Persistent State Machine을 활용한 Sparse Attention – 고희소성 LLM 가속기

Sparse Attention with Persistent State Machines – High‑Sparsity LLM Accelerator