Arthur

Indeed. But as mentioned in the documentation : > The text format for MBart-50 is slightly different from mBART. For MBart-50 the language id token is used as a prefix...

I am not sure I understand. When the `decoder_input_ids` are created from the `labels`, they are a shifted version. Let's use the example: - src_text : `'en_XX UN Chief Says...

Hey! Could you open a new issue with a reproducer for this? 😉

Hey! I'll take this one on as part of #23909, since it is an issue with `rstrip` and `lstrip` being ignored (as the default behaviour if a token is not...

As mentioned, this will take a bit more time, a big refactoring is coming! 🔥

Should be merged this week!

Hey! For anyone wanting to contribute, the best way is to just open a PR and link it here! We don't usually assign issues as they can be taken over...

cc @tomaarsen

Sorry for the late reply, should I have a look?

Alright it's a bit stretched out, sorry to ask you this, could you reformulate in terms of `transformers` what you want to do? Pass a model that is placed on...