What does it work better: one_step mode or not one_step mode?
Closed this issue · 1 comments
avostryakov commented
In one_step mode, you process every utterance in a dialogue separately and pass a previous hidden state to the next initial hidden state of the next utterance.It looks like it is the same like process all utterances in a dialogue in a single sequence.
By the way, can you process several utterances from several dialogues in the utterance encoder? Is there this mode?
P.S. Sorry that I ask by creating an issue. I thought that it appropriate to ask near the code.
julianser commented
one_step mode is designed for processing ONE single token (i.e. the next token in an utterance).
The UtteranceEncoder is designed for processing one utterance at a time, but in principle you can apply to any sequence of words (even a full dialogue).