SIP: Autotuning GPU Native Schedules via Stochastic Instruction Perturbation

Guoliang He, Eiko Yoneki

Large language models (LLMs) have become a significant workload since their appearance. However, they are also computationally expensive as they have billions of parameters and are…