KERNELEVOLVE: SCALING AGENTIC KERNEL CODING FOR HETEROGENEOUS AI ACCELERATORS AT META

Summary from the TLDR AI newsletter

Making deep learning recommendation model (DLRM) training and inference fast and efficient presents three key system challenges: model architecture diversity, kernel primitive diversity, and hardware generation and architecture heterogeneity. KernelEvolve is an agentic kernel coding framework that tackles heterogeneity at scale for DLRM. It is designed to take kernel specifications as input and automate the process of kernel generation and optimization for a recommendation model across heterogeneous hardware architectures. KernelEvolve is designed to optimize a wide variety of production recommendation models across generations of Nvidia and AMD GPUs, as well as Meta's AI accelerators. It significantly mitigates the programmability barrier for new AI hardware.