанки недоступны из-за ошибки</div>")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 207 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 208 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 209 |
|
| 210 |
+
def create_demo_interface(answer_question_func, switch_model_func, current_model, chunk_info=None):
|
| 211 |
with gr.Blocks(title="AIEXP - AI Expert для нормативной документации", theme=gr.themes.Soft()) as demo:
|
| 212 |
|
| 213 |
gr.Markdown("""
|
|
|
|
| 216 |
## Инструмент для работы с нормативной документацией
|
| 217 |
""")
|
| 218 |
|
| 219 |
+
with gr.Tab("Поиск по нормативным документам"):
|
| 220 |
gr.Markdown("### Задайте вопрос по нормативной документации")
|
| 221 |
|
| 222 |
with gr.Row():
|
| 223 |
with gr.Column(scale=2):
|
| 224 |
+
model_dropdown = gr.Dropdown(
|
| 225 |
+
choices=list(AVAILABLE_MODELS.keys()),
|
| 226 |
+
value=current_model,
|
| 227 |
+
label="Выберите языковую модель",
|
| 228 |
+
info="Выберите модель для генерации ответов"
|
| 229 |
)
|
| 230 |
+
with gr.Column(scale=1):
|
| 231 |
+
switch_btn = gr.Button("Переключить модель", variant="secondary")
|
| 232 |
+
model_status = gr.Textbox(
|
| 233 |
+
value=f"Текущая модель: {current_model}",
|
| 234 |
+
label="Статус модели",
|
| 235 |
+
interactive=False
|
| 236 |
+
)
|
| 237 |
+
|
| 238 |
+
with gr.Row():
|
| 239 |
+
with gr.Column(scale=3):
|
| 240 |
+
question_input = gr.Textbox(
|
| 241 |
+
label="Ваш вопрос к базе знаний",
|
| 242 |
+
placeholder="Введите вопрос по нормативным документам...",
|
| 243 |
+
lines=3
|
| 244 |
+
)
|
| 245 |
+
ask_btn = gr.Button("Найти ответ", variant="primary", size="lg")
|
| 246 |
|
| 247 |
gr.Examples(
|
| 248 |
examples=[
|
| 249 |
+
"О чем этот рисунок: ГОСТ Р 50.04.07-2022 Приложение Л. Л.1.5 Рисунок Л.2",
|
| 250 |
+
"Л.9 Формула в ГОСТ Р 50.04.07 - 2022 что и о чем там?",
|
| 251 |
"Какой стандарт устанавливает порядок признания протоколов испытаний продукции в области использования атомной энергии?",
|
| 252 |
"Кто несет ответственность за организацию и проведение признания протоколов испытаний продукции?",
|
| 253 |
+
"В каких случаях могут быть признаны протоколы испытаний, проведенные лабораториями?",
|
| 254 |
+
"В какой таблице можно найти информацию о методы исследований при аттестационных испытаниях технологии термической обработки заготовок из легированных сталей? Какой документ и какой раздел?"
|
| 255 |
],
|
| 256 |
+
inputs=question_input
|
| 257 |
+
)
|
| 258 |
+
|
| 259 |
+
with gr.Row():
|
| 260 |
+
with gr.Column(scale=2):
|
| 261 |
+
answer_output = gr.HTML(
|
| 262 |
+
label="",
|
| 263 |
+
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>",
|
| 264 |
)
|
| 265 |
|
| 266 |
with gr.Column(scale=1):
|
| 267 |
sources_output = gr.HTML(
|
| 268 |
+
label="",
|
| 269 |
+
value="<div style='background-color: #2d3748; color: white; padding: 20px; border-radius: 10px; text-align: center;'>Здесь появятся релевантные чанки...</div>",
|
| 270 |
+
)
|
| 271 |
+
|
| 272 |
+
with gr.Column(scale=1):
|
| 273 |
+
chunks_output = gr.HTML(
|
| 274 |
+
label="Релевантные чанки",
|
| 275 |
+
value="<div style='background-color: #2d3748; color: white; padding: 20px; border-radius: 10px; text-align: center;'>Здесь появятся релевантные чанки...</div>",
|
| 276 |
)
|
| 277 |
|
| 278 |
+
switch_btn.click(
|
| 279 |
+
fn=switch_model_func,
|
| 280 |
+
inputs=[model_dropdown],
|
| 281 |
+
outputs=[model_status]
|
|
|
|
| 282 |
)
|
| 283 |
|
| 284 |
+
ask_btn.click(
|
| 285 |
+
fn=answer_question_func,
|
| 286 |
+
inputs=[question_input],
|
| 287 |
+
outputs=[answer_output, sources_output, chunks_output]
|
| 288 |
)
|
| 289 |
|
| 290 |
+
question_input.submit(
|
| 291 |
+
fn=answer_question_func,
|
| 292 |
+
inputs=[question_input],
|
| 293 |
+
outputs=[answer_output, sources_output, chunks_output]
|
| 294 |
+
)
|
| 295 |
return demo
|
| 296 |
|
| 297 |
+
|
| 298 |
+
query_engine = None
|
| 299 |
+
chunks_df = None
|
| 300 |
+
reranker = None
|
| 301 |
+
vector_index = None
|
| 302 |
+
current_model = DEFAULT_MODEL
|
| 303 |
+
|
| 304 |
+
def main_answer_question(question):
|
| 305 |
+
global query_engine, reranker, current_model, chunks_df
|
| 306 |
+
answer_html, sources_html, chunks_html = answer_question(
|
| 307 |
+
question, query_engine, reranker, current_model, chunks_df
|
| 308 |
+
)
|
| 309 |
+
return answer_html, sources_html, chunks_html
|
| 310 |
+
|
| 311 |
+
def main_switch_model(model_name):
|
| 312 |
+
global query_engine, vector_index, current_model
|
| 313 |
+
|
| 314 |
+
new_query_engine, status_message = switch_model(model_name, vector_index)
|
| 315 |
+
if new_query_engine:
|
| 316 |
+
query_engine = new_query_engine
|
| 317 |
+
current_model = model_name
|
| 318 |
+
|
| 319 |
+
return status_message
|
| 320 |
+
|
| 321 |
+
def main():
|
| 322 |
+
global query_engine, chunks_df, reranker, vector_index, current_model
|
| 323 |
+
|
| 324 |
+
log_message("Запуск AIEXP - AI Expert для нормативной документации")
|
| 325 |
|
| 326 |
+
query_engine, chunks_df, reranker, vector_index, chunk_info = initialize_system(
|
| 327 |
+
repo_id=HF_REPO_ID,
|
| 328 |
+
hf_token=HF_TOKEN,
|
| 329 |
+
download_dir=DOWNLOAD_DIR,
|
| 330 |
+
json_files_dir=JSON_FILES_DIR,
|
| 331 |
+
table_data_dir=TABLE_DATA_DIR,
|
| 332 |
+
image_data_dir=IMAGE_DATA_DIR,
|
| 333 |
+
use_json_instead_csv=True,
|
| 334 |
+
)
|
| 335 |
+
|
| 336 |
+
if query_engine:
|
| 337 |
+
log_message("Запуск веб-интерфейса")
|
| 338 |
+
demo = create_demo_interface(
|
| 339 |
+
answer_question_func=main_answer_question,
|
| 340 |
+
switch_model_func=main_switch_model,
|
| 341 |
+
current_model=current_model,
|
| 342 |
+
chunk_info=chunk_info
|
| 343 |
+
)
|
| 344 |
demo.launch(
|
| 345 |
server_name="0.0.0.0",
|
| 346 |
server_port=7860,
|
|
|
|
| 348 |
debug=False
|
| 349 |
)
|
| 350 |
else:
|
| 351 |
+
log_message("Невозможно запустить приложение из-за ошибки инициализации")
|
| 352 |
+
sys.exit(1)
|
| 353 |
+
|
| 354 |
+
if __name__ == "__main__":
|
| 355 |
+
main()
|