Update app.py

app.py

@@ -82,21 +82,24 @@ def generate(
     )
 
     reformed_query = reformulate_query(prompt, history)
-
     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 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)
-    
+
     sources = "\nSources:\n"
     for doc in docs:
         if 'source' in doc.metadata:
             sources += doc.metadata['source'] + '\n'
-    
+
     output += sources
 
     return output