write-rnn-tensorflow icon indicating copy to clipboard operation
write-rnn-tensorflow copied to clipboard

Published 20 hours ago • verified publisher icon hardmaru

Can you achieve training loss of -1000 nats

Open leinxx opened this issue 7 years ago • 0 comments

Hi, thanks for the nice work. In the paper, the training loss can go down to -1000. I can only get to -5. I wonder how your training loss curve looks like.

Thanks

leinxx avatar Mar 15 '18 13:03 leinxx