Skip to content

咨询问题 #9

Open
Open
@Eric-Chen-007

Description

@Eric-Chen-007

document_bert_architectures.py
class DocumentBertSentenceChunkAttentionLSTM
def forward(...)

bert_output 的size = (batch_size, max_seq_len, num_hiddens),其中有些数据未达到max_seq_len, 会是0,
[
[[x,x,...x],[x,x,...,x]...,[x,x,...,x]], --- 达到max_seq_len的数据,
[[x,x,...x],[x,x,...,x]...,[0,0,...,0]], ----未达到max_seq_len的数据,
]

但经过LSTM之后,相应部分值为非0,
后续这部分非0值还参与了之后的attention score,这部分是不是有问题?不知道我理解的对不对。

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions