--batch_size 4 --num_train_epochs 5 --learning_rate 5e-5 --weight_decay 0.01

initial commit