model           =gpt2
total batch size=40
train num epochs=30
fp16            =True
max seq length  =40
eval_acc = 0.8064516129032258
eval_loss = 1.131058534210728