Force decoding for translation models

[hndrstwn]

Hi,

I am wondering whether it’s possible to run transformers in force decoding mode in order to get the prediction score? In fairseq, I can just use --score-reference flag.

Thanks!