Update app.py

app.py

@@ -17,7 +17,7 @@ def generate_response(message):
     global history
     for _ in range(2):  # Assuming you want two responses, adjust as needed
         tokens = [ord(char) for char in message]  # Convert characters to ASCII values
-        response = llm.generate(tokens=tokens, top_k=50, top_p=0.95, temperature=1.0, repetition_penalty=1.0, last_n_tokens=1)
+        response = llm.generate(tokens=tokens, top_k=50, top_p=0.95, temperature=1.0, repetition_penalty=1.0, last_n_tokens=1,stream=True)
         response_text = ' '.join(map(str, response))
         time.sleep(2)
         corrected_response = correct_words(response_text)