Closed ZichaoHuang closed 7 years ago
Hi no worries. I think it is some poorly designed numerical trick to help preventing numerical instability when calculating the exp
. You can remove the rescale
since I already have tf.clip_by_value
elsewhere.
Thank you so much for helping me improving the quality of the code!
OK, thanks!
Hi, it's me again :) I wonder what's the purpose of rescale in the
sampled_softmax
function?