couldn't you just regularize for this during training? i bet it'd work fine

→ View original post on X — @jxmnop