1 paper
Hongyi Jin, Bohan Hou, Guanjie Wang +18
Modern GPU workloads, especially large language model (LLM) inference, suffer from kernel launch overheads and coarse synchronization that limit inter-kernel parallelism. Recent me…