Toward a Holistic Performance Evaluation of Large Language Models Across Diverse AI Accelerators | IEEE Conference Publication | IEEE Xplore