license: mit
TrainingArgs: lr=2e-5, train-batch-size=16, eval-batch-size=16, num-train-epochs=5, weight-decay=0.01,
