Generally, we choose the most likely word, the id associated with the maximum logit value, as the emitted word (this is the "greedy" behavior). But in some cases, I want to choose the least likely word as the output, how should I modify the code?
I have a similar question. How to select the second, third ... likely word?
Like this issue https://github.com/tensorflow/nmt/issues/219#issue-284092935
I have a similar question. How to select the second, third ... likely word?