#kimi-delta-attention

AI & Machine LearningMay 2, 202611 min

FlashKDA: Moonshot's CUTLASS Kernel for Kimi Linear (2026)

Moonshot open-sourced FlashKDA, a CUTLASS CUDA kernel for Kimi Delta Attention. Drop-in for flash-linear-attention with up to 2.22x prefill speedup on H20 GPUs.