there's more cascades here
opencv_traincascade writes files in the 'new' format (lbp and hog supported, too), opencv_haartraining in the old( haar only ).
(and it does not matter, if it's xml or yml)
so the lbpcascades and the hogcascades in the the link above are in the new format, the haar ones in the old.
if you're looking for faces, there are lbpcascades for that, too (much faster, but not as accurate as haar).