Update app.py

app.py

@@ -1,34 +1,29 @@
 import gradio as gr
 import os
-from langchain.vectorstores import FAISS  # Import für Vektordatenbank FAISS
-from langchain.document_loaders import PyPDFLoader  # PDF-Loader zum Laden der Dokumente
-from langchain.embeddings import HuggingFaceEmbeddings  # Embeddings-Erstellung mit Hugging Face-Modellen
-from langchain.chains import ConversationalRetrievalChain  # Chain für QA-Funktionalität
-from langchain.memory import ConversationBufferMemory  # Speichern des Chat-Verlaufs im Speicher
-from langchain.llms import HuggingFaceHub  # Für das Laden der Modelle von Hugging Face Hub
-from langchain.text_splitter import RecursiveCharacterTextSplitter  # Aufteilen von Dokumenten in Chunks
+from langchain.vectorstores import Chroma  # Chroma als Vektordatenbank
+from langchain.document_loaders import PyPDFLoader
+from langchain.embeddings import HuggingFaceEmbeddings
+from langchain.chains import ConversationalRetrievalChain
+from langchain.memory import ConversationBufferMemory
+from langchain.llms import HuggingFaceHub
+from langchain.text_splitter import RecursiveCharacterTextSplitter
 
-# Liste der LLM-Modelle (leichte CPU-freundliche Modelle)
-list_llm = ["google/flan-t5-small", "distilbert-base-uncased"]
-list_llm_simple = [os.path.basename(llm) for llm in list_llm]
+list_llm = ["google/flan-t5-small", "sentence-transformers/all-MiniLM-L6-v2"]
 
-# PDF-Dokument laden und in Chunks aufteilen
 def load_doc(list_file_path):
     loaders = [PyPDFLoader(x) for x in list_file_path]
     pages = []
     for loader in loaders:
-        pages.extend(loader.load())  # Laden der Seiten aus PDF
-    text_splitter = RecursiveCharacterTextSplitter(chunk_size=512, chunk_overlap=32)  # Chunks für CPU
+        pages.extend(loader.load())
+    text_splitter = RecursiveCharacterTextSplitter(chunk_size=512, chunk_overlap=32)
     doc_splits = text_splitter.split_documents(pages)
     return doc_splits
 
-# Vektordatenbank erstellen
 def create_db(splits):
-    embeddings = HuggingFaceEmbeddings()  # Erstellen der Embeddings mit Hugging Face
-    vectordb = FAISS.from_documents(splits, embeddings)
+    embeddings = HuggingFaceEmbeddings()
+    vectordb = Chroma.from_documents(splits, embeddings)
     return vectordb
 
-# Initialisierung des ConversationalRetrievalChain
 def initialize_llmchain(llm_model, temperature, max_tokens, top_k, vector_db):
     llm = HuggingFaceHub(
         repo_id=llm_model,
@@ -50,64 +45,46 @@ def initialize_llmchain(llm_model, temperature, max_tokens, top_k, vector_db):
     )
     return qa_chain
 
-# Initialisierung der Datenbank
 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_doc(list_file_path)
     vector_db = create_db(doc_splits)
     return vector_db, "Datenbank erfolgreich erstellt!"
 
-# Initialisierung des LLM
 def initialize_LLM(llm_option, llm_temperature, max_tokens, top_k, vector_db):
     llm_name = list_llm[llm_option]
     qa_chain = initialize_llmchain(llm_name, llm_temperature, max_tokens, top_k, vector_db)
     return qa_chain, "LLM erfolgreich initialisiert! Chatbot ist bereit."
 
-# Chat-Historie formatieren
-def format_chat_history(message, chat_history):
-    formatted_chat_history = []
-    for user_message, bot_message in chat_history:
-        formatted_chat_history.append(f"User: {user_message}")
-        formatted_chat_history.append(f"Assistant: {bot_message}")
-    return formatted_chat_history
-
-# Konversationsfunktion
 def conversation(qa_chain, message, history):
-    formatted_chat_history = format_chat_history(message, history)
+    formatted_chat_history = [(f"User: {m}", f"Assistant: {r}") for m, r in history]
     response = qa_chain({"question": message, "chat_history": formatted_chat_history})
     response_answer = response["answer"]
     new_history = history + [(message, response_answer)]
     return qa_chain, gr.update(value=""), new_history
 
-# Gradio-Frontend
 def demo():
     with gr.Blocks() as demo:
         vector_db = gr.State()
         qa_chain = gr.State()
-        gr.HTML("<center><h1>RAG PDF Chatbot</h1></center>")
+        gr.HTML("<center><h1>RAG PDF Chatbot (Kostenlose Version)</h1></center>")
         with gr.Row():
             with gr.Column():
-                gr.Markdown("### Schritt 1: Lade PDF-Dokument hoch")
                 document = gr.Files(height=300, file_count="multiple", file_types=[".pdf"], interactive=True)
                 db_btn = gr.Button("Erstelle Vektordatenbank")
                 db_progress = gr.Textbox(value="Nicht initialisiert", show_label=False)
-                gr.Markdown("### Schritt 2: Wähle LLM und Einstellungen")
-                llm_btn = gr.Radio(list_llm_simple, label="Verfügbare Modelle", value=list_llm_simple[0], type="index")
-                slider_temperature = gr.Slider(0.01, 1.0, value=0.5, step=0.1, label="Temperature")
-                slider_maxtokens = gr.Slider(64, 512, value=256, step=64, label="Max Tokens")
-                slider_topk = gr.Slider(1, 10, value=3, step=1, label="Top-k")
+                llm_btn = gr.Radio(["Flan-T5-Small", "MiniLM"], label="Verfügbare Modelle")
+                slider_temperature = gr.Slider(0.01, 1.0, value=0.5, label="Temperature")
+                slider_maxtokens = gr.Slider(64, 512, value=256, label="Max Tokens")
                 qachain_btn = gr.Button("Initialisiere QA-Chatbot")
-                llm_progress = gr.Textbox(value="Nicht initialisiert", show_label=False)
 
             with gr.Column():
-                gr.Markdown("### Schritt 3: Stelle Fragen an dein Dokument")
                 chatbot = gr.Chatbot(height=400, type="messages")
                 msg = gr.Textbox(placeholder="Frage stellen...")
                 submit_btn = gr.Button("Absenden")
 
         db_btn.click(initialize_database, [document], [vector_db, db_progress])
-        qachain_btn.click(initialize_LLM, [llm_btn, slider_temperature, slider_maxtokens, slider_topk, vector_db], [qa_chain, llm_progress])
-        msg.submit(conversation, [qa_chain, msg, chatbot], [qa_chain, msg, chatbot])
+        qachain_btn.click(initialize_LLM, [llm_btn, slider_temperature, slider_maxtokens, vector_db], [qa_chain])
         submit_btn.click(conversation, [qa_chain, msg, chatbot], [qa_chain, msg, chatbot])
     demo.launch(debug=True)