Results
2
comments of
YanLiang
that is because ur sample size can not evenly divide the batch size, try to set drop_last=True, for your Pytorch Dataloader here: https://github.com/kaushaltrivedi/fast-bert/blob/master/fast_bert/data_cls.py#L460 for all train , validation and test.
is there any existing api or suggestions on how to get it efficiently? thank u.