Update app.py

app.py

@@ -43,7 +43,7 @@ def get_past_key_values(system_prompt):
     
     # assert that this is indeed a prefix (TODO: make sure this is robust)
     messages.append({'role': 'user', 'content': 'TEST'})
-    tokenized_test = tokenizer.apply_chat_template(messages, return_tensors='pt')
+    tokenized_test = pipe.tokenizer.apply_chat_template(messages, return_tensors='pt')
     assert (tokenized_test[:, :tokenized.shape[1]] == tokenized).all().cpu().item()
     
     return model(**tokenized.to(model.device)).past_key_values