alt Hacker Newsone would hope model weights are already high entropy enough and would not be improved by a general-purpose memory compression algo?
one would hope model weights are already high entropy enough and would not be improved by a general-purpose memory compression algo?