Update app.py

app.py

@@ -37,7 +37,68 @@ def respond(
     repeat_penalty,
     selected_topic
 ):
-    # ... [resto del codice della funzione respond rimane invariato]
+    chat_template = MessagesFormatterType.GEMMA_2
+
+    print("HISTORY SO FAR ", history)
+    print("Selected topic:", selected_topic)
+
+    if selected_topic:
+        query_engine = backend.create_index_for_query_engine(documents_paths[selected_topic])
+        message = backend.generate_prompt(query_engine, message)
+        gr.Info(f"Relevant context indexed from {selected_topic} docs...")
+    else:
+        query_engine = backend.load_index_for_query_engine()
+        message = backend.generate_prompt(query_engine, message)
+        gr.Info("Relevant context extracted from db...")
+
+    # Load model only if it's not already loaded or if a new model is selected
+    if backend.llm is None or backend.llm_model != model:
+        try:
+            backend.load_model(model)
+        except Exception as e:
+            return f"Error loading model: {str(e)}"
+
+    provider = LlamaCppPythonProvider(backend.llm)
+
+    agent = LlamaCppAgent(
+        provider,
+        system_prompt=f"{system_message}",
+        predefined_messages_formatter_type=chat_template,
+        debug_output=True
+    )
+
+    settings = provider.get_provider_default_settings()
+    settings.temperature = temperature
+    settings.top_k = top_k
+    settings.top_p = top_p
+    settings.max_tokens = max_tokens
+    settings.repeat_penalty = repeat_penalty
+    settings.stream = True
+
+    messages = BasicChatHistory()
+
+    # add user and assistant messages to the history
+    for msn in history:
+        user = {'role': Roles.user, 'content': msn[0]}
+        assistant = {'role': Roles.assistant, 'content': msn[1]}
+        messages.add_message(user)
+        messages.add_message(assistant)
+
+    try:
+        stream = agent.get_chat_response(
+            message, 
+            llm_sampling_settings=settings,
+            chat_history=messages,
+            returns_streaming_generator=True,
+            print_output=False
+        )
+
+        outputs = ""
+        for output in stream:
+            outputs += output
+            yield outputs
+    except Exception as e:
+        yield f"Error during response generation: {str(e)}"
 
 def select_topic(topic):
     return gr.update(visible=True), topic