Kernel-Smith: A Unified Recipe for Evolutionary Kernel Optimization

TL;DR

Kernel-Smith is a GPU kernel optimization framework that combines evolutionary algorithms with reinforcement learning to achieve state-of-the-art performance across multiple platforms.

Contribution

It introduces a unified evolutionary and RL-based approach for high-performance GPU kernel generation, surpassing existing models on benchmark and production systems.

Findings

Achieves top speedup ratios on KernelBench with Nvidia Triton backend.

Outperforms proprietary models like Gemini-3.0-pro and Claude-4.6-opus.

Successfully adapts across heterogeneous platforms such as MetaX MACA.

Abstract

We present Kernel-Smith, a framework for high-performance GPU kernel and operator generation that combines a stable evaluation-driven evolutionary agent with an evolution-oriented post-training recipe. On the agent side, Kernel-Smith maintains a population of executable candidates and iteratively improves them using an archive of top-performing and diverse programs together with structured execution feedback on compilation, correctness, and speedup. To make this search reliable, we build backend-specific evaluation services for Triton on NVIDIA GPUs and Maca on MetaX GPUs. On the training side, we convert long-horizon evolution trajectories into step-centric supervision and reinforcement learning signals by retaining correctness-preserving, high-gain revisions, so that the model is optimized as a strong local improver inside the evolutionary loop rather than as a one-shot generator. Under a unified evolutionary protocol, Kernel-Smith-235B-RL achieves state-of-the-art overall performance on KernelBench with Nvidia Triton backend, attaining the best average speedup ratio and outperforming frontier proprietary models including Gemini-3.0-pro and Claude-4.6-opus. We further validate the framework on the MetaX MACA backend, where our Kernel-Smith-MACA-30B surpasses large-scale counterparts such as DeepSeek-V3.2-think and Qwen3-235B-2507-think, highlighting potential for seamless adaptation across heterogeneous platforms. Beyond benchmark results, the same workflow produces upstream contributions to production systems including SGLang and LMDeploy, demonstrating that LLM-driven kernel optimization can transfer from controlled evaluation to practical deployment.