Yeah, exactly this. You would really want to pursue orthogonal methods for robust unlearning, so that you can still use quantization to check that the other met

Yeah, exactly this. You would really want to pursue orthogonal methods for robust unlearning, so that you can still use quantization to check that the other methods worked.