Update app.py

app.py

@@ -7,13 +7,19 @@ from langchain_community.vectorstores import FAISS
 from langchain.chains import ConversationalRetrievalChain
 from langchain.memory import ConversationBufferMemory
 
-# API-Token
-api_token = os.getenv("HF_TOKEN")
+# Dein Hugging Face Read Token
+api_token = os.getenv("HF_TOKEN", "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")
 
-# LLM-Optionen
-list_llm = ["google/flan-t5-small", "google/flan-t5-base"]
+# Modelle für Auswahl
+list_llm = [
+    "google/flan-t5-base",  # Leichtes Instruktionsmodell
+    "sentence-transformers/all-MiniLM-L6-v2",  # Embeddings-optimiertes Modell
+    "OpenAssistant/oasst-sft-4-pythia-12b-epoch-3.5",  # Pythia 12B
+    "bigscience/bloom-3b",  # Multilingualer BLOOM
+    "bigscience/bloom-1b7"  # Leichtes BLOOM-Modell
+]
 
-# Dokumente laden und aufteilen
+# Dokumentenverarbeitung
 def load_doc(list_file_path):
     if not list_file_path:
         return [], "Fehler: Keine Dokumente gefunden!"
@@ -21,15 +27,15 @@ def load_doc(list_file_path):
     documents = []
     for loader in loaders:
         documents.extend(loader.load())
-    text_splitter = RecursiveCharacterTextSplitter(chunk_size=1024, chunk_overlap=64)
+    text_splitter = RecursiveCharacterTextSplitter(chunk_size=512, chunk_overlap=32)
     return text_splitter.split_documents(documents)
 
-# Vektor-Datenbank erstellen
+# Erstelle Vektordatenbank
 def create_db(splits):
     embeddings = HuggingFaceEmbeddings(model_name="sentence-transformers/all-MiniLM-L6-v2")
     return FAISS.from_documents(splits, embeddings)
 
-# Datenbank initialisieren
+# Initialisiere Datenbank
 def initialize_database(list_file_obj):
     if not list_file_obj:
         return None, "Fehler: Keine Dateien hochgeladen!"
@@ -38,31 +44,36 @@ def initialize_database(list_file_obj):
     vector_db = create_db(doc_splits)
     return vector_db, "Datenbank erfolgreich erstellt!"
 
-# LLM-Kette initialisieren
-def initialize_llmchain(llm_model, temperature, max_tokens, top_k, vector_db):
+# Initialisiere LLM-Kette
+def initialize_llmchain(llm_model, temperature, max_tokens, vector_db):
     if vector_db is None:
         return None, "Fehler: Keine Vektordatenbank verfügbar."
-    if max_tokens > 250:
-        max_tokens = 250  # Begrenze max_new_tokens, um Fehler zu vermeiden
+    if "pythia" in llm_model or "bloom" in llm_model:
+        max_tokens = min(max_tokens, 2048)
+    else:
+        max_tokens = min(max_tokens, 1024)
+
     llm = HuggingFaceEndpoint(
         repo_id=llm_model,
         huggingfacehub_api_token=api_token,
         temperature=temperature,
-        max_new_tokens=max_tokens,
-        top_k=top_k,
+        max_new_tokens=max_tokens
     )
     memory = ConversationBufferMemory(memory_key="chat_history", output_key="answer", return_messages=True)
     retriever = vector_db.as_retriever()
-    return ConversationalRetrievalChain.from_llm(
+
+    qa_chain = ConversationalRetrievalChain.from_llm(
         llm, retriever=retriever, chain_type="stuff", memory=memory, return_source_documents=True
     )
 
-# LLM initialisieren
-def initialize_LLM(llm_option, llm_temperature, max_tokens, top_k, vector_db):
+    return qa_chain
+
+# Initialisiere LLM
+def initialize_LLM(llm_option, llm_temperature, max_tokens, vector_db):
     if vector_db is None:
         return None, "Fehler: Datenbank wurde nicht erstellt!"
     llm_name = list_llm[llm_option]
-    qa_chain = initialize_llmchain(llm_name, llm_temperature, max_tokens, top_k, vector_db)
+    qa_chain = initialize_llmchain(llm_name, llm_temperature, max_tokens, vector_db)
     return qa_chain, "QA-Kette initialisiert. Chatbot ist bereit!"
 
 # Konversation
@@ -76,23 +87,27 @@ def conversation(qa_chain, message, history):
     formatted_response = history + [{"role": "user", "content": message}, {"role": "assistant", "content": response_text}]
     return qa_chain, formatted_response, formatted_response
 
-# Demo erstellen
+# Gradio UI
 def demo():
     with gr.Blocks() as demo:
         vector_db = gr.State()
         qa_chain = gr.State()
-        gr.Markdown("<center><h1>PDF-Chatbot mit kostenlosen Modellen</h1></center>")
+        gr.Markdown("<center><h1>RAG-Chatbot mit Pythia und BLOOM (CPU-kompatibel)</h1></center>")
 
         with gr.Row():
             with gr.Column():
-                document = gr.Files(label="PDF-Dokument hochladen")
+                document = gr.Files(label="PDF-Dokument hochladen", type="file", file_types=[".pdf"], file_count="multiple")
                 db_btn = gr.Button("Erstelle Vektordatenbank")
                 db_status = gr.Textbox(label="Datenbankstatus", value="Nicht erstellt", interactive=False)
 
-                llm_btn = gr.Radio(["Flan-T5 Small", "Flan-T5 Base"], label="Verfügbare LLMs", value="Flan-T5 Small", type="index")
+                llm_btn = gr.Radio(
+                    ["Flan-T5 Base", "MiniLM", "Pythia 12B", "BLOOM 3B", "BLOOM 1.7B"],
+                    label="Verfügbare LLMs",
+                    value="Flan-T5 Base",
+                    type="index"
+                )
                 slider_temperature = gr.Slider(0.01, 1.0, 0.5, label="Temperature")
-                slider_maxtokens = gr.Slider(1, 250, 128, label="Max Tokens")  # Begrenzung auf 250
-                slider_topk = gr.Slider(1, 10, 3, label="Top-k")
+                slider_maxtokens = gr.Slider(1, 2048, 512, label="Max Tokens")
                 qachain_btn = gr.Button("Initialisiere QA-Chatbot")
                 llm_status = gr.Textbox(label="Chatbot-Status", value="Nicht initialisiert", interactive=False)
 
@@ -101,24 +116,12 @@ def demo():
                 msg = gr.Textbox(label="Frage stellen")
                 submit_btn = gr.Button("Absenden")
 
-        # Event-Handling
-        db_btn.click(
-            initialize_database,
-            inputs=[document],
-            outputs=[vector_db, db_status]
-        )
-        qachain_btn.click(
-            initialize_LLM,
-            inputs=[llm_btn, slider_temperature, slider_maxtokens, slider_topk, vector_db],
-            outputs=[qa_chain, llm_status]
-        )
-        submit_btn.click(
-            conversation,
-            inputs=[qa_chain, msg, chatbot],
-            outputs=[qa_chain, chatbot, chatbot]
-        )
-
-    demo.launch(debug=True)
+        # Events verknüpfen
+        db_btn.click(initialize_database, inputs=[document], outputs=[vector_db, db_status])
+        qachain_btn.click(initialize_LLM, inputs=[llm_btn, slider_temperature, slider_maxtokens, vector_db], outputs=[qa_chain, llm_status])
+        submit_btn.click(conversation, inputs=[qa_chain, msg, chatbot], outputs=[qa_chain, chatbot, chatbot])
+
+    demo.launch(debug=True, share=True)
 
 if __name__ == "__main__":
     demo()