FlashMLA

library
DeepSeek
Highly optimized kernels for Multi-head Latent Attention.
GitHubAnnouncement

Library

Stars 12.7k
GitHub Repository →
infrastructureattention