As described in huggingface's document, BertModel takes at least three arguments for sentence-pair classification.
But I found codes in bert_spc.py not working that way. And it's confusing that input_ids and attention_mask should be the first two arguments, but it seems like you have passed input_ids and token_type_ids for the first two arguments.
As described in huggingface's document, BertModel takes at least three arguments for sentence-pair classification.
But I found codes in bert_spc.py not working that way. And it's confusing that input_ids and attention_mask should be the first two arguments, but it seems like you have passed input_ids and token_type_ids for the first two arguments.