Update app.py

app.py

@@ -82,24 +82,26 @@ def generate(
     )
 
     reformed_query = reformulate_query(prompt, history)
-    formatted_prompt = format_prompt(reformed_query, history)
-
-    text_stream = chat_client.text_generation(formatted_prompt, **generate_kwargs, stream=True, details=True, return_full_text=False)
 
     docs = retriever.get_relevant_documents(reformed_query)
+
     sources = "\nSources:\n"
     for doc in docs:
         if 'source' in doc.metadata:
             sources += doc.metadata['source'] + '\n'
-    
-    output = ""
-    
-    for response in text_stream:
+
+    formatted_prompt = format_prompt(reformed_query, history)
+
+    output = ''
+
+    stream = chat_client.text_generation(formatted_prompt, **generate_kwargs, stream=True, details=True, return_full_text=False)
+
+    for response in stream:
         output += response.token.text
         yield output
 
     output += sources
-
+    yield output 
     return output
 
 demo = gr.ChatInterface(generate,