Hazem Masarani

Hazem Masarani is a graduate researcher at Stony Brook University’s PACE Lab focused on LLM inference optimization. He develops and benchmarks multi-GPU inference for state-space models, including Mamba and Mamba-2, using PyTorch, vLLM, CUDA, and Triton. He co-authored the IEEE ICDCS 2026 paper “Tensor Parallelism for State-Space Models.”

His research focuses on LLM inference optimization, including inference performance, tensor and model parallelism, batching, and speculative decoding.