DeepSeek 开源 FlashMLA，高效 MLA 解码内核
FlashMLA
Why we picked it
DeepSeek 开源 FlashMLA 解决了大模型推理中 MLA 解码的性能瓶颈，做推理优化和模型部署的开发者可以直接拿来用，值得一试。
DeepSeek 开源了 FlashMLA，一个专为英伟达 Hopper GPU 优化的高效 MLA 解码内核。
Body
它针对可变长度序列进行了优化，已在生产中部署。该项目支持 BF16 精度，分页和块大小 64 的块大小，并提供预填充和分页预填充内核。
FlashMLA 通过优化内存访问和计算，显著提升了推理性能。开发者可以直接在 GitHub 上获取代码和文档。
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 DeepSeek, FlashMLA, 开源/仓库, 推理优化.
Primary source verification: published directly by DeepSeek: GitHub 新仓库 with first-party authority.
