Update app.py

app.py

@@ -8,7 +8,15 @@ def predict(inputs, top_p, temperature, openai_api_key, chat_counter, chatbot=[]
     messages = format_messages(chatbot, inputs, chat_counter)
     payload = create_payload(messages, top_p, temperature)
     response = make_request(API_URL, openai_api_key, payload)
-    return process_response(response, history)
+
+    # Processar a resposta
+    chat, new_history, token_counter = process_response(response, history)
+
+    # Atualizar o histórico apenas se houver novos tokens
+    if token_counter > 0:
+        history = new_history
+
+    return chat, history, token_counter
 
 def format_messages(chatbot, inputs, chat_counter):
     messages = []
@@ -46,7 +54,6 @@ def process_response(response, history):
     for chunk in response.iter_lines():
         if chunk:
             chunk_str = chunk.decode('utf-8').lstrip('data: ')
-            # Verificar se a transmissão está concluída
             if "[DONE]" in chunk_str:
                 break
             try:
@@ -64,14 +71,12 @@ def process_response(response, history):
             except json.JSONDecodeError as e:
                 print("Error decoding JSON response:", e)
                 print("Raw chunk:", chunk_str)
-    
-    # Finalizar a construção da resposta
+
     chat = [(history[i], history[i + 1]) for i in range(0, len(history) - 1, 2)]
     return chat, history, token_counter
 
 
 
-
 def setup_ui():
     with gr.Blocks() as demo:
         with gr.Column():