model           =gpt2
total batch size=40
train num epochs=10
fp16            =True
max seq length  =40
eval_acc = 0.836
eval_loss = 1.0890175614092086