transformers

2bbbbee3 - update flex attention to use `return_aux` instead of `return_lse` when torch verison >= 2.9 (#44684)

update flex attention to use `return_aux` instead of `return_lse` when torch verison >= 2.9 (#44684) * update flex attention to use `return_aux` instead of `return_lse` when torch verison >= 2.9 * cleaning things up a bit --------- Co-authored-by: Neil Tenenholtz <neil.tenenholtz@microsoft.com> Co-authored-by: Cyril Vallez <cyril.vallez@huggingface.co>