1 post tagged quantisation.
We're compressing models so aggressively that deployment has become an exercise in reconstructing what the original model was supposed to do.