Lister converges extremely slow

[Apostatee]

trafficstars

Have not changed anything and uses Synthtext for training, loss still stays around 3.0 after 2hours training?

[ccx1997]

Hi,

I have re-implement training using only SynthText, and everything is ok after 2000 steps.

Please refer to attention scaling operation here if it was changed in your case. Or, please provide more details.