Open thisisandreeeee opened 4 months ago
The models are trained to translate sentences and not smaller fragments or larger text snippets. That might be the case here confusing the model and not figuring out how to end the translation. By the way, did you verify that the original marian-nmt model behaves the same or is this a huggingface problem with the converted models in pytorch?
I am trying to translate
缓缓走来
, the correct translation should become slowly
.However, the response I am getting is
Come on, come on, come on, come on, come on, come on, come on, come on, come on, come on, come on, come on, come on, come on, come on, come on, come on, come on, come on, come on, come on, come on, come on, come on, come on, come on, come on, come on.
This is incorrect. It would be acceptable if the translation is
come on
, but the repetition is a little odd. Does anyone know why this is happening?