avg_train_loss: 0.0
optimizer_step: 0
step: 0