Skip to content

想请教下EM特征的问题 #4

@renxingkai

Description

@renxingkai

感谢大佬的开源,我想请教下MTMSN/squad/squad_utils.py 中

for token in query_tokens:

        token_positions.append(len(tokens))

        tokens.append(token)

        segment_ids.append(0)

        if token in sub_doc_tokens and token not in stop_words:

             em_mask.append(1)

        else:

             em_mask.append(0)

        tokens.append("[SEP]")

        segment_ids.append(0)

        em_mask.append(0)

这块的EM特征在BERT模型中如何使用?在BERT Embedding之后如何结合sequence_output 使用呢?看到您并未使用这个文件,谢谢您的解答~

    def forward(self, input_ids, token_type_ids, attention_mask, start_positions=None, end_positions=None):
        all_encoder_layers, _ = self.bert(input_ids, token_type_ids, attention_mask)
        sequence_output = all_encoder_layers[-1]

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions