Moonshot AI has released FlashKDA, an open-source implementation of Kimi Delta Attention. This new kernel achieves up to 2.5 times faster inference speeds on NVIDIA H200 GPUs. It is built using CUTLASS and optimized for variable-length batching, allowing for seamless integration into existing deep learning frameworks. AI
Summary written by gemini-2.5-flash-lite from 2 sources. How we write summaries →
IMPACT Accelerates inference for attention-based models on high-end GPUs, potentially lowering costs and increasing throughput.
RANK_REASON Open-source release of a specialized kernel for attention mechanisms.
