transformers

8598421b - Much more efficient and clear weight initialization and tie weights (#42191)

Much more efficient and clear weight initialization and tie weights (#42191) * everything untilo informer * everything until perceiver * all of them finally * style * replace by transformers init everywhere * use relative import instead * deprecated models * style * start contexts * small fixes * fix modular * remove class switch * do not initialize tied weights * typo * fix * improve * improve comments * improve * improve * fix zamba * fix import * add the post_init * more post_init * fix * protect * more post_init * fix * fixes * fix * fix * switch flag name * more fixes * fixes * fixes * copies * fix * finally find the culprit * style * last small * big bird * better * update init check * final touch * do it everywhere