app.py

import os
from docx import Document
from sentence_transformers import SentenceTransformer
from langchain_community.vectorstores import FAISS
from langchain_community.embeddings import HuggingFaceEmbeddings
from transformers import AutoModelForSeq2SeqLM, AutoTokenizer
from nltk.tokenize import sent_tokenize
import torch
import gradio as gr
import pickle
import nltk

# Download NLTK punkt resource if not already downloaded
try:
    nltk.data.find('tokenizers/punkt')
except LookupError:
    nltk.download('punkt')

# Function to extract text from a PDF file
def extract_text_from_pdf(pdf_path):
    text = ""
    try:
        doc = fitz.open(pdf_path)
        for page_num in range(len(doc)):
            page = doc.load_page(page_num)
            text += page.get_text()
    except Exception as e:
        print(f"Error extracting text from PDF: {e}")
    return text

# Function to extract text from a Word document
def extract_text_from_docx(docx_path):
    text = ""
    try:
        doc = Document(docx_path)
        text = "\n".join([para.text for para in doc.paragraphs])
    except Exception as e:
        print(f"Error extracting text from DOCX: {e}")
    return text

# Initialize the embedding model
embedding_model = SentenceTransformer('all-MiniLM-L6-v2')

# Hugging Face API token
api_token = os.getenv('HUGGINGFACEHUB_API_TOKEN')
if not api_token:
    raise ValueError("HUGGINGFACEHUB_API_TOKEN environment variable is not set")

# Define RAG models
generator_model_name = "facebook/bart-base"
retriever_model_name = "facebook/bart-base"  # Can be the same as generator
generator = AutoModelForSeq2SeqLM.from_pretrained(generator_model_name)
generator_tokenizer = AutoTokenizer.from_pretrained(generator_model_name)
retriever = AutoModelForSeq2SeqLM.from_pretrained(retriever_model_name)
retriever_tokenizer = AutoTokenizer.from_pretrained(retriever_model_name)

# Initialize FAISS index using LangChain
hf_embeddings = HuggingFaceEmbeddings(model_name='sentence-transformers/all-MiniLM-L6-v2')

# Load or create FAISS index
index_path = "faiss_index.pkl"
if os.path.exists(index_path):
    with open(index_path, "rb") as f:
        faiss_index = pickle.load(f)
        print("Loaded FAISS index from faiss_index.pkl")
else:
    faiss_index = FAISS(embedding_function=hf_embeddings)

def preprocess_text(text):
    sentences = sent_tokenize(text)
    return sentences

def upload_files(files):
    global faiss_index
    try:
        for file in files:
            if file.name.endswith('.pdf'):
                text = extract_text_from_pdf(file.name)
            elif file.name.endswith('.docx'):
                text = extract_text_from_docx(file.name)
            else:
                return {"error": "Unsupported file format"}

            # Preprocess text
            sentences = preprocess_text(text)

            # Encode sentences and add to FAISS index
            embeddings = embedding_model.encode(sentences)
            for sentence, embedding in zip(sentences, embeddings):
                faiss_index.add_sentence(sentence, embedding)

        # Save the updated index
        with open(index_path, "wb") as f:
            pickle.dump(faiss_index, f)

        return {"message": "Files processed successfully"}
    except Exception as e:
        print(f"Error processing files: {e}")
        return {"error": str(e)}  # Provide informative error message

def process_and_query(state, files, question):
    if files:
        upload_result = upload_files(files)
        if "error" in upload_result:
            return upload_result

    if question:
        # Preprocess the question
        question_embedding = embedding_model.encode([question])

        # Search the FAISS index for similar passages
        retrieved_results = faiss_index.similarity_search(question, k=5)  # Retrieve top 5 passages
        retrieved_passages = [result['text'] for result in retrieved_results]

        # Use generator model to generate response based on question and retrieved passages
        combined_input = question + " ".join(retrieved_passages)
        inputs = generator_tokenizer(combined_input, return_tensors="pt")
        with torch.no_grad():
            generator_outputs = generator.generate(**inputs)
            generated_text = generator_tokenizer.decode(generator_outputs[0], skip_special_tokens=True)

        # Update conversation history
        state["conversation"].append({"question": question, "answer": generated_text})

        return {"message": generated_text, "conversation": state["conversation"]}

    return {"error": "No question provided"}

# Create Gradio interface
with gr.Blocks() as demo:
    gr.Markdown("## Document Upload and Query System")

    with gr.Tab("Upload Files"):
        upload = gr.File(file_count="multiple", label="Upload PDF or DOCX files")
        upload_button = gr.Button("Upload")
        upload_output = gr.Textbox()
        upload_button.click(fn=upload_files, inputs=upload, outputs=upload_output)

    with gr.Tab("Query"):
        query = gr.Textbox(label="Enter your query")
        query_button = gr.Button("Search")
        query_output = gr.Textbox()
        query_button.click(fn=process_and_query, inputs=[query], outputs=query_output)

demo.launch()