LLaMAX2: Your Translation-Enhanced Model also Performs Well in Reasoning