Spaces:
Build error
Build error
File size: 42,030 Bytes
7c711a3 96fd4b2 3b16e2f 96fd4b2 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 fa9395d 7c711a3 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 |
import os
import subprocess
subprocess.run(["pip", "install", "-q", "pymupdf", "langchain", "langchain_community", "sentence-transformers", "faiss-cpu", "llama-cpp-python", "gradio", "transformers", "rank_bm25"], check=True)
subprocess.run(["curl", "--proto", "=https", "--tlsv1.2", "-sSf", "https://sh.rustup.rs | sh"], check=True)
subprocess.run("source $HOME/.cargo/env", shell=True, check=True)
subprocess.run(["pip", "install", "-q", "git+https://github.com/chroma-core/chroma.git"], check=True)
subprocess.run(["wget", "-q", "-O", "models/mistral-7b-instruct-v0.3.Q8_0.gguf", "https://huggingface.co/MaziyarPanahi/Mistral-7B-Instruct-v0.3-GGUF/resolve/main/Mistral-7B-Instruct-v0.3.Q8_0.gguf"])
os.makedirs("pdfs", exist_ok=True)
os.makedirs("models", exist_ok=True)
import re
import fitz # PyMuPDF
import numpy as np
import gc
import torch
import time
import shutil
import hashlib
import pickle
import traceback
from typing import List, Dict, Any, Tuple, Optional, Union, Generator
from dataclasses import dataclass
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain_community.vectorstores import FAISS
from langchain_community.embeddings import HuggingFaceEmbeddings
from llama_cpp import Llama
import gradio as gr
from rank_bm25 import BM25Okapi
import nltk
from nltk.tokenize import word_tokenize
from nltk.corpus import stopwords
from sentence_transformers import CrossEncoder
# Download nltk resources
try:
nltk.download('punkt', quiet=True)
nltk.download('stopwords', quiet=True)
except:
print("Failed to download NLTK resources, continuing without them")
# === MEMORY MANAGEMENT UTILITIES ===
def clear_memory():
"""Clear memory to prevent OOM errors"""
gc.collect()
torch.cuda.empty_cache() if torch.cuda.is_available() else None
# === PDF PROCESSING ===
@dataclass
class PDFChunk:
"""Class to represent a chunk of text extracted from a PDF"""
text: str
source: str
page_num: int
chunk_id: int
class PDFProcessor:
def __init__(self, pdf_dir: str = "pdfs"):
"""Initialize PDF processor
Args:
pdf_dir: Directory containing PDF files
"""
self.pdf_dir = pdf_dir
# Smaller chunk size with more overlap for better retrieval
self.text_splitter = RecursiveCharacterTextSplitter(
chunk_size=384,
chunk_overlap=288, # 75% overlap for better context preservation
length_function=len,
is_separator_regex=False,
)
# Create cache directory
self.cache_dir = os.path.join(os.getcwd(), "pdf_cache")
os.makedirs(self.cache_dir, exist_ok=True)
def list_pdfs(self) -> List[str]:
"""List all PDF files in the directory"""
if not os.path.exists(self.pdf_dir):
return []
return [f for f in os.listdir(self.pdf_dir) if f.lower().endswith('.pdf')]
def _get_cache_path(self, pdf_path: str) -> str:
"""Get the cache file path for a PDF"""
pdf_hash = hashlib.md5(open(pdf_path, 'rb').read(8192)).hexdigest()
return os.path.join(self.cache_dir, f"{os.path.basename(pdf_path)}_{pdf_hash}.pkl")
def _is_cached(self, pdf_path: str) -> bool:
"""Check if a PDF is cached"""
cache_path = self._get_cache_path(pdf_path)
return os.path.exists(cache_path)
def _load_from_cache(self, pdf_path: str) -> List[PDFChunk]:
"""Load chunks from cache"""
cache_path = self._get_cache_path(pdf_path)
try:
with open(cache_path, 'rb') as f:
return pickle.load(f)
except:
return None
def _save_to_cache(self, pdf_path: str, chunks: List[PDFChunk]) -> None:
"""Save chunks to cache"""
cache_path = self._get_cache_path(pdf_path)
try:
with open(cache_path, 'wb') as f:
pickle.dump(chunks, f)
except Exception as e:
print(f"Warning: Failed to cache PDF {pdf_path}: {str(e)}")
def clean_text(self, text: str) -> str:
"""Clean extracted text"""
# Remove excessive whitespace
text = re.sub(r'\s+', ' ', text).strip()
# Remove header/footer patterns (common in PDFs)
text = re.sub(r'(?<!\w)page \d+(?!\w)', '', text, flags=re.IGNORECASE)
return text
def extract_text_from_pdf(self, pdf_path: str) -> List[PDFChunk]:
"""Extract text content from a PDF file with improved extraction
Args:
pdf_path: Path to the PDF file
Returns:
List of PDFChunk objects extracted from the PDF
"""
# Check cache first
if self._is_cached(pdf_path):
cached_chunks = self._load_from_cache(pdf_path)
if cached_chunks:
print(f"Loaded {len(cached_chunks)} chunks from cache for {os.path.basename(pdf_path)}")
return cached_chunks
try:
doc = fitz.open(pdf_path)
pdf_chunks = []
pdf_name = os.path.basename(pdf_path)
for page_num in range(len(doc)):
page = doc.load_page(page_num)
# Extract text with more options for better quality
page_text = page.get_text("text", sort=True)
# Try to extract text with alternative layout analysis if the text is too short
if len(page_text) < 100:
try:
page_text = page.get_text("dict", sort=True)
# Convert dict to text
if isinstance(page_text, dict) and "blocks" in page_text:
extracted_text = ""
for block in page_text["blocks"]:
if "lines" in block:
for line in block["lines"]:
if "spans" in line:
for span in line["spans"]:
if "text" in span:
extracted_text += span["text"] + " "
page_text = extracted_text
except:
# Fallback to default extraction
page_text = page.get_text("text")
# Clean the text
page_text = self.clean_text(page_text)
# Extract tables
try:
tables = page.find_tables()
if tables and hasattr(tables, "tables"):
for table in tables.tables:
table_text = ""
for i, row in enumerate(table.rows):
row_cells = []
for cell in row.cells:
if hasattr(cell, "rect"):
cell_text = page.get_text("text", clip=cell.rect)
cell_text = self.clean_text(cell_text)
row_cells.append(cell_text)
if row_cells:
table_text += " | ".join(row_cells) + "\n"
# Add table text to page text
if table_text.strip():
page_text += "\n\nTABLE:\n" + table_text
except Exception as table_err:
print(f"Warning: Skipping table extraction for page {page_num}: {str(table_err)}")
# Split the page text into chunks
if page_text.strip():
page_chunks = self.text_splitter.split_text(page_text)
# Create PDFChunk objects
for i, chunk_text in enumerate(page_chunks):
pdf_chunks.append(PDFChunk(
text=chunk_text,
source=pdf_name,
page_num=page_num + 1, # 1-based page numbering for humans
chunk_id=i
))
# Clear memory periodically
if page_num % 10 == 0:
clear_memory()
doc.close()
# Cache the results
self._save_to_cache(pdf_path, pdf_chunks)
return pdf_chunks
except Exception as e:
print(f"Error extracting text from {pdf_path}: {str(e)}")
return []
def process_pdf(self, pdf_name: str) -> List[PDFChunk]:
"""Process a single PDF file and extract chunks
Args:
pdf_name: Name of the PDF file in the pdf_dir
Returns:
List of PDFChunk objects from the PDF
"""
pdf_path = os.path.join(self.pdf_dir, pdf_name)
return self.extract_text_from_pdf(pdf_path)
def process_all_pdfs(self, batch_size: int = 3) -> List[PDFChunk]:
"""Process all PDFs in batches to manage memory
Args:
batch_size: Number of PDFs to process in each batch
Returns:
List of all PDFChunk objects from all PDFs
"""
all_chunks = []
pdf_files = self.list_pdfs()
if not pdf_files:
print("No PDF files found in the directory.")
return []
# Process PDFs in batches
for i in range(0, len(pdf_files), batch_size):
batch = pdf_files[i:i+batch_size]
print(f"Processing batch {i//batch_size + 1}/{(len(pdf_files)-1)//batch_size + 1}")
for pdf_name in batch:
print(f"Processing {pdf_name}")
chunks = self.process_pdf(pdf_name)
all_chunks.extend(chunks)
print(f"Extracted {len(chunks)} chunks from {pdf_name}")
# Clear memory after each batch
clear_memory()
return all_chunks
# === VECTOR DATABASE SETUP ===
class VectorDBManager:
def __init__(self, model_name: str = "sentence-transformers/all-MiniLM-L6-v2"):
"""Initialize vector database manager
Args:
model_name: Name of the embedding model
"""
# Initialize embedding model with normalization
try:
self.embedding_model = HuggingFaceEmbeddings(
model_name=model_name,
model_kwargs={"device": "cpu"},
encode_kwargs={"normalize_embeddings": True}
)
except Exception as e:
print(f"Error initializing embedding model {model_name}: {str(e)}")
print("Falling back to all-MiniLM-L6-v2 model")
self.embedding_model = HuggingFaceEmbeddings(
model_name="sentence-transformers/all-MiniLM-L6-v2",
model_kwargs={"device": "cpu"},
encode_kwargs={"normalize_embeddings": True}
)
# Initialize cross-encoder for re-ranking
try:
self.cross_encoder = CrossEncoder('cross-encoder/ms-marco-MiniLM-L-6-v2')
except Exception as e:
print(f"Error initializing cross-encoder: {str(e)}")
self.cross_encoder = None
self.vectordb = None
# BM25 index for hybrid search
self.bm25_index = None
self.chunks = []
self.tokenized_chunks = []
def _prepare_bm25(self, chunks: List[PDFChunk]):
"""Prepare BM25 index for hybrid search"""
# Tokenize chunks for BM25
try:
tokenized_chunks = []
for chunk in chunks:
# Tokenize and remove stopwords
tokens = word_tokenize(chunk.text.lower())
stop_words = set(stopwords.words('english'))
filtered_tokens = [w for w in tokens if w.isalnum() and w not in stop_words]
tokenized_chunks.append(filtered_tokens)
# Create BM25 index
self.bm25_index = BM25Okapi(tokenized_chunks)
self.tokenized_chunks = tokenized_chunks
except Exception as e:
print(f"Error creating BM25 index: {str(e)}")
print(traceback.format_exc())
self.bm25_index = None
def create_vector_db(self, chunks: List[PDFChunk]) -> None:
"""Create vector database from text chunks
Args:
chunks: List of PDFChunk objects
"""
try:
if not chunks or len(chunks) == 0:
print("ERROR: No chunks provided to create vector database")
return
print(f"Creating vector DB with {len(chunks)} chunks")
# Store chunks for hybrid search
self.chunks = chunks
# Prepare data for vector DB
chunk_texts = [chunk.text for chunk in chunks]
# Create BM25 index for hybrid search
print("Creating BM25 index for hybrid search")
self._prepare_bm25(chunks)
# Process in smaller batches to manage memory
batch_size = 32
all_embeddings = []
for i in range(0, len(chunk_texts), batch_size):
batch = chunk_texts[i:i+batch_size]
print(f"Embedding batch {i//batch_size + 1}/{(len(chunk_texts)-1)//batch_size + 1}")
# Generate embeddings for the batch
batch_embeddings = self.embedding_model.embed_documents(batch)
all_embeddings.extend(batch_embeddings)
# Clear memory after each batch
clear_memory()
# Create FAISS index
print(f"Creating FAISS index with {len(all_embeddings)} embeddings")
self.vectordb = FAISS.from_embeddings(
text_embeddings=list(zip(chunk_texts, all_embeddings)),
embedding=self.embedding_model
)
print(f"Vector database created with {len(chunks)} documents")
except Exception as e:
print(f"Error creating vector database: {str(e)}")
print(traceback.format_exc())
raise
def _format_chunk_with_metadata(self, chunk: PDFChunk) -> str:
"""Format a chunk with its metadata for better context"""
return f"Source: {chunk.source} | Page: {chunk.page_num}\n\n{chunk.text}"
def _rerank_with_cross_encoder(self, query: str, chunks: List[PDFChunk], k: int = 5) -> List[PDFChunk]:
"""Re-rank chunks using cross-encoder
Args:
query: User query
chunks: List of retrieved chunks
k: Number of top chunks to return
Returns:
Re-ranked chunks
"""
if not self.cross_encoder or not chunks:
return chunks[:k] if len(chunks) > k else chunks
try:
# Prepare passage pairs for re-ranking
pairs = [[query, chunk.text] for chunk in chunks]
# Score passages in smaller batches to prevent OOM
batch_size = 16
all_scores = []
for i in range(0, len(pairs), batch_size):
batch_pairs = pairs[i:i+batch_size]
batch_scores = self.cross_encoder.predict(batch_pairs)
all_scores.extend(batch_scores)
# Clear memory
clear_memory()
# Create chunk-score pairs
scored_chunks = list(zip(chunks, all_scores))
# Sort by score
scored_chunks.sort(key=lambda x: x[1], reverse=True)
# Return top k chunks
return [chunk for chunk, score in scored_chunks[:k]]
except Exception as e:
print(f"Error during cross-encoder re-ranking: {str(e)}")
# Fallback to original chunks
return chunks[:k] if len(chunks) > k else chunks
def hybrid_search(self, query: str, k: int = 5, alpha: float = 0.7) -> List[str]:
"""Hybrid search combining vector search and BM25 with cross-encoder re-ranking
Args:
query: Query text
k: Number of results to return
alpha: Weight for vector search (1-alpha for BM25)
Returns:
List of formatted documents
"""
if self.vectordb is None:
print("Vector database not initialized")
return []
try:
# Get vector search results
vector_results = self.vectordb.similarity_search(query, k=k*3) # Get more for re-ranking
vector_texts = [doc.page_content for doc in vector_results]
retrieved_chunks = []
# Combine with BM25 if available
if self.bm25_index is not None:
try:
# Tokenize query for BM25
query_tokens = word_tokenize(query.lower())
stop_words = set(stopwords.words('english'))
filtered_query = [w for w in query_tokens if w.isalnum() and w not in stop_words]
# Get BM25 scores
bm25_scores = self.bm25_index.get_scores(filtered_query)
# Combine scores (normalized)
combined_results = []
seen_texts = set()
# First add vector results with their positions as scores
for i, text in enumerate(vector_texts):
if text not in seen_texts:
seen_texts.add(text)
# Find corresponding chunk
for j, chunk in enumerate(self.chunks):
if chunk.text == text:
# Combine scores: alpha * vector_score + (1-alpha) * bm25_score
# For vector, use inverse of position as score (normalized)
vector_score = 1.0 - (i / len(vector_texts))
# Normalize BM25 score
bm25_score = bm25_scores[j] / max(bm25_scores) if max(bm25_scores) > 0 else 0
combined_score = alpha * vector_score + (1-alpha) * bm25_score
combined_results.append((chunk, combined_score))
break
# Sort by combined score
combined_results.sort(key=lambda x: x[1], reverse=True)
# Get top k*2 results for re-ranking
retrieved_chunks = [item[0] for item in combined_results[:k*2]]
except Exception as e:
print(f"Error in BM25 scoring: {str(e)}")
# Fallback to vector search results
retrieved_chunks = [self.chunks[i] for i, text in enumerate(self.chunks)
if text.text in vector_texts[:k*2]]
else:
# Just use vector search results if BM25 is not available
retrieved_chunks = [self.chunks[i] for i, chunk in enumerate(self.chunks)
if chunk.text in vector_texts[:k*2]]
# Re-rank with cross-encoder
if retrieved_chunks:
reranked_chunks = self._rerank_with_cross_encoder(query, retrieved_chunks, k)
# Format results with metadata
final_results = [self._format_chunk_with_metadata(chunk) for chunk in reranked_chunks]
else:
# Fallback to basic results
final_results = vector_texts[:k]
return final_results
except Exception as e:
print(f"Error during hybrid search: {str(e)}")
return []
# === QUERY EXPANSION ===
class QueryExpander:
def __init__(self, llm_model):
"""Initialize query expander
Args:
llm_model: LLM model for query expansion
"""
self.llm = llm_model
def expand_query(self, query: str) -> str:
"""Expand the query using the LLM to improve retrieval
Args:
query: Original query
Returns:
Expanded query
"""
try:
prompt = f"""<s>[INST] I need to search for documents related to this question: "{query}"
Please help me expand this query by identifying key concepts, synonyms, and related terms that might be used in the documents.
Return only the expanded search query, without any explanations or additional text. [/INST]"""
expanded = self.llm.generate(prompt, max_tokens=100, temperature=0.3)
# Combine original and expanded
combined = f"{query} {expanded}"
# Limit length
if len(combined) > 300:
combined = combined[:300]
return combined
except:
# Return original query if expansion fails
return query
# === LLM SETUP ===
class MistralModel:
def __init__(self, model_path: str = "models/mistral-7b-instruct-v0.3.Q8_0.gguf"):
"""Initialize Mistral model
Args:
model_path: Path to the model file
"""
try:
# Initialize Mistral with llama.cpp
self.llm = Llama(
model_path=model_path,
n_ctx=4096, # Increased context window for better reasoning
n_batch=256, # Batch size to save memory
n_gpu_layers=0, # Run on CPU only for Colab free tier
verbose=False
)
except Exception as e:
print(f"Error initializing Mistral model: {str(e)}")
raise
def generate(self, prompt: str,
max_tokens: int = 512,
temperature: float = 0.7,
top_p: float = 0.9,
stream: bool = False) -> Union[str, Generator[str, None, None]]:
"""Generate text using Mistral
Args:
prompt: Input prompt
max_tokens: Maximum number of tokens to generate
temperature: Sampling temperature
top_p: Top-p sampling parameter
stream: Whether to stream the output
Returns:
Generated text or generator if streaming
"""
try:
if stream:
return self._generate_stream(prompt, max_tokens, temperature, top_p)
else:
output = self.llm(
prompt,
max_tokens=max_tokens,
temperature=temperature,
top_p=top_p,
echo=False
)
return output["choices"][0]["text"].strip()
except Exception as e:
print(f"Error generating text: {str(e)}")
return "Error: Could not generate response."
def _generate_stream(self, prompt: str,
max_tokens: int = 512,
temperature: float = 0.7,
top_p: float = 0.9) -> Generator[str, None, None]:
"""Stream text generation using Mistral
Args:
prompt: Input prompt
max_tokens: Maximum number of tokens to generate
temperature: Sampling temperature
top_p: Top-p sampling parameter
Yields:
Generated text tokens
"""
response = ""
for output in self.llm(
prompt,
max_tokens=max_tokens,
temperature=temperature,
top_p=top_p,
echo=False,
stream=True
):
token = output["choices"][0]["text"]
response += token
yield response
# === SELF-CHECKING ===
class SelfChecker:
def __init__(self, llm_model):
"""Initialize self-checker for improved response quality
Args:
llm_model: LLM model to use for checking
"""
self.llm = llm_model
def check_answer(self, query: str, initial_answer: str, contexts: List[str]) -> str:
"""Check if answer is correct and complete based on the contexts
Args:
query: User query
initial_answer: Initial generated answer
contexts: Retrieved contexts used to generate the answer
Returns:
Improved answer after reflection
"""
# Guard against very long inputs that could cause runtime disconnection
# Limit contexts to prevent excessive token usage
max_contexts_len = 4000
contexts_text = "\n\n".join(contexts)
if len(contexts_text) > max_contexts_len:
# Truncate while keeping as many complete contexts as possible
truncated_contexts = []
current_len = 0
for ctx in contexts:
if current_len + len(ctx) + 2 <= max_contexts_len:
truncated_contexts.append(ctx)
current_len += len(ctx) + 2
else:
break
contexts_text = "\n\n".join(truncated_contexts)
# Check if we should skip reflection to prevent disconnection
if len(initial_answer) + len(contexts_text) + len(query) > 6000:
print("Skipping reflection due to excessive input length")
return initial_answer
try:
prompt = f"""<s>[INST] You're an AI assistant tasked with evaluating and improving an answer to a user query.
QUERY: {query}
INITIAL ANSWER: {initial_answer}
AVAILABLE CONTEXTS:
{contexts_text}
First, carefully check if the initial answer:
1. Is factually accurate based on the provided contexts
2. Addresses all aspects of the user's query
3. Contains any information not supported by the contexts
4. Misses important information from the contexts
Then improve the answer to fix any issues identified. The final answer should:
- Be comprehensive and accurate based ONLY on the contexts
- Not include any unsupported information
- Be well-structured and clear
- Cite specific sources when appropriate (e.g., "According to [Source, Page X]...")
Provide ONLY the improved answer without explanations about your reasoning process. [/INST]"""
# We use slightly lower temperature for more focused reflection
improved_answer = self.llm.generate(
prompt,
max_tokens=1024,
temperature=0.3,
stream=False
)
# If reflection produced nothing useful, return original answer
if not improved_answer or len(improved_answer) < 10:
return initial_answer
return improved_answer
except Exception as e:
# On any error, return the original answer to ensure robustness
print(f"Self-check error: {str(e)}")
return initial_answer
# === RAG SYSTEM ===
class RAGSystem:
def __init__(self, pdf_processor: PDFProcessor,
vector_db: VectorDBManager,
model: MistralModel):
"""Initialize RAG system
Args:
pdf_processor: PDF processor instance
vector_db: Vector database manager instance
model: LLM model instance
"""
self.pdf_processor = pdf_processor
self.vector_db = vector_db
self.model = model
self.query_expander = QueryExpander(model)
self.self_checker = SelfChecker(model)
self.is_initialized = False
def process_documents(self) -> bool:
"""Process all documents and create vector database
Returns:
True if successful, False otherwise
"""
try:
# Process PDFs
chunks = self.pdf_processor.process_all_pdfs()
if not chunks:
print("No chunks were extracted from PDFs")
return False
print(f"Total chunks extracted: {len(chunks)}")
# Create vector database
print("Creating vector database...")
self.vector_db.create_vector_db(chunks)
# Verify success
if self.vector_db.vectordb is None:
print("Failed to create vector database")
return False
# Set initialization flag
self.is_initialized = True
return True
except Exception as e:
print(f"Error processing documents: {str(e)}")
print(traceback.format_exc())
return False
def generate_prompt(self, query: str, contexts: List[str]) -> str:
"""Generate prompt for the LLM with better instructions
Args:
query: User query
contexts: Retrieved contexts
Returns:
Formatted prompt
"""
# Format contexts with numbering for better reference
formatted_contexts = ""
for i, context in enumerate(contexts):
formatted_contexts += f"[CONTEXT {i+1}]\n{context}\n\n"
# Create prompt with Mistral's chat format
prompt = f"""<s>[INST] You are an AI assistant that answers questions based on the provided context information.
User Query: {query}
Below are relevant passages from documents that might help answer the query:
{formatted_contexts}
Using ONLY the information provided in the context above, provide a comprehensive answer to the user's query.
If the provided context doesn't contain relevant information to answer the query, clearly state: "I don't have enough information in the provided context to answer this question."
Do not use any prior knowledge that is not contained in the provided context.
If quoting from the context, mention the source document and page number.
Organize your answer in a clear, coherent manner. [/INST]"""
return prompt
def answer_query(self, query: str, k: int = 5, max_tokens: int = 512,
temperature: float = 0.7, stream: bool = False, enable_reflection: bool = True) -> Union[str, Generator[str, None, None]]:
"""Answer a query using RAG with query expansion and self-checking
Args:
query: User query
k: Number of contexts to retrieve
max_tokens: Maximum number of tokens to generate
temperature: Temperature for generation
stream: Whether to stream the output
enable_reflection: Whether to enable self-reflection for better answers
Returns:
Answer text or generator if streaming
"""
# Check if system is initialized
if not self.is_initialized or self.vector_db.vectordb is None:
return "Error: Documents have not been processed yet. Please process documents first."
try:
# Expand query for better retrieval
expanded_query = self.query_expander.expand_query(query)
print(f"Expanded query: {expanded_query}")
# Retrieve relevant contexts using hybrid search
contexts = self.vector_db.hybrid_search(expanded_query, k=k)
if not contexts:
return "No relevant information found in the documents. Please try a different query or check if documents were processed correctly."
# Generate prompt with improved instructions
prompt = self.generate_prompt(query, contexts)
# For streaming, we can't do self-checking
if stream:
return self.model.generate(
prompt,
max_tokens=max_tokens,
temperature=temperature,
stream=True
)
# Generate initial answer
initial_answer = self.model.generate(
prompt,
max_tokens=max_tokens,
temperature=temperature,
stream=False
)
# Perform self-checking if enabled and initial answer exists
if enable_reflection and initial_answer and len(initial_answer) > 10:
try:
print("Performing self-checking to improve answer quality...")
improved_answer = self.self_checker.check_answer(query, initial_answer, contexts)
return improved_answer
except Exception as e:
print(f"Error during self-checking: {str(e)}")
# Fallback to initial answer if self-checking fails
return initial_answer
else:
return initial_answer
except Exception as e:
print(f"Error answering query: {str(e)}")
print(traceback.format_exc())
return f"Error processing your query: {str(e)}"
# === GRADIO UI ===
class RAGUI:
def __init__(self, rag_system: RAGSystem):
"""Initialize RAG UI
Args:
rag_system: RAG system instance
"""
self.rag_system = rag_system
self.pdf_dir = rag_system.pdf_processor.pdf_dir
self.interface = None
def _list_uploaded_pdfs(self) -> str:
"""List all uploaded PDFs"""
pdfs = self.rag_system.pdf_processor.list_pdfs()
if not pdfs:
return "No PDFs uploaded yet."
return "\n".join([f"- {pdf}" for pdf in pdfs])
def upload_pdf(self, files) -> str:
"""Upload PDF files
Args:
files: File objects
Returns:
Status message
"""
try:
# Create directory if it doesn't exist
os.makedirs(self.pdf_dir, exist_ok=True)
# Copy files to pdf directory
for file in files:
shutil.copy(file.name, os.path.join(self.pdf_dir, os.path.basename(file.name)))
return f"Successfully uploaded {len(files)} file(s). Please process documents to make them searchable."
except Exception as e:
return f"Error uploading files: {str(e)}"
def process_documents(self) -> str:
"""Process documents and create vector database
Returns:
Status message
"""
try:
# Check if there are PDFs
pdf_files = self.rag_system.pdf_processor.list_pdfs()
if not pdf_files:
return "No PDF files uploaded. Please upload PDFs first."
# Process PDFs
start_time = time.time()
success = self.rag_system.process_documents()
process_time = time.time() - start_time
if success:
return f"Successfully processed {len(pdf_files)} PDF file(s) in {process_time:.2f} seconds. You can now ask questions."
else:
return "Failed to process documents. Check the logs for details."
except Exception as e:
return f"Error processing documents: {str(e)}"
def answer_query(self, query: str, stream_output: bool = True,
k: int = 4, temperature: float = 0.7,
enable_reflection: bool = True) -> str:
"""Answer a query using RAG
Args:
query: User query
stream_output: Whether to stream the output
k: Number of contexts to retrieve
temperature: Temperature for text generation
enable_reflection: Whether to use reflection to improve answers
Returns:
Answer text
"""
if not query or query.strip() == "":
return "Please enter a query."
# Check if system is initialized
if not self.rag_system.is_initialized:
return "Documents have not been processed yet. Please process documents first."
try:
# For streaming, we need to handle gradio uniqueness
if stream_output:
# We can't stream with reflection
return self.rag_system.answer_query(
query,
k=k,
max_tokens=1024,
temperature=temperature,
stream=True,
enable_reflection=False
)
else:
return self.rag_system.answer_query(
query,
k=k,
max_tokens=1024,
temperature=temperature,
stream=False,
enable_reflection=enable_reflection
)
except Exception as e:
print(f"Error in answer_query: {str(e)}")
print(traceback.format_exc())
return f"Error processing your query: {str(e)}"
def launch(self):
"""Launch Gradio UI"""
try:
with gr.Blocks(title="Document Q&A System") as self.interface:
gr.Markdown("# PDF Question Answering System")
gr.Markdown("Upload PDF documents and ask questions about their content.")
with gr.Tab("Upload & Process"):
with gr.Row():
with gr.Column():
upload_button = gr.File(
label="Upload PDF Files",
file_count="multiple",
file_types=[".pdf"]
)
upload_output = gr.Textbox(
label="Upload Status",
interactive=False
)
upload_btn = gr.Button("Upload Files")
with gr.Column():
pdf_list = gr.Textbox(
label="Uploaded PDFs",
value=self._list_uploaded_pdfs(),
interactive=False
)
refresh_btn = gr.Button("Refresh List")
process_btn = gr.Button("Process Documents")
process_output = gr.Textbox(
label="Processing Status",
interactive=False
)
with gr.Tab("Ask Questions"):
with gr.Row():
with gr.Column():
query_input = gr.Textbox(
label="Enter your question",
placeholder="What are the main findings of the report?",
lines=2
)
with gr.Row():
k_slider = gr.Slider(
minimum=1,
maximum=10,
value=4,
step=1,
label="Number of contexts to retrieve"
)
temp_slider = gr.Slider(
minimum=0.1,
maximum=1.0,
value=0.7,
step=0.1,
label="Temperature"
)
with gr.Row():
stream_checkbox = gr.Checkbox(
label="Stream output",
value=True
)
reflection_checkbox = gr.Checkbox(
label="Use self-reflection (disables streaming)",
value=True
)
query_btn = gr.Button("Submit Question")
answer_output = gr.Textbox(
label="Answer",
interactive=False,
lines=15
)
# Event handlers
upload_btn.click(
fn=self.upload_pdf,
inputs=[upload_button],
outputs=[upload_output]
)
refresh_btn.click(
fn=lambda: self._list_uploaded_pdfs(),
inputs=[],
outputs=[pdf_list]
)
process_btn.click(
fn=self.process_documents,
inputs=[],
outputs=[process_output]
)
query_btn.click(
fn=self.answer_query,
inputs=[query_input, stream_checkbox, k_slider, temp_slider, reflection_checkbox],
outputs=[answer_output]
)
# Checkbox dependency
def update_stream_state(reflection_enabled):
return not reflection_enabled if reflection_enabled else gr.update()
reflection_checkbox.change(
fn=update_stream_state,
inputs=[reflection_checkbox],
outputs=[stream_checkbox]
)
# Launch UI
self.interface.launch(share=True)
except Exception as e:
print(f"Error launching UI: {str(e)}")
print(traceback.format_exc())
# === MAIN APPLICATION ===
def main():
# Initialize components
print("Initializing PDF processor...")
pdf_processor = PDFProcessor()
print("Initializing vector database manager...")
vector_db = VectorDBManager()
print("Initializing Mistral model...")
model = MistralModel()
print("Initializing RAG system...")
rag_system = RAGSystem(pdf_processor, vector_db, model)
print("Initializing UI...")
ui = RAGUI(rag_system)
print("Launching UI...")
ui.launch()
if __name__ == "__main__":
main() |