Update app.py

app.py

@@ -108,11 +108,12 @@ def generate_text(prompt, system_prompt, temperature, top_p, top_k, repetition_p
     # Start generation in a separate thread
     thread = Thread(target=generator, kwargs=generation_kwargs)
     thread.start()
-
+    outputs = []
     # Yield generated text chunks
     try:
-        for new_text in streamer:
-            yield new_text
+        for chunk in streamer:
+            outputs.append(chunk)
+            yield "".join(outputs)
     finally:
         # Ensure the thread completes
         thread.join()