icefall icon indicating copy to clipboard operation
icefall copied to clipboard

Published 20 hours ago • verified publisher icon k2-fsa

[Transducer Loss] Why not normalize transducer loss

Open ncakhoa opened this issue 1 year ago • 0 comments

Can you explain why you do not normalize transducer loss. And if I increase batch size, it will make gradients to be larger, can model converges.

ncakhoa avatar Aug 19 '24 10:08 ncakhoa