yifanjun233

I even cann't understand what decoder-input refer to...what's this?

i met this problem too and i found the memory-usage is about full but the GPU-util is zero.have you met this?

you may mix the iwslt2014 and test2014,for instance,this program is based on iwslt2016 while its test datas range from test2010 to test2014.maybe early test data is difficult to handle.