learning_rate=2e-5
per_device_train_batch_size=14
per_device_eval_batch_size=14
weight_decay=0.01
save_total_limit=3
num_train_epochs=3
predict_with_generate=True
fp16=True

global_step=7710,
training_loss=2.436398018566087,
metrics={'train_runtime': 30287.1254,
'train_samples_per_second': 3.564,
'train_steps_per_second': 0.255,
'total_flos': 3.1186278368988365e+17,
'train_loss': 2.436398018566087,
'epoch': 3.0}