The Quickstart section of the official document translates incorrectly，The'SENT' and ‘PRED’ are all ‘<unk>’.

[Hsqcsu]

I tried toy-ende translation tasks according to the steps in the official documents, and there were no problems in data preparation and model training. However, when I carried out the third step of inference translation, the scr encoding of the model and the output of the result were UNK . The same error occurred when I was working on translation tasks for other datasets.