* rebase
* Delete shift tokens func
* downsample decoder input seq len for init
* correct attention mask
* add tests
* pt flax cross test
* make fixup
* init file for import
* change pt-flax cross test threshold
* pt-flax test logits only
* move tests
* make repo-consistency
* consistent indentation
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>