pages/llm.py

import streamlit as st
from langchain.embeddings import HuggingFaceInstructEmbeddings
from langchain.vectorstores import FAISS
from langchain.text_splitter import CharacterTextSplitter
from langchain.document_loaders import DirectoryLoader, PyPDFLoader
import os
from PyPDF2 import PdfReader
from transformers import pipeline
from transformers import AutoModel
#from googletrans import Translator
#from transformers import *


###########
#pip install faiss-cpu
#pip install langchain
#pip install pypdf
#pip tiktoken
#pip install InstructorEmbedding
###############


# PDF in String umwandeln
def get_pdf_text(folder_path):
    #translator = Translator()
    text = ""
    # Durchsuche alle Dateien im angegebenen Verzeichnis
    for filename in os.listdir(folder_path):
        filepath = os.path.join(folder_path, filename)

        # Überprüfe, ob die Datei die Erweiterung ".pdf" hat
        if os.path.isfile(filepath) and filename.lower().endswith(".pdf"):
            pdf_reader = PdfReader(filepath)
            for page in pdf_reader.pages:
                text += page.extract_text()
            #text += '\n'
    #text=text.replace("\n", " ")
    #text=text.replace("- ", "")
    return text#translator.translate(text, dest ='en').text

#Chunks erstellen
def get_text_chunks(text):
    #Arbeitsweise Textsplitter definieren
    text_splitter = CharacterTextSplitter(
        separator="\n",
        chunk_size=1000,
        chunk_overlap=200,
        length_function=len
    )
    chunks = text_splitter.split_text(text)
    return chunks

# nur zum Anlegen des lokalen Verzeichnisses "Store" und speichern der Vektor-Datenbank
def create_vectorstore_and_store():
    folder_path = './files'
    pdf_text = get_pdf_text(folder_path)
    text_chunks = get_text_chunks(pdf_text)
    embeddings = HuggingFaceInstructEmbeddings(model_name="hkunlp/instructor-base")
    #embeddings = HuggingFaceInstructEmbeddings(model_name="aari1995/German_Semantic_STS_V2")
    # Initiate Faiss DB
    vectorstoreDB = FAISS.from_texts(texts=text_chunks,embedding=embeddings)#texts=text_chunks,
    # Verzeichnis in dem die VektorDB gespeichert werden soll
    save_directory = "Store"
    #VektorDB lokal speichern
    vectorstoreDB.save_local(save_directory)
    return None
    
########

def get_vectorstore():
    embeddings = HuggingFaceInstructEmbeddings(model_name="hkunlp/instructor-base")
    #embeddings = HuggingFaceInstructEmbeddings(model_name="aari1995/German_Semantic_STS_V2")
    #Abruf lokaler Vektordatenbank
    save_directory = "Store"
    vectorstoreDB = FAISS.load_local(save_directory, embeddings)
    return vectorstoreDB


def get_llm_answer(user_question):
    #if os.path.exists("./Store"): #Nutzereingabe nur eingelesen, wenn vectorstore angelegt
    #user_question = st.text_area("Stell mir eine Frage: ")
            #if os.path.exists("./Store"): #Nutzereingabe nur eingelesen, wenn vectorstore angelegt
    # Retriever sucht passende Textausschnitte in den PDFs (unformatiert)
    #translator = Translator()
    #translator.translate(user_question, dest='en')
    retriever=get_vectorstore().as_retriever()
    retrieved_docs=retriever.invoke(
    user_question
    )
    # Top 3 Suchergebnisse des Retrievers als Context speichern
    context=""+retrieved_docs[0].page_content+retrieved_docs[1].page_content+retrieved_docs[2].page_content
    # Context bereinigen
    context=context.replace("\n", " ")  
    context=context.replace("- ", "")

    # Erstelle die Question Answering-Pipeline für Deutsch
    qa_pipeline = pipeline("question-answering", model="deutsche-telekom/bert-multi-english-german-squad2", tokenizer="deutsche-telekom/bert-multi-english-german-squad2")

    # Frage beantworten mit Q&A Pipeline
    answer = qa_pipeline(question=user_question, context=context, max_length=200)
    
    return answer["answer"]#translator.translate(answer["answer"],dest='de')
   
def main():
    st.set_page_config(
    page_title="Chatbot",
    layout="wide",
    initial_sidebar_state="expanded",
    )
    st.text("Chatbot Rene ist über Telegram erreichbar!")

if __name__ == '__main__':
    main()