app.py

import os
import gradio as gr
from langchain.document_loaders import PyPDFLoader
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain.embeddings import HuggingFaceEmbeddings
from langchain.vectorstores import FAISS
from langchain.chains import ConversationalRetrievalChain
from langchain.memory import ConversationBufferMemory
from langchain.llms import HuggingFacePipeline
from transformers import pipeline

# **Embeddings-Modell (kein API-Key nötig, lokal geladen)**
EMBEDDINGS_MODEL_NAME = "sentence-transformers/all-MiniLM-L6-v2"
LLM_MODEL_NAME = "google/flan-t5-small"  # Alternativ: "google/flan-t5-base", etc.

# **Dokumente laden und aufteilen**
def load_and_split_docs(list_file_path):
    loaders = [PyPDFLoader(x) for x in list_file_path]
    documents = []
    for loader in loaders:
        documents.extend(loader.load())
    text_splitter = RecursiveCharacterTextSplitter(chunk_size=512, chunk_overlap=32)
    doc_splits = text_splitter.split_documents(documents)
    return doc_splits

# **Vektor-Datenbank mit FAISS erstellen**
def create_db(docs):
    embeddings = HuggingFaceEmbeddings(model_name=EMBEDDINGS_MODEL_NAME)
    faiss_index = FAISS.from_documents(docs, embeddings)
    return faiss_index

# **LLM-Kette initialisieren**
def initialize_llm_chain(llm_model, temperature, max_tokens, vector_db):
    # Hugging Face Pipeline lokal verwenden
    local_pipeline = pipeline(
        "text2text-generation",
        model=llm_model,
        max_length=max_tokens,
        temperature=temperature
    )
    llm = HuggingFacePipeline(pipeline=local_pipeline)
    memory = ConversationBufferMemory(memory_key="chat_history", return_messages=True)
    retriever = vector_db.as_retriever()
    
    # Retrieval-Augmented QA-Kette
    qa_chain = ConversationalRetrievalChain.from_llm(
        llm,
        retriever=retriever,
        memory=memory,
        return_source_documents=True
    )
    return qa_chain

# **Datenbank und Kette initialisieren**
def initialize_database(list_file_obj):
    list_file_path = [x.name for x in list_file_obj if x is not None]
    doc_splits = load_and_split_docs(list_file_path)
    vector_db = create_db(doc_splits)
    return vector_db, "Datenbank erfolgreich erstellt!"

def initialize_llm_chain_wrapper(llm_temperature, max_tokens, vector_db):
    qa_chain = initialize_llm_chain(LLM_MODEL_NAME, llm_temperature, max_tokens, vector_db)
    return qa_chain, "QA-Chatbot ist bereit!"

# **Konversation mit QA-Kette führen**
def conversation(qa_chain, message, history):
    response = qa_chain({"question": message, "chat_history": history})
    response_text = response["answer"]
    sources = [doc.metadata["source"] for doc in response["source_documents"]]
    return qa_chain, response_text, history + [(message, response_text)]

# **Gradio-Benutzeroberfläche**
def demo():
    with gr.Blocks() as demo:
        vector_db = gr.State()
        qa_chain = gr.State()

        gr.HTML("<center><h1>RAG Chatbot mit FAISS und lokalen Modellen</h1></center>")
        with gr.Row():
            with gr.Column():
                document = gr.Files(file_types=[".pdf"], label="PDF hochladen")
                db_btn = gr.Button("Erstelle Vektordatenbank")
                db_status = gr.Textbox(value="Status: Nicht initialisiert", show_label=False)
                slider_temperature = gr.Slider(0.01, 1.0, value=0.5, label="Temperature")
                slider_max_tokens = gr.Slider(64, 512, value=256, label="Max Tokens")
                qachain_btn = gr.Button("Initialisiere QA-Chatbot")

            with gr.Column():
                chatbot = gr.Chatbot(height=400)
                msg = gr.Textbox(placeholder="Frage eingeben...")
                submit_btn = gr.Button("Absenden")

        db_btn.click(initialize_database, [document], [vector_db, db_status])
        qachain_btn.click(initialize_llm_chain_wrapper, [slider_temperature, slider_max_tokens, vector_db], [qa_chain])
        submit_btn.click(conversation, [qa_chain, msg, chatbot], [qa_chain, msg, chatbot])

    demo.launch(debug=True)

if __name__ == "__main__":
    demo()