Yea, because the math have stripped down the whole thing to : I have data I do operation on them. while in reality we deal with multi head attention / grouped q

Yea, because the math have stripped down the whole thing to : I have data I do operation on them. while in reality we deal with multi head attention / grouped query and the positional encoding. That’s all without taking into account the broadcasting done on the batch dimension