Update app.py

app.py

@@ -200,14 +200,19 @@ with gr.Blocks(css=css) as demo:
         history_state = gr.State([])  
         last_message_state = gr.State("")  
 
-        def user_interaction(message, history, api_key, max_tokens, top_p, temperature):
-             loading_message = history + [(message, "Loading...", "You", "P-ALPLE", sessions[api_key]["avatar"], ASSISTANT_PIC_PATH)]
-             yield render_message(loading_message), loading_message, ""
-
-             partial_history = history
-             for partial_reply in respond(message, api_key, max_tokens, top_p, temperature):
-                 partial_history = history + [(message, partial_reply, "You", "P-ALPLE", sessions[api_key]["avatar"], ASSISTANT_PIC_PATH)]
-                 yield render_message(partial_history), partial_history, ""
+       def user_interaction(message, history, api_key, max_tokens, top_p, temperature):
+            loading_message = history + [(message, "Loading...", "You", "P-ALPLE", sessions[api_key]["avatar"], ASSISTANT_PIC_PATH)]
+            yield render_message(loading_message), loading_message, ""
+
+            assistant_response = ""
+            for partial_reply in respond(message, api_key, max_tokens, top_p, temperature):
+                assistant_response = partial_reply  # Accumulate streamed response
+                partial_history = history + [(message, assistant_response, "You", "P-ALPLE", sessions[api_key]["avatar"], ASSISTANT_PIC_PATH)]
+                yield render_message(partial_history), partial_history, ""
+
+            final_history = history + [(message, assistant_response, "You", "P-ALPLE", sessions[api_key]["avatar"], ASSISTANT_PIC_PATH)]
+            yield render_message(final_history), final_history, ""
+