Towards Personalized Evaluation of Large Language Models with An Anonymous Crowd-Sourcing Platform