5fc4c351699c9007a727eaaef99413962cf0818f,official/nlp/modeling/layers/self_attention_mask.py,SelfAttentionMask,call,#SelfAttentionMask#Any#,33

Before Change

    to_shape = tf_utils.get_shape_list(to_mask, expected_rank=2)
    to_seq_length = to_shape[1]

    to_mask = tf.cast(
        tf.reshape(to_mask, [batch_size, 1, to_seq_length]),
        dtype=from_tensor.dtype)

    // We don"t assume that `from_tensor` is a mask (although it could be). We
    // don"t actually care if we attend *from* padding tokens (only *to* padding)
    // tokens so we create a tensor of all ones.

After Change

  

  def call(self, inputs):
    if isinstance(inputs, list):
      return super().call(inputs[0], inputs[1])
    else:
      return super().call(inputs)

Instances