Hardware-Aware Optimization of Large Language Model Inference Acceleration. JAAIR 2026, 5 (1).