MKEvolve: A Modular Multi-Agent Framework for Kernel Code Generation
For ML practitioners and systems researchers, MKEvolve addresses the bottleneck of writing performant kernels for hardware accelerators by enabling modular, interpretable, and adaptable kernel generation.
MKEvolve introduces a modular multi-agent framework for kernel code generation that iteratively co-evolves module decomposition and subkernel implementations, achieving up to 35% reduction in LLM token usage while improving correctness and speedup over end-to-end baselines on KernelBench L2 and L3.
Despite rapid progress in LLM-based code generation, writing correct and performant kernels for hardware accelerators remains a key bottleneck in scaling modern ML workloads. We present MKEvolve (Modular Kernel Evolve), a framework that iteratively co-evolves a modular decomposition of complex PyTorch modules and the LLM-generated kernel for each submodule, refining the decomposition by splitting and fusing across iterations while independently improving each subkernel via LLM-driven beam search. The resulting kernels are programmatic compositions of independently verified subkernels, making them configurable (subkernel implementations are swappable), interpretable (errors and speedups are traceable to specific subkernels), and readily adaptable to related model architectures. Experiments with Triton on KernelBench L2 and L3, spanning multi-operator sequences and full model architectures, show that MKEvolve improves both correctness and speedup over end-to-end direct synthesis baselines while reducing LLM token usage by up to 35%.