Update app.py

app.py

@@ -45,6 +45,8 @@ def initialize_llmchain(llm_model, temperature, max_tokens, top_k, vector_db, pr
             temperature=temperature,
             max_new_tokens=max_tokens,
             top_k=top_k,
+            timeout=120,  # Aumentado para 120 segundos
+            max_retries=3  # Tenta até 3 vezes
         )
     else:
         llm = HuggingFaceEndpoint(
@@ -53,6 +55,8 @@ def initialize_llmchain(llm_model, temperature, max_tokens, top_k, vector_db, pr
             temperature=temperature,
             max_new_tokens=max_tokens,
             top_k=top_k,
+            timeout=120,
+            max_retries=3
         )
     memory = ConversationBufferMemory(memory_key="chat_history", output_key='answer', return_messages=True)
     retriever = vector_db.as_retriever()
@@ -88,19 +92,29 @@ def conversation(qa_chain, message, history, language):
     else:
         prompt = f"Answer in English: {message}"
     
-    # Enviar o prompt ajustado ao qa_chain
-    response = qa_chain.invoke({"question": prompt, "chat_history": formatted_chat_history})
-    response_answer = response["answer"]
-    if response_answer.find("Helpful Answer:") != -1:
-        response_answer = response_answer.split("Helpful Answer:")[-1]
+    try:
+        response = qa_chain.invoke({"question": prompt, "chat_history": formatted_chat_history})
+        response_answer = response["answer"]
+        if response_answer.find("Helpful Answer:") != -1:
+            response_answer = response_answer.split("Helpful Answer:")[-1]
+    except Exception as e:
+        if language == "Português":
+            response_answer = f"Erro: Não foi possível obter resposta do modelo devido a problemas no servidor. Tente novamente mais tarde. ({str(e)})"
+        else:
+            response_answer = f"Error: Could not get a response from the model due to server issues. Please try again later. ({str(e)})"
+    
+    try:
+        response_sources = response["source_documents"]
+        response_source1 = response_sources[0].page_content.strip()
+        response_source1_page = response_sources[0].metadata["page"] + 1
+        response_source2 = response_sources[1].page_content.strip()
+        response_source2_page = response_sources[1].metadata["page"] + 1
+        response_source3 = response_sources[2].page_content.strip()
+        response_source3_page = response_sources[2].metadata["page"] + 1
+    except:
+        response_source1 = response_source2 = response_source3 = "N/A"
+        response_source1_page = response_source2_page = response_source3_page = 0
     
-    response_sources = response["source_documents"]
-    response_source1 = response_sources[0].page_content.strip()
-    response_source2 = response_sources[1].page_content.strip()
-    response_source3 = response_sources[2].page_content.strip()
-    response_source1_page = response_sources[0].metadata["page"] + 1
-    response_source2_page = response_sources[1].metadata["page"] + 1
-    response_source3_page = response_sources[2].metadata["page"] + 1
     new_history = history + [(message, response_answer)]
     return qa_chain, gr.update(value=""), new_history, response_source1, response_source1_page, response_source2, response_source2_page, response_source3, response_source3_page