deepseek-ai /
FlashMLA
FlashMLA: Efficient Multi-head Latent Attention Kernels
Updated 15 min ago
+41 stars gained this week
Stars gained each day, last 30 days
Usual pace · 9 / day
Oct 4Oct 19Today
This week Before Usual pace
In short
FlashMLA is DeepSeek's library of optimized attention kernels in C++, for sparse and dense attention used in inference on NVIDIA GPUs and Huawei Ascend NPUs. It targets developers running DeepSeek models.
Summary written by AI from the repository’s description and README.
deepseek-ai / FlashMLA
View on GitHub(opens in new tab)