MaxKernel: Agentic Kernel Generation for TPUs

MaxKernel, a new agentic system presented by Google, uses large language models with real-time compiler feedback to generate high-performance custom kernels for Tensor Processing Units (TPUs), addressing the complexity of hardware-level kernel authoring.

Designing and authoring high-performance custom kernels for accelerators is a complex task that requires deep hardware-level expertise. Large Language Models LLM can be leveraged together with real-time compiler feedback to build agentic systems for kernel generation. In this work, we present MaxK