import streamlit as st import os from streamlit_chat import message import numpy as np import pandas as pd from io import StringIO import io import PyPDF2 import pymupdf import tempfile import base64 from tqdm.auto import tqdm import math from transformers import pipeline import nltk from collections import Counter from nltk.corpus import stopwords from sentence_transformers import SentenceTransformer import torch from langchain_community.llms.ollama import Ollama from langchain.prompts import ChatPromptTemplate from langchain_community.vectorstores import FAISS device = 'cuda' if torch.cuda.is_available() else 'cpu' # if device != 'cuda': # st.markdown(f"you are using {device}. This is much slower than using " # "a CUDA-enabled GPU. If on colab you can change this by " # "clicking Runtime > change runtime type > GPU.") model = SentenceTransformer("sentence-transformers/all-mpnet-base-v2", device=device) def display_title(): selected_value = st.session_state["value"] st.header(f'Vedic Scriptures: {selected_value} :blue[book] :books:') question = "ask anything about scriptures" def open_chat(): question = st.session_state["faq"] if "value" not in st.session_state: st.session_state["value"] = None if "faq" not in st.session_state: st.session_state["faq"] = None st.divider() def highlight_pdf(file_path, text_to_highlight, page_numbers): # Create a temporary file to save the modified PDF # temp_pdf_path = "temp_highlighted_pdf.pdf" # Create a temporary file to save the modified PDF # with tempfile.NamedTemporaryFile(delete=False) as temp_file: # temp_pdf_path = temp_file.name # Open the original PDF doc = pymupdf.open(file_path) pages_to_display = [doc.load_page(page_number) for page_number in page_numbers] print("pages_to_display") print(pages_to_display) # Tokenize the text into words words = text_to_highlight.split() # Remove stopwords stop_words = set(stopwords.words("english")) words = [word for word in words if word.lower() not in stop_words] print(words) # Count the frequency of each word word_counts = Counter(words) # Get the top N most frequent words # top_words = [word for word, _ in word_counts.most_common(5)] # Iterate over each page in the PDF for page in pages_to_display: # Highlight the specified words on the canvas for word in words: highlight_rect = page.search_for(word, quads=True) # Highlight the text # highlight_rect = pymupdf.Rect(word) # highlight_annot = page.add_highlight_annot(highlight_rect) # highlight_annot.set_colors({"stroke": pymupdf.utils.getColor("yellow")}) # highlight_annot.update() page.add_highlight_annot(highlight_rect) # Create a new document with only the specified pages new_doc = pymupdf.open() for page in pages_to_display: new_doc.insert_pdf(doc, from_page=page.number, to_page=page.number) # Save the modified PDF # Save the document to a temporary file with tempfile.NamedTemporaryFile(suffix=".pdf", delete=False) as temp_file: temp_pdf_path = temp_file.name new_doc.save(temp_pdf_path) print(temp_pdf_path) # new_doc.save("example_highlighted.pdf") return temp_pdf_path file_path = "Bhagavad-Gita-As-It-Is.pdf" text_to_highlight = "" sources = [] # Function to display PDF in Streamlit def display_highlighted_pdf(file_path, text_to_highlight, sources): # pdf_path = "../Transformers/Bhagavad-Gita-As-It-Is.pdf" # sources = [7,8] # response_text = "I offer my respectful obeisances unto the lotus feet of my spiritual master and unto the feet of all Vaiñëavas. I offer my respectful" pdf_path = highlight_pdf(file_path=file_path, text_to_highlight=text_to_highlight, page_numbers=sources) with open(pdf_path, "rb") as file: pdf_bytes = file.read() base64_pdf = base64.b64encode(pdf_bytes).decode("utf-8") pdf_display = f'' st.markdown(pdf_display, unsafe_allow_html=True) # Creating a Index(Pinecone Vector Database) import os # import pinecone import pickle def get_faiss_semantic_index(): try: index_path = "./HuggingFaceEmbeddings.pkl" # Load embeddings from the pickle file with open(index_path, "rb") as f: faiss_index = pickle.load(f) st.write("Embeddings loaded successfully.") return faiss_index except Exception as e: st.error(f"Error loading embeddings: {e}") # def promt_engineer(text): PROMPT_TEMPLATE = """ Instructions: -------------------------------------------------------- you're a vedic scriptures AI expert. you shouldnot answer to any other domain specific question. You 1000 Dollars rewards for Before answering questions always try to map the question related to the TITLE > CHAPTER > TEXT > PURPORT. You 1000 Dollars rewards Must provide the Chapter Number and Text number in this format chapter : Text You 1000 Dollars rewards Must provide the Title of the chapter. you also provide source path from where youre answering the question. You 1000 Dollars penality for the relevant questions to answer. Please dont answer from the public sources strictly answer from the context. If the question is not related to the context replay with question doesnot belongs to vedic scriptures or Vedic literature. Answer the question based only on the following context: {context} --- Answer the question based on the above context: {question} """ # # Load the summarization pipeline with the specified model # summarizer = pipeline("summarization", model="facebook/bart-large-cnn") # # Generate the prompt # prompt = prompt_template.format(text=text) # # Generate the summary # summary = summarizer(prompt, max_length=1024, min_length=50)[0]["summary_text"] # with st.sidebar: # st.divider() # st.markdown("*:red[Text Summary Generation]* from above Top 5 **:green[similarity search results]**.") # st.write(summary) # st.divider() def chat_actions(): faiss_index = get_faiss_semantic_index() st.session_state["chat_history"].append( {"role": "user", "content": st.session_state["chat_input"]}, ) # query_embedding = model.encode(st.session_state["chat_input"]) query = st.session_state["chat_input"] docs = faiss_index.similarity_search(query, k=2) for doc in docs: print("\n") print(str(doc.metadata["page"]+1) + ":", doc.page_content) context_text = "\n\n---\n\n".join([doc.page_content for doc in docs]) sources = [doc.metadata.get("page", None) for doc in docs] prompt_template = ChatPromptTemplate.from_template(PROMPT_TEMPLATE) prompt = prompt_template.format(context=context_text, question=query) model = Ollama(model="llama3") response_text = model.invoke(prompt) formatted_response = f"Response: {response_text}\nSources: {sources}" print(formatted_response) st.session_state["chat_history"].append( { "role": "assistant", "content": f"{response_text}", }, # This can be replaced with your chat response logic ) # break; # Example usage file_path = "Bhagavad-Gita-As-It-Is.pdf" text_to_highlight = context_text.strip() display_highlighted_pdf(file_path, response_text, sources) with st.sidebar: option = st.selectbox( "Select Your Favorite Scriptures", ("Bhagvatgeetha", "Bhagavatham", "Ramayanam"), # index=None, # placeholder="Select scriptures...", key="value", on_change=display_title ) st.write("You selected:", option) faq = st.selectbox( "Select Your Favorite Scriptures", ("Why does atheism exist even when all questions are answered in Bhagavad Gita?", "Why don’t all souls surrender to Lord Krishna, although he has demonstrated that everyone is part and parcel of Him, and all can be liberated from all sufferings by surrendering to Him?", "Why do souls misuse their independence by rebelling against Lord Krishna?"), # index=None, # placeholder="Select scriptures...", key="faq", on_change=open_chat ) st.write("You selected:", faq) if "chat_history" not in st.session_state: st.session_state["chat_history"] = [] st.chat_input(question, on_submit=chat_actions, key="chat_input") for i in st.session_state["chat_history"]: with st.chat_message(name=i["role"]): st.write(i["content"])