Num examples = 211225 Num Epochs = 4 Instantaneous batch size per device = 128 Total train batch size (w. parallel, distributed & accumulation) = 128 Gradient Accumulation steps = 1 Total optimization steps = 6604 [6604/6604 53:23, Epoch 4/4] Step Training Loss 500 0.263200 1000 0.156900 1500 0.152500 2000 0.145400 2500 0.140500 3000 0.135900 3500 0.132800 4000 0.129400 4500 0.127200 5000 0.125700 5500 0.124400 6000 0.124100 6500 0.123400