def _attend(self, p):
weight = F.batch_matmul(self.source_hiddens, p)
weight = F.where(self.mask, weight, self.minf)
attention = F.softmax(weight)
return attention
微信
公众号
扫码关注公众号