use old token function

app.py

@@ -39,7 +39,7 @@ def create_response_original(input_str,
   #                                                      num_return_sequences=num_return_sequences)[0])
 
   input_ids = tokenizer.encode(input_str + tokenizer.eos_token, return_tensors="pt")
-  output_ids = original_model.generate(input_ids,
+  output_ids = original_model.generate(**tokenizer(input_str+tokenizer.eos_token,return_tensors="pt",max_length=200),
                                            do_sample=do_sample, 
                                            max_length=100, 
                                            temperature=temperature,