Update app.py

app.py

@@ -84,14 +84,6 @@ def generate(
     reformed_query = reformulate_query(prompt, history)
     formatted_prompt = format_prompt(reformed_query, history)
 
-    stream = chat_client.text_generation(formatted_prompt, **generate_kwargs, stream=True, details=True, return_full_text=False)
-
-    output = ""
-    
-    for response in stream:
-        output += response.token.text
-        yield output
-
     # Get relevant documents
     docs = retriever.get_relevant_documents(prompt)
 
@@ -99,6 +91,14 @@ def generate(
     for doc in docs:
         if 'source' in doc.metadata:
             sources += doc.metadata['source'] + '\n'
+    
+    stream = chat_client.text_generation(formatted_prompt, **generate_kwargs, stream=True, details=True, return_full_text=False)
+
+    output = ""
+    
+    for response in stream:
+        output += response.token.text
+        yield output
 
     output += sources