Two short questions on training


#1

Hi! (This is about opennmt-py)

  1. How could I calculate the number of training “steps” needed for 1 epoch? (An approximate answer is also OK)
  2. How come when I increase the batch size (even by 2x or 4x) the tok/s doesn’t really increase much?

Thank you so much!


(Vincent Nguyen) #2

here

what batch size are you using ?