PKU-SafeRLHF: A Safety Alignment Preference Dataset for Llama Family Models