Closed zaiheshi closed 5 years ago
It is the forward and backward information including the current token.
Thanks! pre_qefv is the model derived feature(z, e) mentioned in the paper. is post_qefv mentioned also?
Thanks! pre_qefv is the model derived feature(z, e) mentioned in the paper. is post_qefv mentioned also?
so is post_qefv mentioned also?you see?
Hello! I am sorry! I have a question, in line 1294 of qe_model.py, why using tf.concat instead of shift_concat or shift_outputs ?