FlashMLA

On February 24th, FlashMLA, the first project of DeepSeek Open Source Week, was officially released. According to the official introduction, FlashMLA is inspired by FlashAttention 2&3 and cutlass project. Specifically, FlashMLA is an efficient MLA (Multi-Head Latent Attention) decoding kernel optimized for Hopper GPUs, supporting variable-length sequence processing, and is now in production use. FlashM...