Do you have the right scissors? Tailoring Pre-trained Language Models via Monte-Carlo Methods