training_args = TrainingArguments( output_dir='pegasus-base-wikilarge-newsela-with-domain-adaptation', num_train_epochs=20, warmup_steps=250, per_device_train_batch_size=BATCH_SIZE, weight_decay=0.01, learning_rate=2e-4, # fp16=True, optim="adafactor", )
Step Training Loss 500 4.391800 1000 3.994400 1500 3.009300 2000 2.596300 2500 2.389600 3000 2.328100 3500 2.272900 4000 2.125300 4500 2.019500 5000 2.005500 5500 1.994400 6000 1.853300 6500 1.838700 7000 1.840200 7500 1.822700 8000 1.716600 8500 1.734100 9000 1.739500 9500 1.696000 10000 1.661700 10500 1.672500 11000 1.666800 11500 1.617700 12000 1.611400 12500 1.616300 13000 1.625800 13500 1.567700 14000 1.584600 14500 1.589800 15000 1.574600 15500 1.548300 16000 1.559800 16500 1.562100 17000 1.541600 17500 1.533500 18000 1.538400 18500 1.545700 19000 1.510100 19500 1.522600 20000 1.529100 20500 1.520900 21000 1.501100 21500 1.508400 22000 1.515400 22500 1.500100 23000 1.496700 23500 1.495900 24000 1.505200 24500 1.484400 25000 1.483000 25500 1.491200 26000 1.491900 26500 1.468600 27000 1.479800 27500 1.479600 28000 1.474400 28500 1.466000 29000 1.471700 29500 1.473100 30000 1.463900 30500 1.459300 31000 1.463200 31500 1.466800 32000 1.455300 32500 1.458600 33000 1.455200 33500 1.455500 34000 1.449100 34500 1.451400 35000 1.451200 35500 1.449000 36000 1.443700 36500 1.447600 37000 1.447300 TrainOutput(global_step=37100, training_loss=1.7110925547967382, metrics={'train_runtime': 33376.534, 'train_samples_per_second': 8.889, 'train_steps_per_second': 1.112, 'total_flos': 0.0, 'train_loss': 1.7110925547967382, 'epoch': 20.0})