deep learning - Modify Decoder Cell Input in seq2seq during Inference in Tensorflow -
i trying implement deep fusion seq2seq architecture presented in paper . architecure nothing vanilla seq2seq + language model in decoder
.
in vanilla seq2seq
during inference, input of decoder cell output of previous decoder cell. , in tensorflow, greedyembeddinghelper
takes care of that.
but in model, input of decoder cell f(prev_decoder_output, lm output)
, f
function, i.e., input of next cell dependent upon output of previous cell lm model @ point.
my question is, how modify input of decoder cell in tensorflow api ?
i using tf 1.2's seq2seq functions.
Comments
Post a Comment