A Practice of Post-Training on Llama-3 70B with Optimal Selection of Additional Language Mixture Ratio