Towards Greener LLMs: Bringing Energy-Efficiency to the Forefront of LLM Inference