bert-large-uncased finetuned on STS-b.
bert-large-uncased
STS-b
batch size is 16, learning rate is 3e-5.
pearson_corr: 0.9024, spearman_corr: 0.8986