Training BART, error when preparing decoder_input_ids. Shape of input_ids?

Beginners
4

No need to change the HF code, just the structure of your own code:

class SampleModule(pl.LightningModule):
    def __init__(kwargs):
         # initialize stuff
         self.model = BartForConditionalGeneration.from_pretrained(kwargs.arch)
         # more stuff

     def forward(kwargs):
          return self.model(kwargs.batch)

     def training_step(kwargs):
           # do all stuff with shifting decoder inputs etc here
           # then call self() as your forward method