add more optimization for the model achieve accuracy 96.42

Create README.md

add bert model fine tuned on hard dataset

initial commit