Update app.py

app.py

@@ -1,88 +1,99 @@
-import gradio as gr
 import os
-from langchain.vectorstores import SimpleVectorStore  # Direkt, ohne zusätzliche Abhängigkeiten
+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 HuggingFaceHub
-from langchain.text_splitter import RecursiveCharacterTextSplitter
+from langchain.llms import HuggingFacePipeline
+from transformers import pipeline
 
-list_llm = ["google/flan-t5-small", "sentence-transformers/all-MiniLM-L6-v2"]
+# **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.
 
-def load_doc(list_file_path):
+# **Dokumente laden und aufteilen**
+def load_and_split_docs(list_file_path):
     loaders = [PyPDFLoader(x) for x in list_file_path]
-    pages = []
+    documents = []
     for loader in loaders:
-        pages.extend(loader.load())
+        documents.extend(loader.load())
     text_splitter = RecursiveCharacterTextSplitter(chunk_size=512, chunk_overlap=32)
-    doc_splits = text_splitter.split_documents(pages)
+    doc_splits = text_splitter.split_documents(documents)
     return doc_splits
 
-def create_db(splits):
-    vectordb = SimpleVectorStore.from_documents(splits)  # Speichern im Speicher
-    return vectordb
+# **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
 
-def initialize_llmchain(llm_model, temperature, max_tokens, top_k, vector_db):
-    llm = HuggingFaceHub(
-        repo_id=llm_model,
-        model_kwargs={
-            "temperature": temperature,
-            "max_length": max_tokens,
-        }
+# **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,
-        chain_type="stuff",
         memory=memory,
-        return_source_documents=True,
+        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_doc(list_file_path)
+    doc_splits = load_and_split_docs(list_file_path)
     vector_db = create_db(doc_splits)
     return vector_db, "Datenbank erfolgreich erstellt!"
 
-def initialize_LLM(llm_option, llm_temperature, max_tokens, vector_db):
-    llm_name = list_llm[llm_option]
-    qa_chain = initialize_llmchain(llm_name, llm_temperature, max_tokens, 3, vector_db)
-    return qa_chain, "Chatbot ist bereit."
+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):
-    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
+    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>PDF QA Chatbot (Kostenlose Version)</h1></center>")
+
+        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"], interactive=True)
+                document = gr.Files(file_types=[".pdf"], label="PDF hochladen")
                 db_btn = gr.Button("Erstelle Vektordatenbank")
-                db_progress = gr.Textbox(value="Nicht initialisiert", show_label=False)
-                llm_btn = gr.Radio(["Flan-T5-Small", "MiniLM"], label="Verfügbare Modelle")
+                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_maxtokens = gr.Slider(64, 512, value=256, label="Max Tokens")
+                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 stellen...")
+                msg = gr.Textbox(placeholder="Frage eingeben...")
                 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, vector_db], [qa_chain])
+        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__":