MIT Press

Skip to Main Content

Table 8

The BERT-L-T model hyperparameters. H is the pre-trained BERT model’s hidden vector size (H_base = 768, H_large = 1,024)

Layer	Input Dimensions	Output Dimensions
BERT Pretrained Encoder	Interview text	H × # Q-A pairs
LSTM^forward ⊕ LSTM^backward	H × Max # Q-A pairs in batch	2H
Linear Output	2H	1

Layer	Input Dimensions	Output Dimensions
BERT Pretrained Encoder	Interview text	H × # Q-A pairs
LSTM^forward ⊕ LSTM^backward	H × Max # Q-A pairs in batch	2H
Linear Output	2H	1