tf.contrib.seq2seq.AttentionMechanism

View source on GitHub

alignments_size

state_size