ModelsOfficial

字节跳动Seed与清华AIR发布CUDA Agent:智能体强化学习生成GPU内核

ByteDance Seed and Tsinghua AIR Introduces CUDA Agent: A Large-Scale Agentic RL System for CUDA Kernel Generation

Why we picked it

字节和清华出的新系统,用强化学习让模型自己写CUDA,KernelBench上Seed1.6有74%通过率,能超编译器。

ByteDance Seed与清华AIR发布CUDA Agent,一个用于CUDA内核生成的智能体强化学习系统。

Image · marktechpost
Body

该系统训练大语言模型写出比编译器更快的GPU内核。现有前沿模型已能生成正确CUDA,但性能不佳。

在KernelBench基准中,基础模型Seed1.6通过率为74.

0%。

Original · marktechpost

ByteDance Seed and Tsinghua AIR Introduces CUDA Agent: A Large-Scale Agentic RL System for CUDA Kernel Generation

ByteDance Seed and Tsinghua AIR have released CUDA Agent, an agentic reinforcement learning system that trains a large language model to write GPU kernels that beat a compiler. The gap it targets is narrow but stubborn: frontier models already produce correct CUDA, they just produce slow CUDA. On KernelBench, the base model Seed1.6 passes 74.0% […] The post ByteDance Seed and Tsinghua AIR Introduces CUDA Agent: A Large-Scale Agentic RL System for CUDA Kernel Generation appeared first on MarkTechPost .

Industry Coordinates & Source Verification

This report is categorized under AITOP's "Models" desk, focusing on China's AI industry and domestic foundation model ecosystem. Within our cross-regional analytical framework, it directly maps to US-China foundation model benchmarks and commercialization comparison. Key entities and technical topics under continuous tracking include CUDA Agent, Seed1.6, KernelBench, ByteDance Seed.

Primary source verification: published directly by marktechpost with first-party authority. Cross-source corroboration: 3 independent sources have covered this development (including Pandaily, IT之家), confirming industry-wide signal strength.