MIT Press

Table 2.

Implementation details of our five basic models.

Model	Learning Rate	Epoch	Dropout [11]	Optimizer
BERT-base+BiLSTM+CRF	5e-5	50	0.3	AdamW [12]
BERT-wwm-ext+BiLSTM+CRF[13]	3e-5	50	0.3	AdamW
RoBERTa-wwm-ext+BiLSTM+CRF[13]	3e-5	50	0.3	AdamW
RoBERTa-wwm-ext-large+BiLSTM+CRF[13]	3e-5	20	0.3	AdamW
RoBERTa-wwm-ext-large+CRF[13]	3e-5	20	0.3	AdamW

Model	Learning Rate	Epoch	Dropout [11]	Optimizer
BERT-base+BiLSTM+CRF	5e-5	50	0.3	AdamW [12]
BERT-wwm-ext+BiLSTM+CRF[13]	3e-5	50	0.3	AdamW
RoBERTa-wwm-ext+BiLSTM+CRF[13]	3e-5	50	0.3	AdamW
RoBERTa-wwm-ext-large+BiLSTM+CRF[13]	3e-5	20	0.3	AdamW
RoBERTa-wwm-ext-large+CRF[13]	3e-5	20	0.3	AdamW