bert-large-uncased finetuned on MRPC.
bert-large-uncased
MRPC
batch size is 16, learning rate is 3e-5.
acc: 0.8922, f1: 0.9225