MKEvolve: A Modular Multi-Agent Framework for Kernel Code Generation

arXiv cs.AI Papers

Summary

Presents MKEvolve, a modular multi-agent framework that iteratively co-evolves modular decomposition and LLM-generated kernels for hardware accelerators, achieving improved correctness and speedup over direct synthesis while reducing token usage.

arXiv:2607.20501v1 Announce Type: new Abstract: 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%.
Original Article
View Cached Full Text

Cached at: 07/24/26, 05:03 AM

# MKEvolve: A Modular Multi-Agent Framework for Kernel Code Generation
Source: [https://arxiv.org/abs/2607.20501](https://arxiv.org/abs/2607.20501)
[View PDF](https://arxiv.org/pdf/2607.20501)

> Abstract: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%\.

## Submission history

From: Jinsoo Yoo \[[view email](https://arxiv.org/show-email/e8583921/2607.20501)\] **\[v1\]**Sat, 20 Jun 2026 23:14:07 UTC \(4,024 KB\)

Similar Articles

MLEvolve: A Self-Evolving Framework for Automated Machine Learning Algorithm Discovery

Hugging Face Daily Papers

MLEvolve is a self-evolving LLM-based multi-agent framework for automated ML algorithm discovery that extends tree search to Progressive MCGS with graph-based cross-branch information flow and retrospective memory. It achieves state-of-the-art performance on MLE-Bench and outperforms AlphaEvolve on mathematical algorithm optimization tasks.