avg_train_loss: 0.0
optimizer_step: 81880
step: 0