High-performance GPU kernels are essential to modern machine learning systems, yet automatically generating kernels that are both correct and efficient remains challenging. Existing LLM-based approaches face two major limitations: the scarcity of high-quality training data aligned with the model's current capabilities, and the inherent trade-off between kernel correctness and performance. To address these challenges, we propose KernelZero, a co-evolution framework that continuously improves GPU kernel generation through two specialized models: a Proposer that generates Torch modules from API sets and a Coder that translates them into CUDA or Triton kernels. KernelZero uses a frontier-driven module generation mechanism to continuously produce capability-aligned training modules based on the Coder's current weaknesses. It further introduces Correctness-Aware Group Relative Policy Optimization (CA-GRPO), which optimizes performance only after correctness becomes sufficiently reliable. By alternating the optimization of the Proposer and Coder, KernelZero forms an automatic curriculum that enables targeted and training-efficient capability improvement. Empirically, KernelZero-7B surpasses Claude-4.5-Sonnet on CUDA and DeepSeek-V4-Pro on Triton. On KernelBench Level 1 and 2, it achieves CUDA pass@1 scores of 75.8% and 69.6%, respectively, with pass@10 reaching 100% and 97%. On Triton, it achieves pass@1 scores of 77.2% and 72.5%, respectively.
KernelOPT is a multi-agent system that treats compiled models as structured artifacts and exclusively targets generated Triton sub-kernels using five profiling-guided LLM agents and achieves geometric mean speedups over PyTorch.
Large language models show promise for code generation, yet their effectiveness in producing high-performance GPU kernels remains insufficiently understood. Recent agentic workflows improve over direct prompting through specialized generation roles and execution feedback, but our characterization of a representative wo...
Yang-Jie Zhou, Xing-Chen Ma, Zi-Han Liu et al.· Proceedings of the 41st IEEE...· 0 citations
Autonomous LLM-driven kernel tuners have proliferated in 2025–2026, yet none produces kernels that match vendor libraries on the operators those libraries target, such as cuBLAS GEMM or cuSPARSELt sparse GEMM on Hopper GPUs. Two critical gaps explain this ceiling. First, every existing system tunes parameters within a...
Heng Shi, Xiao-Feng Guan, Si-Jie Wu et al.· Proceedings of the 2026 ACM...· 0 citations
We study how much the interaction strategy around a fixed, locally hosted 7B-parameter code model (Qwen2.5-Coder-7B-Instruct) affects its ability to replace PyTorch operators with custom CUDA kernels. On 30 KernelBench tasks (levels 1-3, first 10 each, evaluated on a V100), we compare four regimes of increasing structu...
SyclKittens is a hardware-aware tile programming model for Intel GPUs that encodes known-good methods for exploiting its hardware features efficiently, so its operations place operands in matrix-engine layouts, prefetch through the L1 cache, and communicate over the fabric that links GPU stacks into a node.
Ye-Hong Jiang, Sheng Chen, Fang-Wen Fu et al.· 0 citations
As large language models (LLMs) keep growing in size and complexity, their training frameworks evolve at a rapid pace as well. Therefore, continuous integration (CI) is critical for maintaining the quality and stability of these frameworks. However, unlike traditional software, CI for LLM training frameworks relies on...
Tian-Shuo Qiao, Nai-Qian Zheng, Xiao-Peng Liu et al.· 0 citations
Related blog posts
MIT News · Artificial Intelligence· news.mit.eduNov 17, 2026
By studying the computational foundations of language learning, Associate Professor Jacob Andreas builds intelligent systems that can communicate effectively with humans.
Microsoft Research Blog· microsoft.comNov 16, 2026
AI coding agents are increasingly help developers fix bugs and build features largely on their own, but every interaction with their models consumes tokens. In each session, an agent sends instructions, tool descriptions, and session history to its model. Most efforts to reduce this cost focus on individual tasks. Yet developers return to the same […] The post TokenEconomizer: Learning from Experience to Make AI Coding Agents More Efficient appeared first on Microsoft Research.
MIT News · Artificial Intelligence· news.mit.eduNov 13, 2026