@inproceedings{cao-etal-2026-ascendkernelgen,
    title = "{A}scend{K}ernel{G}en: {LLM}-Driven Kernel Generation for {NPU}s",
    author = "Cao, Xinzi  and
      Zhai, Jianyang  and
      Li, Pengfei  and
      Hu, Zhiheng  and
      Yan, Cen  and
      Mubingxu  and
      Fang, Guanghuan  and
      She, Bin  and
      Li, Jiayu  and
      Su, Yihan  and
      Tao, Dongyang  and
      Yang, Feidiao  and
      Wang, Chang-Dong  and
      Lu, Yutong  and
      Xue, Weicheng  and
      Zhou, Bin  and
      Tian, Yonghong",
    editor = "Liakata, Maria  and
      Moreira, Viviane P.  and
      Zhang, Jiajun  and
      Jurgens, David",
    booktitle = "Findings of the {A}ssociation for {C}omputational {L}inguistics: {ACL} 2026",
    month = jul,
    year = "2026",
    address = "San Diego, California, United States",
    publisher = "Association for Computational Linguistics",
    url = "https://preview.aclanthology.org/ingest-acl/2026.findings-acl.1533/",
    pages = "30693--30718",
    ISBN = "979-8-89176-395-1",
    abstract = "Neural Processing Units (NPUs) are critical for AI infrastructure, yet developing kernels remains a bottleneck due to the complexity of vendor-specific Domain-Specific Languages (DSLs). While LLMs excel in general coding, they fail to meet the stringent constraints of NPU development, showing a near-zero success rate on complex kernels in our preliminary study. To address these challenges, we present AscendKernelGen, the first comprehensive framework for NPU kernel development, marking a pioneering effort in this field. This framework consists of three interconnected components: (1) Ascend-CoT, the first dataset in the NPU kernel domain that incorporates chain-of-thought reasoning from real-world kernel implementations; (2) KernelGen-LM, a domain-adaptive model trained on this novel dataset using supervised fine-tuning and reinforcement learning; and (3) NPUKernelBench, the first benchmark platform designed to evaluate the compilation, correctness, and performance of generated NPU kernels. Experimental results demonstrate that our approach dramatically bridges the gap in hardware-specific coding: compilation success on complex Level-2 kernels improves from 0{\%} to 95.5{\%} (Pass@10), with 64{\%} functional correctness. AscendKernGen is available at AscendKernGen and NPUKernelBench."
}