Adds summary about decoder activations. For each of the args, a TensorArray can also be a Tensor representing the stacked array. Args: encoder_outputs: a NestedMap computed by encoder. targets: a NestedMap, usually input_batch.tgt. atten_probs: a TensorArray of max_ta
(self,
encoder_outputs,
targets,
atten_probs,
rnn_outs,
softmax_input,
additional_atten_probs=None,
target_alignments=None)
source not stored for this graph (policy: none)
no test coverage detected